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 static inline void __netif_reschedule(struct Qdisc *q)
1344 struct softnet_data *sd;
1345 unsigned long flags;
1347 local_irq_save(flags);
1348 sd = &__get_cpu_var(softnet_data);
1349 q->next_sched = sd->output_queue;
1350 sd->output_queue = q;
1351 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1352 local_irq_restore(flags);
1355 void __netif_schedule(struct Qdisc *q)
1357 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1358 __netif_reschedule(q);
1360 EXPORT_SYMBOL(__netif_schedule);
1362 void dev_kfree_skb_irq(struct sk_buff *skb)
1364 if (atomic_dec_and_test(&skb->users)) {
1365 struct softnet_data *sd;
1366 unsigned long flags;
1368 local_irq_save(flags);
1369 sd = &__get_cpu_var(softnet_data);
1370 skb->next = sd->completion_queue;
1371 sd->completion_queue = skb;
1372 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1373 local_irq_restore(flags);
1376 EXPORT_SYMBOL(dev_kfree_skb_irq);
1378 void dev_kfree_skb_any(struct sk_buff *skb)
1380 if (in_irq() || irqs_disabled())
1381 dev_kfree_skb_irq(skb);
1385 EXPORT_SYMBOL(dev_kfree_skb_any);
1389 * netif_device_detach - mark device as removed
1390 * @dev: network device
1392 * Mark device as removed from system and therefore no longer available.
1394 void netif_device_detach(struct net_device *dev)
1396 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1397 netif_running(dev)) {
1398 netif_stop_queue(dev);
1401 EXPORT_SYMBOL(netif_device_detach);
1404 * netif_device_attach - mark device as attached
1405 * @dev: network device
1407 * Mark device as attached from system and restart if needed.
1409 void netif_device_attach(struct net_device *dev)
1411 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1412 netif_running(dev)) {
1413 netif_wake_queue(dev);
1414 __netdev_watchdog_up(dev);
1417 EXPORT_SYMBOL(netif_device_attach);
1419 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1421 return ((features & NETIF_F_GEN_CSUM) ||
1422 ((features & NETIF_F_IP_CSUM) &&
1423 protocol == htons(ETH_P_IP)) ||
1424 ((features & NETIF_F_IPV6_CSUM) &&
1425 protocol == htons(ETH_P_IPV6)));
1428 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1430 if (can_checksum_protocol(dev->features, skb->protocol))
1433 if (skb->protocol == htons(ETH_P_8021Q)) {
1434 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1435 if (can_checksum_protocol(dev->features & dev->vlan_features,
1436 veh->h_vlan_encapsulated_proto))
1444 * Invalidate hardware checksum when packet is to be mangled, and
1445 * complete checksum manually on outgoing path.
1447 int skb_checksum_help(struct sk_buff *skb)
1450 int ret = 0, offset;
1452 if (skb->ip_summed == CHECKSUM_COMPLETE)
1453 goto out_set_summed;
1455 if (unlikely(skb_shinfo(skb)->gso_size)) {
1456 /* Let GSO fix up the checksum. */
1457 goto out_set_summed;
1460 offset = skb->csum_start - skb_headroom(skb);
1461 BUG_ON(offset >= skb_headlen(skb));
1462 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1464 offset += skb->csum_offset;
1465 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1467 if (skb_cloned(skb) &&
1468 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1469 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1474 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1476 skb->ip_summed = CHECKSUM_NONE;
1482 * skb_gso_segment - Perform segmentation on skb.
1483 * @skb: buffer to segment
1484 * @features: features for the output path (see dev->features)
1486 * This function segments the given skb and returns a list of segments.
1488 * It may return NULL if the skb requires no segmentation. This is
1489 * only possible when GSO is used for verifying header integrity.
1491 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1493 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1494 struct packet_type *ptype;
1495 __be16 type = skb->protocol;
1498 BUG_ON(skb_shinfo(skb)->frag_list);
1500 skb_reset_mac_header(skb);
1501 skb->mac_len = skb->network_header - skb->mac_header;
1502 __skb_pull(skb, skb->mac_len);
1504 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1505 if (skb_header_cloned(skb) &&
1506 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1507 return ERR_PTR(err);
1511 list_for_each_entry_rcu(ptype,
1512 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1513 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1514 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1515 err = ptype->gso_send_check(skb);
1516 segs = ERR_PTR(err);
1517 if (err || skb_gso_ok(skb, features))
1519 __skb_push(skb, (skb->data -
1520 skb_network_header(skb)));
1522 segs = ptype->gso_segment(skb, features);
1528 __skb_push(skb, skb->data - skb_mac_header(skb));
1533 EXPORT_SYMBOL(skb_gso_segment);
1535 /* Take action when hardware reception checksum errors are detected. */
1537 void netdev_rx_csum_fault(struct net_device *dev)
1539 if (net_ratelimit()) {
1540 printk(KERN_ERR "%s: hw csum failure.\n",
1541 dev ? dev->name : "<unknown>");
1545 EXPORT_SYMBOL(netdev_rx_csum_fault);
1548 /* Actually, we should eliminate this check as soon as we know, that:
1549 * 1. IOMMU is present and allows to map all the memory.
1550 * 2. No high memory really exists on this machine.
1553 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1555 #ifdef CONFIG_HIGHMEM
1558 if (dev->features & NETIF_F_HIGHDMA)
1561 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1562 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1570 void (*destructor)(struct sk_buff *skb);
1573 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1575 static void dev_gso_skb_destructor(struct sk_buff *skb)
1577 struct dev_gso_cb *cb;
1580 struct sk_buff *nskb = skb->next;
1582 skb->next = nskb->next;
1585 } while (skb->next);
1587 cb = DEV_GSO_CB(skb);
1589 cb->destructor(skb);
1593 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1594 * @skb: buffer to segment
1596 * This function segments the given skb and stores the list of segments
1599 static int dev_gso_segment(struct sk_buff *skb)
1601 struct net_device *dev = skb->dev;
1602 struct sk_buff *segs;
1603 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1606 segs = skb_gso_segment(skb, features);
1608 /* Verifying header integrity only. */
1613 return PTR_ERR(segs);
1616 DEV_GSO_CB(skb)->destructor = skb->destructor;
1617 skb->destructor = dev_gso_skb_destructor;
1622 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1623 struct netdev_queue *txq)
1625 if (likely(!skb->next)) {
1626 if (!list_empty(&ptype_all))
1627 dev_queue_xmit_nit(skb, dev);
1629 if (netif_needs_gso(dev, skb)) {
1630 if (unlikely(dev_gso_segment(skb)))
1636 return dev->hard_start_xmit(skb, dev);
1641 struct sk_buff *nskb = skb->next;
1644 skb->next = nskb->next;
1646 rc = dev->hard_start_xmit(nskb, dev);
1648 nskb->next = skb->next;
1652 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1653 return NETDEV_TX_BUSY;
1654 } while (skb->next);
1656 skb->destructor = DEV_GSO_CB(skb)->destructor;
1663 static u32 simple_tx_hashrnd;
1664 static int simple_tx_hashrnd_initialized = 0;
1666 static u16 simple_tx_hash(struct net_device *dev, struct sk_buff *skb)
1668 u32 addr1, addr2, ports;
1672 if (unlikely(!simple_tx_hashrnd_initialized)) {
1673 get_random_bytes(&simple_tx_hashrnd, 4);
1674 simple_tx_hashrnd_initialized = 1;
1677 switch (skb->protocol) {
1678 case __constant_htons(ETH_P_IP):
1679 ip_proto = ip_hdr(skb)->protocol;
1680 addr1 = ip_hdr(skb)->saddr;
1681 addr2 = ip_hdr(skb)->daddr;
1682 ihl = ip_hdr(skb)->ihl;
1684 case __constant_htons(ETH_P_IPV6):
1685 ip_proto = ipv6_hdr(skb)->nexthdr;
1686 addr1 = ipv6_hdr(skb)->saddr.s6_addr32[3];
1687 addr2 = ipv6_hdr(skb)->daddr.s6_addr32[3];
1702 case IPPROTO_UDPLITE:
1703 ports = *((u32 *) (skb_network_header(skb) + (ihl * 4)));
1711 hash = jhash_3words(addr1, addr2, ports, simple_tx_hashrnd);
1713 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1716 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1717 struct sk_buff *skb)
1719 u16 queue_index = 0;
1721 if (dev->select_queue)
1722 queue_index = dev->select_queue(dev, skb);
1723 else if (dev->real_num_tx_queues > 1)
1724 queue_index = simple_tx_hash(dev, skb);
1726 skb_set_queue_mapping(skb, queue_index);
1727 return netdev_get_tx_queue(dev, queue_index);
1731 * dev_queue_xmit - transmit a buffer
1732 * @skb: buffer to transmit
1734 * Queue a buffer for transmission to a network device. The caller must
1735 * have set the device and priority and built the buffer before calling
1736 * this function. The function can be called from an interrupt.
1738 * A negative errno code is returned on a failure. A success does not
1739 * guarantee the frame will be transmitted as it may be dropped due
1740 * to congestion or traffic shaping.
1742 * -----------------------------------------------------------------------------------
1743 * I notice this method can also return errors from the queue disciplines,
1744 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1747 * Regardless of the return value, the skb is consumed, so it is currently
1748 * difficult to retry a send to this method. (You can bump the ref count
1749 * before sending to hold a reference for retry if you are careful.)
1751 * When calling this method, interrupts MUST be enabled. This is because
1752 * the BH enable code must have IRQs enabled so that it will not deadlock.
1755 int dev_queue_xmit(struct sk_buff *skb)
1757 struct net_device *dev = skb->dev;
1758 struct netdev_queue *txq;
1762 /* GSO will handle the following emulations directly. */
1763 if (netif_needs_gso(dev, skb))
1766 if (skb_shinfo(skb)->frag_list &&
1767 !(dev->features & NETIF_F_FRAGLIST) &&
1768 __skb_linearize(skb))
1771 /* Fragmented skb is linearized if device does not support SG,
1772 * or if at least one of fragments is in highmem and device
1773 * does not support DMA from it.
1775 if (skb_shinfo(skb)->nr_frags &&
1776 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1777 __skb_linearize(skb))
1780 /* If packet is not checksummed and device does not support
1781 * checksumming for this protocol, complete checksumming here.
1783 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1784 skb_set_transport_header(skb, skb->csum_start -
1786 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1791 /* Disable soft irqs for various locks below. Also
1792 * stops preemption for RCU.
1796 txq = dev_pick_tx(dev, skb);
1797 q = rcu_dereference(txq->qdisc);
1799 #ifdef CONFIG_NET_CLS_ACT
1800 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1803 spinlock_t *root_lock = qdisc_lock(q);
1805 spin_lock(root_lock);
1807 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1811 rc = qdisc_enqueue_root(skb, q);
1814 spin_unlock(root_lock);
1819 /* The device has no queue. Common case for software devices:
1820 loopback, all the sorts of tunnels...
1822 Really, it is unlikely that netif_tx_lock protection is necessary
1823 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1825 However, it is possible, that they rely on protection
1828 Check this and shot the lock. It is not prone from deadlocks.
1829 Either shot noqueue qdisc, it is even simpler 8)
1831 if (dev->flags & IFF_UP) {
1832 int cpu = smp_processor_id(); /* ok because BHs are off */
1834 if (txq->xmit_lock_owner != cpu) {
1836 HARD_TX_LOCK(dev, txq, cpu);
1838 if (!netif_tx_queue_stopped(txq)) {
1840 if (!dev_hard_start_xmit(skb, dev, txq)) {
1841 HARD_TX_UNLOCK(dev, txq);
1845 HARD_TX_UNLOCK(dev, txq);
1846 if (net_ratelimit())
1847 printk(KERN_CRIT "Virtual device %s asks to "
1848 "queue packet!\n", dev->name);
1850 /* Recursion is detected! It is possible,
1852 if (net_ratelimit())
1853 printk(KERN_CRIT "Dead loop on virtual device "
1854 "%s, fix it urgently!\n", dev->name);
1859 rcu_read_unlock_bh();
1865 rcu_read_unlock_bh();
1870 /*=======================================================================
1872 =======================================================================*/
1874 int netdev_max_backlog __read_mostly = 1000;
1875 int netdev_budget __read_mostly = 300;
1876 int weight_p __read_mostly = 64; /* old backlog weight */
1878 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1882 * netif_rx - post buffer to the network code
1883 * @skb: buffer to post
1885 * This function receives a packet from a device driver and queues it for
1886 * the upper (protocol) levels to process. It always succeeds. The buffer
1887 * may be dropped during processing for congestion control or by the
1891 * NET_RX_SUCCESS (no congestion)
1892 * NET_RX_DROP (packet was dropped)
1896 int netif_rx(struct sk_buff *skb)
1898 struct softnet_data *queue;
1899 unsigned long flags;
1901 /* if netpoll wants it, pretend we never saw it */
1902 if (netpoll_rx(skb))
1905 if (!skb->tstamp.tv64)
1909 * The code is rearranged so that the path is the most
1910 * short when CPU is congested, but is still operating.
1912 local_irq_save(flags);
1913 queue = &__get_cpu_var(softnet_data);
1915 __get_cpu_var(netdev_rx_stat).total++;
1916 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1917 if (queue->input_pkt_queue.qlen) {
1919 __skb_queue_tail(&queue->input_pkt_queue, skb);
1920 local_irq_restore(flags);
1921 return NET_RX_SUCCESS;
1924 napi_schedule(&queue->backlog);
1928 __get_cpu_var(netdev_rx_stat).dropped++;
1929 local_irq_restore(flags);
1935 int netif_rx_ni(struct sk_buff *skb)
1940 err = netif_rx(skb);
1941 if (local_softirq_pending())
1948 EXPORT_SYMBOL(netif_rx_ni);
1950 static void net_tx_action(struct softirq_action *h)
1952 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1954 if (sd->completion_queue) {
1955 struct sk_buff *clist;
1957 local_irq_disable();
1958 clist = sd->completion_queue;
1959 sd->completion_queue = NULL;
1963 struct sk_buff *skb = clist;
1964 clist = clist->next;
1966 WARN_ON(atomic_read(&skb->users));
1971 if (sd->output_queue) {
1974 local_irq_disable();
1975 head = sd->output_queue;
1976 sd->output_queue = NULL;
1980 struct Qdisc *q = head;
1981 spinlock_t *root_lock;
1983 head = head->next_sched;
1985 root_lock = qdisc_lock(q);
1986 if (spin_trylock(root_lock)) {
1987 smp_mb__before_clear_bit();
1988 clear_bit(__QDISC_STATE_SCHED,
1991 spin_unlock(root_lock);
1993 __netif_reschedule(q);
1999 static inline int deliver_skb(struct sk_buff *skb,
2000 struct packet_type *pt_prev,
2001 struct net_device *orig_dev)
2003 atomic_inc(&skb->users);
2004 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2007 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2008 /* These hooks defined here for ATM */
2010 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2011 unsigned char *addr);
2012 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
2015 * If bridge module is loaded call bridging hook.
2016 * returns NULL if packet was consumed.
2018 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2019 struct sk_buff *skb) __read_mostly;
2020 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2021 struct packet_type **pt_prev, int *ret,
2022 struct net_device *orig_dev)
2024 struct net_bridge_port *port;
2026 if (skb->pkt_type == PACKET_LOOPBACK ||
2027 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2031 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2035 return br_handle_frame_hook(port, skb);
2038 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2041 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2042 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2043 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2045 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2046 struct packet_type **pt_prev,
2048 struct net_device *orig_dev)
2050 if (skb->dev->macvlan_port == NULL)
2054 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2057 return macvlan_handle_frame_hook(skb);
2060 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2063 #ifdef CONFIG_NET_CLS_ACT
2064 /* TODO: Maybe we should just force sch_ingress to be compiled in
2065 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2066 * a compare and 2 stores extra right now if we dont have it on
2067 * but have CONFIG_NET_CLS_ACT
2068 * NOTE: This doesnt stop any functionality; if you dont have
2069 * the ingress scheduler, you just cant add policies on ingress.
2072 static int ing_filter(struct sk_buff *skb)
2074 struct net_device *dev = skb->dev;
2075 u32 ttl = G_TC_RTTL(skb->tc_verd);
2076 struct netdev_queue *rxq;
2077 int result = TC_ACT_OK;
2080 if (MAX_RED_LOOP < ttl++) {
2082 "Redir loop detected Dropping packet (%d->%d)\n",
2083 skb->iif, dev->ifindex);
2087 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2088 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2090 rxq = &dev->rx_queue;
2093 if (q != &noop_qdisc) {
2094 spin_lock(qdisc_lock(q));
2095 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2096 result = qdisc_enqueue_root(skb, q);
2097 spin_unlock(qdisc_lock(q));
2103 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2104 struct packet_type **pt_prev,
2105 int *ret, struct net_device *orig_dev)
2107 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2111 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2114 /* Huh? Why does turning on AF_PACKET affect this? */
2115 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2118 switch (ing_filter(skb)) {
2132 * netif_nit_deliver - deliver received packets to network taps
2135 * This function is used to deliver incoming packets to network
2136 * taps. It should be used when the normal netif_receive_skb path
2137 * is bypassed, for example because of VLAN acceleration.
2139 void netif_nit_deliver(struct sk_buff *skb)
2141 struct packet_type *ptype;
2143 if (list_empty(&ptype_all))
2146 skb_reset_network_header(skb);
2147 skb_reset_transport_header(skb);
2148 skb->mac_len = skb->network_header - skb->mac_header;
2151 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2152 if (!ptype->dev || ptype->dev == skb->dev)
2153 deliver_skb(skb, ptype, skb->dev);
2159 * netif_receive_skb - process receive buffer from network
2160 * @skb: buffer to process
2162 * netif_receive_skb() is the main receive data processing function.
2163 * It always succeeds. The buffer may be dropped during processing
2164 * for congestion control or by the protocol layers.
2166 * This function may only be called from softirq context and interrupts
2167 * should be enabled.
2169 * Return values (usually ignored):
2170 * NET_RX_SUCCESS: no congestion
2171 * NET_RX_DROP: packet was dropped
2173 int netif_receive_skb(struct sk_buff *skb)
2175 struct packet_type *ptype, *pt_prev;
2176 struct net_device *orig_dev;
2177 struct net_device *null_or_orig;
2178 int ret = NET_RX_DROP;
2181 /* if we've gotten here through NAPI, check netpoll */
2182 if (netpoll_receive_skb(skb))
2185 if (!skb->tstamp.tv64)
2189 skb->iif = skb->dev->ifindex;
2191 null_or_orig = NULL;
2192 orig_dev = skb->dev;
2193 if (orig_dev->master) {
2194 if (skb_bond_should_drop(skb))
2195 null_or_orig = orig_dev; /* deliver only exact match */
2197 skb->dev = orig_dev->master;
2200 __get_cpu_var(netdev_rx_stat).total++;
2202 skb_reset_network_header(skb);
2203 skb_reset_transport_header(skb);
2204 skb->mac_len = skb->network_header - skb->mac_header;
2210 /* Don't receive packets in an exiting network namespace */
2211 if (!net_alive(dev_net(skb->dev)))
2214 #ifdef CONFIG_NET_CLS_ACT
2215 if (skb->tc_verd & TC_NCLS) {
2216 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2221 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2222 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2223 ptype->dev == orig_dev) {
2225 ret = deliver_skb(skb, pt_prev, orig_dev);
2230 #ifdef CONFIG_NET_CLS_ACT
2231 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2237 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2240 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2244 type = skb->protocol;
2245 list_for_each_entry_rcu(ptype,
2246 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2247 if (ptype->type == type &&
2248 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2249 ptype->dev == orig_dev)) {
2251 ret = deliver_skb(skb, pt_prev, orig_dev);
2257 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2260 /* Jamal, now you will not able to escape explaining
2261 * me how you were going to use this. :-)
2271 /* Network device is going away, flush any packets still pending */
2272 static void flush_backlog(void *arg)
2274 struct net_device *dev = arg;
2275 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2276 struct sk_buff *skb, *tmp;
2278 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2279 if (skb->dev == dev) {
2280 __skb_unlink(skb, &queue->input_pkt_queue);
2285 static int process_backlog(struct napi_struct *napi, int quota)
2288 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2289 unsigned long start_time = jiffies;
2291 napi->weight = weight_p;
2293 struct sk_buff *skb;
2295 local_irq_disable();
2296 skb = __skb_dequeue(&queue->input_pkt_queue);
2298 __napi_complete(napi);
2304 netif_receive_skb(skb);
2305 } while (++work < quota && jiffies == start_time);
2311 * __napi_schedule - schedule for receive
2312 * @n: entry to schedule
2314 * The entry's receive function will be scheduled to run
2316 void __napi_schedule(struct napi_struct *n)
2318 unsigned long flags;
2320 local_irq_save(flags);
2321 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2322 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2323 local_irq_restore(flags);
2325 EXPORT_SYMBOL(__napi_schedule);
2328 static void net_rx_action(struct softirq_action *h)
2330 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2331 unsigned long start_time = jiffies;
2332 int budget = netdev_budget;
2335 local_irq_disable();
2337 while (!list_empty(list)) {
2338 struct napi_struct *n;
2341 /* If softirq window is exhuasted then punt.
2343 * Note that this is a slight policy change from the
2344 * previous NAPI code, which would allow up to 2
2345 * jiffies to pass before breaking out. The test
2346 * used to be "jiffies - start_time > 1".
2348 if (unlikely(budget <= 0 || jiffies != start_time))
2353 /* Even though interrupts have been re-enabled, this
2354 * access is safe because interrupts can only add new
2355 * entries to the tail of this list, and only ->poll()
2356 * calls can remove this head entry from the list.
2358 n = list_entry(list->next, struct napi_struct, poll_list);
2360 have = netpoll_poll_lock(n);
2364 /* This NAPI_STATE_SCHED test is for avoiding a race
2365 * with netpoll's poll_napi(). Only the entity which
2366 * obtains the lock and sees NAPI_STATE_SCHED set will
2367 * actually make the ->poll() call. Therefore we avoid
2368 * accidently calling ->poll() when NAPI is not scheduled.
2371 if (test_bit(NAPI_STATE_SCHED, &n->state))
2372 work = n->poll(n, weight);
2374 WARN_ON_ONCE(work > weight);
2378 local_irq_disable();
2380 /* Drivers must not modify the NAPI state if they
2381 * consume the entire weight. In such cases this code
2382 * still "owns" the NAPI instance and therefore can
2383 * move the instance around on the list at-will.
2385 if (unlikely(work == weight)) {
2386 if (unlikely(napi_disable_pending(n)))
2389 list_move_tail(&n->poll_list, list);
2392 netpoll_poll_unlock(have);
2397 #ifdef CONFIG_NET_DMA
2399 * There may not be any more sk_buffs coming right now, so push
2400 * any pending DMA copies to hardware
2402 if (!cpus_empty(net_dma.channel_mask)) {
2404 for_each_cpu_mask_nr(chan_idx, net_dma.channel_mask) {
2405 struct dma_chan *chan = net_dma.channels[chan_idx];
2407 dma_async_memcpy_issue_pending(chan);
2415 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2416 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2420 static gifconf_func_t * gifconf_list [NPROTO];
2423 * register_gifconf - register a SIOCGIF handler
2424 * @family: Address family
2425 * @gifconf: Function handler
2427 * Register protocol dependent address dumping routines. The handler
2428 * that is passed must not be freed or reused until it has been replaced
2429 * by another handler.
2431 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2433 if (family >= NPROTO)
2435 gifconf_list[family] = gifconf;
2441 * Map an interface index to its name (SIOCGIFNAME)
2445 * We need this ioctl for efficient implementation of the
2446 * if_indextoname() function required by the IPv6 API. Without
2447 * it, we would have to search all the interfaces to find a
2451 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2453 struct net_device *dev;
2457 * Fetch the caller's info block.
2460 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2463 read_lock(&dev_base_lock);
2464 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2466 read_unlock(&dev_base_lock);
2470 strcpy(ifr.ifr_name, dev->name);
2471 read_unlock(&dev_base_lock);
2473 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2479 * Perform a SIOCGIFCONF call. This structure will change
2480 * size eventually, and there is nothing I can do about it.
2481 * Thus we will need a 'compatibility mode'.
2484 static int dev_ifconf(struct net *net, char __user *arg)
2487 struct net_device *dev;
2494 * Fetch the caller's info block.
2497 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2504 * Loop over the interfaces, and write an info block for each.
2508 for_each_netdev(net, dev) {
2509 for (i = 0; i < NPROTO; i++) {
2510 if (gifconf_list[i]) {
2513 done = gifconf_list[i](dev, NULL, 0);
2515 done = gifconf_list[i](dev, pos + total,
2525 * All done. Write the updated control block back to the caller.
2527 ifc.ifc_len = total;
2530 * Both BSD and Solaris return 0 here, so we do too.
2532 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2535 #ifdef CONFIG_PROC_FS
2537 * This is invoked by the /proc filesystem handler to display a device
2540 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2541 __acquires(dev_base_lock)
2543 struct net *net = seq_file_net(seq);
2545 struct net_device *dev;
2547 read_lock(&dev_base_lock);
2549 return SEQ_START_TOKEN;
2552 for_each_netdev(net, dev)
2559 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2561 struct net *net = seq_file_net(seq);
2563 return v == SEQ_START_TOKEN ?
2564 first_net_device(net) : next_net_device((struct net_device *)v);
2567 void dev_seq_stop(struct seq_file *seq, void *v)
2568 __releases(dev_base_lock)
2570 read_unlock(&dev_base_lock);
2573 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2575 struct net_device_stats *stats = dev->get_stats(dev);
2577 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2578 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2579 dev->name, stats->rx_bytes, stats->rx_packets,
2581 stats->rx_dropped + stats->rx_missed_errors,
2582 stats->rx_fifo_errors,
2583 stats->rx_length_errors + stats->rx_over_errors +
2584 stats->rx_crc_errors + stats->rx_frame_errors,
2585 stats->rx_compressed, stats->multicast,
2586 stats->tx_bytes, stats->tx_packets,
2587 stats->tx_errors, stats->tx_dropped,
2588 stats->tx_fifo_errors, stats->collisions,
2589 stats->tx_carrier_errors +
2590 stats->tx_aborted_errors +
2591 stats->tx_window_errors +
2592 stats->tx_heartbeat_errors,
2593 stats->tx_compressed);
2597 * Called from the PROCfs module. This now uses the new arbitrary sized
2598 * /proc/net interface to create /proc/net/dev
2600 static int dev_seq_show(struct seq_file *seq, void *v)
2602 if (v == SEQ_START_TOKEN)
2603 seq_puts(seq, "Inter-| Receive "
2605 " face |bytes packets errs drop fifo frame "
2606 "compressed multicast|bytes packets errs "
2607 "drop fifo colls carrier compressed\n");
2609 dev_seq_printf_stats(seq, v);
2613 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2615 struct netif_rx_stats *rc = NULL;
2617 while (*pos < nr_cpu_ids)
2618 if (cpu_online(*pos)) {
2619 rc = &per_cpu(netdev_rx_stat, *pos);
2626 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2628 return softnet_get_online(pos);
2631 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2634 return softnet_get_online(pos);
2637 static void softnet_seq_stop(struct seq_file *seq, void *v)
2641 static int softnet_seq_show(struct seq_file *seq, void *v)
2643 struct netif_rx_stats *s = v;
2645 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2646 s->total, s->dropped, s->time_squeeze, 0,
2647 0, 0, 0, 0, /* was fastroute */
2652 static const struct seq_operations dev_seq_ops = {
2653 .start = dev_seq_start,
2654 .next = dev_seq_next,
2655 .stop = dev_seq_stop,
2656 .show = dev_seq_show,
2659 static int dev_seq_open(struct inode *inode, struct file *file)
2661 return seq_open_net(inode, file, &dev_seq_ops,
2662 sizeof(struct seq_net_private));
2665 static const struct file_operations dev_seq_fops = {
2666 .owner = THIS_MODULE,
2667 .open = dev_seq_open,
2669 .llseek = seq_lseek,
2670 .release = seq_release_net,
2673 static const struct seq_operations softnet_seq_ops = {
2674 .start = softnet_seq_start,
2675 .next = softnet_seq_next,
2676 .stop = softnet_seq_stop,
2677 .show = softnet_seq_show,
2680 static int softnet_seq_open(struct inode *inode, struct file *file)
2682 return seq_open(file, &softnet_seq_ops);
2685 static const struct file_operations softnet_seq_fops = {
2686 .owner = THIS_MODULE,
2687 .open = softnet_seq_open,
2689 .llseek = seq_lseek,
2690 .release = seq_release,
2693 static void *ptype_get_idx(loff_t pos)
2695 struct packet_type *pt = NULL;
2699 list_for_each_entry_rcu(pt, &ptype_all, list) {
2705 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
2706 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2715 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2719 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2722 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2724 struct packet_type *pt;
2725 struct list_head *nxt;
2729 if (v == SEQ_START_TOKEN)
2730 return ptype_get_idx(0);
2733 nxt = pt->list.next;
2734 if (pt->type == htons(ETH_P_ALL)) {
2735 if (nxt != &ptype_all)
2738 nxt = ptype_base[0].next;
2740 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
2742 while (nxt == &ptype_base[hash]) {
2743 if (++hash >= PTYPE_HASH_SIZE)
2745 nxt = ptype_base[hash].next;
2748 return list_entry(nxt, struct packet_type, list);
2751 static void ptype_seq_stop(struct seq_file *seq, void *v)
2757 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2759 #ifdef CONFIG_KALLSYMS
2760 unsigned long offset = 0, symsize;
2761 const char *symname;
2765 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2772 modname = delim = "";
2773 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2779 seq_printf(seq, "[%p]", sym);
2782 static int ptype_seq_show(struct seq_file *seq, void *v)
2784 struct packet_type *pt = v;
2786 if (v == SEQ_START_TOKEN)
2787 seq_puts(seq, "Type Device Function\n");
2788 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
2789 if (pt->type == htons(ETH_P_ALL))
2790 seq_puts(seq, "ALL ");
2792 seq_printf(seq, "%04x", ntohs(pt->type));
2794 seq_printf(seq, " %-8s ",
2795 pt->dev ? pt->dev->name : "");
2796 ptype_seq_decode(seq, pt->func);
2797 seq_putc(seq, '\n');
2803 static const struct seq_operations ptype_seq_ops = {
2804 .start = ptype_seq_start,
2805 .next = ptype_seq_next,
2806 .stop = ptype_seq_stop,
2807 .show = ptype_seq_show,
2810 static int ptype_seq_open(struct inode *inode, struct file *file)
2812 return seq_open_net(inode, file, &ptype_seq_ops,
2813 sizeof(struct seq_net_private));
2816 static const struct file_operations ptype_seq_fops = {
2817 .owner = THIS_MODULE,
2818 .open = ptype_seq_open,
2820 .llseek = seq_lseek,
2821 .release = seq_release_net,
2825 static int __net_init dev_proc_net_init(struct net *net)
2829 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2831 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2833 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2836 if (wext_proc_init(net))
2842 proc_net_remove(net, "ptype");
2844 proc_net_remove(net, "softnet_stat");
2846 proc_net_remove(net, "dev");
2850 static void __net_exit dev_proc_net_exit(struct net *net)
2852 wext_proc_exit(net);
2854 proc_net_remove(net, "ptype");
2855 proc_net_remove(net, "softnet_stat");
2856 proc_net_remove(net, "dev");
2859 static struct pernet_operations __net_initdata dev_proc_ops = {
2860 .init = dev_proc_net_init,
2861 .exit = dev_proc_net_exit,
2864 static int __init dev_proc_init(void)
2866 return register_pernet_subsys(&dev_proc_ops);
2869 #define dev_proc_init() 0
2870 #endif /* CONFIG_PROC_FS */
2874 * netdev_set_master - set up master/slave pair
2875 * @slave: slave device
2876 * @master: new master device
2878 * Changes the master device of the slave. Pass %NULL to break the
2879 * bonding. The caller must hold the RTNL semaphore. On a failure
2880 * a negative errno code is returned. On success the reference counts
2881 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2882 * function returns zero.
2884 int netdev_set_master(struct net_device *slave, struct net_device *master)
2886 struct net_device *old = slave->master;
2896 slave->master = master;
2904 slave->flags |= IFF_SLAVE;
2906 slave->flags &= ~IFF_SLAVE;
2908 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2912 static int __dev_set_promiscuity(struct net_device *dev, int inc)
2914 unsigned short old_flags = dev->flags;
2918 dev->flags |= IFF_PROMISC;
2919 dev->promiscuity += inc;
2920 if (dev->promiscuity == 0) {
2923 * If inc causes overflow, untouch promisc and return error.
2926 dev->flags &= ~IFF_PROMISC;
2928 dev->promiscuity -= inc;
2929 printk(KERN_WARNING "%s: promiscuity touches roof, "
2930 "set promiscuity failed, promiscuity feature "
2931 "of device might be broken.\n", dev->name);
2935 if (dev->flags != old_flags) {
2936 printk(KERN_INFO "device %s %s promiscuous mode\n",
2937 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2940 audit_log(current->audit_context, GFP_ATOMIC,
2941 AUDIT_ANOM_PROMISCUOUS,
2942 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2943 dev->name, (dev->flags & IFF_PROMISC),
2944 (old_flags & IFF_PROMISC),
2945 audit_get_loginuid(current),
2946 current->uid, current->gid,
2947 audit_get_sessionid(current));
2949 if (dev->change_rx_flags)
2950 dev->change_rx_flags(dev, IFF_PROMISC);
2956 * dev_set_promiscuity - update promiscuity count on a device
2960 * Add or remove promiscuity from a device. While the count in the device
2961 * remains above zero the interface remains promiscuous. Once it hits zero
2962 * the device reverts back to normal filtering operation. A negative inc
2963 * value is used to drop promiscuity on the device.
2964 * Return 0 if successful or a negative errno code on error.
2966 int dev_set_promiscuity(struct net_device *dev, int inc)
2968 unsigned short old_flags = dev->flags;
2971 err = __dev_set_promiscuity(dev, inc);
2974 if (dev->flags != old_flags)
2975 dev_set_rx_mode(dev);
2980 * dev_set_allmulti - update allmulti count on a device
2984 * Add or remove reception of all multicast frames to a device. While the
2985 * count in the device remains above zero the interface remains listening
2986 * to all interfaces. Once it hits zero the device reverts back to normal
2987 * filtering operation. A negative @inc value is used to drop the counter
2988 * when releasing a resource needing all multicasts.
2989 * Return 0 if successful or a negative errno code on error.
2992 int dev_set_allmulti(struct net_device *dev, int inc)
2994 unsigned short old_flags = dev->flags;
2998 dev->flags |= IFF_ALLMULTI;
2999 dev->allmulti += inc;
3000 if (dev->allmulti == 0) {
3003 * If inc causes overflow, untouch allmulti and return error.
3006 dev->flags &= ~IFF_ALLMULTI;
3008 dev->allmulti -= inc;
3009 printk(KERN_WARNING "%s: allmulti touches roof, "
3010 "set allmulti failed, allmulti feature of "
3011 "device might be broken.\n", dev->name);
3015 if (dev->flags ^ old_flags) {
3016 if (dev->change_rx_flags)
3017 dev->change_rx_flags(dev, IFF_ALLMULTI);
3018 dev_set_rx_mode(dev);
3024 * Upload unicast and multicast address lists to device and
3025 * configure RX filtering. When the device doesn't support unicast
3026 * filtering it is put in promiscuous mode while unicast addresses
3029 void __dev_set_rx_mode(struct net_device *dev)
3031 /* dev_open will call this function so the list will stay sane. */
3032 if (!(dev->flags&IFF_UP))
3035 if (!netif_device_present(dev))
3038 if (dev->set_rx_mode)
3039 dev->set_rx_mode(dev);
3041 /* Unicast addresses changes may only happen under the rtnl,
3042 * therefore calling __dev_set_promiscuity here is safe.
3044 if (dev->uc_count > 0 && !dev->uc_promisc) {
3045 __dev_set_promiscuity(dev, 1);
3046 dev->uc_promisc = 1;
3047 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3048 __dev_set_promiscuity(dev, -1);
3049 dev->uc_promisc = 0;
3052 if (dev->set_multicast_list)
3053 dev->set_multicast_list(dev);
3057 void dev_set_rx_mode(struct net_device *dev)
3059 netif_addr_lock_bh(dev);
3060 __dev_set_rx_mode(dev);
3061 netif_addr_unlock_bh(dev);
3064 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3065 void *addr, int alen, int glbl)
3067 struct dev_addr_list *da;
3069 for (; (da = *list) != NULL; list = &da->next) {
3070 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3071 alen == da->da_addrlen) {
3073 int old_glbl = da->da_gusers;
3090 int __dev_addr_add(struct dev_addr_list **list, int *count,
3091 void *addr, int alen, int glbl)
3093 struct dev_addr_list *da;
3095 for (da = *list; da != NULL; da = da->next) {
3096 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3097 da->da_addrlen == alen) {
3099 int old_glbl = da->da_gusers;
3109 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3112 memcpy(da->da_addr, addr, alen);
3113 da->da_addrlen = alen;
3115 da->da_gusers = glbl ? 1 : 0;
3123 * dev_unicast_delete - Release secondary unicast address.
3125 * @addr: address to delete
3126 * @alen: length of @addr
3128 * Release reference to a secondary unicast address and remove it
3129 * from the device if the reference count drops to zero.
3131 * The caller must hold the rtnl_mutex.
3133 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3139 netif_addr_lock_bh(dev);
3140 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3142 __dev_set_rx_mode(dev);
3143 netif_addr_unlock_bh(dev);
3146 EXPORT_SYMBOL(dev_unicast_delete);
3149 * dev_unicast_add - add a secondary unicast address
3151 * @addr: address to add
3152 * @alen: length of @addr
3154 * Add a secondary unicast address to the device or increase
3155 * the reference count if it already exists.
3157 * The caller must hold the rtnl_mutex.
3159 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3165 netif_addr_lock_bh(dev);
3166 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3168 __dev_set_rx_mode(dev);
3169 netif_addr_unlock_bh(dev);
3172 EXPORT_SYMBOL(dev_unicast_add);
3174 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3175 struct dev_addr_list **from, int *from_count)
3177 struct dev_addr_list *da, *next;
3181 while (da != NULL) {
3183 if (!da->da_synced) {
3184 err = __dev_addr_add(to, to_count,
3185 da->da_addr, da->da_addrlen, 0);
3190 } else if (da->da_users == 1) {
3191 __dev_addr_delete(to, to_count,
3192 da->da_addr, da->da_addrlen, 0);
3193 __dev_addr_delete(from, from_count,
3194 da->da_addr, da->da_addrlen, 0);
3201 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3202 struct dev_addr_list **from, int *from_count)
3204 struct dev_addr_list *da, *next;
3207 while (da != NULL) {
3209 if (da->da_synced) {
3210 __dev_addr_delete(to, to_count,
3211 da->da_addr, da->da_addrlen, 0);
3213 __dev_addr_delete(from, from_count,
3214 da->da_addr, da->da_addrlen, 0);
3221 * dev_unicast_sync - Synchronize device's unicast list to another device
3222 * @to: destination device
3223 * @from: source device
3225 * Add newly added addresses to the destination device and release
3226 * addresses that have no users left. The source device must be
3227 * locked by netif_tx_lock_bh.
3229 * This function is intended to be called from the dev->set_rx_mode
3230 * function of layered software devices.
3232 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3236 netif_addr_lock_bh(to);
3237 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3238 &from->uc_list, &from->uc_count);
3240 __dev_set_rx_mode(to);
3241 netif_addr_unlock_bh(to);
3244 EXPORT_SYMBOL(dev_unicast_sync);
3247 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3248 * @to: destination device
3249 * @from: source device
3251 * Remove all addresses that were added to the destination device by
3252 * dev_unicast_sync(). This function is intended to be called from the
3253 * dev->stop function of layered software devices.
3255 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3257 netif_addr_lock_bh(from);
3258 netif_addr_lock(to);
3260 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3261 &from->uc_list, &from->uc_count);
3262 __dev_set_rx_mode(to);
3264 netif_addr_unlock(to);
3265 netif_addr_unlock_bh(from);
3267 EXPORT_SYMBOL(dev_unicast_unsync);
3269 static void __dev_addr_discard(struct dev_addr_list **list)
3271 struct dev_addr_list *tmp;
3273 while (*list != NULL) {
3276 if (tmp->da_users > tmp->da_gusers)
3277 printk("__dev_addr_discard: address leakage! "
3278 "da_users=%d\n", tmp->da_users);
3283 static void dev_addr_discard(struct net_device *dev)
3285 netif_addr_lock_bh(dev);
3287 __dev_addr_discard(&dev->uc_list);
3290 __dev_addr_discard(&dev->mc_list);
3293 netif_addr_unlock_bh(dev);
3296 unsigned dev_get_flags(const struct net_device *dev)
3300 flags = (dev->flags & ~(IFF_PROMISC |
3305 (dev->gflags & (IFF_PROMISC |
3308 if (netif_running(dev)) {
3309 if (netif_oper_up(dev))
3310 flags |= IFF_RUNNING;
3311 if (netif_carrier_ok(dev))
3312 flags |= IFF_LOWER_UP;
3313 if (netif_dormant(dev))
3314 flags |= IFF_DORMANT;
3320 int dev_change_flags(struct net_device *dev, unsigned flags)
3323 int old_flags = dev->flags;
3328 * Set the flags on our device.
3331 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3332 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3334 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3338 * Load in the correct multicast list now the flags have changed.
3341 if (dev->change_rx_flags && (old_flags ^ flags) & IFF_MULTICAST)
3342 dev->change_rx_flags(dev, IFF_MULTICAST);
3344 dev_set_rx_mode(dev);
3347 * Have we downed the interface. We handle IFF_UP ourselves
3348 * according to user attempts to set it, rather than blindly
3353 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3354 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3357 dev_set_rx_mode(dev);
3360 if (dev->flags & IFF_UP &&
3361 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3363 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3365 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3366 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3367 dev->gflags ^= IFF_PROMISC;
3368 dev_set_promiscuity(dev, inc);
3371 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3372 is important. Some (broken) drivers set IFF_PROMISC, when
3373 IFF_ALLMULTI is requested not asking us and not reporting.
3375 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3376 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3377 dev->gflags ^= IFF_ALLMULTI;
3378 dev_set_allmulti(dev, inc);
3381 /* Exclude state transition flags, already notified */
3382 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3384 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3389 int dev_set_mtu(struct net_device *dev, int new_mtu)
3393 if (new_mtu == dev->mtu)
3396 /* MTU must be positive. */
3400 if (!netif_device_present(dev))
3404 if (dev->change_mtu)
3405 err = dev->change_mtu(dev, new_mtu);
3408 if (!err && dev->flags & IFF_UP)
3409 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3413 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3417 if (!dev->set_mac_address)
3419 if (sa->sa_family != dev->type)
3421 if (!netif_device_present(dev))
3423 err = dev->set_mac_address(dev, sa);
3425 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3430 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3432 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3435 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3441 case SIOCGIFFLAGS: /* Get interface flags */
3442 ifr->ifr_flags = dev_get_flags(dev);
3445 case SIOCGIFMETRIC: /* Get the metric on the interface
3446 (currently unused) */
3447 ifr->ifr_metric = 0;
3450 case SIOCGIFMTU: /* Get the MTU of a device */
3451 ifr->ifr_mtu = dev->mtu;
3456 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3458 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3459 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3460 ifr->ifr_hwaddr.sa_family = dev->type;
3468 ifr->ifr_map.mem_start = dev->mem_start;
3469 ifr->ifr_map.mem_end = dev->mem_end;
3470 ifr->ifr_map.base_addr = dev->base_addr;
3471 ifr->ifr_map.irq = dev->irq;
3472 ifr->ifr_map.dma = dev->dma;
3473 ifr->ifr_map.port = dev->if_port;
3477 ifr->ifr_ifindex = dev->ifindex;
3481 ifr->ifr_qlen = dev->tx_queue_len;
3485 /* dev_ioctl() should ensure this case
3497 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3499 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3502 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3508 case SIOCSIFFLAGS: /* Set interface flags */
3509 return dev_change_flags(dev, ifr->ifr_flags);
3511 case SIOCSIFMETRIC: /* Set the metric on the interface
3512 (currently unused) */
3515 case SIOCSIFMTU: /* Set the MTU of a device */
3516 return dev_set_mtu(dev, ifr->ifr_mtu);
3519 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3521 case SIOCSIFHWBROADCAST:
3522 if (ifr->ifr_hwaddr.sa_family != dev->type)
3524 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3525 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3526 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3530 if (dev->set_config) {
3531 if (!netif_device_present(dev))
3533 return dev->set_config(dev, &ifr->ifr_map);
3538 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3539 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3541 if (!netif_device_present(dev))
3543 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3547 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3548 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3550 if (!netif_device_present(dev))
3552 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3556 if (ifr->ifr_qlen < 0)
3558 dev->tx_queue_len = ifr->ifr_qlen;
3562 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3563 return dev_change_name(dev, ifr->ifr_newname);
3566 * Unknown or private ioctl
3570 if ((cmd >= SIOCDEVPRIVATE &&
3571 cmd <= SIOCDEVPRIVATE + 15) ||
3572 cmd == SIOCBONDENSLAVE ||
3573 cmd == SIOCBONDRELEASE ||
3574 cmd == SIOCBONDSETHWADDR ||
3575 cmd == SIOCBONDSLAVEINFOQUERY ||
3576 cmd == SIOCBONDINFOQUERY ||
3577 cmd == SIOCBONDCHANGEACTIVE ||
3578 cmd == SIOCGMIIPHY ||
3579 cmd == SIOCGMIIREG ||
3580 cmd == SIOCSMIIREG ||
3581 cmd == SIOCBRADDIF ||
3582 cmd == SIOCBRDELIF ||
3583 cmd == SIOCWANDEV) {
3585 if (dev->do_ioctl) {
3586 if (netif_device_present(dev))
3587 err = dev->do_ioctl(dev, ifr,
3600 * This function handles all "interface"-type I/O control requests. The actual
3601 * 'doing' part of this is dev_ifsioc above.
3605 * dev_ioctl - network device ioctl
3606 * @net: the applicable net namespace
3607 * @cmd: command to issue
3608 * @arg: pointer to a struct ifreq in user space
3610 * Issue ioctl functions to devices. This is normally called by the
3611 * user space syscall interfaces but can sometimes be useful for
3612 * other purposes. The return value is the return from the syscall if
3613 * positive or a negative errno code on error.
3616 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3622 /* One special case: SIOCGIFCONF takes ifconf argument
3623 and requires shared lock, because it sleeps writing
3627 if (cmd == SIOCGIFCONF) {
3629 ret = dev_ifconf(net, (char __user *) arg);
3633 if (cmd == SIOCGIFNAME)
3634 return dev_ifname(net, (struct ifreq __user *)arg);
3636 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3639 ifr.ifr_name[IFNAMSIZ-1] = 0;
3641 colon = strchr(ifr.ifr_name, ':');
3646 * See which interface the caller is talking about.
3651 * These ioctl calls:
3652 * - can be done by all.
3653 * - atomic and do not require locking.
3664 dev_load(net, ifr.ifr_name);
3665 read_lock(&dev_base_lock);
3666 ret = dev_ifsioc_locked(net, &ifr, cmd);
3667 read_unlock(&dev_base_lock);
3671 if (copy_to_user(arg, &ifr,
3672 sizeof(struct ifreq)))
3678 dev_load(net, ifr.ifr_name);
3680 ret = dev_ethtool(net, &ifr);
3685 if (copy_to_user(arg, &ifr,
3686 sizeof(struct ifreq)))
3692 * These ioctl calls:
3693 * - require superuser power.
3694 * - require strict serialization.
3700 if (!capable(CAP_NET_ADMIN))
3702 dev_load(net, ifr.ifr_name);
3704 ret = dev_ifsioc(net, &ifr, cmd);
3709 if (copy_to_user(arg, &ifr,
3710 sizeof(struct ifreq)))
3716 * These ioctl calls:
3717 * - require superuser power.
3718 * - require strict serialization.
3719 * - do not return a value
3729 case SIOCSIFHWBROADCAST:
3732 case SIOCBONDENSLAVE:
3733 case SIOCBONDRELEASE:
3734 case SIOCBONDSETHWADDR:
3735 case SIOCBONDCHANGEACTIVE:
3738 if (!capable(CAP_NET_ADMIN))
3741 case SIOCBONDSLAVEINFOQUERY:
3742 case SIOCBONDINFOQUERY:
3743 dev_load(net, ifr.ifr_name);
3745 ret = dev_ifsioc(net, &ifr, cmd);
3750 /* Get the per device memory space. We can add this but
3751 * currently do not support it */
3753 /* Set the per device memory buffer space.
3754 * Not applicable in our case */
3759 * Unknown or private ioctl.
3762 if (cmd == SIOCWANDEV ||
3763 (cmd >= SIOCDEVPRIVATE &&
3764 cmd <= SIOCDEVPRIVATE + 15)) {
3765 dev_load(net, ifr.ifr_name);
3767 ret = dev_ifsioc(net, &ifr, cmd);
3769 if (!ret && copy_to_user(arg, &ifr,
3770 sizeof(struct ifreq)))
3774 /* Take care of Wireless Extensions */
3775 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3776 return wext_handle_ioctl(net, &ifr, cmd, arg);
3783 * dev_new_index - allocate an ifindex
3784 * @net: the applicable net namespace
3786 * Returns a suitable unique value for a new device interface
3787 * number. The caller must hold the rtnl semaphore or the
3788 * dev_base_lock to be sure it remains unique.
3790 static int dev_new_index(struct net *net)
3796 if (!__dev_get_by_index(net, ifindex))
3801 /* Delayed registration/unregisteration */
3802 static DEFINE_SPINLOCK(net_todo_list_lock);
3803 static LIST_HEAD(net_todo_list);
3805 static void net_set_todo(struct net_device *dev)
3807 spin_lock(&net_todo_list_lock);
3808 list_add_tail(&dev->todo_list, &net_todo_list);
3809 spin_unlock(&net_todo_list_lock);
3812 static void rollback_registered(struct net_device *dev)
3814 BUG_ON(dev_boot_phase);
3817 /* Some devices call without registering for initialization unwind. */
3818 if (dev->reg_state == NETREG_UNINITIALIZED) {
3819 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3820 "was registered\n", dev->name, dev);
3826 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3828 /* If device is running, close it first. */
3831 /* And unlink it from device chain. */
3832 unlist_netdevice(dev);
3834 dev->reg_state = NETREG_UNREGISTERING;
3838 /* Shutdown queueing discipline. */
3842 /* Notify protocols, that we are about to destroy
3843 this device. They should clean all the things.
3845 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3848 * Flush the unicast and multicast chains
3850 dev_addr_discard(dev);
3855 /* Notifier chain MUST detach us from master device. */
3856 WARN_ON(dev->master);
3858 /* Remove entries from kobject tree */
3859 netdev_unregister_kobject(dev);
3866 static void __netdev_init_queue_locks_one(struct net_device *dev,
3867 struct netdev_queue *dev_queue,
3870 spin_lock_init(&dev_queue->_xmit_lock);
3871 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
3872 dev_queue->xmit_lock_owner = -1;
3875 static void netdev_init_queue_locks(struct net_device *dev)
3877 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
3878 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
3882 * register_netdevice - register a network device
3883 * @dev: device to register
3885 * Take a completed network device structure and add it to the kernel
3886 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3887 * chain. 0 is returned on success. A negative errno code is returned
3888 * on a failure to set up the device, or if the name is a duplicate.
3890 * Callers must hold the rtnl semaphore. You may want
3891 * register_netdev() instead of this.
3894 * The locking appears insufficient to guarantee two parallel registers
3895 * will not get the same name.
3898 int register_netdevice(struct net_device *dev)
3900 struct hlist_head *head;
3901 struct hlist_node *p;
3905 BUG_ON(dev_boot_phase);
3910 /* When net_device's are persistent, this will be fatal. */
3911 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3912 BUG_ON(!dev_net(dev));
3915 spin_lock_init(&dev->addr_list_lock);
3916 netdev_set_addr_lockdep_class(dev);
3917 netdev_init_queue_locks(dev);
3921 /* Init, if this function is available */
3923 ret = dev->init(dev);
3931 if (!dev_valid_name(dev->name)) {
3936 dev->ifindex = dev_new_index(net);
3937 if (dev->iflink == -1)
3938 dev->iflink = dev->ifindex;
3940 /* Check for existence of name */
3941 head = dev_name_hash(net, dev->name);
3942 hlist_for_each(p, head) {
3943 struct net_device *d
3944 = hlist_entry(p, struct net_device, name_hlist);
3945 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3951 /* Fix illegal checksum combinations */
3952 if ((dev->features & NETIF_F_HW_CSUM) &&
3953 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3954 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3956 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3959 if ((dev->features & NETIF_F_NO_CSUM) &&
3960 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3961 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3963 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3967 /* Fix illegal SG+CSUM combinations. */
3968 if ((dev->features & NETIF_F_SG) &&
3969 !(dev->features & NETIF_F_ALL_CSUM)) {
3970 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3972 dev->features &= ~NETIF_F_SG;
3975 /* TSO requires that SG is present as well. */
3976 if ((dev->features & NETIF_F_TSO) &&
3977 !(dev->features & NETIF_F_SG)) {
3978 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3980 dev->features &= ~NETIF_F_TSO;
3982 if (dev->features & NETIF_F_UFO) {
3983 if (!(dev->features & NETIF_F_HW_CSUM)) {
3984 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3985 "NETIF_F_HW_CSUM feature.\n",
3987 dev->features &= ~NETIF_F_UFO;
3989 if (!(dev->features & NETIF_F_SG)) {
3990 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3991 "NETIF_F_SG feature.\n",
3993 dev->features &= ~NETIF_F_UFO;
3997 /* Enable software GSO if SG is supported. */
3998 if (dev->features & NETIF_F_SG)
3999 dev->features |= NETIF_F_GSO;
4001 netdev_initialize_kobject(dev);
4002 ret = netdev_register_kobject(dev);
4005 dev->reg_state = NETREG_REGISTERED;
4008 * Default initial state at registry is that the
4009 * device is present.
4012 set_bit(__LINK_STATE_PRESENT, &dev->state);
4014 dev_init_scheduler(dev);
4016 list_netdevice(dev);
4018 /* Notify protocols, that a new device appeared. */
4019 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4020 ret = notifier_to_errno(ret);
4022 rollback_registered(dev);
4023 dev->reg_state = NETREG_UNREGISTERED;
4036 * register_netdev - register a network device
4037 * @dev: device to register
4039 * Take a completed network device structure and add it to the kernel
4040 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4041 * chain. 0 is returned on success. A negative errno code is returned
4042 * on a failure to set up the device, or if the name is a duplicate.
4044 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4045 * and expands the device name if you passed a format string to
4048 int register_netdev(struct net_device *dev)
4055 * If the name is a format string the caller wants us to do a
4058 if (strchr(dev->name, '%')) {
4059 err = dev_alloc_name(dev, dev->name);
4064 err = register_netdevice(dev);
4069 EXPORT_SYMBOL(register_netdev);
4072 * netdev_wait_allrefs - wait until all references are gone.
4074 * This is called when unregistering network devices.
4076 * Any protocol or device that holds a reference should register
4077 * for netdevice notification, and cleanup and put back the
4078 * reference if they receive an UNREGISTER event.
4079 * We can get stuck here if buggy protocols don't correctly
4082 static void netdev_wait_allrefs(struct net_device *dev)
4084 unsigned long rebroadcast_time, warning_time;
4086 rebroadcast_time = warning_time = jiffies;
4087 while (atomic_read(&dev->refcnt) != 0) {
4088 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4091 /* Rebroadcast unregister notification */
4092 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4094 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4096 /* We must not have linkwatch events
4097 * pending on unregister. If this
4098 * happens, we simply run the queue
4099 * unscheduled, resulting in a noop
4102 linkwatch_run_queue();
4107 rebroadcast_time = jiffies;
4112 if (time_after(jiffies, warning_time + 10 * HZ)) {
4113 printk(KERN_EMERG "unregister_netdevice: "
4114 "waiting for %s to become free. Usage "
4116 dev->name, atomic_read(&dev->refcnt));
4117 warning_time = jiffies;
4126 * register_netdevice(x1);
4127 * register_netdevice(x2);
4129 * unregister_netdevice(y1);
4130 * unregister_netdevice(y2);
4136 * We are invoked by rtnl_unlock() after it drops the semaphore.
4137 * This allows us to deal with problems:
4138 * 1) We can delete sysfs objects which invoke hotplug
4139 * without deadlocking with linkwatch via keventd.
4140 * 2) Since we run with the RTNL semaphore not held, we can sleep
4141 * safely in order to wait for the netdev refcnt to drop to zero.
4143 static DEFINE_MUTEX(net_todo_run_mutex);
4144 void netdev_run_todo(void)
4146 struct list_head list;
4148 /* Need to guard against multiple cpu's getting out of order. */
4149 mutex_lock(&net_todo_run_mutex);
4151 /* Not safe to do outside the semaphore. We must not return
4152 * until all unregister events invoked by the local processor
4153 * have been completed (either by this todo run, or one on
4156 if (list_empty(&net_todo_list))
4159 /* Snapshot list, allow later requests */
4160 spin_lock(&net_todo_list_lock);
4161 list_replace_init(&net_todo_list, &list);
4162 spin_unlock(&net_todo_list_lock);
4164 while (!list_empty(&list)) {
4165 struct net_device *dev
4166 = list_entry(list.next, struct net_device, todo_list);
4167 list_del(&dev->todo_list);
4169 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4170 printk(KERN_ERR "network todo '%s' but state %d\n",
4171 dev->name, dev->reg_state);
4176 dev->reg_state = NETREG_UNREGISTERED;
4178 on_each_cpu(flush_backlog, dev, 1);
4180 netdev_wait_allrefs(dev);
4183 BUG_ON(atomic_read(&dev->refcnt));
4184 WARN_ON(dev->ip_ptr);
4185 WARN_ON(dev->ip6_ptr);
4186 WARN_ON(dev->dn_ptr);
4188 if (dev->destructor)
4189 dev->destructor(dev);
4191 /* Free network device */
4192 kobject_put(&dev->dev.kobj);
4196 mutex_unlock(&net_todo_run_mutex);
4199 static struct net_device_stats *internal_stats(struct net_device *dev)
4204 static void netdev_init_one_queue(struct net_device *dev,
4205 struct netdev_queue *queue,
4211 static void netdev_init_queues(struct net_device *dev)
4213 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4214 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4215 spin_lock_init(&dev->tx_global_lock);
4219 * alloc_netdev_mq - allocate network device
4220 * @sizeof_priv: size of private data to allocate space for
4221 * @name: device name format string
4222 * @setup: callback to initialize device
4223 * @queue_count: the number of subqueues to allocate
4225 * Allocates a struct net_device with private data area for driver use
4226 * and performs basic initialization. Also allocates subquue structs
4227 * for each queue on the device at the end of the netdevice.
4229 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4230 void (*setup)(struct net_device *), unsigned int queue_count)
4232 struct netdev_queue *tx;
4233 struct net_device *dev;
4237 BUG_ON(strlen(name) >= sizeof(dev->name));
4239 alloc_size = sizeof(struct net_device);
4241 /* ensure 32-byte alignment of private area */
4242 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4243 alloc_size += sizeof_priv;
4245 /* ensure 32-byte alignment of whole construct */
4246 alloc_size += NETDEV_ALIGN_CONST;
4248 p = kzalloc(alloc_size, GFP_KERNEL);
4250 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4254 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
4256 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4262 dev = (struct net_device *)
4263 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4264 dev->padded = (char *)dev - (char *)p;
4265 dev_net_set(dev, &init_net);
4268 dev->num_tx_queues = queue_count;
4269 dev->real_num_tx_queues = queue_count;
4272 dev->priv = ((char *)dev +
4273 ((sizeof(struct net_device) + NETDEV_ALIGN_CONST)
4274 & ~NETDEV_ALIGN_CONST));
4277 dev->gso_max_size = GSO_MAX_SIZE;
4279 netdev_init_queues(dev);
4281 dev->get_stats = internal_stats;
4282 netpoll_netdev_init(dev);
4284 strcpy(dev->name, name);
4287 EXPORT_SYMBOL(alloc_netdev_mq);
4290 * free_netdev - free network device
4293 * This function does the last stage of destroying an allocated device
4294 * interface. The reference to the device object is released.
4295 * If this is the last reference then it will be freed.
4297 void free_netdev(struct net_device *dev)
4299 release_net(dev_net(dev));
4303 /* Compatibility with error handling in drivers */
4304 if (dev->reg_state == NETREG_UNINITIALIZED) {
4305 kfree((char *)dev - dev->padded);
4309 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4310 dev->reg_state = NETREG_RELEASED;
4312 /* will free via device release */
4313 put_device(&dev->dev);
4316 /* Synchronize with packet receive processing. */
4317 void synchronize_net(void)
4324 * unregister_netdevice - remove device from the kernel
4327 * This function shuts down a device interface and removes it
4328 * from the kernel tables.
4330 * Callers must hold the rtnl semaphore. You may want
4331 * unregister_netdev() instead of this.
4334 void unregister_netdevice(struct net_device *dev)
4338 rollback_registered(dev);
4339 /* Finish processing unregister after unlock */
4344 * unregister_netdev - remove device from the kernel
4347 * This function shuts down a device interface and removes it
4348 * from the kernel tables.
4350 * This is just a wrapper for unregister_netdevice that takes
4351 * the rtnl semaphore. In general you want to use this and not
4352 * unregister_netdevice.
4354 void unregister_netdev(struct net_device *dev)
4357 unregister_netdevice(dev);
4361 EXPORT_SYMBOL(unregister_netdev);
4364 * dev_change_net_namespace - move device to different nethost namespace
4366 * @net: network namespace
4367 * @pat: If not NULL name pattern to try if the current device name
4368 * is already taken in the destination network namespace.
4370 * This function shuts down a device interface and moves it
4371 * to a new network namespace. On success 0 is returned, on
4372 * a failure a netagive errno code is returned.
4374 * Callers must hold the rtnl semaphore.
4377 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4380 const char *destname;
4385 /* Don't allow namespace local devices to be moved. */
4387 if (dev->features & NETIF_F_NETNS_LOCAL)
4390 /* Ensure the device has been registrered */
4392 if (dev->reg_state != NETREG_REGISTERED)
4395 /* Get out if there is nothing todo */
4397 if (net_eq(dev_net(dev), net))
4400 /* Pick the destination device name, and ensure
4401 * we can use it in the destination network namespace.
4404 destname = dev->name;
4405 if (__dev_get_by_name(net, destname)) {
4406 /* We get here if we can't use the current device name */
4409 if (!dev_valid_name(pat))
4411 if (strchr(pat, '%')) {
4412 if (__dev_alloc_name(net, pat, buf) < 0)
4417 if (__dev_get_by_name(net, destname))
4422 * And now a mini version of register_netdevice unregister_netdevice.
4425 /* If device is running close it first. */
4428 /* And unlink it from device chain */
4430 unlist_netdevice(dev);
4434 /* Shutdown queueing discipline. */
4437 /* Notify protocols, that we are about to destroy
4438 this device. They should clean all the things.
4440 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4443 * Flush the unicast and multicast chains
4445 dev_addr_discard(dev);
4447 /* Actually switch the network namespace */
4448 dev_net_set(dev, net);
4450 /* Assign the new device name */
4451 if (destname != dev->name)
4452 strcpy(dev->name, destname);
4454 /* If there is an ifindex conflict assign a new one */
4455 if (__dev_get_by_index(net, dev->ifindex)) {
4456 int iflink = (dev->iflink == dev->ifindex);
4457 dev->ifindex = dev_new_index(net);
4459 dev->iflink = dev->ifindex;
4462 /* Fixup kobjects */
4463 netdev_unregister_kobject(dev);
4464 err = netdev_register_kobject(dev);
4467 /* Add the device back in the hashes */
4468 list_netdevice(dev);
4470 /* Notify protocols, that a new device appeared. */
4471 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4479 static int dev_cpu_callback(struct notifier_block *nfb,
4480 unsigned long action,
4483 struct sk_buff **list_skb;
4484 struct Qdisc **list_net;
4485 struct sk_buff *skb;
4486 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4487 struct softnet_data *sd, *oldsd;
4489 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4492 local_irq_disable();
4493 cpu = smp_processor_id();
4494 sd = &per_cpu(softnet_data, cpu);
4495 oldsd = &per_cpu(softnet_data, oldcpu);
4497 /* Find end of our completion_queue. */
4498 list_skb = &sd->completion_queue;
4500 list_skb = &(*list_skb)->next;
4501 /* Append completion queue from offline CPU. */
4502 *list_skb = oldsd->completion_queue;
4503 oldsd->completion_queue = NULL;
4505 /* Find end of our output_queue. */
4506 list_net = &sd->output_queue;
4508 list_net = &(*list_net)->next_sched;
4509 /* Append output queue from offline CPU. */
4510 *list_net = oldsd->output_queue;
4511 oldsd->output_queue = NULL;
4513 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4516 /* Process offline CPU's input_pkt_queue */
4517 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4523 #ifdef CONFIG_NET_DMA
4525 * net_dma_rebalance - try to maintain one DMA channel per CPU
4526 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4528 * This is called when the number of channels allocated to the net_dma client
4529 * changes. The net_dma client tries to have one DMA channel per CPU.
4532 static void net_dma_rebalance(struct net_dma *net_dma)
4534 unsigned int cpu, i, n, chan_idx;
4535 struct dma_chan *chan;
4537 if (cpus_empty(net_dma->channel_mask)) {
4538 for_each_online_cpu(cpu)
4539 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4544 cpu = first_cpu(cpu_online_map);
4546 for_each_cpu_mask_nr(chan_idx, net_dma->channel_mask) {
4547 chan = net_dma->channels[chan_idx];
4549 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4550 + (i < (num_online_cpus() %
4551 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4554 per_cpu(softnet_data, cpu).net_dma = chan;
4555 cpu = next_cpu(cpu, cpu_online_map);
4563 * netdev_dma_event - event callback for the net_dma_client
4564 * @client: should always be net_dma_client
4565 * @chan: DMA channel for the event
4566 * @state: DMA state to be handled
4568 static enum dma_state_client
4569 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4570 enum dma_state state)
4572 int i, found = 0, pos = -1;
4573 struct net_dma *net_dma =
4574 container_of(client, struct net_dma, client);
4575 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4577 spin_lock(&net_dma->lock);
4579 case DMA_RESOURCE_AVAILABLE:
4580 for (i = 0; i < nr_cpu_ids; i++)
4581 if (net_dma->channels[i] == chan) {
4584 } else if (net_dma->channels[i] == NULL && pos < 0)
4587 if (!found && pos >= 0) {
4589 net_dma->channels[pos] = chan;
4590 cpu_set(pos, net_dma->channel_mask);
4591 net_dma_rebalance(net_dma);
4594 case DMA_RESOURCE_REMOVED:
4595 for (i = 0; i < nr_cpu_ids; i++)
4596 if (net_dma->channels[i] == chan) {
4604 cpu_clear(pos, net_dma->channel_mask);
4605 net_dma->channels[i] = NULL;
4606 net_dma_rebalance(net_dma);
4612 spin_unlock(&net_dma->lock);
4618 * netdev_dma_regiser - register the networking subsystem as a DMA client
4620 static int __init netdev_dma_register(void)
4622 net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma),
4624 if (unlikely(!net_dma.channels)) {
4626 "netdev_dma: no memory for net_dma.channels\n");
4629 spin_lock_init(&net_dma.lock);
4630 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4631 dma_async_client_register(&net_dma.client);
4632 dma_async_client_chan_request(&net_dma.client);
4637 static int __init netdev_dma_register(void) { return -ENODEV; }
4638 #endif /* CONFIG_NET_DMA */
4641 * netdev_compute_feature - compute conjunction of two feature sets
4642 * @all: first feature set
4643 * @one: second feature set
4645 * Computes a new feature set after adding a device with feature set
4646 * @one to the master device with current feature set @all. Returns
4647 * the new feature set.
4649 int netdev_compute_features(unsigned long all, unsigned long one)
4651 /* if device needs checksumming, downgrade to hw checksumming */
4652 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4653 all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
4655 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4656 if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
4657 all ^= NETIF_F_HW_CSUM
4658 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4660 if (one & NETIF_F_GSO)
4661 one |= NETIF_F_GSO_SOFTWARE;
4664 /* If even one device supports robust GSO, enable it for all. */
4665 if (one & NETIF_F_GSO_ROBUST)
4666 all |= NETIF_F_GSO_ROBUST;
4668 all &= one | NETIF_F_LLTX;
4670 if (!(all & NETIF_F_ALL_CSUM))
4672 if (!(all & NETIF_F_SG))
4673 all &= ~NETIF_F_GSO_MASK;
4677 EXPORT_SYMBOL(netdev_compute_features);
4679 static struct hlist_head *netdev_create_hash(void)
4682 struct hlist_head *hash;
4684 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4686 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4687 INIT_HLIST_HEAD(&hash[i]);
4692 /* Initialize per network namespace state */
4693 static int __net_init netdev_init(struct net *net)
4695 INIT_LIST_HEAD(&net->dev_base_head);
4697 net->dev_name_head = netdev_create_hash();
4698 if (net->dev_name_head == NULL)
4701 net->dev_index_head = netdev_create_hash();
4702 if (net->dev_index_head == NULL)
4708 kfree(net->dev_name_head);
4713 char *netdev_drivername(struct net_device *dev, char *buffer, int len)
4715 struct device_driver *driver;
4716 struct device *parent;
4718 if (len <= 0 || !buffer)
4722 parent = dev->dev.parent;
4727 driver = parent->driver;
4728 if (driver && driver->name)
4729 strlcpy(buffer, driver->name, len);
4733 static void __net_exit netdev_exit(struct net *net)
4735 kfree(net->dev_name_head);
4736 kfree(net->dev_index_head);
4739 static struct pernet_operations __net_initdata netdev_net_ops = {
4740 .init = netdev_init,
4741 .exit = netdev_exit,
4744 static void __net_exit default_device_exit(struct net *net)
4746 struct net_device *dev, *next;
4748 * Push all migratable of the network devices back to the
4749 * initial network namespace
4752 for_each_netdev_safe(net, dev, next) {
4754 char fb_name[IFNAMSIZ];
4756 /* Ignore unmoveable devices (i.e. loopback) */
4757 if (dev->features & NETIF_F_NETNS_LOCAL)
4760 /* Push remaing network devices to init_net */
4761 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
4762 err = dev_change_net_namespace(dev, &init_net, fb_name);
4764 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
4765 __func__, dev->name, err);
4772 static struct pernet_operations __net_initdata default_device_ops = {
4773 .exit = default_device_exit,
4777 * Initialize the DEV module. At boot time this walks the device list and
4778 * unhooks any devices that fail to initialise (normally hardware not
4779 * present) and leaves us with a valid list of present and active devices.
4784 * This is called single threaded during boot, so no need
4785 * to take the rtnl semaphore.
4787 static int __init net_dev_init(void)
4789 int i, rc = -ENOMEM;
4791 BUG_ON(!dev_boot_phase);
4793 if (dev_proc_init())
4796 if (netdev_kobject_init())
4799 INIT_LIST_HEAD(&ptype_all);
4800 for (i = 0; i < PTYPE_HASH_SIZE; i++)
4801 INIT_LIST_HEAD(&ptype_base[i]);
4803 if (register_pernet_subsys(&netdev_net_ops))
4806 if (register_pernet_device(&default_device_ops))
4810 * Initialise the packet receive queues.
4813 for_each_possible_cpu(i) {
4814 struct softnet_data *queue;
4816 queue = &per_cpu(softnet_data, i);
4817 skb_queue_head_init(&queue->input_pkt_queue);
4818 queue->completion_queue = NULL;
4819 INIT_LIST_HEAD(&queue->poll_list);
4821 queue->backlog.poll = process_backlog;
4822 queue->backlog.weight = weight_p;
4825 netdev_dma_register();
4829 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
4830 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
4832 hotcpu_notifier(dev_cpu_callback, 0);
4840 subsys_initcall(net_dev_init);
4842 EXPORT_SYMBOL(__dev_get_by_index);
4843 EXPORT_SYMBOL(__dev_get_by_name);
4844 EXPORT_SYMBOL(__dev_remove_pack);
4845 EXPORT_SYMBOL(dev_valid_name);
4846 EXPORT_SYMBOL(dev_add_pack);
4847 EXPORT_SYMBOL(dev_alloc_name);
4848 EXPORT_SYMBOL(dev_close);
4849 EXPORT_SYMBOL(dev_get_by_flags);
4850 EXPORT_SYMBOL(dev_get_by_index);
4851 EXPORT_SYMBOL(dev_get_by_name);
4852 EXPORT_SYMBOL(dev_open);
4853 EXPORT_SYMBOL(dev_queue_xmit);
4854 EXPORT_SYMBOL(dev_remove_pack);
4855 EXPORT_SYMBOL(dev_set_allmulti);
4856 EXPORT_SYMBOL(dev_set_promiscuity);
4857 EXPORT_SYMBOL(dev_change_flags);
4858 EXPORT_SYMBOL(dev_set_mtu);
4859 EXPORT_SYMBOL(dev_set_mac_address);
4860 EXPORT_SYMBOL(free_netdev);
4861 EXPORT_SYMBOL(netdev_boot_setup_check);
4862 EXPORT_SYMBOL(netdev_set_master);
4863 EXPORT_SYMBOL(netdev_state_change);
4864 EXPORT_SYMBOL(netif_receive_skb);
4865 EXPORT_SYMBOL(netif_rx);
4866 EXPORT_SYMBOL(register_gifconf);
4867 EXPORT_SYMBOL(register_netdevice);
4868 EXPORT_SYMBOL(register_netdevice_notifier);
4869 EXPORT_SYMBOL(skb_checksum_help);
4870 EXPORT_SYMBOL(synchronize_net);
4871 EXPORT_SYMBOL(unregister_netdevice);
4872 EXPORT_SYMBOL(unregister_netdevice_notifier);
4873 EXPORT_SYMBOL(net_enable_timestamp);
4874 EXPORT_SYMBOL(net_disable_timestamp);
4875 EXPORT_SYMBOL(dev_get_flags);
4877 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4878 EXPORT_SYMBOL(br_handle_frame_hook);
4879 EXPORT_SYMBOL(br_fdb_get_hook);
4880 EXPORT_SYMBOL(br_fdb_put_hook);
4884 EXPORT_SYMBOL(dev_load);
4887 EXPORT_PER_CPU_SYMBOL(softnet_data);