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);
1808 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1812 /* The device has no queue. Common case for software devices:
1813 loopback, all the sorts of tunnels...
1815 Really, it is unlikely that netif_tx_lock protection is necessary
1816 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1818 However, it is possible, that they rely on protection
1821 Check this and shot the lock. It is not prone from deadlocks.
1822 Either shot noqueue qdisc, it is even simpler 8)
1824 if (dev->flags & IFF_UP) {
1825 int cpu = smp_processor_id(); /* ok because BHs are off */
1827 if (txq->xmit_lock_owner != cpu) {
1829 HARD_TX_LOCK(dev, txq, cpu);
1831 if (!netif_tx_queue_stopped(txq)) {
1833 if (!dev_hard_start_xmit(skb, dev, txq)) {
1834 HARD_TX_UNLOCK(dev, txq);
1838 HARD_TX_UNLOCK(dev, txq);
1839 if (net_ratelimit())
1840 printk(KERN_CRIT "Virtual device %s asks to "
1841 "queue packet!\n", dev->name);
1843 /* Recursion is detected! It is possible,
1845 if (net_ratelimit())
1846 printk(KERN_CRIT "Dead loop on virtual device "
1847 "%s, fix it urgently!\n", dev->name);
1852 rcu_read_unlock_bh();
1858 rcu_read_unlock_bh();
1863 /*=======================================================================
1865 =======================================================================*/
1867 int netdev_max_backlog __read_mostly = 1000;
1868 int netdev_budget __read_mostly = 300;
1869 int weight_p __read_mostly = 64; /* old backlog weight */
1871 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1875 * netif_rx - post buffer to the network code
1876 * @skb: buffer to post
1878 * This function receives a packet from a device driver and queues it for
1879 * the upper (protocol) levels to process. It always succeeds. The buffer
1880 * may be dropped during processing for congestion control or by the
1884 * NET_RX_SUCCESS (no congestion)
1885 * NET_RX_DROP (packet was dropped)
1889 int netif_rx(struct sk_buff *skb)
1891 struct softnet_data *queue;
1892 unsigned long flags;
1894 /* if netpoll wants it, pretend we never saw it */
1895 if (netpoll_rx(skb))
1898 if (!skb->tstamp.tv64)
1902 * The code is rearranged so that the path is the most
1903 * short when CPU is congested, but is still operating.
1905 local_irq_save(flags);
1906 queue = &__get_cpu_var(softnet_data);
1908 __get_cpu_var(netdev_rx_stat).total++;
1909 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1910 if (queue->input_pkt_queue.qlen) {
1912 __skb_queue_tail(&queue->input_pkt_queue, skb);
1913 local_irq_restore(flags);
1914 return NET_RX_SUCCESS;
1917 napi_schedule(&queue->backlog);
1921 __get_cpu_var(netdev_rx_stat).dropped++;
1922 local_irq_restore(flags);
1928 int netif_rx_ni(struct sk_buff *skb)
1933 err = netif_rx(skb);
1934 if (local_softirq_pending())
1941 EXPORT_SYMBOL(netif_rx_ni);
1943 static inline struct net_device *skb_bond(struct sk_buff *skb)
1945 struct net_device *dev = skb->dev;
1948 if (skb_bond_should_drop(skb)) {
1952 skb->dev = dev->master;
1959 static void net_tx_action(struct softirq_action *h)
1961 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1963 if (sd->completion_queue) {
1964 struct sk_buff *clist;
1966 local_irq_disable();
1967 clist = sd->completion_queue;
1968 sd->completion_queue = NULL;
1972 struct sk_buff *skb = clist;
1973 clist = clist->next;
1975 WARN_ON(atomic_read(&skb->users));
1980 if (sd->output_queue) {
1983 local_irq_disable();
1984 head = sd->output_queue;
1985 sd->output_queue = NULL;
1989 struct Qdisc *q = head;
1990 spinlock_t *root_lock;
1992 head = head->next_sched;
1994 smp_mb__before_clear_bit();
1995 clear_bit(__QDISC_STATE_SCHED, &q->state);
1997 root_lock = qdisc_lock(q);
1998 if (spin_trylock(root_lock)) {
2000 spin_unlock(root_lock);
2002 __netif_schedule(q);
2008 static inline int deliver_skb(struct sk_buff *skb,
2009 struct packet_type *pt_prev,
2010 struct net_device *orig_dev)
2012 atomic_inc(&skb->users);
2013 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2016 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2017 /* These hooks defined here for ATM */
2019 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2020 unsigned char *addr);
2021 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
2024 * If bridge module is loaded call bridging hook.
2025 * returns NULL if packet was consumed.
2027 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2028 struct sk_buff *skb) __read_mostly;
2029 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2030 struct packet_type **pt_prev, int *ret,
2031 struct net_device *orig_dev)
2033 struct net_bridge_port *port;
2035 if (skb->pkt_type == PACKET_LOOPBACK ||
2036 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2040 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2044 return br_handle_frame_hook(port, skb);
2047 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2050 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2051 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2052 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2054 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2055 struct packet_type **pt_prev,
2057 struct net_device *orig_dev)
2059 if (skb->dev->macvlan_port == NULL)
2063 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2066 return macvlan_handle_frame_hook(skb);
2069 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2072 #ifdef CONFIG_NET_CLS_ACT
2073 /* TODO: Maybe we should just force sch_ingress to be compiled in
2074 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2075 * a compare and 2 stores extra right now if we dont have it on
2076 * but have CONFIG_NET_CLS_ACT
2077 * NOTE: This doesnt stop any functionality; if you dont have
2078 * the ingress scheduler, you just cant add policies on ingress.
2081 static int ing_filter(struct sk_buff *skb)
2083 struct net_device *dev = skb->dev;
2084 u32 ttl = G_TC_RTTL(skb->tc_verd);
2085 struct netdev_queue *rxq;
2086 int result = TC_ACT_OK;
2089 if (MAX_RED_LOOP < ttl++) {
2091 "Redir loop detected Dropping packet (%d->%d)\n",
2092 skb->iif, dev->ifindex);
2096 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2097 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2099 rxq = &dev->rx_queue;
2102 if (q != &noop_qdisc) {
2103 spin_lock(qdisc_lock(q));
2104 result = qdisc_enqueue_root(skb, q);
2105 spin_unlock(qdisc_lock(q));
2111 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2112 struct packet_type **pt_prev,
2113 int *ret, struct net_device *orig_dev)
2115 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2119 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2122 /* Huh? Why does turning on AF_PACKET affect this? */
2123 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2126 switch (ing_filter(skb)) {
2140 * netif_nit_deliver - deliver received packets to network taps
2143 * This function is used to deliver incoming packets to network
2144 * taps. It should be used when the normal netif_receive_skb path
2145 * is bypassed, for example because of VLAN acceleration.
2147 void netif_nit_deliver(struct sk_buff *skb)
2149 struct packet_type *ptype;
2151 if (list_empty(&ptype_all))
2154 skb_reset_network_header(skb);
2155 skb_reset_transport_header(skb);
2156 skb->mac_len = skb->network_header - skb->mac_header;
2159 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2160 if (!ptype->dev || ptype->dev == skb->dev)
2161 deliver_skb(skb, ptype, skb->dev);
2167 * netif_receive_skb - process receive buffer from network
2168 * @skb: buffer to process
2170 * netif_receive_skb() is the main receive data processing function.
2171 * It always succeeds. The buffer may be dropped during processing
2172 * for congestion control or by the protocol layers.
2174 * This function may only be called from softirq context and interrupts
2175 * should be enabled.
2177 * Return values (usually ignored):
2178 * NET_RX_SUCCESS: no congestion
2179 * NET_RX_DROP: packet was dropped
2181 int netif_receive_skb(struct sk_buff *skb)
2183 struct packet_type *ptype, *pt_prev;
2184 struct net_device *orig_dev;
2185 int ret = NET_RX_DROP;
2188 /* if we've gotten here through NAPI, check netpoll */
2189 if (netpoll_receive_skb(skb))
2192 if (!skb->tstamp.tv64)
2196 skb->iif = skb->dev->ifindex;
2198 orig_dev = skb_bond(skb);
2203 __get_cpu_var(netdev_rx_stat).total++;
2205 skb_reset_network_header(skb);
2206 skb_reset_transport_header(skb);
2207 skb->mac_len = skb->network_header - skb->mac_header;
2213 /* Don't receive packets in an exiting network namespace */
2214 if (!net_alive(dev_net(skb->dev)))
2217 #ifdef CONFIG_NET_CLS_ACT
2218 if (skb->tc_verd & TC_NCLS) {
2219 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2224 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2225 if (!ptype->dev || ptype->dev == skb->dev) {
2227 ret = deliver_skb(skb, pt_prev, orig_dev);
2232 #ifdef CONFIG_NET_CLS_ACT
2233 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2239 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2242 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2246 type = skb->protocol;
2247 list_for_each_entry_rcu(ptype,
2248 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2249 if (ptype->type == type &&
2250 (!ptype->dev || ptype->dev == skb->dev)) {
2252 ret = deliver_skb(skb, pt_prev, orig_dev);
2258 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2261 /* Jamal, now you will not able to escape explaining
2262 * me how you were going to use this. :-)
2272 /* Network device is going away, flush any packets still pending */
2273 static void flush_backlog(void *arg)
2275 struct net_device *dev = arg;
2276 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2277 struct sk_buff *skb, *tmp;
2279 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2280 if (skb->dev == dev) {
2281 __skb_unlink(skb, &queue->input_pkt_queue);
2286 static int process_backlog(struct napi_struct *napi, int quota)
2289 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2290 unsigned long start_time = jiffies;
2292 napi->weight = weight_p;
2294 struct sk_buff *skb;
2296 local_irq_disable();
2297 skb = __skb_dequeue(&queue->input_pkt_queue);
2299 __napi_complete(napi);
2305 netif_receive_skb(skb);
2306 } while (++work < quota && jiffies == start_time);
2312 * __napi_schedule - schedule for receive
2313 * @n: entry to schedule
2315 * The entry's receive function will be scheduled to run
2317 void __napi_schedule(struct napi_struct *n)
2319 unsigned long flags;
2321 local_irq_save(flags);
2322 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2323 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2324 local_irq_restore(flags);
2326 EXPORT_SYMBOL(__napi_schedule);
2329 static void net_rx_action(struct softirq_action *h)
2331 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2332 unsigned long start_time = jiffies;
2333 int budget = netdev_budget;
2336 local_irq_disable();
2338 while (!list_empty(list)) {
2339 struct napi_struct *n;
2342 /* If softirq window is exhuasted then punt.
2344 * Note that this is a slight policy change from the
2345 * previous NAPI code, which would allow up to 2
2346 * jiffies to pass before breaking out. The test
2347 * used to be "jiffies - start_time > 1".
2349 if (unlikely(budget <= 0 || jiffies != start_time))
2354 /* Even though interrupts have been re-enabled, this
2355 * access is safe because interrupts can only add new
2356 * entries to the tail of this list, and only ->poll()
2357 * calls can remove this head entry from the list.
2359 n = list_entry(list->next, struct napi_struct, poll_list);
2361 have = netpoll_poll_lock(n);
2365 /* This NAPI_STATE_SCHED test is for avoiding a race
2366 * with netpoll's poll_napi(). Only the entity which
2367 * obtains the lock and sees NAPI_STATE_SCHED set will
2368 * actually make the ->poll() call. Therefore we avoid
2369 * accidently calling ->poll() when NAPI is not scheduled.
2372 if (test_bit(NAPI_STATE_SCHED, &n->state))
2373 work = n->poll(n, weight);
2375 WARN_ON_ONCE(work > weight);
2379 local_irq_disable();
2381 /* Drivers must not modify the NAPI state if they
2382 * consume the entire weight. In such cases this code
2383 * still "owns" the NAPI instance and therefore can
2384 * move the instance around on the list at-will.
2386 if (unlikely(work == weight)) {
2387 if (unlikely(napi_disable_pending(n)))
2390 list_move_tail(&n->poll_list, list);
2393 netpoll_poll_unlock(have);
2398 #ifdef CONFIG_NET_DMA
2400 * There may not be any more sk_buffs coming right now, so push
2401 * any pending DMA copies to hardware
2403 if (!cpus_empty(net_dma.channel_mask)) {
2405 for_each_cpu_mask_nr(chan_idx, net_dma.channel_mask) {
2406 struct dma_chan *chan = net_dma.channels[chan_idx];
2408 dma_async_memcpy_issue_pending(chan);
2416 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2417 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2421 static gifconf_func_t * gifconf_list [NPROTO];
2424 * register_gifconf - register a SIOCGIF handler
2425 * @family: Address family
2426 * @gifconf: Function handler
2428 * Register protocol dependent address dumping routines. The handler
2429 * that is passed must not be freed or reused until it has been replaced
2430 * by another handler.
2432 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2434 if (family >= NPROTO)
2436 gifconf_list[family] = gifconf;
2442 * Map an interface index to its name (SIOCGIFNAME)
2446 * We need this ioctl for efficient implementation of the
2447 * if_indextoname() function required by the IPv6 API. Without
2448 * it, we would have to search all the interfaces to find a
2452 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2454 struct net_device *dev;
2458 * Fetch the caller's info block.
2461 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2464 read_lock(&dev_base_lock);
2465 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2467 read_unlock(&dev_base_lock);
2471 strcpy(ifr.ifr_name, dev->name);
2472 read_unlock(&dev_base_lock);
2474 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2480 * Perform a SIOCGIFCONF call. This structure will change
2481 * size eventually, and there is nothing I can do about it.
2482 * Thus we will need a 'compatibility mode'.
2485 static int dev_ifconf(struct net *net, char __user *arg)
2488 struct net_device *dev;
2495 * Fetch the caller's info block.
2498 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2505 * Loop over the interfaces, and write an info block for each.
2509 for_each_netdev(net, dev) {
2510 for (i = 0; i < NPROTO; i++) {
2511 if (gifconf_list[i]) {
2514 done = gifconf_list[i](dev, NULL, 0);
2516 done = gifconf_list[i](dev, pos + total,
2526 * All done. Write the updated control block back to the caller.
2528 ifc.ifc_len = total;
2531 * Both BSD and Solaris return 0 here, so we do too.
2533 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2536 #ifdef CONFIG_PROC_FS
2538 * This is invoked by the /proc filesystem handler to display a device
2541 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2542 __acquires(dev_base_lock)
2544 struct net *net = seq_file_net(seq);
2546 struct net_device *dev;
2548 read_lock(&dev_base_lock);
2550 return SEQ_START_TOKEN;
2553 for_each_netdev(net, dev)
2560 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2562 struct net *net = seq_file_net(seq);
2564 return v == SEQ_START_TOKEN ?
2565 first_net_device(net) : next_net_device((struct net_device *)v);
2568 void dev_seq_stop(struct seq_file *seq, void *v)
2569 __releases(dev_base_lock)
2571 read_unlock(&dev_base_lock);
2574 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2576 struct net_device_stats *stats = dev->get_stats(dev);
2578 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2579 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2580 dev->name, stats->rx_bytes, stats->rx_packets,
2582 stats->rx_dropped + stats->rx_missed_errors,
2583 stats->rx_fifo_errors,
2584 stats->rx_length_errors + stats->rx_over_errors +
2585 stats->rx_crc_errors + stats->rx_frame_errors,
2586 stats->rx_compressed, stats->multicast,
2587 stats->tx_bytes, stats->tx_packets,
2588 stats->tx_errors, stats->tx_dropped,
2589 stats->tx_fifo_errors, stats->collisions,
2590 stats->tx_carrier_errors +
2591 stats->tx_aborted_errors +
2592 stats->tx_window_errors +
2593 stats->tx_heartbeat_errors,
2594 stats->tx_compressed);
2598 * Called from the PROCfs module. This now uses the new arbitrary sized
2599 * /proc/net interface to create /proc/net/dev
2601 static int dev_seq_show(struct seq_file *seq, void *v)
2603 if (v == SEQ_START_TOKEN)
2604 seq_puts(seq, "Inter-| Receive "
2606 " face |bytes packets errs drop fifo frame "
2607 "compressed multicast|bytes packets errs "
2608 "drop fifo colls carrier compressed\n");
2610 dev_seq_printf_stats(seq, v);
2614 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2616 struct netif_rx_stats *rc = NULL;
2618 while (*pos < nr_cpu_ids)
2619 if (cpu_online(*pos)) {
2620 rc = &per_cpu(netdev_rx_stat, *pos);
2627 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2629 return softnet_get_online(pos);
2632 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2635 return softnet_get_online(pos);
2638 static void softnet_seq_stop(struct seq_file *seq, void *v)
2642 static int softnet_seq_show(struct seq_file *seq, void *v)
2644 struct netif_rx_stats *s = v;
2646 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2647 s->total, s->dropped, s->time_squeeze, 0,
2648 0, 0, 0, 0, /* was fastroute */
2653 static const struct seq_operations dev_seq_ops = {
2654 .start = dev_seq_start,
2655 .next = dev_seq_next,
2656 .stop = dev_seq_stop,
2657 .show = dev_seq_show,
2660 static int dev_seq_open(struct inode *inode, struct file *file)
2662 return seq_open_net(inode, file, &dev_seq_ops,
2663 sizeof(struct seq_net_private));
2666 static const struct file_operations dev_seq_fops = {
2667 .owner = THIS_MODULE,
2668 .open = dev_seq_open,
2670 .llseek = seq_lseek,
2671 .release = seq_release_net,
2674 static const struct seq_operations softnet_seq_ops = {
2675 .start = softnet_seq_start,
2676 .next = softnet_seq_next,
2677 .stop = softnet_seq_stop,
2678 .show = softnet_seq_show,
2681 static int softnet_seq_open(struct inode *inode, struct file *file)
2683 return seq_open(file, &softnet_seq_ops);
2686 static const struct file_operations softnet_seq_fops = {
2687 .owner = THIS_MODULE,
2688 .open = softnet_seq_open,
2690 .llseek = seq_lseek,
2691 .release = seq_release,
2694 static void *ptype_get_idx(loff_t pos)
2696 struct packet_type *pt = NULL;
2700 list_for_each_entry_rcu(pt, &ptype_all, list) {
2706 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
2707 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2716 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2720 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2723 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2725 struct packet_type *pt;
2726 struct list_head *nxt;
2730 if (v == SEQ_START_TOKEN)
2731 return ptype_get_idx(0);
2734 nxt = pt->list.next;
2735 if (pt->type == htons(ETH_P_ALL)) {
2736 if (nxt != &ptype_all)
2739 nxt = ptype_base[0].next;
2741 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
2743 while (nxt == &ptype_base[hash]) {
2744 if (++hash >= PTYPE_HASH_SIZE)
2746 nxt = ptype_base[hash].next;
2749 return list_entry(nxt, struct packet_type, list);
2752 static void ptype_seq_stop(struct seq_file *seq, void *v)
2758 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2760 #ifdef CONFIG_KALLSYMS
2761 unsigned long offset = 0, symsize;
2762 const char *symname;
2766 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2773 modname = delim = "";
2774 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2780 seq_printf(seq, "[%p]", sym);
2783 static int ptype_seq_show(struct seq_file *seq, void *v)
2785 struct packet_type *pt = v;
2787 if (v == SEQ_START_TOKEN)
2788 seq_puts(seq, "Type Device Function\n");
2789 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
2790 if (pt->type == htons(ETH_P_ALL))
2791 seq_puts(seq, "ALL ");
2793 seq_printf(seq, "%04x", ntohs(pt->type));
2795 seq_printf(seq, " %-8s ",
2796 pt->dev ? pt->dev->name : "");
2797 ptype_seq_decode(seq, pt->func);
2798 seq_putc(seq, '\n');
2804 static const struct seq_operations ptype_seq_ops = {
2805 .start = ptype_seq_start,
2806 .next = ptype_seq_next,
2807 .stop = ptype_seq_stop,
2808 .show = ptype_seq_show,
2811 static int ptype_seq_open(struct inode *inode, struct file *file)
2813 return seq_open_net(inode, file, &ptype_seq_ops,
2814 sizeof(struct seq_net_private));
2817 static const struct file_operations ptype_seq_fops = {
2818 .owner = THIS_MODULE,
2819 .open = ptype_seq_open,
2821 .llseek = seq_lseek,
2822 .release = seq_release_net,
2826 static int __net_init dev_proc_net_init(struct net *net)
2830 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2832 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2834 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2837 if (wext_proc_init(net))
2843 proc_net_remove(net, "ptype");
2845 proc_net_remove(net, "softnet_stat");
2847 proc_net_remove(net, "dev");
2851 static void __net_exit dev_proc_net_exit(struct net *net)
2853 wext_proc_exit(net);
2855 proc_net_remove(net, "ptype");
2856 proc_net_remove(net, "softnet_stat");
2857 proc_net_remove(net, "dev");
2860 static struct pernet_operations __net_initdata dev_proc_ops = {
2861 .init = dev_proc_net_init,
2862 .exit = dev_proc_net_exit,
2865 static int __init dev_proc_init(void)
2867 return register_pernet_subsys(&dev_proc_ops);
2870 #define dev_proc_init() 0
2871 #endif /* CONFIG_PROC_FS */
2875 * netdev_set_master - set up master/slave pair
2876 * @slave: slave device
2877 * @master: new master device
2879 * Changes the master device of the slave. Pass %NULL to break the
2880 * bonding. The caller must hold the RTNL semaphore. On a failure
2881 * a negative errno code is returned. On success the reference counts
2882 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2883 * function returns zero.
2885 int netdev_set_master(struct net_device *slave, struct net_device *master)
2887 struct net_device *old = slave->master;
2897 slave->master = master;
2905 slave->flags |= IFF_SLAVE;
2907 slave->flags &= ~IFF_SLAVE;
2909 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2913 static int __dev_set_promiscuity(struct net_device *dev, int inc)
2915 unsigned short old_flags = dev->flags;
2919 dev->flags |= IFF_PROMISC;
2920 dev->promiscuity += inc;
2921 if (dev->promiscuity == 0) {
2924 * If inc causes overflow, untouch promisc and return error.
2927 dev->flags &= ~IFF_PROMISC;
2929 dev->promiscuity -= inc;
2930 printk(KERN_WARNING "%s: promiscuity touches roof, "
2931 "set promiscuity failed, promiscuity feature "
2932 "of device might be broken.\n", dev->name);
2936 if (dev->flags != old_flags) {
2937 printk(KERN_INFO "device %s %s promiscuous mode\n",
2938 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2941 audit_log(current->audit_context, GFP_ATOMIC,
2942 AUDIT_ANOM_PROMISCUOUS,
2943 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2944 dev->name, (dev->flags & IFF_PROMISC),
2945 (old_flags & IFF_PROMISC),
2946 audit_get_loginuid(current),
2947 current->uid, current->gid,
2948 audit_get_sessionid(current));
2950 if (dev->change_rx_flags)
2951 dev->change_rx_flags(dev, IFF_PROMISC);
2957 * dev_set_promiscuity - update promiscuity count on a device
2961 * Add or remove promiscuity from a device. While the count in the device
2962 * remains above zero the interface remains promiscuous. Once it hits zero
2963 * the device reverts back to normal filtering operation. A negative inc
2964 * value is used to drop promiscuity on the device.
2965 * Return 0 if successful or a negative errno code on error.
2967 int dev_set_promiscuity(struct net_device *dev, int inc)
2969 unsigned short old_flags = dev->flags;
2972 err = __dev_set_promiscuity(dev, inc);
2975 if (dev->flags != old_flags)
2976 dev_set_rx_mode(dev);
2981 * dev_set_allmulti - update allmulti count on a device
2985 * Add or remove reception of all multicast frames to a device. While the
2986 * count in the device remains above zero the interface remains listening
2987 * to all interfaces. Once it hits zero the device reverts back to normal
2988 * filtering operation. A negative @inc value is used to drop the counter
2989 * when releasing a resource needing all multicasts.
2990 * Return 0 if successful or a negative errno code on error.
2993 int dev_set_allmulti(struct net_device *dev, int inc)
2995 unsigned short old_flags = dev->flags;
2999 dev->flags |= IFF_ALLMULTI;
3000 dev->allmulti += inc;
3001 if (dev->allmulti == 0) {
3004 * If inc causes overflow, untouch allmulti and return error.
3007 dev->flags &= ~IFF_ALLMULTI;
3009 dev->allmulti -= inc;
3010 printk(KERN_WARNING "%s: allmulti touches roof, "
3011 "set allmulti failed, allmulti feature of "
3012 "device might be broken.\n", dev->name);
3016 if (dev->flags ^ old_flags) {
3017 if (dev->change_rx_flags)
3018 dev->change_rx_flags(dev, IFF_ALLMULTI);
3019 dev_set_rx_mode(dev);
3025 * Upload unicast and multicast address lists to device and
3026 * configure RX filtering. When the device doesn't support unicast
3027 * filtering it is put in promiscuous mode while unicast addresses
3030 void __dev_set_rx_mode(struct net_device *dev)
3032 /* dev_open will call this function so the list will stay sane. */
3033 if (!(dev->flags&IFF_UP))
3036 if (!netif_device_present(dev))
3039 if (dev->set_rx_mode)
3040 dev->set_rx_mode(dev);
3042 /* Unicast addresses changes may only happen under the rtnl,
3043 * therefore calling __dev_set_promiscuity here is safe.
3045 if (dev->uc_count > 0 && !dev->uc_promisc) {
3046 __dev_set_promiscuity(dev, 1);
3047 dev->uc_promisc = 1;
3048 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3049 __dev_set_promiscuity(dev, -1);
3050 dev->uc_promisc = 0;
3053 if (dev->set_multicast_list)
3054 dev->set_multicast_list(dev);
3058 void dev_set_rx_mode(struct net_device *dev)
3060 netif_addr_lock_bh(dev);
3061 __dev_set_rx_mode(dev);
3062 netif_addr_unlock_bh(dev);
3065 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3066 void *addr, int alen, int glbl)
3068 struct dev_addr_list *da;
3070 for (; (da = *list) != NULL; list = &da->next) {
3071 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3072 alen == da->da_addrlen) {
3074 int old_glbl = da->da_gusers;
3091 int __dev_addr_add(struct dev_addr_list **list, int *count,
3092 void *addr, int alen, int glbl)
3094 struct dev_addr_list *da;
3096 for (da = *list; da != NULL; da = da->next) {
3097 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3098 da->da_addrlen == alen) {
3100 int old_glbl = da->da_gusers;
3110 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3113 memcpy(da->da_addr, addr, alen);
3114 da->da_addrlen = alen;
3116 da->da_gusers = glbl ? 1 : 0;
3124 * dev_unicast_delete - Release secondary unicast address.
3126 * @addr: address to delete
3127 * @alen: length of @addr
3129 * Release reference to a secondary unicast address and remove it
3130 * from the device if the reference count drops to zero.
3132 * The caller must hold the rtnl_mutex.
3134 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3140 netif_addr_lock_bh(dev);
3141 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3143 __dev_set_rx_mode(dev);
3144 netif_addr_unlock_bh(dev);
3147 EXPORT_SYMBOL(dev_unicast_delete);
3150 * dev_unicast_add - add a secondary unicast address
3152 * @addr: address to add
3153 * @alen: length of @addr
3155 * Add a secondary unicast address to the device or increase
3156 * the reference count if it already exists.
3158 * The caller must hold the rtnl_mutex.
3160 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3166 netif_addr_lock_bh(dev);
3167 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3169 __dev_set_rx_mode(dev);
3170 netif_addr_unlock_bh(dev);
3173 EXPORT_SYMBOL(dev_unicast_add);
3175 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3176 struct dev_addr_list **from, int *from_count)
3178 struct dev_addr_list *da, *next;
3182 while (da != NULL) {
3184 if (!da->da_synced) {
3185 err = __dev_addr_add(to, to_count,
3186 da->da_addr, da->da_addrlen, 0);
3191 } else if (da->da_users == 1) {
3192 __dev_addr_delete(to, to_count,
3193 da->da_addr, da->da_addrlen, 0);
3194 __dev_addr_delete(from, from_count,
3195 da->da_addr, da->da_addrlen, 0);
3202 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3203 struct dev_addr_list **from, int *from_count)
3205 struct dev_addr_list *da, *next;
3208 while (da != NULL) {
3210 if (da->da_synced) {
3211 __dev_addr_delete(to, to_count,
3212 da->da_addr, da->da_addrlen, 0);
3214 __dev_addr_delete(from, from_count,
3215 da->da_addr, da->da_addrlen, 0);
3222 * dev_unicast_sync - Synchronize device's unicast list to another device
3223 * @to: destination device
3224 * @from: source device
3226 * Add newly added addresses to the destination device and release
3227 * addresses that have no users left. The source device must be
3228 * locked by netif_tx_lock_bh.
3230 * This function is intended to be called from the dev->set_rx_mode
3231 * function of layered software devices.
3233 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3237 netif_addr_lock_bh(to);
3238 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3239 &from->uc_list, &from->uc_count);
3241 __dev_set_rx_mode(to);
3242 netif_addr_unlock_bh(to);
3245 EXPORT_SYMBOL(dev_unicast_sync);
3248 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3249 * @to: destination device
3250 * @from: source device
3252 * Remove all addresses that were added to the destination device by
3253 * dev_unicast_sync(). This function is intended to be called from the
3254 * dev->stop function of layered software devices.
3256 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3258 netif_addr_lock_bh(from);
3259 netif_addr_lock(to);
3261 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3262 &from->uc_list, &from->uc_count);
3263 __dev_set_rx_mode(to);
3265 netif_addr_unlock(to);
3266 netif_addr_unlock_bh(from);
3268 EXPORT_SYMBOL(dev_unicast_unsync);
3270 static void __dev_addr_discard(struct dev_addr_list **list)
3272 struct dev_addr_list *tmp;
3274 while (*list != NULL) {
3277 if (tmp->da_users > tmp->da_gusers)
3278 printk("__dev_addr_discard: address leakage! "
3279 "da_users=%d\n", tmp->da_users);
3284 static void dev_addr_discard(struct net_device *dev)
3286 netif_addr_lock_bh(dev);
3288 __dev_addr_discard(&dev->uc_list);
3291 __dev_addr_discard(&dev->mc_list);
3294 netif_addr_unlock_bh(dev);
3297 unsigned dev_get_flags(const struct net_device *dev)
3301 flags = (dev->flags & ~(IFF_PROMISC |
3306 (dev->gflags & (IFF_PROMISC |
3309 if (netif_running(dev)) {
3310 if (netif_oper_up(dev))
3311 flags |= IFF_RUNNING;
3312 if (netif_carrier_ok(dev))
3313 flags |= IFF_LOWER_UP;
3314 if (netif_dormant(dev))
3315 flags |= IFF_DORMANT;
3321 int dev_change_flags(struct net_device *dev, unsigned flags)
3324 int old_flags = dev->flags;
3329 * Set the flags on our device.
3332 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3333 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3335 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3339 * Load in the correct multicast list now the flags have changed.
3342 if (dev->change_rx_flags && (old_flags ^ flags) & IFF_MULTICAST)
3343 dev->change_rx_flags(dev, IFF_MULTICAST);
3345 dev_set_rx_mode(dev);
3348 * Have we downed the interface. We handle IFF_UP ourselves
3349 * according to user attempts to set it, rather than blindly
3354 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3355 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3358 dev_set_rx_mode(dev);
3361 if (dev->flags & IFF_UP &&
3362 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3364 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3366 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3367 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3368 dev->gflags ^= IFF_PROMISC;
3369 dev_set_promiscuity(dev, inc);
3372 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3373 is important. Some (broken) drivers set IFF_PROMISC, when
3374 IFF_ALLMULTI is requested not asking us and not reporting.
3376 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3377 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3378 dev->gflags ^= IFF_ALLMULTI;
3379 dev_set_allmulti(dev, inc);
3382 /* Exclude state transition flags, already notified */
3383 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3385 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3390 int dev_set_mtu(struct net_device *dev, int new_mtu)
3394 if (new_mtu == dev->mtu)
3397 /* MTU must be positive. */
3401 if (!netif_device_present(dev))
3405 if (dev->change_mtu)
3406 err = dev->change_mtu(dev, new_mtu);
3409 if (!err && dev->flags & IFF_UP)
3410 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3414 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3418 if (!dev->set_mac_address)
3420 if (sa->sa_family != dev->type)
3422 if (!netif_device_present(dev))
3424 err = dev->set_mac_address(dev, sa);
3426 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3431 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3433 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3436 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3442 case SIOCGIFFLAGS: /* Get interface flags */
3443 ifr->ifr_flags = dev_get_flags(dev);
3446 case SIOCGIFMETRIC: /* Get the metric on the interface
3447 (currently unused) */
3448 ifr->ifr_metric = 0;
3451 case SIOCGIFMTU: /* Get the MTU of a device */
3452 ifr->ifr_mtu = dev->mtu;
3457 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3459 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3460 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3461 ifr->ifr_hwaddr.sa_family = dev->type;
3469 ifr->ifr_map.mem_start = dev->mem_start;
3470 ifr->ifr_map.mem_end = dev->mem_end;
3471 ifr->ifr_map.base_addr = dev->base_addr;
3472 ifr->ifr_map.irq = dev->irq;
3473 ifr->ifr_map.dma = dev->dma;
3474 ifr->ifr_map.port = dev->if_port;
3478 ifr->ifr_ifindex = dev->ifindex;
3482 ifr->ifr_qlen = dev->tx_queue_len;
3486 /* dev_ioctl() should ensure this case
3498 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3500 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3503 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3509 case SIOCSIFFLAGS: /* Set interface flags */
3510 return dev_change_flags(dev, ifr->ifr_flags);
3512 case SIOCSIFMETRIC: /* Set the metric on the interface
3513 (currently unused) */
3516 case SIOCSIFMTU: /* Set the MTU of a device */
3517 return dev_set_mtu(dev, ifr->ifr_mtu);
3520 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3522 case SIOCSIFHWBROADCAST:
3523 if (ifr->ifr_hwaddr.sa_family != dev->type)
3525 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3526 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3527 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3531 if (dev->set_config) {
3532 if (!netif_device_present(dev))
3534 return dev->set_config(dev, &ifr->ifr_map);
3539 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3540 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3542 if (!netif_device_present(dev))
3544 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3548 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3549 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3551 if (!netif_device_present(dev))
3553 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3557 if (ifr->ifr_qlen < 0)
3559 dev->tx_queue_len = ifr->ifr_qlen;
3563 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3564 return dev_change_name(dev, ifr->ifr_newname);
3567 * Unknown or private ioctl
3571 if ((cmd >= SIOCDEVPRIVATE &&
3572 cmd <= SIOCDEVPRIVATE + 15) ||
3573 cmd == SIOCBONDENSLAVE ||
3574 cmd == SIOCBONDRELEASE ||
3575 cmd == SIOCBONDSETHWADDR ||
3576 cmd == SIOCBONDSLAVEINFOQUERY ||
3577 cmd == SIOCBONDINFOQUERY ||
3578 cmd == SIOCBONDCHANGEACTIVE ||
3579 cmd == SIOCGMIIPHY ||
3580 cmd == SIOCGMIIREG ||
3581 cmd == SIOCSMIIREG ||
3582 cmd == SIOCBRADDIF ||
3583 cmd == SIOCBRDELIF ||
3584 cmd == SIOCWANDEV) {
3586 if (dev->do_ioctl) {
3587 if (netif_device_present(dev))
3588 err = dev->do_ioctl(dev, ifr,
3601 * This function handles all "interface"-type I/O control requests. The actual
3602 * 'doing' part of this is dev_ifsioc above.
3606 * dev_ioctl - network device ioctl
3607 * @net: the applicable net namespace
3608 * @cmd: command to issue
3609 * @arg: pointer to a struct ifreq in user space
3611 * Issue ioctl functions to devices. This is normally called by the
3612 * user space syscall interfaces but can sometimes be useful for
3613 * other purposes. The return value is the return from the syscall if
3614 * positive or a negative errno code on error.
3617 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3623 /* One special case: SIOCGIFCONF takes ifconf argument
3624 and requires shared lock, because it sleeps writing
3628 if (cmd == SIOCGIFCONF) {
3630 ret = dev_ifconf(net, (char __user *) arg);
3634 if (cmd == SIOCGIFNAME)
3635 return dev_ifname(net, (struct ifreq __user *)arg);
3637 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3640 ifr.ifr_name[IFNAMSIZ-1] = 0;
3642 colon = strchr(ifr.ifr_name, ':');
3647 * See which interface the caller is talking about.
3652 * These ioctl calls:
3653 * - can be done by all.
3654 * - atomic and do not require locking.
3665 dev_load(net, ifr.ifr_name);
3666 read_lock(&dev_base_lock);
3667 ret = dev_ifsioc_locked(net, &ifr, cmd);
3668 read_unlock(&dev_base_lock);
3672 if (copy_to_user(arg, &ifr,
3673 sizeof(struct ifreq)))
3679 dev_load(net, ifr.ifr_name);
3681 ret = dev_ethtool(net, &ifr);
3686 if (copy_to_user(arg, &ifr,
3687 sizeof(struct ifreq)))
3693 * These ioctl calls:
3694 * - require superuser power.
3695 * - require strict serialization.
3701 if (!capable(CAP_NET_ADMIN))
3703 dev_load(net, ifr.ifr_name);
3705 ret = dev_ifsioc(net, &ifr, cmd);
3710 if (copy_to_user(arg, &ifr,
3711 sizeof(struct ifreq)))
3717 * These ioctl calls:
3718 * - require superuser power.
3719 * - require strict serialization.
3720 * - do not return a value
3730 case SIOCSIFHWBROADCAST:
3733 case SIOCBONDENSLAVE:
3734 case SIOCBONDRELEASE:
3735 case SIOCBONDSETHWADDR:
3736 case SIOCBONDCHANGEACTIVE:
3739 if (!capable(CAP_NET_ADMIN))
3742 case SIOCBONDSLAVEINFOQUERY:
3743 case SIOCBONDINFOQUERY:
3744 dev_load(net, ifr.ifr_name);
3746 ret = dev_ifsioc(net, &ifr, cmd);
3751 /* Get the per device memory space. We can add this but
3752 * currently do not support it */
3754 /* Set the per device memory buffer space.
3755 * Not applicable in our case */
3760 * Unknown or private ioctl.
3763 if (cmd == SIOCWANDEV ||
3764 (cmd >= SIOCDEVPRIVATE &&
3765 cmd <= SIOCDEVPRIVATE + 15)) {
3766 dev_load(net, ifr.ifr_name);
3768 ret = dev_ifsioc(net, &ifr, cmd);
3770 if (!ret && copy_to_user(arg, &ifr,
3771 sizeof(struct ifreq)))
3775 /* Take care of Wireless Extensions */
3776 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3777 return wext_handle_ioctl(net, &ifr, cmd, arg);
3784 * dev_new_index - allocate an ifindex
3785 * @net: the applicable net namespace
3787 * Returns a suitable unique value for a new device interface
3788 * number. The caller must hold the rtnl semaphore or the
3789 * dev_base_lock to be sure it remains unique.
3791 static int dev_new_index(struct net *net)
3797 if (!__dev_get_by_index(net, ifindex))
3802 /* Delayed registration/unregisteration */
3803 static DEFINE_SPINLOCK(net_todo_list_lock);
3804 static LIST_HEAD(net_todo_list);
3806 static void net_set_todo(struct net_device *dev)
3808 spin_lock(&net_todo_list_lock);
3809 list_add_tail(&dev->todo_list, &net_todo_list);
3810 spin_unlock(&net_todo_list_lock);
3813 static void rollback_registered(struct net_device *dev)
3815 BUG_ON(dev_boot_phase);
3818 /* Some devices call without registering for initialization unwind. */
3819 if (dev->reg_state == NETREG_UNINITIALIZED) {
3820 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3821 "was registered\n", dev->name, dev);
3827 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3829 /* If device is running, close it first. */
3832 /* And unlink it from device chain. */
3833 unlist_netdevice(dev);
3835 dev->reg_state = NETREG_UNREGISTERING;
3839 /* Shutdown queueing discipline. */
3843 /* Notify protocols, that we are about to destroy
3844 this device. They should clean all the things.
3846 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3849 * Flush the unicast and multicast chains
3851 dev_addr_discard(dev);
3856 /* Notifier chain MUST detach us from master device. */
3857 WARN_ON(dev->master);
3859 /* Remove entries from kobject tree */
3860 netdev_unregister_kobject(dev);
3867 static void __netdev_init_queue_locks_one(struct net_device *dev,
3868 struct netdev_queue *dev_queue,
3871 spin_lock_init(&dev_queue->_xmit_lock);
3872 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
3873 dev_queue->xmit_lock_owner = -1;
3876 static void netdev_init_queue_locks(struct net_device *dev)
3878 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
3879 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
3883 * register_netdevice - register a network device
3884 * @dev: device to register
3886 * Take a completed network device structure and add it to the kernel
3887 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3888 * chain. 0 is returned on success. A negative errno code is returned
3889 * on a failure to set up the device, or if the name is a duplicate.
3891 * Callers must hold the rtnl semaphore. You may want
3892 * register_netdev() instead of this.
3895 * The locking appears insufficient to guarantee two parallel registers
3896 * will not get the same name.
3899 int register_netdevice(struct net_device *dev)
3901 struct hlist_head *head;
3902 struct hlist_node *p;
3906 BUG_ON(dev_boot_phase);
3911 /* When net_device's are persistent, this will be fatal. */
3912 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3913 BUG_ON(!dev_net(dev));
3916 spin_lock_init(&dev->addr_list_lock);
3917 netdev_set_addr_lockdep_class(dev);
3918 netdev_init_queue_locks(dev);
3922 /* Init, if this function is available */
3924 ret = dev->init(dev);
3932 if (!dev_valid_name(dev->name)) {
3937 dev->ifindex = dev_new_index(net);
3938 if (dev->iflink == -1)
3939 dev->iflink = dev->ifindex;
3941 /* Check for existence of name */
3942 head = dev_name_hash(net, dev->name);
3943 hlist_for_each(p, head) {
3944 struct net_device *d
3945 = hlist_entry(p, struct net_device, name_hlist);
3946 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3952 /* Fix illegal checksum combinations */
3953 if ((dev->features & NETIF_F_HW_CSUM) &&
3954 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3955 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3957 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3960 if ((dev->features & NETIF_F_NO_CSUM) &&
3961 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3962 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3964 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3968 /* Fix illegal SG+CSUM combinations. */
3969 if ((dev->features & NETIF_F_SG) &&
3970 !(dev->features & NETIF_F_ALL_CSUM)) {
3971 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3973 dev->features &= ~NETIF_F_SG;
3976 /* TSO requires that SG is present as well. */
3977 if ((dev->features & NETIF_F_TSO) &&
3978 !(dev->features & NETIF_F_SG)) {
3979 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3981 dev->features &= ~NETIF_F_TSO;
3983 if (dev->features & NETIF_F_UFO) {
3984 if (!(dev->features & NETIF_F_HW_CSUM)) {
3985 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3986 "NETIF_F_HW_CSUM feature.\n",
3988 dev->features &= ~NETIF_F_UFO;
3990 if (!(dev->features & NETIF_F_SG)) {
3991 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3992 "NETIF_F_SG feature.\n",
3994 dev->features &= ~NETIF_F_UFO;
3998 /* Enable software GSO if SG is supported. */
3999 if (dev->features & NETIF_F_SG)
4000 dev->features |= NETIF_F_GSO;
4002 netdev_initialize_kobject(dev);
4003 ret = netdev_register_kobject(dev);
4006 dev->reg_state = NETREG_REGISTERED;
4009 * Default initial state at registry is that the
4010 * device is present.
4013 set_bit(__LINK_STATE_PRESENT, &dev->state);
4015 dev_init_scheduler(dev);
4017 list_netdevice(dev);
4019 /* Notify protocols, that a new device appeared. */
4020 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4021 ret = notifier_to_errno(ret);
4023 rollback_registered(dev);
4024 dev->reg_state = NETREG_UNREGISTERED;
4037 * register_netdev - register a network device
4038 * @dev: device to register
4040 * Take a completed network device structure and add it to the kernel
4041 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4042 * chain. 0 is returned on success. A negative errno code is returned
4043 * on a failure to set up the device, or if the name is a duplicate.
4045 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4046 * and expands the device name if you passed a format string to
4049 int register_netdev(struct net_device *dev)
4056 * If the name is a format string the caller wants us to do a
4059 if (strchr(dev->name, '%')) {
4060 err = dev_alloc_name(dev, dev->name);
4065 err = register_netdevice(dev);
4070 EXPORT_SYMBOL(register_netdev);
4073 * netdev_wait_allrefs - wait until all references are gone.
4075 * This is called when unregistering network devices.
4077 * Any protocol or device that holds a reference should register
4078 * for netdevice notification, and cleanup and put back the
4079 * reference if they receive an UNREGISTER event.
4080 * We can get stuck here if buggy protocols don't correctly
4083 static void netdev_wait_allrefs(struct net_device *dev)
4085 unsigned long rebroadcast_time, warning_time;
4087 rebroadcast_time = warning_time = jiffies;
4088 while (atomic_read(&dev->refcnt) != 0) {
4089 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4092 /* Rebroadcast unregister notification */
4093 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4095 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4097 /* We must not have linkwatch events
4098 * pending on unregister. If this
4099 * happens, we simply run the queue
4100 * unscheduled, resulting in a noop
4103 linkwatch_run_queue();
4108 rebroadcast_time = jiffies;
4113 if (time_after(jiffies, warning_time + 10 * HZ)) {
4114 printk(KERN_EMERG "unregister_netdevice: "
4115 "waiting for %s to become free. Usage "
4117 dev->name, atomic_read(&dev->refcnt));
4118 warning_time = jiffies;
4127 * register_netdevice(x1);
4128 * register_netdevice(x2);
4130 * unregister_netdevice(y1);
4131 * unregister_netdevice(y2);
4137 * We are invoked by rtnl_unlock() after it drops the semaphore.
4138 * This allows us to deal with problems:
4139 * 1) We can delete sysfs objects which invoke hotplug
4140 * without deadlocking with linkwatch via keventd.
4141 * 2) Since we run with the RTNL semaphore not held, we can sleep
4142 * safely in order to wait for the netdev refcnt to drop to zero.
4144 static DEFINE_MUTEX(net_todo_run_mutex);
4145 void netdev_run_todo(void)
4147 struct list_head list;
4149 /* Need to guard against multiple cpu's getting out of order. */
4150 mutex_lock(&net_todo_run_mutex);
4152 /* Not safe to do outside the semaphore. We must not return
4153 * until all unregister events invoked by the local processor
4154 * have been completed (either by this todo run, or one on
4157 if (list_empty(&net_todo_list))
4160 /* Snapshot list, allow later requests */
4161 spin_lock(&net_todo_list_lock);
4162 list_replace_init(&net_todo_list, &list);
4163 spin_unlock(&net_todo_list_lock);
4165 while (!list_empty(&list)) {
4166 struct net_device *dev
4167 = list_entry(list.next, struct net_device, todo_list);
4168 list_del(&dev->todo_list);
4170 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4171 printk(KERN_ERR "network todo '%s' but state %d\n",
4172 dev->name, dev->reg_state);
4177 dev->reg_state = NETREG_UNREGISTERED;
4179 on_each_cpu(flush_backlog, dev, 1);
4181 netdev_wait_allrefs(dev);
4184 BUG_ON(atomic_read(&dev->refcnt));
4185 WARN_ON(dev->ip_ptr);
4186 WARN_ON(dev->ip6_ptr);
4187 WARN_ON(dev->dn_ptr);
4189 if (dev->destructor)
4190 dev->destructor(dev);
4192 /* Free network device */
4193 kobject_put(&dev->dev.kobj);
4197 mutex_unlock(&net_todo_run_mutex);
4200 static struct net_device_stats *internal_stats(struct net_device *dev)
4205 static void netdev_init_one_queue(struct net_device *dev,
4206 struct netdev_queue *queue,
4212 static void netdev_init_queues(struct net_device *dev)
4214 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4215 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4216 spin_lock_init(&dev->tx_global_lock);
4220 * alloc_netdev_mq - allocate network device
4221 * @sizeof_priv: size of private data to allocate space for
4222 * @name: device name format string
4223 * @setup: callback to initialize device
4224 * @queue_count: the number of subqueues to allocate
4226 * Allocates a struct net_device with private data area for driver use
4227 * and performs basic initialization. Also allocates subquue structs
4228 * for each queue on the device at the end of the netdevice.
4230 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4231 void (*setup)(struct net_device *), unsigned int queue_count)
4233 struct netdev_queue *tx;
4234 struct net_device *dev;
4238 BUG_ON(strlen(name) >= sizeof(dev->name));
4240 alloc_size = sizeof(struct net_device);
4242 /* ensure 32-byte alignment of private area */
4243 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4244 alloc_size += sizeof_priv;
4246 /* ensure 32-byte alignment of whole construct */
4247 alloc_size += NETDEV_ALIGN_CONST;
4249 p = kzalloc(alloc_size, GFP_KERNEL);
4251 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4255 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
4257 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4263 dev = (struct net_device *)
4264 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4265 dev->padded = (char *)dev - (char *)p;
4266 dev_net_set(dev, &init_net);
4269 dev->num_tx_queues = queue_count;
4270 dev->real_num_tx_queues = queue_count;
4273 dev->priv = ((char *)dev +
4274 ((sizeof(struct net_device) + NETDEV_ALIGN_CONST)
4275 & ~NETDEV_ALIGN_CONST));
4278 dev->gso_max_size = GSO_MAX_SIZE;
4280 netdev_init_queues(dev);
4282 dev->get_stats = internal_stats;
4283 netpoll_netdev_init(dev);
4285 strcpy(dev->name, name);
4288 EXPORT_SYMBOL(alloc_netdev_mq);
4291 * free_netdev - free network device
4294 * This function does the last stage of destroying an allocated device
4295 * interface. The reference to the device object is released.
4296 * If this is the last reference then it will be freed.
4298 void free_netdev(struct net_device *dev)
4300 release_net(dev_net(dev));
4304 /* Compatibility with error handling in drivers */
4305 if (dev->reg_state == NETREG_UNINITIALIZED) {
4306 kfree((char *)dev - dev->padded);
4310 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4311 dev->reg_state = NETREG_RELEASED;
4313 /* will free via device release */
4314 put_device(&dev->dev);
4317 /* Synchronize with packet receive processing. */
4318 void synchronize_net(void)
4325 * unregister_netdevice - remove device from the kernel
4328 * This function shuts down a device interface and removes it
4329 * from the kernel tables.
4331 * Callers must hold the rtnl semaphore. You may want
4332 * unregister_netdev() instead of this.
4335 void unregister_netdevice(struct net_device *dev)
4339 rollback_registered(dev);
4340 /* Finish processing unregister after unlock */
4345 * unregister_netdev - remove device from the kernel
4348 * This function shuts down a device interface and removes it
4349 * from the kernel tables.
4351 * This is just a wrapper for unregister_netdevice that takes
4352 * the rtnl semaphore. In general you want to use this and not
4353 * unregister_netdevice.
4355 void unregister_netdev(struct net_device *dev)
4358 unregister_netdevice(dev);
4362 EXPORT_SYMBOL(unregister_netdev);
4365 * dev_change_net_namespace - move device to different nethost namespace
4367 * @net: network namespace
4368 * @pat: If not NULL name pattern to try if the current device name
4369 * is already taken in the destination network namespace.
4371 * This function shuts down a device interface and moves it
4372 * to a new network namespace. On success 0 is returned, on
4373 * a failure a netagive errno code is returned.
4375 * Callers must hold the rtnl semaphore.
4378 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4381 const char *destname;
4386 /* Don't allow namespace local devices to be moved. */
4388 if (dev->features & NETIF_F_NETNS_LOCAL)
4391 /* Ensure the device has been registrered */
4393 if (dev->reg_state != NETREG_REGISTERED)
4396 /* Get out if there is nothing todo */
4398 if (net_eq(dev_net(dev), net))
4401 /* Pick the destination device name, and ensure
4402 * we can use it in the destination network namespace.
4405 destname = dev->name;
4406 if (__dev_get_by_name(net, destname)) {
4407 /* We get here if we can't use the current device name */
4410 if (!dev_valid_name(pat))
4412 if (strchr(pat, '%')) {
4413 if (__dev_alloc_name(net, pat, buf) < 0)
4418 if (__dev_get_by_name(net, destname))
4423 * And now a mini version of register_netdevice unregister_netdevice.
4426 /* If device is running close it first. */
4429 /* And unlink it from device chain */
4431 unlist_netdevice(dev);
4435 /* Shutdown queueing discipline. */
4438 /* Notify protocols, that we are about to destroy
4439 this device. They should clean all the things.
4441 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4444 * Flush the unicast and multicast chains
4446 dev_addr_discard(dev);
4448 /* Actually switch the network namespace */
4449 dev_net_set(dev, net);
4451 /* Assign the new device name */
4452 if (destname != dev->name)
4453 strcpy(dev->name, destname);
4455 /* If there is an ifindex conflict assign a new one */
4456 if (__dev_get_by_index(net, dev->ifindex)) {
4457 int iflink = (dev->iflink == dev->ifindex);
4458 dev->ifindex = dev_new_index(net);
4460 dev->iflink = dev->ifindex;
4463 /* Fixup kobjects */
4464 netdev_unregister_kobject(dev);
4465 err = netdev_register_kobject(dev);
4468 /* Add the device back in the hashes */
4469 list_netdevice(dev);
4471 /* Notify protocols, that a new device appeared. */
4472 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4480 static int dev_cpu_callback(struct notifier_block *nfb,
4481 unsigned long action,
4484 struct sk_buff **list_skb;
4485 struct Qdisc **list_net;
4486 struct sk_buff *skb;
4487 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4488 struct softnet_data *sd, *oldsd;
4490 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4493 local_irq_disable();
4494 cpu = smp_processor_id();
4495 sd = &per_cpu(softnet_data, cpu);
4496 oldsd = &per_cpu(softnet_data, oldcpu);
4498 /* Find end of our completion_queue. */
4499 list_skb = &sd->completion_queue;
4501 list_skb = &(*list_skb)->next;
4502 /* Append completion queue from offline CPU. */
4503 *list_skb = oldsd->completion_queue;
4504 oldsd->completion_queue = NULL;
4506 /* Find end of our output_queue. */
4507 list_net = &sd->output_queue;
4509 list_net = &(*list_net)->next_sched;
4510 /* Append output queue from offline CPU. */
4511 *list_net = oldsd->output_queue;
4512 oldsd->output_queue = NULL;
4514 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4517 /* Process offline CPU's input_pkt_queue */
4518 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4524 #ifdef CONFIG_NET_DMA
4526 * net_dma_rebalance - try to maintain one DMA channel per CPU
4527 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4529 * This is called when the number of channels allocated to the net_dma client
4530 * changes. The net_dma client tries to have one DMA channel per CPU.
4533 static void net_dma_rebalance(struct net_dma *net_dma)
4535 unsigned int cpu, i, n, chan_idx;
4536 struct dma_chan *chan;
4538 if (cpus_empty(net_dma->channel_mask)) {
4539 for_each_online_cpu(cpu)
4540 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4545 cpu = first_cpu(cpu_online_map);
4547 for_each_cpu_mask_nr(chan_idx, net_dma->channel_mask) {
4548 chan = net_dma->channels[chan_idx];
4550 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4551 + (i < (num_online_cpus() %
4552 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4555 per_cpu(softnet_data, cpu).net_dma = chan;
4556 cpu = next_cpu(cpu, cpu_online_map);
4564 * netdev_dma_event - event callback for the net_dma_client
4565 * @client: should always be net_dma_client
4566 * @chan: DMA channel for the event
4567 * @state: DMA state to be handled
4569 static enum dma_state_client
4570 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4571 enum dma_state state)
4573 int i, found = 0, pos = -1;
4574 struct net_dma *net_dma =
4575 container_of(client, struct net_dma, client);
4576 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4578 spin_lock(&net_dma->lock);
4580 case DMA_RESOURCE_AVAILABLE:
4581 for (i = 0; i < nr_cpu_ids; i++)
4582 if (net_dma->channels[i] == chan) {
4585 } else if (net_dma->channels[i] == NULL && pos < 0)
4588 if (!found && pos >= 0) {
4590 net_dma->channels[pos] = chan;
4591 cpu_set(pos, net_dma->channel_mask);
4592 net_dma_rebalance(net_dma);
4595 case DMA_RESOURCE_REMOVED:
4596 for (i = 0; i < nr_cpu_ids; i++)
4597 if (net_dma->channels[i] == chan) {
4605 cpu_clear(pos, net_dma->channel_mask);
4606 net_dma->channels[i] = NULL;
4607 net_dma_rebalance(net_dma);
4613 spin_unlock(&net_dma->lock);
4619 * netdev_dma_regiser - register the networking subsystem as a DMA client
4621 static int __init netdev_dma_register(void)
4623 net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma),
4625 if (unlikely(!net_dma.channels)) {
4627 "netdev_dma: no memory for net_dma.channels\n");
4630 spin_lock_init(&net_dma.lock);
4631 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4632 dma_async_client_register(&net_dma.client);
4633 dma_async_client_chan_request(&net_dma.client);
4638 static int __init netdev_dma_register(void) { return -ENODEV; }
4639 #endif /* CONFIG_NET_DMA */
4642 * netdev_compute_feature - compute conjunction of two feature sets
4643 * @all: first feature set
4644 * @one: second feature set
4646 * Computes a new feature set after adding a device with feature set
4647 * @one to the master device with current feature set @all. Returns
4648 * the new feature set.
4650 int netdev_compute_features(unsigned long all, unsigned long one)
4652 /* if device needs checksumming, downgrade to hw checksumming */
4653 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4654 all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
4656 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4657 if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
4658 all ^= NETIF_F_HW_CSUM
4659 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4661 if (one & NETIF_F_GSO)
4662 one |= NETIF_F_GSO_SOFTWARE;
4665 /* If even one device supports robust GSO, enable it for all. */
4666 if (one & NETIF_F_GSO_ROBUST)
4667 all |= NETIF_F_GSO_ROBUST;
4669 all &= one | NETIF_F_LLTX;
4671 if (!(all & NETIF_F_ALL_CSUM))
4673 if (!(all & NETIF_F_SG))
4674 all &= ~NETIF_F_GSO_MASK;
4678 EXPORT_SYMBOL(netdev_compute_features);
4680 static struct hlist_head *netdev_create_hash(void)
4683 struct hlist_head *hash;
4685 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4687 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4688 INIT_HLIST_HEAD(&hash[i]);
4693 /* Initialize per network namespace state */
4694 static int __net_init netdev_init(struct net *net)
4696 INIT_LIST_HEAD(&net->dev_base_head);
4698 net->dev_name_head = netdev_create_hash();
4699 if (net->dev_name_head == NULL)
4702 net->dev_index_head = netdev_create_hash();
4703 if (net->dev_index_head == NULL)
4709 kfree(net->dev_name_head);
4714 char *netdev_drivername(struct net_device *dev, char *buffer, int len)
4716 struct device_driver *driver;
4717 struct device *parent;
4719 if (len <= 0 || !buffer)
4723 parent = dev->dev.parent;
4728 driver = parent->driver;
4729 if (driver && driver->name)
4730 strlcpy(buffer, driver->name, len);
4734 static void __net_exit netdev_exit(struct net *net)
4736 kfree(net->dev_name_head);
4737 kfree(net->dev_index_head);
4740 static struct pernet_operations __net_initdata netdev_net_ops = {
4741 .init = netdev_init,
4742 .exit = netdev_exit,
4745 static void __net_exit default_device_exit(struct net *net)
4747 struct net_device *dev, *next;
4749 * Push all migratable of the network devices back to the
4750 * initial network namespace
4753 for_each_netdev_safe(net, dev, next) {
4755 char fb_name[IFNAMSIZ];
4757 /* Ignore unmoveable devices (i.e. loopback) */
4758 if (dev->features & NETIF_F_NETNS_LOCAL)
4761 /* Push remaing network devices to init_net */
4762 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
4763 err = dev_change_net_namespace(dev, &init_net, fb_name);
4765 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
4766 __func__, dev->name, err);
4773 static struct pernet_operations __net_initdata default_device_ops = {
4774 .exit = default_device_exit,
4778 * Initialize the DEV module. At boot time this walks the device list and
4779 * unhooks any devices that fail to initialise (normally hardware not
4780 * present) and leaves us with a valid list of present and active devices.
4785 * This is called single threaded during boot, so no need
4786 * to take the rtnl semaphore.
4788 static int __init net_dev_init(void)
4790 int i, rc = -ENOMEM;
4792 BUG_ON(!dev_boot_phase);
4794 if (dev_proc_init())
4797 if (netdev_kobject_init())
4800 INIT_LIST_HEAD(&ptype_all);
4801 for (i = 0; i < PTYPE_HASH_SIZE; i++)
4802 INIT_LIST_HEAD(&ptype_base[i]);
4804 if (register_pernet_subsys(&netdev_net_ops))
4807 if (register_pernet_device(&default_device_ops))
4811 * Initialise the packet receive queues.
4814 for_each_possible_cpu(i) {
4815 struct softnet_data *queue;
4817 queue = &per_cpu(softnet_data, i);
4818 skb_queue_head_init(&queue->input_pkt_queue);
4819 queue->completion_queue = NULL;
4820 INIT_LIST_HEAD(&queue->poll_list);
4822 queue->backlog.poll = process_backlog;
4823 queue->backlog.weight = weight_p;
4826 netdev_dma_register();
4830 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
4831 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
4833 hotcpu_notifier(dev_cpu_callback, 0);
4841 subsys_initcall(net_dev_init);
4843 EXPORT_SYMBOL(__dev_get_by_index);
4844 EXPORT_SYMBOL(__dev_get_by_name);
4845 EXPORT_SYMBOL(__dev_remove_pack);
4846 EXPORT_SYMBOL(dev_valid_name);
4847 EXPORT_SYMBOL(dev_add_pack);
4848 EXPORT_SYMBOL(dev_alloc_name);
4849 EXPORT_SYMBOL(dev_close);
4850 EXPORT_SYMBOL(dev_get_by_flags);
4851 EXPORT_SYMBOL(dev_get_by_index);
4852 EXPORT_SYMBOL(dev_get_by_name);
4853 EXPORT_SYMBOL(dev_open);
4854 EXPORT_SYMBOL(dev_queue_xmit);
4855 EXPORT_SYMBOL(dev_remove_pack);
4856 EXPORT_SYMBOL(dev_set_allmulti);
4857 EXPORT_SYMBOL(dev_set_promiscuity);
4858 EXPORT_SYMBOL(dev_change_flags);
4859 EXPORT_SYMBOL(dev_set_mtu);
4860 EXPORT_SYMBOL(dev_set_mac_address);
4861 EXPORT_SYMBOL(free_netdev);
4862 EXPORT_SYMBOL(netdev_boot_setup_check);
4863 EXPORT_SYMBOL(netdev_set_master);
4864 EXPORT_SYMBOL(netdev_state_change);
4865 EXPORT_SYMBOL(netif_receive_skb);
4866 EXPORT_SYMBOL(netif_rx);
4867 EXPORT_SYMBOL(register_gifconf);
4868 EXPORT_SYMBOL(register_netdevice);
4869 EXPORT_SYMBOL(register_netdevice_notifier);
4870 EXPORT_SYMBOL(skb_checksum_help);
4871 EXPORT_SYMBOL(synchronize_net);
4872 EXPORT_SYMBOL(unregister_netdevice);
4873 EXPORT_SYMBOL(unregister_netdevice_notifier);
4874 EXPORT_SYMBOL(net_enable_timestamp);
4875 EXPORT_SYMBOL(net_disable_timestamp);
4876 EXPORT_SYMBOL(dev_get_flags);
4878 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4879 EXPORT_SYMBOL(br_handle_frame_hook);
4880 EXPORT_SYMBOL(br_fdb_get_hook);
4881 EXPORT_SYMBOL(br_fdb_put_hook);
4885 EXPORT_SYMBOL(dev_load);
4888 EXPORT_PER_CPU_SYMBOL(softnet_data);