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>
124 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
130 #include "net-sysfs.h"
133 * The list of packet types we will receive (as opposed to discard)
134 * and the routines to invoke.
136 * Why 16. Because with 16 the only overlap we get on a hash of the
137 * low nibble of the protocol value is RARP/SNAP/X.25.
139 * NOTE: That is no longer true with the addition of VLAN tags. Not
140 * sure which should go first, but I bet it won't make much
141 * difference if we are running VLANs. The good news is that
142 * this protocol won't be in the list unless compiled in, so
143 * the average user (w/out VLANs) will not be adversely affected.
160 #define PTYPE_HASH_SIZE (16)
161 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
163 static DEFINE_SPINLOCK(ptype_lock);
164 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
165 static struct list_head ptype_all __read_mostly; /* Taps */
167 #ifdef CONFIG_NET_DMA
169 struct dma_client client;
171 cpumask_t channel_mask;
172 struct dma_chan **channels;
175 static enum dma_state_client
176 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
177 enum dma_state state);
179 static struct net_dma net_dma = {
181 .event_callback = netdev_dma_event,
187 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
190 * Pure readers hold dev_base_lock for reading.
192 * Writers must hold the rtnl semaphore while they loop through the
193 * dev_base_head list, and hold dev_base_lock for writing when they do the
194 * actual updates. This allows pure readers to access the list even
195 * while a writer is preparing to update it.
197 * To put it another way, dev_base_lock is held for writing only to
198 * protect against pure readers; the rtnl semaphore provides the
199 * protection against other writers.
201 * See, for example usages, register_netdevice() and
202 * unregister_netdevice(), which must be called with the rtnl
205 DEFINE_RWLOCK(dev_base_lock);
207 EXPORT_SYMBOL(dev_base_lock);
209 #define NETDEV_HASHBITS 8
210 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
212 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
214 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
215 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
218 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
220 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
223 /* Device list insertion */
224 static int list_netdevice(struct net_device *dev)
226 struct net *net = dev_net(dev);
230 write_lock_bh(&dev_base_lock);
231 list_add_tail(&dev->dev_list, &net->dev_base_head);
232 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
233 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
234 write_unlock_bh(&dev_base_lock);
238 /* Device list removal */
239 static void unlist_netdevice(struct net_device *dev)
243 /* Unlink dev from the device chain */
244 write_lock_bh(&dev_base_lock);
245 list_del(&dev->dev_list);
246 hlist_del(&dev->name_hlist);
247 hlist_del(&dev->index_hlist);
248 write_unlock_bh(&dev_base_lock);
255 static RAW_NOTIFIER_HEAD(netdev_chain);
258 * Device drivers call our routines to queue packets here. We empty the
259 * queue in the local softnet handler.
262 DEFINE_PER_CPU(struct softnet_data, softnet_data);
264 #ifdef CONFIG_LOCKDEP
266 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
267 * according to dev->type
269 static const unsigned short netdev_lock_type[] =
270 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
271 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
272 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
273 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
274 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
275 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
276 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
277 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
278 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
279 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
280 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
281 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
282 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
283 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
286 static const char *netdev_lock_name[] =
287 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
288 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
289 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
290 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
291 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
292 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
293 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
294 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
295 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
296 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
297 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
298 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
299 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
300 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
303 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
304 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
306 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
310 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
311 if (netdev_lock_type[i] == dev_type)
313 /* the last key is used by default */
314 return ARRAY_SIZE(netdev_lock_type) - 1;
317 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
318 unsigned short dev_type)
322 i = netdev_lock_pos(dev_type);
323 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
324 netdev_lock_name[i]);
327 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
331 i = netdev_lock_pos(dev->type);
332 lockdep_set_class_and_name(&dev->addr_list_lock,
333 &netdev_addr_lock_key[i],
334 netdev_lock_name[i]);
337 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
338 unsigned short dev_type)
341 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
346 /*******************************************************************************
348 Protocol management and registration routines
350 *******************************************************************************/
353 * Add a protocol ID to the list. Now that the input handler is
354 * smarter we can dispense with all the messy stuff that used to be
357 * BEWARE!!! Protocol handlers, mangling input packets,
358 * MUST BE last in hash buckets and checking protocol handlers
359 * MUST start from promiscuous ptype_all chain in net_bh.
360 * It is true now, do not change it.
361 * Explanation follows: if protocol handler, mangling packet, will
362 * be the first on list, it is not able to sense, that packet
363 * is cloned and should be copied-on-write, so that it will
364 * change it and subsequent readers will get broken packet.
369 * dev_add_pack - add packet handler
370 * @pt: packet type declaration
372 * Add a protocol handler to the networking stack. The passed &packet_type
373 * is linked into kernel lists and may not be freed until it has been
374 * removed from the kernel lists.
376 * This call does not sleep therefore it can not
377 * guarantee all CPU's that are in middle of receiving packets
378 * will see the new packet type (until the next received packet).
381 void dev_add_pack(struct packet_type *pt)
385 spin_lock_bh(&ptype_lock);
386 if (pt->type == htons(ETH_P_ALL))
387 list_add_rcu(&pt->list, &ptype_all);
389 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
390 list_add_rcu(&pt->list, &ptype_base[hash]);
392 spin_unlock_bh(&ptype_lock);
396 * __dev_remove_pack - remove packet handler
397 * @pt: packet type declaration
399 * Remove a protocol handler that was previously added to the kernel
400 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
401 * from the kernel lists and can be freed or reused once this function
404 * The packet type might still be in use by receivers
405 * and must not be freed until after all the CPU's have gone
406 * through a quiescent state.
408 void __dev_remove_pack(struct packet_type *pt)
410 struct list_head *head;
411 struct packet_type *pt1;
413 spin_lock_bh(&ptype_lock);
415 if (pt->type == htons(ETH_P_ALL))
418 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
420 list_for_each_entry(pt1, head, list) {
422 list_del_rcu(&pt->list);
427 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
429 spin_unlock_bh(&ptype_lock);
432 * dev_remove_pack - remove packet handler
433 * @pt: packet type declaration
435 * Remove a protocol handler that was previously added to the kernel
436 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
437 * from the kernel lists and can be freed or reused once this function
440 * This call sleeps to guarantee that no CPU is looking at the packet
443 void dev_remove_pack(struct packet_type *pt)
445 __dev_remove_pack(pt);
450 /******************************************************************************
452 Device Boot-time Settings Routines
454 *******************************************************************************/
456 /* Boot time configuration table */
457 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
460 * netdev_boot_setup_add - add new setup entry
461 * @name: name of the device
462 * @map: configured settings for the device
464 * Adds new setup entry to the dev_boot_setup list. The function
465 * returns 0 on error and 1 on success. This is a generic routine to
468 static int netdev_boot_setup_add(char *name, struct ifmap *map)
470 struct netdev_boot_setup *s;
474 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
475 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
476 memset(s[i].name, 0, sizeof(s[i].name));
477 strlcpy(s[i].name, name, IFNAMSIZ);
478 memcpy(&s[i].map, map, sizeof(s[i].map));
483 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
487 * netdev_boot_setup_check - check boot time settings
488 * @dev: the netdevice
490 * Check boot time settings for the device.
491 * The found settings are set for the device to be used
492 * later in the device probing.
493 * Returns 0 if no settings found, 1 if they are.
495 int netdev_boot_setup_check(struct net_device *dev)
497 struct netdev_boot_setup *s = dev_boot_setup;
500 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
501 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
502 !strcmp(dev->name, s[i].name)) {
503 dev->irq = s[i].map.irq;
504 dev->base_addr = s[i].map.base_addr;
505 dev->mem_start = s[i].map.mem_start;
506 dev->mem_end = s[i].map.mem_end;
515 * netdev_boot_base - get address from boot time settings
516 * @prefix: prefix for network device
517 * @unit: id for network device
519 * Check boot time settings for the base address of device.
520 * The found settings are set for the device to be used
521 * later in the device probing.
522 * Returns 0 if no settings found.
524 unsigned long netdev_boot_base(const char *prefix, int unit)
526 const struct netdev_boot_setup *s = dev_boot_setup;
530 sprintf(name, "%s%d", prefix, unit);
533 * If device already registered then return base of 1
534 * to indicate not to probe for this interface
536 if (__dev_get_by_name(&init_net, name))
539 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
540 if (!strcmp(name, s[i].name))
541 return s[i].map.base_addr;
546 * Saves at boot time configured settings for any netdevice.
548 int __init netdev_boot_setup(char *str)
553 str = get_options(str, ARRAY_SIZE(ints), ints);
558 memset(&map, 0, sizeof(map));
562 map.base_addr = ints[2];
564 map.mem_start = ints[3];
566 map.mem_end = ints[4];
568 /* Add new entry to the list */
569 return netdev_boot_setup_add(str, &map);
572 __setup("netdev=", netdev_boot_setup);
574 /*******************************************************************************
576 Device Interface Subroutines
578 *******************************************************************************/
581 * __dev_get_by_name - find a device by its name
582 * @net: the applicable net namespace
583 * @name: name to find
585 * Find an interface by name. Must be called under RTNL semaphore
586 * or @dev_base_lock. If the name is found a pointer to the device
587 * is returned. If the name is not found then %NULL is returned. The
588 * reference counters are not incremented so the caller must be
589 * careful with locks.
592 struct net_device *__dev_get_by_name(struct net *net, const char *name)
594 struct hlist_node *p;
596 hlist_for_each(p, dev_name_hash(net, name)) {
597 struct net_device *dev
598 = hlist_entry(p, struct net_device, name_hlist);
599 if (!strncmp(dev->name, name, IFNAMSIZ))
606 * dev_get_by_name - find a device by its name
607 * @net: the applicable net namespace
608 * @name: name to find
610 * Find an interface by name. This can be called from any
611 * context and does its own locking. The returned handle has
612 * the usage count incremented and the caller must use dev_put() to
613 * release it when it is no longer needed. %NULL is returned if no
614 * matching device is found.
617 struct net_device *dev_get_by_name(struct net *net, const char *name)
619 struct net_device *dev;
621 read_lock(&dev_base_lock);
622 dev = __dev_get_by_name(net, name);
625 read_unlock(&dev_base_lock);
630 * __dev_get_by_index - find a device by its ifindex
631 * @net: the applicable net namespace
632 * @ifindex: index of device
634 * Search for an interface by index. Returns %NULL if the device
635 * is not found or a pointer to the device. The device has not
636 * had its reference counter increased so the caller must be careful
637 * about locking. The caller must hold either the RTNL semaphore
641 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
643 struct hlist_node *p;
645 hlist_for_each(p, dev_index_hash(net, ifindex)) {
646 struct net_device *dev
647 = hlist_entry(p, struct net_device, index_hlist);
648 if (dev->ifindex == ifindex)
656 * dev_get_by_index - find a device by its ifindex
657 * @net: the applicable net namespace
658 * @ifindex: index of device
660 * Search for an interface by index. Returns NULL if the device
661 * is not found or a pointer to the device. The device returned has
662 * had a reference added and the pointer is safe until the user calls
663 * dev_put to indicate they have finished with it.
666 struct net_device *dev_get_by_index(struct net *net, int ifindex)
668 struct net_device *dev;
670 read_lock(&dev_base_lock);
671 dev = __dev_get_by_index(net, ifindex);
674 read_unlock(&dev_base_lock);
679 * dev_getbyhwaddr - find a device by its hardware address
680 * @net: the applicable net namespace
681 * @type: media type of device
682 * @ha: hardware address
684 * Search for an interface by MAC address. Returns NULL if the device
685 * is not found or a pointer to the device. The caller must hold the
686 * rtnl semaphore. The returned device has not had its ref count increased
687 * and the caller must therefore be careful about locking
690 * If the API was consistent this would be __dev_get_by_hwaddr
693 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
695 struct net_device *dev;
699 for_each_netdev(net, dev)
700 if (dev->type == type &&
701 !memcmp(dev->dev_addr, ha, dev->addr_len))
707 EXPORT_SYMBOL(dev_getbyhwaddr);
709 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
711 struct net_device *dev;
714 for_each_netdev(net, dev)
715 if (dev->type == type)
721 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
723 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
725 struct net_device *dev;
728 dev = __dev_getfirstbyhwtype(net, type);
735 EXPORT_SYMBOL(dev_getfirstbyhwtype);
738 * dev_get_by_flags - find any device with given flags
739 * @net: the applicable net namespace
740 * @if_flags: IFF_* values
741 * @mask: bitmask of bits in if_flags to check
743 * Search for any interface with the given flags. Returns NULL if a device
744 * is not found or a pointer to the device. The device returned has
745 * had a reference added and the pointer is safe until the user calls
746 * dev_put to indicate they have finished with it.
749 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
751 struct net_device *dev, *ret;
754 read_lock(&dev_base_lock);
755 for_each_netdev(net, dev) {
756 if (((dev->flags ^ if_flags) & mask) == 0) {
762 read_unlock(&dev_base_lock);
767 * dev_valid_name - check if name is okay for network device
770 * Network device names need to be valid file names to
771 * to allow sysfs to work. We also disallow any kind of
774 int dev_valid_name(const char *name)
778 if (strlen(name) >= IFNAMSIZ)
780 if (!strcmp(name, ".") || !strcmp(name, ".."))
784 if (*name == '/' || isspace(*name))
792 * __dev_alloc_name - allocate a name for a device
793 * @net: network namespace to allocate the device name in
794 * @name: name format string
795 * @buf: scratch buffer and result name string
797 * Passed a format string - eg "lt%d" it will try and find a suitable
798 * id. It scans list of devices to build up a free map, then chooses
799 * the first empty slot. The caller must hold the dev_base or rtnl lock
800 * while allocating the name and adding the device in order to avoid
802 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
803 * Returns the number of the unit assigned or a negative errno code.
806 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
810 const int max_netdevices = 8*PAGE_SIZE;
811 unsigned long *inuse;
812 struct net_device *d;
814 p = strnchr(name, IFNAMSIZ-1, '%');
817 * Verify the string as this thing may have come from
818 * the user. There must be either one "%d" and no other "%"
821 if (p[1] != 'd' || strchr(p + 2, '%'))
824 /* Use one page as a bit array of possible slots */
825 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
829 for_each_netdev(net, d) {
830 if (!sscanf(d->name, name, &i))
832 if (i < 0 || i >= max_netdevices)
835 /* avoid cases where sscanf is not exact inverse of printf */
836 snprintf(buf, IFNAMSIZ, name, i);
837 if (!strncmp(buf, d->name, IFNAMSIZ))
841 i = find_first_zero_bit(inuse, max_netdevices);
842 free_page((unsigned long) inuse);
845 snprintf(buf, IFNAMSIZ, name, i);
846 if (!__dev_get_by_name(net, buf))
849 /* It is possible to run out of possible slots
850 * when the name is long and there isn't enough space left
851 * for the digits, or if all bits are used.
857 * dev_alloc_name - allocate a name for a device
859 * @name: name format string
861 * Passed a format string - eg "lt%d" it will try and find a suitable
862 * id. It scans list of devices to build up a free map, then chooses
863 * the first empty slot. The caller must hold the dev_base or rtnl lock
864 * while allocating the name and adding the device in order to avoid
866 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
867 * Returns the number of the unit assigned or a negative errno code.
870 int dev_alloc_name(struct net_device *dev, const char *name)
876 BUG_ON(!dev_net(dev));
878 ret = __dev_alloc_name(net, name, buf);
880 strlcpy(dev->name, buf, IFNAMSIZ);
886 * dev_change_name - change name of a device
888 * @newname: name (or format string) must be at least IFNAMSIZ
890 * Change name of a device, can pass format strings "eth%d".
893 int dev_change_name(struct net_device *dev, char *newname)
895 char oldname[IFNAMSIZ];
901 BUG_ON(!dev_net(dev));
904 if (dev->flags & IFF_UP)
907 if (!dev_valid_name(newname))
910 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
913 memcpy(oldname, dev->name, IFNAMSIZ);
915 if (strchr(newname, '%')) {
916 err = dev_alloc_name(dev, newname);
919 strcpy(newname, dev->name);
921 else if (__dev_get_by_name(net, newname))
924 strlcpy(dev->name, newname, IFNAMSIZ);
927 err = device_rename(&dev->dev, dev->name);
929 memcpy(dev->name, oldname, IFNAMSIZ);
933 write_lock_bh(&dev_base_lock);
934 hlist_del(&dev->name_hlist);
935 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
936 write_unlock_bh(&dev_base_lock);
938 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
939 ret = notifier_to_errno(ret);
944 "%s: name change rollback failed: %d.\n",
948 memcpy(dev->name, oldname, IFNAMSIZ);
957 * netdev_features_change - device changes features
958 * @dev: device to cause notification
960 * Called to indicate a device has changed features.
962 void netdev_features_change(struct net_device *dev)
964 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
966 EXPORT_SYMBOL(netdev_features_change);
969 * netdev_state_change - device changes state
970 * @dev: device to cause notification
972 * Called to indicate a device has changed state. This function calls
973 * the notifier chains for netdev_chain and sends a NEWLINK message
974 * to the routing socket.
976 void netdev_state_change(struct net_device *dev)
978 if (dev->flags & IFF_UP) {
979 call_netdevice_notifiers(NETDEV_CHANGE, dev);
980 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
984 void netdev_bonding_change(struct net_device *dev)
986 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev);
988 EXPORT_SYMBOL(netdev_bonding_change);
991 * dev_load - load a network module
992 * @net: the applicable net namespace
993 * @name: name of interface
995 * If a network interface is not present and the process has suitable
996 * privileges this function loads the module. If module loading is not
997 * available in this kernel then it becomes a nop.
1000 void dev_load(struct net *net, const char *name)
1002 struct net_device *dev;
1004 read_lock(&dev_base_lock);
1005 dev = __dev_get_by_name(net, name);
1006 read_unlock(&dev_base_lock);
1008 if (!dev && capable(CAP_SYS_MODULE))
1009 request_module("%s", name);
1013 * dev_open - prepare an interface for use.
1014 * @dev: device to open
1016 * Takes a device from down to up state. The device's private open
1017 * function is invoked and then the multicast lists are loaded. Finally
1018 * the device is moved into the up state and a %NETDEV_UP message is
1019 * sent to the netdev notifier chain.
1021 * Calling this function on an active interface is a nop. On a failure
1022 * a negative errno code is returned.
1024 int dev_open(struct net_device *dev)
1034 if (dev->flags & IFF_UP)
1038 * Is it even present?
1040 if (!netif_device_present(dev))
1044 * Call device private open method
1046 set_bit(__LINK_STATE_START, &dev->state);
1048 if (dev->validate_addr)
1049 ret = dev->validate_addr(dev);
1051 if (!ret && dev->open)
1052 ret = dev->open(dev);
1055 * If it went open OK then:
1059 clear_bit(__LINK_STATE_START, &dev->state);
1064 dev->flags |= IFF_UP;
1067 * Initialize multicasting status
1069 dev_set_rx_mode(dev);
1072 * Wakeup transmit queue engine
1077 * ... and announce new interface.
1079 call_netdevice_notifiers(NETDEV_UP, dev);
1086 * dev_close - shutdown an interface.
1087 * @dev: device to shutdown
1089 * This function moves an active device into down state. A
1090 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1091 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1094 int dev_close(struct net_device *dev)
1100 if (!(dev->flags & IFF_UP))
1104 * Tell people we are going down, so that they can
1105 * prepare to death, when device is still operating.
1107 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1109 clear_bit(__LINK_STATE_START, &dev->state);
1111 /* Synchronize to scheduled poll. We cannot touch poll list,
1112 * it can be even on different cpu. So just clear netif_running().
1114 * dev->stop() will invoke napi_disable() on all of it's
1115 * napi_struct instances on this device.
1117 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1119 dev_deactivate(dev);
1122 * Call the device specific close. This cannot fail.
1123 * Only if device is UP
1125 * We allow it to be called even after a DETACH hot-plug
1132 * Device is now down.
1135 dev->flags &= ~IFF_UP;
1138 * Tell people we are down
1140 call_netdevice_notifiers(NETDEV_DOWN, dev);
1147 * dev_disable_lro - disable Large Receive Offload on a device
1150 * Disable Large Receive Offload (LRO) on a net device. Must be
1151 * called under RTNL. This is needed if received packets may be
1152 * forwarded to another interface.
1154 void dev_disable_lro(struct net_device *dev)
1156 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1157 dev->ethtool_ops->set_flags) {
1158 u32 flags = dev->ethtool_ops->get_flags(dev);
1159 if (flags & ETH_FLAG_LRO) {
1160 flags &= ~ETH_FLAG_LRO;
1161 dev->ethtool_ops->set_flags(dev, flags);
1164 WARN_ON(dev->features & NETIF_F_LRO);
1166 EXPORT_SYMBOL(dev_disable_lro);
1169 static int dev_boot_phase = 1;
1172 * Device change register/unregister. These are not inline or static
1173 * as we export them to the world.
1177 * register_netdevice_notifier - register a network notifier block
1180 * Register a notifier to be called when network device events occur.
1181 * The notifier passed is linked into the kernel structures and must
1182 * not be reused until it has been unregistered. A negative errno code
1183 * is returned on a failure.
1185 * When registered all registration and up events are replayed
1186 * to the new notifier to allow device to have a race free
1187 * view of the network device list.
1190 int register_netdevice_notifier(struct notifier_block *nb)
1192 struct net_device *dev;
1193 struct net_device *last;
1198 err = raw_notifier_chain_register(&netdev_chain, nb);
1204 for_each_netdev(net, dev) {
1205 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1206 err = notifier_to_errno(err);
1210 if (!(dev->flags & IFF_UP))
1213 nb->notifier_call(nb, NETDEV_UP, dev);
1224 for_each_netdev(net, dev) {
1228 if (dev->flags & IFF_UP) {
1229 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1230 nb->notifier_call(nb, NETDEV_DOWN, dev);
1232 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1236 raw_notifier_chain_unregister(&netdev_chain, nb);
1241 * unregister_netdevice_notifier - unregister a network notifier block
1244 * Unregister a notifier previously registered by
1245 * register_netdevice_notifier(). The notifier is unlinked into the
1246 * kernel structures and may then be reused. A negative errno code
1247 * is returned on a failure.
1250 int unregister_netdevice_notifier(struct notifier_block *nb)
1255 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1261 * call_netdevice_notifiers - call all network notifier blocks
1262 * @val: value passed unmodified to notifier function
1263 * @dev: net_device pointer passed unmodified to notifier function
1265 * Call all network notifier blocks. Parameters and return value
1266 * are as for raw_notifier_call_chain().
1269 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1271 return raw_notifier_call_chain(&netdev_chain, val, dev);
1274 /* When > 0 there are consumers of rx skb time stamps */
1275 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1277 void net_enable_timestamp(void)
1279 atomic_inc(&netstamp_needed);
1282 void net_disable_timestamp(void)
1284 atomic_dec(&netstamp_needed);
1287 static inline void net_timestamp(struct sk_buff *skb)
1289 if (atomic_read(&netstamp_needed))
1290 __net_timestamp(skb);
1292 skb->tstamp.tv64 = 0;
1296 * Support routine. Sends outgoing frames to any network
1297 * taps currently in use.
1300 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1302 struct packet_type *ptype;
1307 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1308 /* Never send packets back to the socket
1309 * they originated from - MvS (miquels@drinkel.ow.org)
1311 if ((ptype->dev == dev || !ptype->dev) &&
1312 (ptype->af_packet_priv == NULL ||
1313 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1314 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1318 /* skb->nh should be correctly
1319 set by sender, so that the second statement is
1320 just protection against buggy protocols.
1322 skb_reset_mac_header(skb2);
1324 if (skb_network_header(skb2) < skb2->data ||
1325 skb2->network_header > skb2->tail) {
1326 if (net_ratelimit())
1327 printk(KERN_CRIT "protocol %04x is "
1329 skb2->protocol, dev->name);
1330 skb_reset_network_header(skb2);
1333 skb2->transport_header = skb2->network_header;
1334 skb2->pkt_type = PACKET_OUTGOING;
1335 ptype->func(skb2, skb->dev, ptype, skb->dev);
1342 void __netif_schedule(struct Qdisc *q)
1344 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) {
1345 struct softnet_data *sd;
1346 unsigned long flags;
1348 local_irq_save(flags);
1349 sd = &__get_cpu_var(softnet_data);
1350 q->next_sched = sd->output_queue;
1351 sd->output_queue = q;
1352 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1353 local_irq_restore(flags);
1356 EXPORT_SYMBOL(__netif_schedule);
1358 void dev_kfree_skb_irq(struct sk_buff *skb)
1360 if (atomic_dec_and_test(&skb->users)) {
1361 struct softnet_data *sd;
1362 unsigned long flags;
1364 local_irq_save(flags);
1365 sd = &__get_cpu_var(softnet_data);
1366 skb->next = sd->completion_queue;
1367 sd->completion_queue = skb;
1368 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1369 local_irq_restore(flags);
1372 EXPORT_SYMBOL(dev_kfree_skb_irq);
1374 void dev_kfree_skb_any(struct sk_buff *skb)
1376 if (in_irq() || irqs_disabled())
1377 dev_kfree_skb_irq(skb);
1381 EXPORT_SYMBOL(dev_kfree_skb_any);
1385 * netif_device_detach - mark device as removed
1386 * @dev: network device
1388 * Mark device as removed from system and therefore no longer available.
1390 void netif_device_detach(struct net_device *dev)
1392 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1393 netif_running(dev)) {
1394 netif_stop_queue(dev);
1397 EXPORT_SYMBOL(netif_device_detach);
1400 * netif_device_attach - mark device as attached
1401 * @dev: network device
1403 * Mark device as attached from system and restart if needed.
1405 void netif_device_attach(struct net_device *dev)
1407 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1408 netif_running(dev)) {
1409 netif_wake_queue(dev);
1410 __netdev_watchdog_up(dev);
1413 EXPORT_SYMBOL(netif_device_attach);
1415 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1417 return ((features & NETIF_F_GEN_CSUM) ||
1418 ((features & NETIF_F_IP_CSUM) &&
1419 protocol == htons(ETH_P_IP)) ||
1420 ((features & NETIF_F_IPV6_CSUM) &&
1421 protocol == htons(ETH_P_IPV6)));
1424 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1426 if (can_checksum_protocol(dev->features, skb->protocol))
1429 if (skb->protocol == htons(ETH_P_8021Q)) {
1430 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1431 if (can_checksum_protocol(dev->features & dev->vlan_features,
1432 veh->h_vlan_encapsulated_proto))
1440 * Invalidate hardware checksum when packet is to be mangled, and
1441 * complete checksum manually on outgoing path.
1443 int skb_checksum_help(struct sk_buff *skb)
1446 int ret = 0, offset;
1448 if (skb->ip_summed == CHECKSUM_COMPLETE)
1449 goto out_set_summed;
1451 if (unlikely(skb_shinfo(skb)->gso_size)) {
1452 /* Let GSO fix up the checksum. */
1453 goto out_set_summed;
1456 offset = skb->csum_start - skb_headroom(skb);
1457 BUG_ON(offset >= skb_headlen(skb));
1458 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1460 offset += skb->csum_offset;
1461 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1463 if (skb_cloned(skb) &&
1464 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1465 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1470 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1472 skb->ip_summed = CHECKSUM_NONE;
1478 * skb_gso_segment - Perform segmentation on skb.
1479 * @skb: buffer to segment
1480 * @features: features for the output path (see dev->features)
1482 * This function segments the given skb and returns a list of segments.
1484 * It may return NULL if the skb requires no segmentation. This is
1485 * only possible when GSO is used for verifying header integrity.
1487 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1489 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1490 struct packet_type *ptype;
1491 __be16 type = skb->protocol;
1494 BUG_ON(skb_shinfo(skb)->frag_list);
1496 skb_reset_mac_header(skb);
1497 skb->mac_len = skb->network_header - skb->mac_header;
1498 __skb_pull(skb, skb->mac_len);
1500 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1501 if (skb_header_cloned(skb) &&
1502 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1503 return ERR_PTR(err);
1507 list_for_each_entry_rcu(ptype,
1508 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1509 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1510 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1511 err = ptype->gso_send_check(skb);
1512 segs = ERR_PTR(err);
1513 if (err || skb_gso_ok(skb, features))
1515 __skb_push(skb, (skb->data -
1516 skb_network_header(skb)));
1518 segs = ptype->gso_segment(skb, features);
1524 __skb_push(skb, skb->data - skb_mac_header(skb));
1529 EXPORT_SYMBOL(skb_gso_segment);
1531 /* Take action when hardware reception checksum errors are detected. */
1533 void netdev_rx_csum_fault(struct net_device *dev)
1535 if (net_ratelimit()) {
1536 printk(KERN_ERR "%s: hw csum failure.\n",
1537 dev ? dev->name : "<unknown>");
1541 EXPORT_SYMBOL(netdev_rx_csum_fault);
1544 /* Actually, we should eliminate this check as soon as we know, that:
1545 * 1. IOMMU is present and allows to map all the memory.
1546 * 2. No high memory really exists on this machine.
1549 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1551 #ifdef CONFIG_HIGHMEM
1554 if (dev->features & NETIF_F_HIGHDMA)
1557 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1558 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1566 void (*destructor)(struct sk_buff *skb);
1569 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1571 static void dev_gso_skb_destructor(struct sk_buff *skb)
1573 struct dev_gso_cb *cb;
1576 struct sk_buff *nskb = skb->next;
1578 skb->next = nskb->next;
1581 } while (skb->next);
1583 cb = DEV_GSO_CB(skb);
1585 cb->destructor(skb);
1589 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1590 * @skb: buffer to segment
1592 * This function segments the given skb and stores the list of segments
1595 static int dev_gso_segment(struct sk_buff *skb)
1597 struct net_device *dev = skb->dev;
1598 struct sk_buff *segs;
1599 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1602 segs = skb_gso_segment(skb, features);
1604 /* Verifying header integrity only. */
1609 return PTR_ERR(segs);
1612 DEV_GSO_CB(skb)->destructor = skb->destructor;
1613 skb->destructor = dev_gso_skb_destructor;
1618 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1619 struct netdev_queue *txq)
1621 if (likely(!skb->next)) {
1622 if (!list_empty(&ptype_all))
1623 dev_queue_xmit_nit(skb, dev);
1625 if (netif_needs_gso(dev, skb)) {
1626 if (unlikely(dev_gso_segment(skb)))
1632 return dev->hard_start_xmit(skb, dev);
1637 struct sk_buff *nskb = skb->next;
1640 skb->next = nskb->next;
1642 rc = dev->hard_start_xmit(nskb, dev);
1644 nskb->next = skb->next;
1648 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1649 return NETDEV_TX_BUSY;
1650 } while (skb->next);
1652 skb->destructor = DEV_GSO_CB(skb)->destructor;
1659 static u32 simple_tx_hashrnd;
1660 static int simple_tx_hashrnd_initialized = 0;
1662 static u16 simple_tx_hash(struct net_device *dev, struct sk_buff *skb)
1664 u32 addr1, addr2, ports;
1668 if (unlikely(!simple_tx_hashrnd_initialized)) {
1669 get_random_bytes(&simple_tx_hashrnd, 4);
1670 simple_tx_hashrnd_initialized = 1;
1673 switch (skb->protocol) {
1674 case __constant_htons(ETH_P_IP):
1675 ip_proto = ip_hdr(skb)->protocol;
1676 addr1 = ip_hdr(skb)->saddr;
1677 addr2 = ip_hdr(skb)->daddr;
1678 ihl = ip_hdr(skb)->ihl;
1680 case __constant_htons(ETH_P_IPV6):
1681 ip_proto = ipv6_hdr(skb)->nexthdr;
1682 addr1 = ipv6_hdr(skb)->saddr.s6_addr32[3];
1683 addr2 = ipv6_hdr(skb)->daddr.s6_addr32[3];
1698 case IPPROTO_UDPLITE:
1699 ports = *((u32 *) (skb_network_header(skb) + (ihl * 4)));
1707 hash = jhash_3words(addr1, addr2, ports, simple_tx_hashrnd);
1709 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1712 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1713 struct sk_buff *skb)
1715 u16 queue_index = 0;
1717 if (dev->select_queue)
1718 queue_index = dev->select_queue(dev, skb);
1719 else if (dev->real_num_tx_queues > 1)
1720 queue_index = simple_tx_hash(dev, skb);
1722 skb_set_queue_mapping(skb, queue_index);
1723 return netdev_get_tx_queue(dev, queue_index);
1727 * dev_queue_xmit - transmit a buffer
1728 * @skb: buffer to transmit
1730 * Queue a buffer for transmission to a network device. The caller must
1731 * have set the device and priority and built the buffer before calling
1732 * this function. The function can be called from an interrupt.
1734 * A negative errno code is returned on a failure. A success does not
1735 * guarantee the frame will be transmitted as it may be dropped due
1736 * to congestion or traffic shaping.
1738 * -----------------------------------------------------------------------------------
1739 * I notice this method can also return errors from the queue disciplines,
1740 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1743 * Regardless of the return value, the skb is consumed, so it is currently
1744 * difficult to retry a send to this method. (You can bump the ref count
1745 * before sending to hold a reference for retry if you are careful.)
1747 * When calling this method, interrupts MUST be enabled. This is because
1748 * the BH enable code must have IRQs enabled so that it will not deadlock.
1751 int dev_queue_xmit(struct sk_buff *skb)
1753 struct net_device *dev = skb->dev;
1754 struct netdev_queue *txq;
1758 /* GSO will handle the following emulations directly. */
1759 if (netif_needs_gso(dev, skb))
1762 if (skb_shinfo(skb)->frag_list &&
1763 !(dev->features & NETIF_F_FRAGLIST) &&
1764 __skb_linearize(skb))
1767 /* Fragmented skb is linearized if device does not support SG,
1768 * or if at least one of fragments is in highmem and device
1769 * does not support DMA from it.
1771 if (skb_shinfo(skb)->nr_frags &&
1772 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1773 __skb_linearize(skb))
1776 /* If packet is not checksummed and device does not support
1777 * checksumming for this protocol, complete checksumming here.
1779 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1780 skb_set_transport_header(skb, skb->csum_start -
1782 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1787 /* Disable soft irqs for various locks below. Also
1788 * stops preemption for RCU.
1792 txq = dev_pick_tx(dev, skb);
1793 q = rcu_dereference(txq->qdisc);
1795 #ifdef CONFIG_NET_CLS_ACT
1796 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1799 spinlock_t *root_lock = qdisc_lock(q);
1801 spin_lock(root_lock);
1803 rc = qdisc_enqueue_root(skb, q);
1806 spin_unlock(root_lock);
1811 /* The device has no queue. Common case for software devices:
1812 loopback, all the sorts of tunnels...
1814 Really, it is unlikely that netif_tx_lock protection is necessary
1815 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1817 However, it is possible, that they rely on protection
1820 Check this and shot the lock. It is not prone from deadlocks.
1821 Either shot noqueue qdisc, it is even simpler 8)
1823 if (dev->flags & IFF_UP) {
1824 int cpu = smp_processor_id(); /* ok because BHs are off */
1826 if (txq->xmit_lock_owner != cpu) {
1828 HARD_TX_LOCK(dev, txq, cpu);
1830 if (!netif_tx_queue_stopped(txq)) {
1832 if (!dev_hard_start_xmit(skb, dev, txq)) {
1833 HARD_TX_UNLOCK(dev, txq);
1837 HARD_TX_UNLOCK(dev, txq);
1838 if (net_ratelimit())
1839 printk(KERN_CRIT "Virtual device %s asks to "
1840 "queue packet!\n", dev->name);
1842 /* Recursion is detected! It is possible,
1844 if (net_ratelimit())
1845 printk(KERN_CRIT "Dead loop on virtual device "
1846 "%s, fix it urgently!\n", dev->name);
1851 rcu_read_unlock_bh();
1857 rcu_read_unlock_bh();
1862 /*=======================================================================
1864 =======================================================================*/
1866 int netdev_max_backlog __read_mostly = 1000;
1867 int netdev_budget __read_mostly = 300;
1868 int weight_p __read_mostly = 64; /* old backlog weight */
1870 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1874 * netif_rx - post buffer to the network code
1875 * @skb: buffer to post
1877 * This function receives a packet from a device driver and queues it for
1878 * the upper (protocol) levels to process. It always succeeds. The buffer
1879 * may be dropped during processing for congestion control or by the
1883 * NET_RX_SUCCESS (no congestion)
1884 * NET_RX_DROP (packet was dropped)
1888 int netif_rx(struct sk_buff *skb)
1890 struct softnet_data *queue;
1891 unsigned long flags;
1893 /* if netpoll wants it, pretend we never saw it */
1894 if (netpoll_rx(skb))
1897 if (!skb->tstamp.tv64)
1901 * The code is rearranged so that the path is the most
1902 * short when CPU is congested, but is still operating.
1904 local_irq_save(flags);
1905 queue = &__get_cpu_var(softnet_data);
1907 __get_cpu_var(netdev_rx_stat).total++;
1908 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1909 if (queue->input_pkt_queue.qlen) {
1911 __skb_queue_tail(&queue->input_pkt_queue, skb);
1912 local_irq_restore(flags);
1913 return NET_RX_SUCCESS;
1916 napi_schedule(&queue->backlog);
1920 __get_cpu_var(netdev_rx_stat).dropped++;
1921 local_irq_restore(flags);
1927 int netif_rx_ni(struct sk_buff *skb)
1932 err = netif_rx(skb);
1933 if (local_softirq_pending())
1940 EXPORT_SYMBOL(netif_rx_ni);
1942 static void net_tx_action(struct softirq_action *h)
1944 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1946 if (sd->completion_queue) {
1947 struct sk_buff *clist;
1949 local_irq_disable();
1950 clist = sd->completion_queue;
1951 sd->completion_queue = NULL;
1955 struct sk_buff *skb = clist;
1956 clist = clist->next;
1958 WARN_ON(atomic_read(&skb->users));
1963 if (sd->output_queue) {
1966 local_irq_disable();
1967 head = sd->output_queue;
1968 sd->output_queue = NULL;
1972 struct Qdisc *q = head;
1973 spinlock_t *root_lock;
1975 head = head->next_sched;
1977 smp_mb__before_clear_bit();
1978 clear_bit(__QDISC_STATE_SCHED, &q->state);
1980 root_lock = qdisc_lock(q);
1981 if (spin_trylock(root_lock)) {
1983 spin_unlock(root_lock);
1985 __netif_schedule(q);
1991 static inline int deliver_skb(struct sk_buff *skb,
1992 struct packet_type *pt_prev,
1993 struct net_device *orig_dev)
1995 atomic_inc(&skb->users);
1996 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1999 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2000 /* These hooks defined here for ATM */
2002 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2003 unsigned char *addr);
2004 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
2007 * If bridge module is loaded call bridging hook.
2008 * returns NULL if packet was consumed.
2010 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2011 struct sk_buff *skb) __read_mostly;
2012 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2013 struct packet_type **pt_prev, int *ret,
2014 struct net_device *orig_dev)
2016 struct net_bridge_port *port;
2018 if (skb->pkt_type == PACKET_LOOPBACK ||
2019 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2023 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2027 return br_handle_frame_hook(port, skb);
2030 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2033 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2034 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2035 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2037 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2038 struct packet_type **pt_prev,
2040 struct net_device *orig_dev)
2042 if (skb->dev->macvlan_port == NULL)
2046 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2049 return macvlan_handle_frame_hook(skb);
2052 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2055 #ifdef CONFIG_NET_CLS_ACT
2056 /* TODO: Maybe we should just force sch_ingress to be compiled in
2057 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2058 * a compare and 2 stores extra right now if we dont have it on
2059 * but have CONFIG_NET_CLS_ACT
2060 * NOTE: This doesnt stop any functionality; if you dont have
2061 * the ingress scheduler, you just cant add policies on ingress.
2064 static int ing_filter(struct sk_buff *skb)
2066 struct net_device *dev = skb->dev;
2067 u32 ttl = G_TC_RTTL(skb->tc_verd);
2068 struct netdev_queue *rxq;
2069 int result = TC_ACT_OK;
2072 if (MAX_RED_LOOP < ttl++) {
2074 "Redir loop detected Dropping packet (%d->%d)\n",
2075 skb->iif, dev->ifindex);
2079 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2080 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2082 rxq = &dev->rx_queue;
2085 if (q != &noop_qdisc) {
2086 spin_lock(qdisc_lock(q));
2087 result = qdisc_enqueue_root(skb, q);
2088 spin_unlock(qdisc_lock(q));
2094 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2095 struct packet_type **pt_prev,
2096 int *ret, struct net_device *orig_dev)
2098 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2102 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2105 /* Huh? Why does turning on AF_PACKET affect this? */
2106 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2109 switch (ing_filter(skb)) {
2123 * netif_nit_deliver - deliver received packets to network taps
2126 * This function is used to deliver incoming packets to network
2127 * taps. It should be used when the normal netif_receive_skb path
2128 * is bypassed, for example because of VLAN acceleration.
2130 void netif_nit_deliver(struct sk_buff *skb)
2132 struct packet_type *ptype;
2134 if (list_empty(&ptype_all))
2137 skb_reset_network_header(skb);
2138 skb_reset_transport_header(skb);
2139 skb->mac_len = skb->network_header - skb->mac_header;
2142 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2143 if (!ptype->dev || ptype->dev == skb->dev)
2144 deliver_skb(skb, ptype, skb->dev);
2150 * netif_receive_skb - process receive buffer from network
2151 * @skb: buffer to process
2153 * netif_receive_skb() is the main receive data processing function.
2154 * It always succeeds. The buffer may be dropped during processing
2155 * for congestion control or by the protocol layers.
2157 * This function may only be called from softirq context and interrupts
2158 * should be enabled.
2160 * Return values (usually ignored):
2161 * NET_RX_SUCCESS: no congestion
2162 * NET_RX_DROP: packet was dropped
2164 int netif_receive_skb(struct sk_buff *skb)
2166 struct packet_type *ptype, *pt_prev;
2167 struct net_device *orig_dev;
2168 int ret = NET_RX_DROP;
2171 /* if we've gotten here through NAPI, check netpoll */
2172 if (netpoll_receive_skb(skb))
2175 if (!skb->tstamp.tv64)
2179 skb->iif = skb->dev->ifindex;
2181 orig_dev = skb->dev;
2182 if (orig_dev->master) {
2183 if (skb_bond_should_drop(skb)) {
2187 skb->dev = orig_dev->master;
2190 __get_cpu_var(netdev_rx_stat).total++;
2192 skb_reset_network_header(skb);
2193 skb_reset_transport_header(skb);
2194 skb->mac_len = skb->network_header - skb->mac_header;
2200 /* Don't receive packets in an exiting network namespace */
2201 if (!net_alive(dev_net(skb->dev)))
2204 #ifdef CONFIG_NET_CLS_ACT
2205 if (skb->tc_verd & TC_NCLS) {
2206 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2211 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2212 if (!ptype->dev || ptype->dev == skb->dev) {
2214 ret = deliver_skb(skb, pt_prev, orig_dev);
2219 #ifdef CONFIG_NET_CLS_ACT
2220 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2226 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2229 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2233 type = skb->protocol;
2234 list_for_each_entry_rcu(ptype,
2235 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2236 if (ptype->type == type &&
2237 (!ptype->dev || ptype->dev == skb->dev)) {
2239 ret = deliver_skb(skb, pt_prev, orig_dev);
2245 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2248 /* Jamal, now you will not able to escape explaining
2249 * me how you were going to use this. :-)
2259 /* Network device is going away, flush any packets still pending */
2260 static void flush_backlog(void *arg)
2262 struct net_device *dev = arg;
2263 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2264 struct sk_buff *skb, *tmp;
2266 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2267 if (skb->dev == dev) {
2268 __skb_unlink(skb, &queue->input_pkt_queue);
2273 static int process_backlog(struct napi_struct *napi, int quota)
2276 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2277 unsigned long start_time = jiffies;
2279 napi->weight = weight_p;
2281 struct sk_buff *skb;
2283 local_irq_disable();
2284 skb = __skb_dequeue(&queue->input_pkt_queue);
2286 __napi_complete(napi);
2292 netif_receive_skb(skb);
2293 } while (++work < quota && jiffies == start_time);
2299 * __napi_schedule - schedule for receive
2300 * @n: entry to schedule
2302 * The entry's receive function will be scheduled to run
2304 void __napi_schedule(struct napi_struct *n)
2306 unsigned long flags;
2308 local_irq_save(flags);
2309 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2310 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2311 local_irq_restore(flags);
2313 EXPORT_SYMBOL(__napi_schedule);
2316 static void net_rx_action(struct softirq_action *h)
2318 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2319 unsigned long start_time = jiffies;
2320 int budget = netdev_budget;
2323 local_irq_disable();
2325 while (!list_empty(list)) {
2326 struct napi_struct *n;
2329 /* If softirq window is exhuasted then punt.
2331 * Note that this is a slight policy change from the
2332 * previous NAPI code, which would allow up to 2
2333 * jiffies to pass before breaking out. The test
2334 * used to be "jiffies - start_time > 1".
2336 if (unlikely(budget <= 0 || jiffies != start_time))
2341 /* Even though interrupts have been re-enabled, this
2342 * access is safe because interrupts can only add new
2343 * entries to the tail of this list, and only ->poll()
2344 * calls can remove this head entry from the list.
2346 n = list_entry(list->next, struct napi_struct, poll_list);
2348 have = netpoll_poll_lock(n);
2352 /* This NAPI_STATE_SCHED test is for avoiding a race
2353 * with netpoll's poll_napi(). Only the entity which
2354 * obtains the lock and sees NAPI_STATE_SCHED set will
2355 * actually make the ->poll() call. Therefore we avoid
2356 * accidently calling ->poll() when NAPI is not scheduled.
2359 if (test_bit(NAPI_STATE_SCHED, &n->state))
2360 work = n->poll(n, weight);
2362 WARN_ON_ONCE(work > weight);
2366 local_irq_disable();
2368 /* Drivers must not modify the NAPI state if they
2369 * consume the entire weight. In such cases this code
2370 * still "owns" the NAPI instance and therefore can
2371 * move the instance around on the list at-will.
2373 if (unlikely(work == weight)) {
2374 if (unlikely(napi_disable_pending(n)))
2377 list_move_tail(&n->poll_list, list);
2380 netpoll_poll_unlock(have);
2385 #ifdef CONFIG_NET_DMA
2387 * There may not be any more sk_buffs coming right now, so push
2388 * any pending DMA copies to hardware
2390 if (!cpus_empty(net_dma.channel_mask)) {
2392 for_each_cpu_mask_nr(chan_idx, net_dma.channel_mask) {
2393 struct dma_chan *chan = net_dma.channels[chan_idx];
2395 dma_async_memcpy_issue_pending(chan);
2403 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2404 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2408 static gifconf_func_t * gifconf_list [NPROTO];
2411 * register_gifconf - register a SIOCGIF handler
2412 * @family: Address family
2413 * @gifconf: Function handler
2415 * Register protocol dependent address dumping routines. The handler
2416 * that is passed must not be freed or reused until it has been replaced
2417 * by another handler.
2419 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2421 if (family >= NPROTO)
2423 gifconf_list[family] = gifconf;
2429 * Map an interface index to its name (SIOCGIFNAME)
2433 * We need this ioctl for efficient implementation of the
2434 * if_indextoname() function required by the IPv6 API. Without
2435 * it, we would have to search all the interfaces to find a
2439 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2441 struct net_device *dev;
2445 * Fetch the caller's info block.
2448 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2451 read_lock(&dev_base_lock);
2452 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2454 read_unlock(&dev_base_lock);
2458 strcpy(ifr.ifr_name, dev->name);
2459 read_unlock(&dev_base_lock);
2461 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2467 * Perform a SIOCGIFCONF call. This structure will change
2468 * size eventually, and there is nothing I can do about it.
2469 * Thus we will need a 'compatibility mode'.
2472 static int dev_ifconf(struct net *net, char __user *arg)
2475 struct net_device *dev;
2482 * Fetch the caller's info block.
2485 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2492 * Loop over the interfaces, and write an info block for each.
2496 for_each_netdev(net, dev) {
2497 for (i = 0; i < NPROTO; i++) {
2498 if (gifconf_list[i]) {
2501 done = gifconf_list[i](dev, NULL, 0);
2503 done = gifconf_list[i](dev, pos + total,
2513 * All done. Write the updated control block back to the caller.
2515 ifc.ifc_len = total;
2518 * Both BSD and Solaris return 0 here, so we do too.
2520 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2523 #ifdef CONFIG_PROC_FS
2525 * This is invoked by the /proc filesystem handler to display a device
2528 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2529 __acquires(dev_base_lock)
2531 struct net *net = seq_file_net(seq);
2533 struct net_device *dev;
2535 read_lock(&dev_base_lock);
2537 return SEQ_START_TOKEN;
2540 for_each_netdev(net, dev)
2547 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2549 struct net *net = seq_file_net(seq);
2551 return v == SEQ_START_TOKEN ?
2552 first_net_device(net) : next_net_device((struct net_device *)v);
2555 void dev_seq_stop(struct seq_file *seq, void *v)
2556 __releases(dev_base_lock)
2558 read_unlock(&dev_base_lock);
2561 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2563 struct net_device_stats *stats = dev->get_stats(dev);
2565 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2566 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2567 dev->name, stats->rx_bytes, stats->rx_packets,
2569 stats->rx_dropped + stats->rx_missed_errors,
2570 stats->rx_fifo_errors,
2571 stats->rx_length_errors + stats->rx_over_errors +
2572 stats->rx_crc_errors + stats->rx_frame_errors,
2573 stats->rx_compressed, stats->multicast,
2574 stats->tx_bytes, stats->tx_packets,
2575 stats->tx_errors, stats->tx_dropped,
2576 stats->tx_fifo_errors, stats->collisions,
2577 stats->tx_carrier_errors +
2578 stats->tx_aborted_errors +
2579 stats->tx_window_errors +
2580 stats->tx_heartbeat_errors,
2581 stats->tx_compressed);
2585 * Called from the PROCfs module. This now uses the new arbitrary sized
2586 * /proc/net interface to create /proc/net/dev
2588 static int dev_seq_show(struct seq_file *seq, void *v)
2590 if (v == SEQ_START_TOKEN)
2591 seq_puts(seq, "Inter-| Receive "
2593 " face |bytes packets errs drop fifo frame "
2594 "compressed multicast|bytes packets errs "
2595 "drop fifo colls carrier compressed\n");
2597 dev_seq_printf_stats(seq, v);
2601 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2603 struct netif_rx_stats *rc = NULL;
2605 while (*pos < nr_cpu_ids)
2606 if (cpu_online(*pos)) {
2607 rc = &per_cpu(netdev_rx_stat, *pos);
2614 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2616 return softnet_get_online(pos);
2619 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2622 return softnet_get_online(pos);
2625 static void softnet_seq_stop(struct seq_file *seq, void *v)
2629 static int softnet_seq_show(struct seq_file *seq, void *v)
2631 struct netif_rx_stats *s = v;
2633 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2634 s->total, s->dropped, s->time_squeeze, 0,
2635 0, 0, 0, 0, /* was fastroute */
2640 static const struct seq_operations dev_seq_ops = {
2641 .start = dev_seq_start,
2642 .next = dev_seq_next,
2643 .stop = dev_seq_stop,
2644 .show = dev_seq_show,
2647 static int dev_seq_open(struct inode *inode, struct file *file)
2649 return seq_open_net(inode, file, &dev_seq_ops,
2650 sizeof(struct seq_net_private));
2653 static const struct file_operations dev_seq_fops = {
2654 .owner = THIS_MODULE,
2655 .open = dev_seq_open,
2657 .llseek = seq_lseek,
2658 .release = seq_release_net,
2661 static const struct seq_operations softnet_seq_ops = {
2662 .start = softnet_seq_start,
2663 .next = softnet_seq_next,
2664 .stop = softnet_seq_stop,
2665 .show = softnet_seq_show,
2668 static int softnet_seq_open(struct inode *inode, struct file *file)
2670 return seq_open(file, &softnet_seq_ops);
2673 static const struct file_operations softnet_seq_fops = {
2674 .owner = THIS_MODULE,
2675 .open = softnet_seq_open,
2677 .llseek = seq_lseek,
2678 .release = seq_release,
2681 static void *ptype_get_idx(loff_t pos)
2683 struct packet_type *pt = NULL;
2687 list_for_each_entry_rcu(pt, &ptype_all, list) {
2693 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
2694 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2703 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2707 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2710 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2712 struct packet_type *pt;
2713 struct list_head *nxt;
2717 if (v == SEQ_START_TOKEN)
2718 return ptype_get_idx(0);
2721 nxt = pt->list.next;
2722 if (pt->type == htons(ETH_P_ALL)) {
2723 if (nxt != &ptype_all)
2726 nxt = ptype_base[0].next;
2728 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
2730 while (nxt == &ptype_base[hash]) {
2731 if (++hash >= PTYPE_HASH_SIZE)
2733 nxt = ptype_base[hash].next;
2736 return list_entry(nxt, struct packet_type, list);
2739 static void ptype_seq_stop(struct seq_file *seq, void *v)
2745 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2747 #ifdef CONFIG_KALLSYMS
2748 unsigned long offset = 0, symsize;
2749 const char *symname;
2753 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2760 modname = delim = "";
2761 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2767 seq_printf(seq, "[%p]", sym);
2770 static int ptype_seq_show(struct seq_file *seq, void *v)
2772 struct packet_type *pt = v;
2774 if (v == SEQ_START_TOKEN)
2775 seq_puts(seq, "Type Device Function\n");
2776 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
2777 if (pt->type == htons(ETH_P_ALL))
2778 seq_puts(seq, "ALL ");
2780 seq_printf(seq, "%04x", ntohs(pt->type));
2782 seq_printf(seq, " %-8s ",
2783 pt->dev ? pt->dev->name : "");
2784 ptype_seq_decode(seq, pt->func);
2785 seq_putc(seq, '\n');
2791 static const struct seq_operations ptype_seq_ops = {
2792 .start = ptype_seq_start,
2793 .next = ptype_seq_next,
2794 .stop = ptype_seq_stop,
2795 .show = ptype_seq_show,
2798 static int ptype_seq_open(struct inode *inode, struct file *file)
2800 return seq_open_net(inode, file, &ptype_seq_ops,
2801 sizeof(struct seq_net_private));
2804 static const struct file_operations ptype_seq_fops = {
2805 .owner = THIS_MODULE,
2806 .open = ptype_seq_open,
2808 .llseek = seq_lseek,
2809 .release = seq_release_net,
2813 static int __net_init dev_proc_net_init(struct net *net)
2817 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2819 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2821 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2824 if (wext_proc_init(net))
2830 proc_net_remove(net, "ptype");
2832 proc_net_remove(net, "softnet_stat");
2834 proc_net_remove(net, "dev");
2838 static void __net_exit dev_proc_net_exit(struct net *net)
2840 wext_proc_exit(net);
2842 proc_net_remove(net, "ptype");
2843 proc_net_remove(net, "softnet_stat");
2844 proc_net_remove(net, "dev");
2847 static struct pernet_operations __net_initdata dev_proc_ops = {
2848 .init = dev_proc_net_init,
2849 .exit = dev_proc_net_exit,
2852 static int __init dev_proc_init(void)
2854 return register_pernet_subsys(&dev_proc_ops);
2857 #define dev_proc_init() 0
2858 #endif /* CONFIG_PROC_FS */
2862 * netdev_set_master - set up master/slave pair
2863 * @slave: slave device
2864 * @master: new master device
2866 * Changes the master device of the slave. Pass %NULL to break the
2867 * bonding. The caller must hold the RTNL semaphore. On a failure
2868 * a negative errno code is returned. On success the reference counts
2869 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2870 * function returns zero.
2872 int netdev_set_master(struct net_device *slave, struct net_device *master)
2874 struct net_device *old = slave->master;
2884 slave->master = master;
2892 slave->flags |= IFF_SLAVE;
2894 slave->flags &= ~IFF_SLAVE;
2896 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2900 static int __dev_set_promiscuity(struct net_device *dev, int inc)
2902 unsigned short old_flags = dev->flags;
2906 dev->flags |= IFF_PROMISC;
2907 dev->promiscuity += inc;
2908 if (dev->promiscuity == 0) {
2911 * If inc causes overflow, untouch promisc and return error.
2914 dev->flags &= ~IFF_PROMISC;
2916 dev->promiscuity -= inc;
2917 printk(KERN_WARNING "%s: promiscuity touches roof, "
2918 "set promiscuity failed, promiscuity feature "
2919 "of device might be broken.\n", dev->name);
2923 if (dev->flags != old_flags) {
2924 printk(KERN_INFO "device %s %s promiscuous mode\n",
2925 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2928 audit_log(current->audit_context, GFP_ATOMIC,
2929 AUDIT_ANOM_PROMISCUOUS,
2930 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2931 dev->name, (dev->flags & IFF_PROMISC),
2932 (old_flags & IFF_PROMISC),
2933 audit_get_loginuid(current),
2934 current->uid, current->gid,
2935 audit_get_sessionid(current));
2937 if (dev->change_rx_flags)
2938 dev->change_rx_flags(dev, IFF_PROMISC);
2944 * dev_set_promiscuity - update promiscuity count on a device
2948 * Add or remove promiscuity from a device. While the count in the device
2949 * remains above zero the interface remains promiscuous. Once it hits zero
2950 * the device reverts back to normal filtering operation. A negative inc
2951 * value is used to drop promiscuity on the device.
2952 * Return 0 if successful or a negative errno code on error.
2954 int dev_set_promiscuity(struct net_device *dev, int inc)
2956 unsigned short old_flags = dev->flags;
2959 err = __dev_set_promiscuity(dev, inc);
2962 if (dev->flags != old_flags)
2963 dev_set_rx_mode(dev);
2968 * dev_set_allmulti - update allmulti count on a device
2972 * Add or remove reception of all multicast frames to a device. While the
2973 * count in the device remains above zero the interface remains listening
2974 * to all interfaces. Once it hits zero the device reverts back to normal
2975 * filtering operation. A negative @inc value is used to drop the counter
2976 * when releasing a resource needing all multicasts.
2977 * Return 0 if successful or a negative errno code on error.
2980 int dev_set_allmulti(struct net_device *dev, int inc)
2982 unsigned short old_flags = dev->flags;
2986 dev->flags |= IFF_ALLMULTI;
2987 dev->allmulti += inc;
2988 if (dev->allmulti == 0) {
2991 * If inc causes overflow, untouch allmulti and return error.
2994 dev->flags &= ~IFF_ALLMULTI;
2996 dev->allmulti -= inc;
2997 printk(KERN_WARNING "%s: allmulti touches roof, "
2998 "set allmulti failed, allmulti feature of "
2999 "device might be broken.\n", dev->name);
3003 if (dev->flags ^ old_flags) {
3004 if (dev->change_rx_flags)
3005 dev->change_rx_flags(dev, IFF_ALLMULTI);
3006 dev_set_rx_mode(dev);
3012 * Upload unicast and multicast address lists to device and
3013 * configure RX filtering. When the device doesn't support unicast
3014 * filtering it is put in promiscuous mode while unicast addresses
3017 void __dev_set_rx_mode(struct net_device *dev)
3019 /* dev_open will call this function so the list will stay sane. */
3020 if (!(dev->flags&IFF_UP))
3023 if (!netif_device_present(dev))
3026 if (dev->set_rx_mode)
3027 dev->set_rx_mode(dev);
3029 /* Unicast addresses changes may only happen under the rtnl,
3030 * therefore calling __dev_set_promiscuity here is safe.
3032 if (dev->uc_count > 0 && !dev->uc_promisc) {
3033 __dev_set_promiscuity(dev, 1);
3034 dev->uc_promisc = 1;
3035 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3036 __dev_set_promiscuity(dev, -1);
3037 dev->uc_promisc = 0;
3040 if (dev->set_multicast_list)
3041 dev->set_multicast_list(dev);
3045 void dev_set_rx_mode(struct net_device *dev)
3047 netif_addr_lock_bh(dev);
3048 __dev_set_rx_mode(dev);
3049 netif_addr_unlock_bh(dev);
3052 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3053 void *addr, int alen, int glbl)
3055 struct dev_addr_list *da;
3057 for (; (da = *list) != NULL; list = &da->next) {
3058 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3059 alen == da->da_addrlen) {
3061 int old_glbl = da->da_gusers;
3078 int __dev_addr_add(struct dev_addr_list **list, int *count,
3079 void *addr, int alen, int glbl)
3081 struct dev_addr_list *da;
3083 for (da = *list; da != NULL; da = da->next) {
3084 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3085 da->da_addrlen == alen) {
3087 int old_glbl = da->da_gusers;
3097 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3100 memcpy(da->da_addr, addr, alen);
3101 da->da_addrlen = alen;
3103 da->da_gusers = glbl ? 1 : 0;
3111 * dev_unicast_delete - Release secondary unicast address.
3113 * @addr: address to delete
3114 * @alen: length of @addr
3116 * Release reference to a secondary unicast address and remove it
3117 * from the device if the reference count drops to zero.
3119 * The caller must hold the rtnl_mutex.
3121 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3127 netif_addr_lock_bh(dev);
3128 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3130 __dev_set_rx_mode(dev);
3131 netif_addr_unlock_bh(dev);
3134 EXPORT_SYMBOL(dev_unicast_delete);
3137 * dev_unicast_add - add a secondary unicast address
3139 * @addr: address to add
3140 * @alen: length of @addr
3142 * Add a secondary unicast address to the device or increase
3143 * the reference count if it already exists.
3145 * The caller must hold the rtnl_mutex.
3147 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3153 netif_addr_lock_bh(dev);
3154 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3156 __dev_set_rx_mode(dev);
3157 netif_addr_unlock_bh(dev);
3160 EXPORT_SYMBOL(dev_unicast_add);
3162 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3163 struct dev_addr_list **from, int *from_count)
3165 struct dev_addr_list *da, *next;
3169 while (da != NULL) {
3171 if (!da->da_synced) {
3172 err = __dev_addr_add(to, to_count,
3173 da->da_addr, da->da_addrlen, 0);
3178 } else if (da->da_users == 1) {
3179 __dev_addr_delete(to, to_count,
3180 da->da_addr, da->da_addrlen, 0);
3181 __dev_addr_delete(from, from_count,
3182 da->da_addr, da->da_addrlen, 0);
3189 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3190 struct dev_addr_list **from, int *from_count)
3192 struct dev_addr_list *da, *next;
3195 while (da != NULL) {
3197 if (da->da_synced) {
3198 __dev_addr_delete(to, to_count,
3199 da->da_addr, da->da_addrlen, 0);
3201 __dev_addr_delete(from, from_count,
3202 da->da_addr, da->da_addrlen, 0);
3209 * dev_unicast_sync - Synchronize device's unicast list to another device
3210 * @to: destination device
3211 * @from: source device
3213 * Add newly added addresses to the destination device and release
3214 * addresses that have no users left. The source device must be
3215 * locked by netif_tx_lock_bh.
3217 * This function is intended to be called from the dev->set_rx_mode
3218 * function of layered software devices.
3220 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3224 netif_addr_lock_bh(to);
3225 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3226 &from->uc_list, &from->uc_count);
3228 __dev_set_rx_mode(to);
3229 netif_addr_unlock_bh(to);
3232 EXPORT_SYMBOL(dev_unicast_sync);
3235 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3236 * @to: destination device
3237 * @from: source device
3239 * Remove all addresses that were added to the destination device by
3240 * dev_unicast_sync(). This function is intended to be called from the
3241 * dev->stop function of layered software devices.
3243 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3245 netif_addr_lock_bh(from);
3246 netif_addr_lock(to);
3248 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3249 &from->uc_list, &from->uc_count);
3250 __dev_set_rx_mode(to);
3252 netif_addr_unlock(to);
3253 netif_addr_unlock_bh(from);
3255 EXPORT_SYMBOL(dev_unicast_unsync);
3257 static void __dev_addr_discard(struct dev_addr_list **list)
3259 struct dev_addr_list *tmp;
3261 while (*list != NULL) {
3264 if (tmp->da_users > tmp->da_gusers)
3265 printk("__dev_addr_discard: address leakage! "
3266 "da_users=%d\n", tmp->da_users);
3271 static void dev_addr_discard(struct net_device *dev)
3273 netif_addr_lock_bh(dev);
3275 __dev_addr_discard(&dev->uc_list);
3278 __dev_addr_discard(&dev->mc_list);
3281 netif_addr_unlock_bh(dev);
3284 unsigned dev_get_flags(const struct net_device *dev)
3288 flags = (dev->flags & ~(IFF_PROMISC |
3293 (dev->gflags & (IFF_PROMISC |
3296 if (netif_running(dev)) {
3297 if (netif_oper_up(dev))
3298 flags |= IFF_RUNNING;
3299 if (netif_carrier_ok(dev))
3300 flags |= IFF_LOWER_UP;
3301 if (netif_dormant(dev))
3302 flags |= IFF_DORMANT;
3308 int dev_change_flags(struct net_device *dev, unsigned flags)
3311 int old_flags = dev->flags;
3316 * Set the flags on our device.
3319 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3320 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3322 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3326 * Load in the correct multicast list now the flags have changed.
3329 if (dev->change_rx_flags && (old_flags ^ flags) & IFF_MULTICAST)
3330 dev->change_rx_flags(dev, IFF_MULTICAST);
3332 dev_set_rx_mode(dev);
3335 * Have we downed the interface. We handle IFF_UP ourselves
3336 * according to user attempts to set it, rather than blindly
3341 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3342 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3345 dev_set_rx_mode(dev);
3348 if (dev->flags & IFF_UP &&
3349 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3351 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3353 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3354 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3355 dev->gflags ^= IFF_PROMISC;
3356 dev_set_promiscuity(dev, inc);
3359 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3360 is important. Some (broken) drivers set IFF_PROMISC, when
3361 IFF_ALLMULTI is requested not asking us and not reporting.
3363 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3364 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3365 dev->gflags ^= IFF_ALLMULTI;
3366 dev_set_allmulti(dev, inc);
3369 /* Exclude state transition flags, already notified */
3370 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3372 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3377 int dev_set_mtu(struct net_device *dev, int new_mtu)
3381 if (new_mtu == dev->mtu)
3384 /* MTU must be positive. */
3388 if (!netif_device_present(dev))
3392 if (dev->change_mtu)
3393 err = dev->change_mtu(dev, new_mtu);
3396 if (!err && dev->flags & IFF_UP)
3397 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3401 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3405 if (!dev->set_mac_address)
3407 if (sa->sa_family != dev->type)
3409 if (!netif_device_present(dev))
3411 err = dev->set_mac_address(dev, sa);
3413 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3418 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3420 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3423 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3429 case SIOCGIFFLAGS: /* Get interface flags */
3430 ifr->ifr_flags = dev_get_flags(dev);
3433 case SIOCGIFMETRIC: /* Get the metric on the interface
3434 (currently unused) */
3435 ifr->ifr_metric = 0;
3438 case SIOCGIFMTU: /* Get the MTU of a device */
3439 ifr->ifr_mtu = dev->mtu;
3444 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3446 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3447 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3448 ifr->ifr_hwaddr.sa_family = dev->type;
3456 ifr->ifr_map.mem_start = dev->mem_start;
3457 ifr->ifr_map.mem_end = dev->mem_end;
3458 ifr->ifr_map.base_addr = dev->base_addr;
3459 ifr->ifr_map.irq = dev->irq;
3460 ifr->ifr_map.dma = dev->dma;
3461 ifr->ifr_map.port = dev->if_port;
3465 ifr->ifr_ifindex = dev->ifindex;
3469 ifr->ifr_qlen = dev->tx_queue_len;
3473 /* dev_ioctl() should ensure this case
3485 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3487 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3490 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3496 case SIOCSIFFLAGS: /* Set interface flags */
3497 return dev_change_flags(dev, ifr->ifr_flags);
3499 case SIOCSIFMETRIC: /* Set the metric on the interface
3500 (currently unused) */
3503 case SIOCSIFMTU: /* Set the MTU of a device */
3504 return dev_set_mtu(dev, ifr->ifr_mtu);
3507 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3509 case SIOCSIFHWBROADCAST:
3510 if (ifr->ifr_hwaddr.sa_family != dev->type)
3512 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3513 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3514 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3518 if (dev->set_config) {
3519 if (!netif_device_present(dev))
3521 return dev->set_config(dev, &ifr->ifr_map);
3526 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3527 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3529 if (!netif_device_present(dev))
3531 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3535 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3536 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3538 if (!netif_device_present(dev))
3540 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3544 if (ifr->ifr_qlen < 0)
3546 dev->tx_queue_len = ifr->ifr_qlen;
3550 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3551 return dev_change_name(dev, ifr->ifr_newname);
3554 * Unknown or private ioctl
3558 if ((cmd >= SIOCDEVPRIVATE &&
3559 cmd <= SIOCDEVPRIVATE + 15) ||
3560 cmd == SIOCBONDENSLAVE ||
3561 cmd == SIOCBONDRELEASE ||
3562 cmd == SIOCBONDSETHWADDR ||
3563 cmd == SIOCBONDSLAVEINFOQUERY ||
3564 cmd == SIOCBONDINFOQUERY ||
3565 cmd == SIOCBONDCHANGEACTIVE ||
3566 cmd == SIOCGMIIPHY ||
3567 cmd == SIOCGMIIREG ||
3568 cmd == SIOCSMIIREG ||
3569 cmd == SIOCBRADDIF ||
3570 cmd == SIOCBRDELIF ||
3571 cmd == SIOCWANDEV) {
3573 if (dev->do_ioctl) {
3574 if (netif_device_present(dev))
3575 err = dev->do_ioctl(dev, ifr,
3588 * This function handles all "interface"-type I/O control requests. The actual
3589 * 'doing' part of this is dev_ifsioc above.
3593 * dev_ioctl - network device ioctl
3594 * @net: the applicable net namespace
3595 * @cmd: command to issue
3596 * @arg: pointer to a struct ifreq in user space
3598 * Issue ioctl functions to devices. This is normally called by the
3599 * user space syscall interfaces but can sometimes be useful for
3600 * other purposes. The return value is the return from the syscall if
3601 * positive or a negative errno code on error.
3604 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3610 /* One special case: SIOCGIFCONF takes ifconf argument
3611 and requires shared lock, because it sleeps writing
3615 if (cmd == SIOCGIFCONF) {
3617 ret = dev_ifconf(net, (char __user *) arg);
3621 if (cmd == SIOCGIFNAME)
3622 return dev_ifname(net, (struct ifreq __user *)arg);
3624 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3627 ifr.ifr_name[IFNAMSIZ-1] = 0;
3629 colon = strchr(ifr.ifr_name, ':');
3634 * See which interface the caller is talking about.
3639 * These ioctl calls:
3640 * - can be done by all.
3641 * - atomic and do not require locking.
3652 dev_load(net, ifr.ifr_name);
3653 read_lock(&dev_base_lock);
3654 ret = dev_ifsioc_locked(net, &ifr, cmd);
3655 read_unlock(&dev_base_lock);
3659 if (copy_to_user(arg, &ifr,
3660 sizeof(struct ifreq)))
3666 dev_load(net, ifr.ifr_name);
3668 ret = dev_ethtool(net, &ifr);
3673 if (copy_to_user(arg, &ifr,
3674 sizeof(struct ifreq)))
3680 * These ioctl calls:
3681 * - require superuser power.
3682 * - require strict serialization.
3688 if (!capable(CAP_NET_ADMIN))
3690 dev_load(net, ifr.ifr_name);
3692 ret = dev_ifsioc(net, &ifr, cmd);
3697 if (copy_to_user(arg, &ifr,
3698 sizeof(struct ifreq)))
3704 * These ioctl calls:
3705 * - require superuser power.
3706 * - require strict serialization.
3707 * - do not return a value
3717 case SIOCSIFHWBROADCAST:
3720 case SIOCBONDENSLAVE:
3721 case SIOCBONDRELEASE:
3722 case SIOCBONDSETHWADDR:
3723 case SIOCBONDCHANGEACTIVE:
3726 if (!capable(CAP_NET_ADMIN))
3729 case SIOCBONDSLAVEINFOQUERY:
3730 case SIOCBONDINFOQUERY:
3731 dev_load(net, ifr.ifr_name);
3733 ret = dev_ifsioc(net, &ifr, cmd);
3738 /* Get the per device memory space. We can add this but
3739 * currently do not support it */
3741 /* Set the per device memory buffer space.
3742 * Not applicable in our case */
3747 * Unknown or private ioctl.
3750 if (cmd == SIOCWANDEV ||
3751 (cmd >= SIOCDEVPRIVATE &&
3752 cmd <= SIOCDEVPRIVATE + 15)) {
3753 dev_load(net, ifr.ifr_name);
3755 ret = dev_ifsioc(net, &ifr, cmd);
3757 if (!ret && copy_to_user(arg, &ifr,
3758 sizeof(struct ifreq)))
3762 /* Take care of Wireless Extensions */
3763 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3764 return wext_handle_ioctl(net, &ifr, cmd, arg);
3771 * dev_new_index - allocate an ifindex
3772 * @net: the applicable net namespace
3774 * Returns a suitable unique value for a new device interface
3775 * number. The caller must hold the rtnl semaphore or the
3776 * dev_base_lock to be sure it remains unique.
3778 static int dev_new_index(struct net *net)
3784 if (!__dev_get_by_index(net, ifindex))
3789 /* Delayed registration/unregisteration */
3790 static DEFINE_SPINLOCK(net_todo_list_lock);
3791 static LIST_HEAD(net_todo_list);
3793 static void net_set_todo(struct net_device *dev)
3795 spin_lock(&net_todo_list_lock);
3796 list_add_tail(&dev->todo_list, &net_todo_list);
3797 spin_unlock(&net_todo_list_lock);
3800 static void rollback_registered(struct net_device *dev)
3802 BUG_ON(dev_boot_phase);
3805 /* Some devices call without registering for initialization unwind. */
3806 if (dev->reg_state == NETREG_UNINITIALIZED) {
3807 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3808 "was registered\n", dev->name, dev);
3814 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3816 /* If device is running, close it first. */
3819 /* And unlink it from device chain. */
3820 unlist_netdevice(dev);
3822 dev->reg_state = NETREG_UNREGISTERING;
3826 /* Shutdown queueing discipline. */
3830 /* Notify protocols, that we are about to destroy
3831 this device. They should clean all the things.
3833 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3836 * Flush the unicast and multicast chains
3838 dev_addr_discard(dev);
3843 /* Notifier chain MUST detach us from master device. */
3844 WARN_ON(dev->master);
3846 /* Remove entries from kobject tree */
3847 netdev_unregister_kobject(dev);
3854 static void __netdev_init_queue_locks_one(struct net_device *dev,
3855 struct netdev_queue *dev_queue,
3858 spin_lock_init(&dev_queue->_xmit_lock);
3859 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
3860 dev_queue->xmit_lock_owner = -1;
3863 static void netdev_init_queue_locks(struct net_device *dev)
3865 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
3866 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
3870 * register_netdevice - register a network device
3871 * @dev: device to register
3873 * Take a completed network device structure and add it to the kernel
3874 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3875 * chain. 0 is returned on success. A negative errno code is returned
3876 * on a failure to set up the device, or if the name is a duplicate.
3878 * Callers must hold the rtnl semaphore. You may want
3879 * register_netdev() instead of this.
3882 * The locking appears insufficient to guarantee two parallel registers
3883 * will not get the same name.
3886 int register_netdevice(struct net_device *dev)
3888 struct hlist_head *head;
3889 struct hlist_node *p;
3893 BUG_ON(dev_boot_phase);
3898 /* When net_device's are persistent, this will be fatal. */
3899 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3900 BUG_ON(!dev_net(dev));
3903 spin_lock_init(&dev->addr_list_lock);
3904 netdev_set_addr_lockdep_class(dev);
3905 netdev_init_queue_locks(dev);
3909 /* Init, if this function is available */
3911 ret = dev->init(dev);
3919 if (!dev_valid_name(dev->name)) {
3924 dev->ifindex = dev_new_index(net);
3925 if (dev->iflink == -1)
3926 dev->iflink = dev->ifindex;
3928 /* Check for existence of name */
3929 head = dev_name_hash(net, dev->name);
3930 hlist_for_each(p, head) {
3931 struct net_device *d
3932 = hlist_entry(p, struct net_device, name_hlist);
3933 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3939 /* Fix illegal checksum combinations */
3940 if ((dev->features & NETIF_F_HW_CSUM) &&
3941 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3942 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3944 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3947 if ((dev->features & NETIF_F_NO_CSUM) &&
3948 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3949 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3951 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3955 /* Fix illegal SG+CSUM combinations. */
3956 if ((dev->features & NETIF_F_SG) &&
3957 !(dev->features & NETIF_F_ALL_CSUM)) {
3958 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3960 dev->features &= ~NETIF_F_SG;
3963 /* TSO requires that SG is present as well. */
3964 if ((dev->features & NETIF_F_TSO) &&
3965 !(dev->features & NETIF_F_SG)) {
3966 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3968 dev->features &= ~NETIF_F_TSO;
3970 if (dev->features & NETIF_F_UFO) {
3971 if (!(dev->features & NETIF_F_HW_CSUM)) {
3972 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3973 "NETIF_F_HW_CSUM feature.\n",
3975 dev->features &= ~NETIF_F_UFO;
3977 if (!(dev->features & NETIF_F_SG)) {
3978 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3979 "NETIF_F_SG feature.\n",
3981 dev->features &= ~NETIF_F_UFO;
3985 /* Enable software GSO if SG is supported. */
3986 if (dev->features & NETIF_F_SG)
3987 dev->features |= NETIF_F_GSO;
3989 netdev_initialize_kobject(dev);
3990 ret = netdev_register_kobject(dev);
3993 dev->reg_state = NETREG_REGISTERED;
3996 * Default initial state at registry is that the
3997 * device is present.
4000 set_bit(__LINK_STATE_PRESENT, &dev->state);
4002 dev_init_scheduler(dev);
4004 list_netdevice(dev);
4006 /* Notify protocols, that a new device appeared. */
4007 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4008 ret = notifier_to_errno(ret);
4010 rollback_registered(dev);
4011 dev->reg_state = NETREG_UNREGISTERED;
4024 * register_netdev - register a network device
4025 * @dev: device to register
4027 * Take a completed network device structure and add it to the kernel
4028 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4029 * chain. 0 is returned on success. A negative errno code is returned
4030 * on a failure to set up the device, or if the name is a duplicate.
4032 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4033 * and expands the device name if you passed a format string to
4036 int register_netdev(struct net_device *dev)
4043 * If the name is a format string the caller wants us to do a
4046 if (strchr(dev->name, '%')) {
4047 err = dev_alloc_name(dev, dev->name);
4052 err = register_netdevice(dev);
4057 EXPORT_SYMBOL(register_netdev);
4060 * netdev_wait_allrefs - wait until all references are gone.
4062 * This is called when unregistering network devices.
4064 * Any protocol or device that holds a reference should register
4065 * for netdevice notification, and cleanup and put back the
4066 * reference if they receive an UNREGISTER event.
4067 * We can get stuck here if buggy protocols don't correctly
4070 static void netdev_wait_allrefs(struct net_device *dev)
4072 unsigned long rebroadcast_time, warning_time;
4074 rebroadcast_time = warning_time = jiffies;
4075 while (atomic_read(&dev->refcnt) != 0) {
4076 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4079 /* Rebroadcast unregister notification */
4080 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4082 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4084 /* We must not have linkwatch events
4085 * pending on unregister. If this
4086 * happens, we simply run the queue
4087 * unscheduled, resulting in a noop
4090 linkwatch_run_queue();
4095 rebroadcast_time = jiffies;
4100 if (time_after(jiffies, warning_time + 10 * HZ)) {
4101 printk(KERN_EMERG "unregister_netdevice: "
4102 "waiting for %s to become free. Usage "
4104 dev->name, atomic_read(&dev->refcnt));
4105 warning_time = jiffies;
4114 * register_netdevice(x1);
4115 * register_netdevice(x2);
4117 * unregister_netdevice(y1);
4118 * unregister_netdevice(y2);
4124 * We are invoked by rtnl_unlock() after it drops the semaphore.
4125 * This allows us to deal with problems:
4126 * 1) We can delete sysfs objects which invoke hotplug
4127 * without deadlocking with linkwatch via keventd.
4128 * 2) Since we run with the RTNL semaphore not held, we can sleep
4129 * safely in order to wait for the netdev refcnt to drop to zero.
4131 static DEFINE_MUTEX(net_todo_run_mutex);
4132 void netdev_run_todo(void)
4134 struct list_head list;
4136 /* Need to guard against multiple cpu's getting out of order. */
4137 mutex_lock(&net_todo_run_mutex);
4139 /* Not safe to do outside the semaphore. We must not return
4140 * until all unregister events invoked by the local processor
4141 * have been completed (either by this todo run, or one on
4144 if (list_empty(&net_todo_list))
4147 /* Snapshot list, allow later requests */
4148 spin_lock(&net_todo_list_lock);
4149 list_replace_init(&net_todo_list, &list);
4150 spin_unlock(&net_todo_list_lock);
4152 while (!list_empty(&list)) {
4153 struct net_device *dev
4154 = list_entry(list.next, struct net_device, todo_list);
4155 list_del(&dev->todo_list);
4157 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4158 printk(KERN_ERR "network todo '%s' but state %d\n",
4159 dev->name, dev->reg_state);
4164 dev->reg_state = NETREG_UNREGISTERED;
4166 on_each_cpu(flush_backlog, dev, 1);
4168 netdev_wait_allrefs(dev);
4171 BUG_ON(atomic_read(&dev->refcnt));
4172 WARN_ON(dev->ip_ptr);
4173 WARN_ON(dev->ip6_ptr);
4174 WARN_ON(dev->dn_ptr);
4176 if (dev->destructor)
4177 dev->destructor(dev);
4179 /* Free network device */
4180 kobject_put(&dev->dev.kobj);
4184 mutex_unlock(&net_todo_run_mutex);
4187 static struct net_device_stats *internal_stats(struct net_device *dev)
4192 static void netdev_init_one_queue(struct net_device *dev,
4193 struct netdev_queue *queue,
4199 static void netdev_init_queues(struct net_device *dev)
4201 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4202 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4203 spin_lock_init(&dev->tx_global_lock);
4207 * alloc_netdev_mq - allocate network device
4208 * @sizeof_priv: size of private data to allocate space for
4209 * @name: device name format string
4210 * @setup: callback to initialize device
4211 * @queue_count: the number of subqueues to allocate
4213 * Allocates a struct net_device with private data area for driver use
4214 * and performs basic initialization. Also allocates subquue structs
4215 * for each queue on the device at the end of the netdevice.
4217 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4218 void (*setup)(struct net_device *), unsigned int queue_count)
4220 struct netdev_queue *tx;
4221 struct net_device *dev;
4225 BUG_ON(strlen(name) >= sizeof(dev->name));
4227 alloc_size = sizeof(struct net_device);
4229 /* ensure 32-byte alignment of private area */
4230 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4231 alloc_size += sizeof_priv;
4233 /* ensure 32-byte alignment of whole construct */
4234 alloc_size += NETDEV_ALIGN_CONST;
4236 p = kzalloc(alloc_size, GFP_KERNEL);
4238 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4242 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
4244 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4250 dev = (struct net_device *)
4251 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4252 dev->padded = (char *)dev - (char *)p;
4253 dev_net_set(dev, &init_net);
4256 dev->num_tx_queues = queue_count;
4257 dev->real_num_tx_queues = queue_count;
4260 dev->priv = ((char *)dev +
4261 ((sizeof(struct net_device) + NETDEV_ALIGN_CONST)
4262 & ~NETDEV_ALIGN_CONST));
4265 dev->gso_max_size = GSO_MAX_SIZE;
4267 netdev_init_queues(dev);
4269 dev->get_stats = internal_stats;
4270 netpoll_netdev_init(dev);
4272 strcpy(dev->name, name);
4275 EXPORT_SYMBOL(alloc_netdev_mq);
4278 * free_netdev - free network device
4281 * This function does the last stage of destroying an allocated device
4282 * interface. The reference to the device object is released.
4283 * If this is the last reference then it will be freed.
4285 void free_netdev(struct net_device *dev)
4287 release_net(dev_net(dev));
4291 /* Compatibility with error handling in drivers */
4292 if (dev->reg_state == NETREG_UNINITIALIZED) {
4293 kfree((char *)dev - dev->padded);
4297 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4298 dev->reg_state = NETREG_RELEASED;
4300 /* will free via device release */
4301 put_device(&dev->dev);
4304 /* Synchronize with packet receive processing. */
4305 void synchronize_net(void)
4312 * unregister_netdevice - remove device from the kernel
4315 * This function shuts down a device interface and removes it
4316 * from the kernel tables.
4318 * Callers must hold the rtnl semaphore. You may want
4319 * unregister_netdev() instead of this.
4322 void unregister_netdevice(struct net_device *dev)
4326 rollback_registered(dev);
4327 /* Finish processing unregister after unlock */
4332 * unregister_netdev - remove device from the kernel
4335 * This function shuts down a device interface and removes it
4336 * from the kernel tables.
4338 * This is just a wrapper for unregister_netdevice that takes
4339 * the rtnl semaphore. In general you want to use this and not
4340 * unregister_netdevice.
4342 void unregister_netdev(struct net_device *dev)
4345 unregister_netdevice(dev);
4349 EXPORT_SYMBOL(unregister_netdev);
4352 * dev_change_net_namespace - move device to different nethost namespace
4354 * @net: network namespace
4355 * @pat: If not NULL name pattern to try if the current device name
4356 * is already taken in the destination network namespace.
4358 * This function shuts down a device interface and moves it
4359 * to a new network namespace. On success 0 is returned, on
4360 * a failure a netagive errno code is returned.
4362 * Callers must hold the rtnl semaphore.
4365 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4368 const char *destname;
4373 /* Don't allow namespace local devices to be moved. */
4375 if (dev->features & NETIF_F_NETNS_LOCAL)
4378 /* Ensure the device has been registrered */
4380 if (dev->reg_state != NETREG_REGISTERED)
4383 /* Get out if there is nothing todo */
4385 if (net_eq(dev_net(dev), net))
4388 /* Pick the destination device name, and ensure
4389 * we can use it in the destination network namespace.
4392 destname = dev->name;
4393 if (__dev_get_by_name(net, destname)) {
4394 /* We get here if we can't use the current device name */
4397 if (!dev_valid_name(pat))
4399 if (strchr(pat, '%')) {
4400 if (__dev_alloc_name(net, pat, buf) < 0)
4405 if (__dev_get_by_name(net, destname))
4410 * And now a mini version of register_netdevice unregister_netdevice.
4413 /* If device is running close it first. */
4416 /* And unlink it from device chain */
4418 unlist_netdevice(dev);
4422 /* Shutdown queueing discipline. */
4425 /* Notify protocols, that we are about to destroy
4426 this device. They should clean all the things.
4428 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4431 * Flush the unicast and multicast chains
4433 dev_addr_discard(dev);
4435 /* Actually switch the network namespace */
4436 dev_net_set(dev, net);
4438 /* Assign the new device name */
4439 if (destname != dev->name)
4440 strcpy(dev->name, destname);
4442 /* If there is an ifindex conflict assign a new one */
4443 if (__dev_get_by_index(net, dev->ifindex)) {
4444 int iflink = (dev->iflink == dev->ifindex);
4445 dev->ifindex = dev_new_index(net);
4447 dev->iflink = dev->ifindex;
4450 /* Fixup kobjects */
4451 netdev_unregister_kobject(dev);
4452 err = netdev_register_kobject(dev);
4455 /* Add the device back in the hashes */
4456 list_netdevice(dev);
4458 /* Notify protocols, that a new device appeared. */
4459 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4467 static int dev_cpu_callback(struct notifier_block *nfb,
4468 unsigned long action,
4471 struct sk_buff **list_skb;
4472 struct Qdisc **list_net;
4473 struct sk_buff *skb;
4474 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4475 struct softnet_data *sd, *oldsd;
4477 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4480 local_irq_disable();
4481 cpu = smp_processor_id();
4482 sd = &per_cpu(softnet_data, cpu);
4483 oldsd = &per_cpu(softnet_data, oldcpu);
4485 /* Find end of our completion_queue. */
4486 list_skb = &sd->completion_queue;
4488 list_skb = &(*list_skb)->next;
4489 /* Append completion queue from offline CPU. */
4490 *list_skb = oldsd->completion_queue;
4491 oldsd->completion_queue = NULL;
4493 /* Find end of our output_queue. */
4494 list_net = &sd->output_queue;
4496 list_net = &(*list_net)->next_sched;
4497 /* Append output queue from offline CPU. */
4498 *list_net = oldsd->output_queue;
4499 oldsd->output_queue = NULL;
4501 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4504 /* Process offline CPU's input_pkt_queue */
4505 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4511 #ifdef CONFIG_NET_DMA
4513 * net_dma_rebalance - try to maintain one DMA channel per CPU
4514 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4516 * This is called when the number of channels allocated to the net_dma client
4517 * changes. The net_dma client tries to have one DMA channel per CPU.
4520 static void net_dma_rebalance(struct net_dma *net_dma)
4522 unsigned int cpu, i, n, chan_idx;
4523 struct dma_chan *chan;
4525 if (cpus_empty(net_dma->channel_mask)) {
4526 for_each_online_cpu(cpu)
4527 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4532 cpu = first_cpu(cpu_online_map);
4534 for_each_cpu_mask_nr(chan_idx, net_dma->channel_mask) {
4535 chan = net_dma->channels[chan_idx];
4537 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4538 + (i < (num_online_cpus() %
4539 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4542 per_cpu(softnet_data, cpu).net_dma = chan;
4543 cpu = next_cpu(cpu, cpu_online_map);
4551 * netdev_dma_event - event callback for the net_dma_client
4552 * @client: should always be net_dma_client
4553 * @chan: DMA channel for the event
4554 * @state: DMA state to be handled
4556 static enum dma_state_client
4557 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4558 enum dma_state state)
4560 int i, found = 0, pos = -1;
4561 struct net_dma *net_dma =
4562 container_of(client, struct net_dma, client);
4563 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4565 spin_lock(&net_dma->lock);
4567 case DMA_RESOURCE_AVAILABLE:
4568 for (i = 0; i < nr_cpu_ids; i++)
4569 if (net_dma->channels[i] == chan) {
4572 } else if (net_dma->channels[i] == NULL && pos < 0)
4575 if (!found && pos >= 0) {
4577 net_dma->channels[pos] = chan;
4578 cpu_set(pos, net_dma->channel_mask);
4579 net_dma_rebalance(net_dma);
4582 case DMA_RESOURCE_REMOVED:
4583 for (i = 0; i < nr_cpu_ids; i++)
4584 if (net_dma->channels[i] == chan) {
4592 cpu_clear(pos, net_dma->channel_mask);
4593 net_dma->channels[i] = NULL;
4594 net_dma_rebalance(net_dma);
4600 spin_unlock(&net_dma->lock);
4606 * netdev_dma_regiser - register the networking subsystem as a DMA client
4608 static int __init netdev_dma_register(void)
4610 net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma),
4612 if (unlikely(!net_dma.channels)) {
4614 "netdev_dma: no memory for net_dma.channels\n");
4617 spin_lock_init(&net_dma.lock);
4618 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4619 dma_async_client_register(&net_dma.client);
4620 dma_async_client_chan_request(&net_dma.client);
4625 static int __init netdev_dma_register(void) { return -ENODEV; }
4626 #endif /* CONFIG_NET_DMA */
4629 * netdev_compute_feature - compute conjunction of two feature sets
4630 * @all: first feature set
4631 * @one: second feature set
4633 * Computes a new feature set after adding a device with feature set
4634 * @one to the master device with current feature set @all. Returns
4635 * the new feature set.
4637 int netdev_compute_features(unsigned long all, unsigned long one)
4639 /* if device needs checksumming, downgrade to hw checksumming */
4640 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4641 all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
4643 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4644 if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
4645 all ^= NETIF_F_HW_CSUM
4646 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4648 if (one & NETIF_F_GSO)
4649 one |= NETIF_F_GSO_SOFTWARE;
4652 /* If even one device supports robust GSO, enable it for all. */
4653 if (one & NETIF_F_GSO_ROBUST)
4654 all |= NETIF_F_GSO_ROBUST;
4656 all &= one | NETIF_F_LLTX;
4658 if (!(all & NETIF_F_ALL_CSUM))
4660 if (!(all & NETIF_F_SG))
4661 all &= ~NETIF_F_GSO_MASK;
4665 EXPORT_SYMBOL(netdev_compute_features);
4667 static struct hlist_head *netdev_create_hash(void)
4670 struct hlist_head *hash;
4672 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4674 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4675 INIT_HLIST_HEAD(&hash[i]);
4680 /* Initialize per network namespace state */
4681 static int __net_init netdev_init(struct net *net)
4683 INIT_LIST_HEAD(&net->dev_base_head);
4685 net->dev_name_head = netdev_create_hash();
4686 if (net->dev_name_head == NULL)
4689 net->dev_index_head = netdev_create_hash();
4690 if (net->dev_index_head == NULL)
4696 kfree(net->dev_name_head);
4701 char *netdev_drivername(struct net_device *dev, char *buffer, int len)
4703 struct device_driver *driver;
4704 struct device *parent;
4706 if (len <= 0 || !buffer)
4710 parent = dev->dev.parent;
4715 driver = parent->driver;
4716 if (driver && driver->name)
4717 strlcpy(buffer, driver->name, len);
4721 static void __net_exit netdev_exit(struct net *net)
4723 kfree(net->dev_name_head);
4724 kfree(net->dev_index_head);
4727 static struct pernet_operations __net_initdata netdev_net_ops = {
4728 .init = netdev_init,
4729 .exit = netdev_exit,
4732 static void __net_exit default_device_exit(struct net *net)
4734 struct net_device *dev, *next;
4736 * Push all migratable of the network devices back to the
4737 * initial network namespace
4740 for_each_netdev_safe(net, dev, next) {
4742 char fb_name[IFNAMSIZ];
4744 /* Ignore unmoveable devices (i.e. loopback) */
4745 if (dev->features & NETIF_F_NETNS_LOCAL)
4748 /* Push remaing network devices to init_net */
4749 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
4750 err = dev_change_net_namespace(dev, &init_net, fb_name);
4752 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
4753 __func__, dev->name, err);
4760 static struct pernet_operations __net_initdata default_device_ops = {
4761 .exit = default_device_exit,
4765 * Initialize the DEV module. At boot time this walks the device list and
4766 * unhooks any devices that fail to initialise (normally hardware not
4767 * present) and leaves us with a valid list of present and active devices.
4772 * This is called single threaded during boot, so no need
4773 * to take the rtnl semaphore.
4775 static int __init net_dev_init(void)
4777 int i, rc = -ENOMEM;
4779 BUG_ON(!dev_boot_phase);
4781 if (dev_proc_init())
4784 if (netdev_kobject_init())
4787 INIT_LIST_HEAD(&ptype_all);
4788 for (i = 0; i < PTYPE_HASH_SIZE; i++)
4789 INIT_LIST_HEAD(&ptype_base[i]);
4791 if (register_pernet_subsys(&netdev_net_ops))
4794 if (register_pernet_device(&default_device_ops))
4798 * Initialise the packet receive queues.
4801 for_each_possible_cpu(i) {
4802 struct softnet_data *queue;
4804 queue = &per_cpu(softnet_data, i);
4805 skb_queue_head_init(&queue->input_pkt_queue);
4806 queue->completion_queue = NULL;
4807 INIT_LIST_HEAD(&queue->poll_list);
4809 queue->backlog.poll = process_backlog;
4810 queue->backlog.weight = weight_p;
4813 netdev_dma_register();
4817 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
4818 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
4820 hotcpu_notifier(dev_cpu_callback, 0);
4828 subsys_initcall(net_dev_init);
4830 EXPORT_SYMBOL(__dev_get_by_index);
4831 EXPORT_SYMBOL(__dev_get_by_name);
4832 EXPORT_SYMBOL(__dev_remove_pack);
4833 EXPORT_SYMBOL(dev_valid_name);
4834 EXPORT_SYMBOL(dev_add_pack);
4835 EXPORT_SYMBOL(dev_alloc_name);
4836 EXPORT_SYMBOL(dev_close);
4837 EXPORT_SYMBOL(dev_get_by_flags);
4838 EXPORT_SYMBOL(dev_get_by_index);
4839 EXPORT_SYMBOL(dev_get_by_name);
4840 EXPORT_SYMBOL(dev_open);
4841 EXPORT_SYMBOL(dev_queue_xmit);
4842 EXPORT_SYMBOL(dev_remove_pack);
4843 EXPORT_SYMBOL(dev_set_allmulti);
4844 EXPORT_SYMBOL(dev_set_promiscuity);
4845 EXPORT_SYMBOL(dev_change_flags);
4846 EXPORT_SYMBOL(dev_set_mtu);
4847 EXPORT_SYMBOL(dev_set_mac_address);
4848 EXPORT_SYMBOL(free_netdev);
4849 EXPORT_SYMBOL(netdev_boot_setup_check);
4850 EXPORT_SYMBOL(netdev_set_master);
4851 EXPORT_SYMBOL(netdev_state_change);
4852 EXPORT_SYMBOL(netif_receive_skb);
4853 EXPORT_SYMBOL(netif_rx);
4854 EXPORT_SYMBOL(register_gifconf);
4855 EXPORT_SYMBOL(register_netdevice);
4856 EXPORT_SYMBOL(register_netdevice_notifier);
4857 EXPORT_SYMBOL(skb_checksum_help);
4858 EXPORT_SYMBOL(synchronize_net);
4859 EXPORT_SYMBOL(unregister_netdevice);
4860 EXPORT_SYMBOL(unregister_netdevice_notifier);
4861 EXPORT_SYMBOL(net_enable_timestamp);
4862 EXPORT_SYMBOL(net_disable_timestamp);
4863 EXPORT_SYMBOL(dev_get_flags);
4865 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4866 EXPORT_SYMBOL(br_handle_frame_hook);
4867 EXPORT_SYMBOL(br_fdb_get_hook);
4868 EXPORT_SYMBOL(br_fdb_put_hook);
4872 EXPORT_SYMBOL(dev_load);
4875 EXPORT_PER_CPU_SYMBOL(softnet_data);