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 if (!test_bit(__QDISC_STATE_DEACTIVATED,
1995 __netif_reschedule(q);
2001 static inline int deliver_skb(struct sk_buff *skb,
2002 struct packet_type *pt_prev,
2003 struct net_device *orig_dev)
2005 atomic_inc(&skb->users);
2006 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2009 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2010 /* These hooks defined here for ATM */
2012 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2013 unsigned char *addr);
2014 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
2017 * If bridge module is loaded call bridging hook.
2018 * returns NULL if packet was consumed.
2020 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2021 struct sk_buff *skb) __read_mostly;
2022 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2023 struct packet_type **pt_prev, int *ret,
2024 struct net_device *orig_dev)
2026 struct net_bridge_port *port;
2028 if (skb->pkt_type == PACKET_LOOPBACK ||
2029 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2033 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2037 return br_handle_frame_hook(port, skb);
2040 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2043 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2044 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2045 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2047 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2048 struct packet_type **pt_prev,
2050 struct net_device *orig_dev)
2052 if (skb->dev->macvlan_port == NULL)
2056 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2059 return macvlan_handle_frame_hook(skb);
2062 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2065 #ifdef CONFIG_NET_CLS_ACT
2066 /* TODO: Maybe we should just force sch_ingress to be compiled in
2067 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2068 * a compare and 2 stores extra right now if we dont have it on
2069 * but have CONFIG_NET_CLS_ACT
2070 * NOTE: This doesnt stop any functionality; if you dont have
2071 * the ingress scheduler, you just cant add policies on ingress.
2074 static int ing_filter(struct sk_buff *skb)
2076 struct net_device *dev = skb->dev;
2077 u32 ttl = G_TC_RTTL(skb->tc_verd);
2078 struct netdev_queue *rxq;
2079 int result = TC_ACT_OK;
2082 if (MAX_RED_LOOP < ttl++) {
2084 "Redir loop detected Dropping packet (%d->%d)\n",
2085 skb->iif, dev->ifindex);
2089 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2090 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2092 rxq = &dev->rx_queue;
2095 if (q != &noop_qdisc) {
2096 spin_lock(qdisc_lock(q));
2097 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2098 result = qdisc_enqueue_root(skb, q);
2099 spin_unlock(qdisc_lock(q));
2105 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2106 struct packet_type **pt_prev,
2107 int *ret, struct net_device *orig_dev)
2109 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2113 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2116 /* Huh? Why does turning on AF_PACKET affect this? */
2117 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2120 switch (ing_filter(skb)) {
2134 * netif_nit_deliver - deliver received packets to network taps
2137 * This function is used to deliver incoming packets to network
2138 * taps. It should be used when the normal netif_receive_skb path
2139 * is bypassed, for example because of VLAN acceleration.
2141 void netif_nit_deliver(struct sk_buff *skb)
2143 struct packet_type *ptype;
2145 if (list_empty(&ptype_all))
2148 skb_reset_network_header(skb);
2149 skb_reset_transport_header(skb);
2150 skb->mac_len = skb->network_header - skb->mac_header;
2153 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2154 if (!ptype->dev || ptype->dev == skb->dev)
2155 deliver_skb(skb, ptype, skb->dev);
2161 * netif_receive_skb - process receive buffer from network
2162 * @skb: buffer to process
2164 * netif_receive_skb() is the main receive data processing function.
2165 * It always succeeds. The buffer may be dropped during processing
2166 * for congestion control or by the protocol layers.
2168 * This function may only be called from softirq context and interrupts
2169 * should be enabled.
2171 * Return values (usually ignored):
2172 * NET_RX_SUCCESS: no congestion
2173 * NET_RX_DROP: packet was dropped
2175 int netif_receive_skb(struct sk_buff *skb)
2177 struct packet_type *ptype, *pt_prev;
2178 struct net_device *orig_dev;
2179 struct net_device *null_or_orig;
2180 int ret = NET_RX_DROP;
2183 /* if we've gotten here through NAPI, check netpoll */
2184 if (netpoll_receive_skb(skb))
2187 if (!skb->tstamp.tv64)
2191 skb->iif = skb->dev->ifindex;
2193 null_or_orig = NULL;
2194 orig_dev = skb->dev;
2195 if (orig_dev->master) {
2196 if (skb_bond_should_drop(skb))
2197 null_or_orig = orig_dev; /* deliver only exact match */
2199 skb->dev = orig_dev->master;
2202 __get_cpu_var(netdev_rx_stat).total++;
2204 skb_reset_network_header(skb);
2205 skb_reset_transport_header(skb);
2206 skb->mac_len = skb->network_header - skb->mac_header;
2212 /* Don't receive packets in an exiting network namespace */
2213 if (!net_alive(dev_net(skb->dev)))
2216 #ifdef CONFIG_NET_CLS_ACT
2217 if (skb->tc_verd & TC_NCLS) {
2218 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2223 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2224 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2225 ptype->dev == orig_dev) {
2227 ret = deliver_skb(skb, pt_prev, orig_dev);
2232 #ifdef CONFIG_NET_CLS_ACT
2233 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2239 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2242 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2246 type = skb->protocol;
2247 list_for_each_entry_rcu(ptype,
2248 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2249 if (ptype->type == type &&
2250 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2251 ptype->dev == orig_dev)) {
2253 ret = deliver_skb(skb, pt_prev, orig_dev);
2259 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2262 /* Jamal, now you will not able to escape explaining
2263 * me how you were going to use this. :-)
2273 /* Network device is going away, flush any packets still pending */
2274 static void flush_backlog(void *arg)
2276 struct net_device *dev = arg;
2277 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2278 struct sk_buff *skb, *tmp;
2280 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2281 if (skb->dev == dev) {
2282 __skb_unlink(skb, &queue->input_pkt_queue);
2287 static int process_backlog(struct napi_struct *napi, int quota)
2290 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2291 unsigned long start_time = jiffies;
2293 napi->weight = weight_p;
2295 struct sk_buff *skb;
2297 local_irq_disable();
2298 skb = __skb_dequeue(&queue->input_pkt_queue);
2300 __napi_complete(napi);
2306 netif_receive_skb(skb);
2307 } while (++work < quota && jiffies == start_time);
2313 * __napi_schedule - schedule for receive
2314 * @n: entry to schedule
2316 * The entry's receive function will be scheduled to run
2318 void __napi_schedule(struct napi_struct *n)
2320 unsigned long flags;
2322 local_irq_save(flags);
2323 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2324 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2325 local_irq_restore(flags);
2327 EXPORT_SYMBOL(__napi_schedule);
2330 static void net_rx_action(struct softirq_action *h)
2332 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2333 unsigned long start_time = jiffies;
2334 int budget = netdev_budget;
2337 local_irq_disable();
2339 while (!list_empty(list)) {
2340 struct napi_struct *n;
2343 /* If softirq window is exhuasted then punt.
2345 * Note that this is a slight policy change from the
2346 * previous NAPI code, which would allow up to 2
2347 * jiffies to pass before breaking out. The test
2348 * used to be "jiffies - start_time > 1".
2350 if (unlikely(budget <= 0 || jiffies != start_time))
2355 /* Even though interrupts have been re-enabled, this
2356 * access is safe because interrupts can only add new
2357 * entries to the tail of this list, and only ->poll()
2358 * calls can remove this head entry from the list.
2360 n = list_entry(list->next, struct napi_struct, poll_list);
2362 have = netpoll_poll_lock(n);
2366 /* This NAPI_STATE_SCHED test is for avoiding a race
2367 * with netpoll's poll_napi(). Only the entity which
2368 * obtains the lock and sees NAPI_STATE_SCHED set will
2369 * actually make the ->poll() call. Therefore we avoid
2370 * accidently calling ->poll() when NAPI is not scheduled.
2373 if (test_bit(NAPI_STATE_SCHED, &n->state))
2374 work = n->poll(n, weight);
2376 WARN_ON_ONCE(work > weight);
2380 local_irq_disable();
2382 /* Drivers must not modify the NAPI state if they
2383 * consume the entire weight. In such cases this code
2384 * still "owns" the NAPI instance and therefore can
2385 * move the instance around on the list at-will.
2387 if (unlikely(work == weight)) {
2388 if (unlikely(napi_disable_pending(n)))
2391 list_move_tail(&n->poll_list, list);
2394 netpoll_poll_unlock(have);
2399 #ifdef CONFIG_NET_DMA
2401 * There may not be any more sk_buffs coming right now, so push
2402 * any pending DMA copies to hardware
2404 if (!cpus_empty(net_dma.channel_mask)) {
2406 for_each_cpu_mask_nr(chan_idx, net_dma.channel_mask) {
2407 struct dma_chan *chan = net_dma.channels[chan_idx];
2409 dma_async_memcpy_issue_pending(chan);
2417 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2418 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2422 static gifconf_func_t * gifconf_list [NPROTO];
2425 * register_gifconf - register a SIOCGIF handler
2426 * @family: Address family
2427 * @gifconf: Function handler
2429 * Register protocol dependent address dumping routines. The handler
2430 * that is passed must not be freed or reused until it has been replaced
2431 * by another handler.
2433 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2435 if (family >= NPROTO)
2437 gifconf_list[family] = gifconf;
2443 * Map an interface index to its name (SIOCGIFNAME)
2447 * We need this ioctl for efficient implementation of the
2448 * if_indextoname() function required by the IPv6 API. Without
2449 * it, we would have to search all the interfaces to find a
2453 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2455 struct net_device *dev;
2459 * Fetch the caller's info block.
2462 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2465 read_lock(&dev_base_lock);
2466 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2468 read_unlock(&dev_base_lock);
2472 strcpy(ifr.ifr_name, dev->name);
2473 read_unlock(&dev_base_lock);
2475 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2481 * Perform a SIOCGIFCONF call. This structure will change
2482 * size eventually, and there is nothing I can do about it.
2483 * Thus we will need a 'compatibility mode'.
2486 static int dev_ifconf(struct net *net, char __user *arg)
2489 struct net_device *dev;
2496 * Fetch the caller's info block.
2499 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2506 * Loop over the interfaces, and write an info block for each.
2510 for_each_netdev(net, dev) {
2511 for (i = 0; i < NPROTO; i++) {
2512 if (gifconf_list[i]) {
2515 done = gifconf_list[i](dev, NULL, 0);
2517 done = gifconf_list[i](dev, pos + total,
2527 * All done. Write the updated control block back to the caller.
2529 ifc.ifc_len = total;
2532 * Both BSD and Solaris return 0 here, so we do too.
2534 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2537 #ifdef CONFIG_PROC_FS
2539 * This is invoked by the /proc filesystem handler to display a device
2542 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2543 __acquires(dev_base_lock)
2545 struct net *net = seq_file_net(seq);
2547 struct net_device *dev;
2549 read_lock(&dev_base_lock);
2551 return SEQ_START_TOKEN;
2554 for_each_netdev(net, dev)
2561 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2563 struct net *net = seq_file_net(seq);
2565 return v == SEQ_START_TOKEN ?
2566 first_net_device(net) : next_net_device((struct net_device *)v);
2569 void dev_seq_stop(struct seq_file *seq, void *v)
2570 __releases(dev_base_lock)
2572 read_unlock(&dev_base_lock);
2575 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2577 struct net_device_stats *stats = dev->get_stats(dev);
2579 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2580 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2581 dev->name, stats->rx_bytes, stats->rx_packets,
2583 stats->rx_dropped + stats->rx_missed_errors,
2584 stats->rx_fifo_errors,
2585 stats->rx_length_errors + stats->rx_over_errors +
2586 stats->rx_crc_errors + stats->rx_frame_errors,
2587 stats->rx_compressed, stats->multicast,
2588 stats->tx_bytes, stats->tx_packets,
2589 stats->tx_errors, stats->tx_dropped,
2590 stats->tx_fifo_errors, stats->collisions,
2591 stats->tx_carrier_errors +
2592 stats->tx_aborted_errors +
2593 stats->tx_window_errors +
2594 stats->tx_heartbeat_errors,
2595 stats->tx_compressed);
2599 * Called from the PROCfs module. This now uses the new arbitrary sized
2600 * /proc/net interface to create /proc/net/dev
2602 static int dev_seq_show(struct seq_file *seq, void *v)
2604 if (v == SEQ_START_TOKEN)
2605 seq_puts(seq, "Inter-| Receive "
2607 " face |bytes packets errs drop fifo frame "
2608 "compressed multicast|bytes packets errs "
2609 "drop fifo colls carrier compressed\n");
2611 dev_seq_printf_stats(seq, v);
2615 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2617 struct netif_rx_stats *rc = NULL;
2619 while (*pos < nr_cpu_ids)
2620 if (cpu_online(*pos)) {
2621 rc = &per_cpu(netdev_rx_stat, *pos);
2628 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2630 return softnet_get_online(pos);
2633 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2636 return softnet_get_online(pos);
2639 static void softnet_seq_stop(struct seq_file *seq, void *v)
2643 static int softnet_seq_show(struct seq_file *seq, void *v)
2645 struct netif_rx_stats *s = v;
2647 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2648 s->total, s->dropped, s->time_squeeze, 0,
2649 0, 0, 0, 0, /* was fastroute */
2654 static const struct seq_operations dev_seq_ops = {
2655 .start = dev_seq_start,
2656 .next = dev_seq_next,
2657 .stop = dev_seq_stop,
2658 .show = dev_seq_show,
2661 static int dev_seq_open(struct inode *inode, struct file *file)
2663 return seq_open_net(inode, file, &dev_seq_ops,
2664 sizeof(struct seq_net_private));
2667 static const struct file_operations dev_seq_fops = {
2668 .owner = THIS_MODULE,
2669 .open = dev_seq_open,
2671 .llseek = seq_lseek,
2672 .release = seq_release_net,
2675 static const struct seq_operations softnet_seq_ops = {
2676 .start = softnet_seq_start,
2677 .next = softnet_seq_next,
2678 .stop = softnet_seq_stop,
2679 .show = softnet_seq_show,
2682 static int softnet_seq_open(struct inode *inode, struct file *file)
2684 return seq_open(file, &softnet_seq_ops);
2687 static const struct file_operations softnet_seq_fops = {
2688 .owner = THIS_MODULE,
2689 .open = softnet_seq_open,
2691 .llseek = seq_lseek,
2692 .release = seq_release,
2695 static void *ptype_get_idx(loff_t pos)
2697 struct packet_type *pt = NULL;
2701 list_for_each_entry_rcu(pt, &ptype_all, list) {
2707 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
2708 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2717 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2721 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2724 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2726 struct packet_type *pt;
2727 struct list_head *nxt;
2731 if (v == SEQ_START_TOKEN)
2732 return ptype_get_idx(0);
2735 nxt = pt->list.next;
2736 if (pt->type == htons(ETH_P_ALL)) {
2737 if (nxt != &ptype_all)
2740 nxt = ptype_base[0].next;
2742 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
2744 while (nxt == &ptype_base[hash]) {
2745 if (++hash >= PTYPE_HASH_SIZE)
2747 nxt = ptype_base[hash].next;
2750 return list_entry(nxt, struct packet_type, list);
2753 static void ptype_seq_stop(struct seq_file *seq, void *v)
2759 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2761 #ifdef CONFIG_KALLSYMS
2762 unsigned long offset = 0, symsize;
2763 const char *symname;
2767 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2774 modname = delim = "";
2775 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2781 seq_printf(seq, "[%p]", sym);
2784 static int ptype_seq_show(struct seq_file *seq, void *v)
2786 struct packet_type *pt = v;
2788 if (v == SEQ_START_TOKEN)
2789 seq_puts(seq, "Type Device Function\n");
2790 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
2791 if (pt->type == htons(ETH_P_ALL))
2792 seq_puts(seq, "ALL ");
2794 seq_printf(seq, "%04x", ntohs(pt->type));
2796 seq_printf(seq, " %-8s ",
2797 pt->dev ? pt->dev->name : "");
2798 ptype_seq_decode(seq, pt->func);
2799 seq_putc(seq, '\n');
2805 static const struct seq_operations ptype_seq_ops = {
2806 .start = ptype_seq_start,
2807 .next = ptype_seq_next,
2808 .stop = ptype_seq_stop,
2809 .show = ptype_seq_show,
2812 static int ptype_seq_open(struct inode *inode, struct file *file)
2814 return seq_open_net(inode, file, &ptype_seq_ops,
2815 sizeof(struct seq_net_private));
2818 static const struct file_operations ptype_seq_fops = {
2819 .owner = THIS_MODULE,
2820 .open = ptype_seq_open,
2822 .llseek = seq_lseek,
2823 .release = seq_release_net,
2827 static int __net_init dev_proc_net_init(struct net *net)
2831 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2833 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2835 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2838 if (wext_proc_init(net))
2844 proc_net_remove(net, "ptype");
2846 proc_net_remove(net, "softnet_stat");
2848 proc_net_remove(net, "dev");
2852 static void __net_exit dev_proc_net_exit(struct net *net)
2854 wext_proc_exit(net);
2856 proc_net_remove(net, "ptype");
2857 proc_net_remove(net, "softnet_stat");
2858 proc_net_remove(net, "dev");
2861 static struct pernet_operations __net_initdata dev_proc_ops = {
2862 .init = dev_proc_net_init,
2863 .exit = dev_proc_net_exit,
2866 static int __init dev_proc_init(void)
2868 return register_pernet_subsys(&dev_proc_ops);
2871 #define dev_proc_init() 0
2872 #endif /* CONFIG_PROC_FS */
2876 * netdev_set_master - set up master/slave pair
2877 * @slave: slave device
2878 * @master: new master device
2880 * Changes the master device of the slave. Pass %NULL to break the
2881 * bonding. The caller must hold the RTNL semaphore. On a failure
2882 * a negative errno code is returned. On success the reference counts
2883 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2884 * function returns zero.
2886 int netdev_set_master(struct net_device *slave, struct net_device *master)
2888 struct net_device *old = slave->master;
2898 slave->master = master;
2906 slave->flags |= IFF_SLAVE;
2908 slave->flags &= ~IFF_SLAVE;
2910 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2914 static int __dev_set_promiscuity(struct net_device *dev, int inc)
2916 unsigned short old_flags = dev->flags;
2920 dev->flags |= IFF_PROMISC;
2921 dev->promiscuity += inc;
2922 if (dev->promiscuity == 0) {
2925 * If inc causes overflow, untouch promisc and return error.
2928 dev->flags &= ~IFF_PROMISC;
2930 dev->promiscuity -= inc;
2931 printk(KERN_WARNING "%s: promiscuity touches roof, "
2932 "set promiscuity failed, promiscuity feature "
2933 "of device might be broken.\n", dev->name);
2937 if (dev->flags != old_flags) {
2938 printk(KERN_INFO "device %s %s promiscuous mode\n",
2939 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2942 audit_log(current->audit_context, GFP_ATOMIC,
2943 AUDIT_ANOM_PROMISCUOUS,
2944 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2945 dev->name, (dev->flags & IFF_PROMISC),
2946 (old_flags & IFF_PROMISC),
2947 audit_get_loginuid(current),
2948 current->uid, current->gid,
2949 audit_get_sessionid(current));
2951 if (dev->change_rx_flags)
2952 dev->change_rx_flags(dev, IFF_PROMISC);
2958 * dev_set_promiscuity - update promiscuity count on a device
2962 * Add or remove promiscuity from a device. While the count in the device
2963 * remains above zero the interface remains promiscuous. Once it hits zero
2964 * the device reverts back to normal filtering operation. A negative inc
2965 * value is used to drop promiscuity on the device.
2966 * Return 0 if successful or a negative errno code on error.
2968 int dev_set_promiscuity(struct net_device *dev, int inc)
2970 unsigned short old_flags = dev->flags;
2973 err = __dev_set_promiscuity(dev, inc);
2976 if (dev->flags != old_flags)
2977 dev_set_rx_mode(dev);
2982 * dev_set_allmulti - update allmulti count on a device
2986 * Add or remove reception of all multicast frames to a device. While the
2987 * count in the device remains above zero the interface remains listening
2988 * to all interfaces. Once it hits zero the device reverts back to normal
2989 * filtering operation. A negative @inc value is used to drop the counter
2990 * when releasing a resource needing all multicasts.
2991 * Return 0 if successful or a negative errno code on error.
2994 int dev_set_allmulti(struct net_device *dev, int inc)
2996 unsigned short old_flags = dev->flags;
3000 dev->flags |= IFF_ALLMULTI;
3001 dev->allmulti += inc;
3002 if (dev->allmulti == 0) {
3005 * If inc causes overflow, untouch allmulti and return error.
3008 dev->flags &= ~IFF_ALLMULTI;
3010 dev->allmulti -= inc;
3011 printk(KERN_WARNING "%s: allmulti touches roof, "
3012 "set allmulti failed, allmulti feature of "
3013 "device might be broken.\n", dev->name);
3017 if (dev->flags ^ old_flags) {
3018 if (dev->change_rx_flags)
3019 dev->change_rx_flags(dev, IFF_ALLMULTI);
3020 dev_set_rx_mode(dev);
3026 * Upload unicast and multicast address lists to device and
3027 * configure RX filtering. When the device doesn't support unicast
3028 * filtering it is put in promiscuous mode while unicast addresses
3031 void __dev_set_rx_mode(struct net_device *dev)
3033 /* dev_open will call this function so the list will stay sane. */
3034 if (!(dev->flags&IFF_UP))
3037 if (!netif_device_present(dev))
3040 if (dev->set_rx_mode)
3041 dev->set_rx_mode(dev);
3043 /* Unicast addresses changes may only happen under the rtnl,
3044 * therefore calling __dev_set_promiscuity here is safe.
3046 if (dev->uc_count > 0 && !dev->uc_promisc) {
3047 __dev_set_promiscuity(dev, 1);
3048 dev->uc_promisc = 1;
3049 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3050 __dev_set_promiscuity(dev, -1);
3051 dev->uc_promisc = 0;
3054 if (dev->set_multicast_list)
3055 dev->set_multicast_list(dev);
3059 void dev_set_rx_mode(struct net_device *dev)
3061 netif_addr_lock_bh(dev);
3062 __dev_set_rx_mode(dev);
3063 netif_addr_unlock_bh(dev);
3066 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3067 void *addr, int alen, int glbl)
3069 struct dev_addr_list *da;
3071 for (; (da = *list) != NULL; list = &da->next) {
3072 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3073 alen == da->da_addrlen) {
3075 int old_glbl = da->da_gusers;
3092 int __dev_addr_add(struct dev_addr_list **list, int *count,
3093 void *addr, int alen, int glbl)
3095 struct dev_addr_list *da;
3097 for (da = *list; da != NULL; da = da->next) {
3098 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3099 da->da_addrlen == alen) {
3101 int old_glbl = da->da_gusers;
3111 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3114 memcpy(da->da_addr, addr, alen);
3115 da->da_addrlen = alen;
3117 da->da_gusers = glbl ? 1 : 0;
3125 * dev_unicast_delete - Release secondary unicast address.
3127 * @addr: address to delete
3128 * @alen: length of @addr
3130 * Release reference to a secondary unicast address and remove it
3131 * from the device if the reference count drops to zero.
3133 * The caller must hold the rtnl_mutex.
3135 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3141 netif_addr_lock_bh(dev);
3142 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3144 __dev_set_rx_mode(dev);
3145 netif_addr_unlock_bh(dev);
3148 EXPORT_SYMBOL(dev_unicast_delete);
3151 * dev_unicast_add - add a secondary unicast address
3153 * @addr: address to add
3154 * @alen: length of @addr
3156 * Add a secondary unicast address to the device or increase
3157 * the reference count if it already exists.
3159 * The caller must hold the rtnl_mutex.
3161 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3167 netif_addr_lock_bh(dev);
3168 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3170 __dev_set_rx_mode(dev);
3171 netif_addr_unlock_bh(dev);
3174 EXPORT_SYMBOL(dev_unicast_add);
3176 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3177 struct dev_addr_list **from, int *from_count)
3179 struct dev_addr_list *da, *next;
3183 while (da != NULL) {
3185 if (!da->da_synced) {
3186 err = __dev_addr_add(to, to_count,
3187 da->da_addr, da->da_addrlen, 0);
3192 } else if (da->da_users == 1) {
3193 __dev_addr_delete(to, to_count,
3194 da->da_addr, da->da_addrlen, 0);
3195 __dev_addr_delete(from, from_count,
3196 da->da_addr, da->da_addrlen, 0);
3203 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3204 struct dev_addr_list **from, int *from_count)
3206 struct dev_addr_list *da, *next;
3209 while (da != NULL) {
3211 if (da->da_synced) {
3212 __dev_addr_delete(to, to_count,
3213 da->da_addr, da->da_addrlen, 0);
3215 __dev_addr_delete(from, from_count,
3216 da->da_addr, da->da_addrlen, 0);
3223 * dev_unicast_sync - Synchronize device's unicast list to another device
3224 * @to: destination device
3225 * @from: source device
3227 * Add newly added addresses to the destination device and release
3228 * addresses that have no users left. The source device must be
3229 * locked by netif_tx_lock_bh.
3231 * This function is intended to be called from the dev->set_rx_mode
3232 * function of layered software devices.
3234 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3238 netif_addr_lock_bh(to);
3239 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3240 &from->uc_list, &from->uc_count);
3242 __dev_set_rx_mode(to);
3243 netif_addr_unlock_bh(to);
3246 EXPORT_SYMBOL(dev_unicast_sync);
3249 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3250 * @to: destination device
3251 * @from: source device
3253 * Remove all addresses that were added to the destination device by
3254 * dev_unicast_sync(). This function is intended to be called from the
3255 * dev->stop function of layered software devices.
3257 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3259 netif_addr_lock_bh(from);
3260 netif_addr_lock(to);
3262 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3263 &from->uc_list, &from->uc_count);
3264 __dev_set_rx_mode(to);
3266 netif_addr_unlock(to);
3267 netif_addr_unlock_bh(from);
3269 EXPORT_SYMBOL(dev_unicast_unsync);
3271 static void __dev_addr_discard(struct dev_addr_list **list)
3273 struct dev_addr_list *tmp;
3275 while (*list != NULL) {
3278 if (tmp->da_users > tmp->da_gusers)
3279 printk("__dev_addr_discard: address leakage! "
3280 "da_users=%d\n", tmp->da_users);
3285 static void dev_addr_discard(struct net_device *dev)
3287 netif_addr_lock_bh(dev);
3289 __dev_addr_discard(&dev->uc_list);
3292 __dev_addr_discard(&dev->mc_list);
3295 netif_addr_unlock_bh(dev);
3298 unsigned dev_get_flags(const struct net_device *dev)
3302 flags = (dev->flags & ~(IFF_PROMISC |
3307 (dev->gflags & (IFF_PROMISC |
3310 if (netif_running(dev)) {
3311 if (netif_oper_up(dev))
3312 flags |= IFF_RUNNING;
3313 if (netif_carrier_ok(dev))
3314 flags |= IFF_LOWER_UP;
3315 if (netif_dormant(dev))
3316 flags |= IFF_DORMANT;
3322 int dev_change_flags(struct net_device *dev, unsigned flags)
3325 int old_flags = dev->flags;
3330 * Set the flags on our device.
3333 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3334 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3336 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3340 * Load in the correct multicast list now the flags have changed.
3343 if (dev->change_rx_flags && (old_flags ^ flags) & IFF_MULTICAST)
3344 dev->change_rx_flags(dev, IFF_MULTICAST);
3346 dev_set_rx_mode(dev);
3349 * Have we downed the interface. We handle IFF_UP ourselves
3350 * according to user attempts to set it, rather than blindly
3355 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3356 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3359 dev_set_rx_mode(dev);
3362 if (dev->flags & IFF_UP &&
3363 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3365 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3367 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3368 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3369 dev->gflags ^= IFF_PROMISC;
3370 dev_set_promiscuity(dev, inc);
3373 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3374 is important. Some (broken) drivers set IFF_PROMISC, when
3375 IFF_ALLMULTI is requested not asking us and not reporting.
3377 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3378 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3379 dev->gflags ^= IFF_ALLMULTI;
3380 dev_set_allmulti(dev, inc);
3383 /* Exclude state transition flags, already notified */
3384 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3386 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3391 int dev_set_mtu(struct net_device *dev, int new_mtu)
3395 if (new_mtu == dev->mtu)
3398 /* MTU must be positive. */
3402 if (!netif_device_present(dev))
3406 if (dev->change_mtu)
3407 err = dev->change_mtu(dev, new_mtu);
3410 if (!err && dev->flags & IFF_UP)
3411 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3415 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3419 if (!dev->set_mac_address)
3421 if (sa->sa_family != dev->type)
3423 if (!netif_device_present(dev))
3425 err = dev->set_mac_address(dev, sa);
3427 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3432 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3434 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3437 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3443 case SIOCGIFFLAGS: /* Get interface flags */
3444 ifr->ifr_flags = dev_get_flags(dev);
3447 case SIOCGIFMETRIC: /* Get the metric on the interface
3448 (currently unused) */
3449 ifr->ifr_metric = 0;
3452 case SIOCGIFMTU: /* Get the MTU of a device */
3453 ifr->ifr_mtu = dev->mtu;
3458 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3460 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3461 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3462 ifr->ifr_hwaddr.sa_family = dev->type;
3470 ifr->ifr_map.mem_start = dev->mem_start;
3471 ifr->ifr_map.mem_end = dev->mem_end;
3472 ifr->ifr_map.base_addr = dev->base_addr;
3473 ifr->ifr_map.irq = dev->irq;
3474 ifr->ifr_map.dma = dev->dma;
3475 ifr->ifr_map.port = dev->if_port;
3479 ifr->ifr_ifindex = dev->ifindex;
3483 ifr->ifr_qlen = dev->tx_queue_len;
3487 /* dev_ioctl() should ensure this case
3499 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3501 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3504 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3510 case SIOCSIFFLAGS: /* Set interface flags */
3511 return dev_change_flags(dev, ifr->ifr_flags);
3513 case SIOCSIFMETRIC: /* Set the metric on the interface
3514 (currently unused) */
3517 case SIOCSIFMTU: /* Set the MTU of a device */
3518 return dev_set_mtu(dev, ifr->ifr_mtu);
3521 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3523 case SIOCSIFHWBROADCAST:
3524 if (ifr->ifr_hwaddr.sa_family != dev->type)
3526 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3527 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3528 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3532 if (dev->set_config) {
3533 if (!netif_device_present(dev))
3535 return dev->set_config(dev, &ifr->ifr_map);
3540 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3541 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3543 if (!netif_device_present(dev))
3545 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3549 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3550 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3552 if (!netif_device_present(dev))
3554 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3558 if (ifr->ifr_qlen < 0)
3560 dev->tx_queue_len = ifr->ifr_qlen;
3564 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3565 return dev_change_name(dev, ifr->ifr_newname);
3568 * Unknown or private ioctl
3572 if ((cmd >= SIOCDEVPRIVATE &&
3573 cmd <= SIOCDEVPRIVATE + 15) ||
3574 cmd == SIOCBONDENSLAVE ||
3575 cmd == SIOCBONDRELEASE ||
3576 cmd == SIOCBONDSETHWADDR ||
3577 cmd == SIOCBONDSLAVEINFOQUERY ||
3578 cmd == SIOCBONDINFOQUERY ||
3579 cmd == SIOCBONDCHANGEACTIVE ||
3580 cmd == SIOCGMIIPHY ||
3581 cmd == SIOCGMIIREG ||
3582 cmd == SIOCSMIIREG ||
3583 cmd == SIOCBRADDIF ||
3584 cmd == SIOCBRDELIF ||
3585 cmd == SIOCWANDEV) {
3587 if (dev->do_ioctl) {
3588 if (netif_device_present(dev))
3589 err = dev->do_ioctl(dev, ifr,
3602 * This function handles all "interface"-type I/O control requests. The actual
3603 * 'doing' part of this is dev_ifsioc above.
3607 * dev_ioctl - network device ioctl
3608 * @net: the applicable net namespace
3609 * @cmd: command to issue
3610 * @arg: pointer to a struct ifreq in user space
3612 * Issue ioctl functions to devices. This is normally called by the
3613 * user space syscall interfaces but can sometimes be useful for
3614 * other purposes. The return value is the return from the syscall if
3615 * positive or a negative errno code on error.
3618 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3624 /* One special case: SIOCGIFCONF takes ifconf argument
3625 and requires shared lock, because it sleeps writing
3629 if (cmd == SIOCGIFCONF) {
3631 ret = dev_ifconf(net, (char __user *) arg);
3635 if (cmd == SIOCGIFNAME)
3636 return dev_ifname(net, (struct ifreq __user *)arg);
3638 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3641 ifr.ifr_name[IFNAMSIZ-1] = 0;
3643 colon = strchr(ifr.ifr_name, ':');
3648 * See which interface the caller is talking about.
3653 * These ioctl calls:
3654 * - can be done by all.
3655 * - atomic and do not require locking.
3666 dev_load(net, ifr.ifr_name);
3667 read_lock(&dev_base_lock);
3668 ret = dev_ifsioc_locked(net, &ifr, cmd);
3669 read_unlock(&dev_base_lock);
3673 if (copy_to_user(arg, &ifr,
3674 sizeof(struct ifreq)))
3680 dev_load(net, ifr.ifr_name);
3682 ret = dev_ethtool(net, &ifr);
3687 if (copy_to_user(arg, &ifr,
3688 sizeof(struct ifreq)))
3694 * These ioctl calls:
3695 * - require superuser power.
3696 * - require strict serialization.
3702 if (!capable(CAP_NET_ADMIN))
3704 dev_load(net, ifr.ifr_name);
3706 ret = dev_ifsioc(net, &ifr, cmd);
3711 if (copy_to_user(arg, &ifr,
3712 sizeof(struct ifreq)))
3718 * These ioctl calls:
3719 * - require superuser power.
3720 * - require strict serialization.
3721 * - do not return a value
3731 case SIOCSIFHWBROADCAST:
3734 case SIOCBONDENSLAVE:
3735 case SIOCBONDRELEASE:
3736 case SIOCBONDSETHWADDR:
3737 case SIOCBONDCHANGEACTIVE:
3740 if (!capable(CAP_NET_ADMIN))
3743 case SIOCBONDSLAVEINFOQUERY:
3744 case SIOCBONDINFOQUERY:
3745 dev_load(net, ifr.ifr_name);
3747 ret = dev_ifsioc(net, &ifr, cmd);
3752 /* Get the per device memory space. We can add this but
3753 * currently do not support it */
3755 /* Set the per device memory buffer space.
3756 * Not applicable in our case */
3761 * Unknown or private ioctl.
3764 if (cmd == SIOCWANDEV ||
3765 (cmd >= SIOCDEVPRIVATE &&
3766 cmd <= SIOCDEVPRIVATE + 15)) {
3767 dev_load(net, ifr.ifr_name);
3769 ret = dev_ifsioc(net, &ifr, cmd);
3771 if (!ret && copy_to_user(arg, &ifr,
3772 sizeof(struct ifreq)))
3776 /* Take care of Wireless Extensions */
3777 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3778 return wext_handle_ioctl(net, &ifr, cmd, arg);
3785 * dev_new_index - allocate an ifindex
3786 * @net: the applicable net namespace
3788 * Returns a suitable unique value for a new device interface
3789 * number. The caller must hold the rtnl semaphore or the
3790 * dev_base_lock to be sure it remains unique.
3792 static int dev_new_index(struct net *net)
3798 if (!__dev_get_by_index(net, ifindex))
3803 /* Delayed registration/unregisteration */
3804 static DEFINE_SPINLOCK(net_todo_list_lock);
3805 static LIST_HEAD(net_todo_list);
3807 static void net_set_todo(struct net_device *dev)
3809 spin_lock(&net_todo_list_lock);
3810 list_add_tail(&dev->todo_list, &net_todo_list);
3811 spin_unlock(&net_todo_list_lock);
3814 static void rollback_registered(struct net_device *dev)
3816 BUG_ON(dev_boot_phase);
3819 /* Some devices call without registering for initialization unwind. */
3820 if (dev->reg_state == NETREG_UNINITIALIZED) {
3821 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3822 "was registered\n", dev->name, dev);
3828 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3830 /* If device is running, close it first. */
3833 /* And unlink it from device chain. */
3834 unlist_netdevice(dev);
3836 dev->reg_state = NETREG_UNREGISTERING;
3840 /* Shutdown queueing discipline. */
3844 /* Notify protocols, that we are about to destroy
3845 this device. They should clean all the things.
3847 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3850 * Flush the unicast and multicast chains
3852 dev_addr_discard(dev);
3857 /* Notifier chain MUST detach us from master device. */
3858 WARN_ON(dev->master);
3860 /* Remove entries from kobject tree */
3861 netdev_unregister_kobject(dev);
3868 static void __netdev_init_queue_locks_one(struct net_device *dev,
3869 struct netdev_queue *dev_queue,
3872 spin_lock_init(&dev_queue->_xmit_lock);
3873 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
3874 dev_queue->xmit_lock_owner = -1;
3877 static void netdev_init_queue_locks(struct net_device *dev)
3879 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
3880 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
3884 * register_netdevice - register a network device
3885 * @dev: device to register
3887 * Take a completed network device structure and add it to the kernel
3888 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3889 * chain. 0 is returned on success. A negative errno code is returned
3890 * on a failure to set up the device, or if the name is a duplicate.
3892 * Callers must hold the rtnl semaphore. You may want
3893 * register_netdev() instead of this.
3896 * The locking appears insufficient to guarantee two parallel registers
3897 * will not get the same name.
3900 int register_netdevice(struct net_device *dev)
3902 struct hlist_head *head;
3903 struct hlist_node *p;
3907 BUG_ON(dev_boot_phase);
3912 /* When net_device's are persistent, this will be fatal. */
3913 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3914 BUG_ON(!dev_net(dev));
3917 spin_lock_init(&dev->addr_list_lock);
3918 netdev_set_addr_lockdep_class(dev);
3919 netdev_init_queue_locks(dev);
3923 /* Init, if this function is available */
3925 ret = dev->init(dev);
3933 if (!dev_valid_name(dev->name)) {
3938 dev->ifindex = dev_new_index(net);
3939 if (dev->iflink == -1)
3940 dev->iflink = dev->ifindex;
3942 /* Check for existence of name */
3943 head = dev_name_hash(net, dev->name);
3944 hlist_for_each(p, head) {
3945 struct net_device *d
3946 = hlist_entry(p, struct net_device, name_hlist);
3947 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3953 /* Fix illegal checksum combinations */
3954 if ((dev->features & NETIF_F_HW_CSUM) &&
3955 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3956 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3958 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3961 if ((dev->features & NETIF_F_NO_CSUM) &&
3962 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3963 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3965 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3969 /* Fix illegal SG+CSUM combinations. */
3970 if ((dev->features & NETIF_F_SG) &&
3971 !(dev->features & NETIF_F_ALL_CSUM)) {
3972 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3974 dev->features &= ~NETIF_F_SG;
3977 /* TSO requires that SG is present as well. */
3978 if ((dev->features & NETIF_F_TSO) &&
3979 !(dev->features & NETIF_F_SG)) {
3980 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3982 dev->features &= ~NETIF_F_TSO;
3984 if (dev->features & NETIF_F_UFO) {
3985 if (!(dev->features & NETIF_F_HW_CSUM)) {
3986 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3987 "NETIF_F_HW_CSUM feature.\n",
3989 dev->features &= ~NETIF_F_UFO;
3991 if (!(dev->features & NETIF_F_SG)) {
3992 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3993 "NETIF_F_SG feature.\n",
3995 dev->features &= ~NETIF_F_UFO;
3999 /* Enable software GSO if SG is supported. */
4000 if (dev->features & NETIF_F_SG)
4001 dev->features |= NETIF_F_GSO;
4003 netdev_initialize_kobject(dev);
4004 ret = netdev_register_kobject(dev);
4007 dev->reg_state = NETREG_REGISTERED;
4010 * Default initial state at registry is that the
4011 * device is present.
4014 set_bit(__LINK_STATE_PRESENT, &dev->state);
4016 dev_init_scheduler(dev);
4018 list_netdevice(dev);
4020 /* Notify protocols, that a new device appeared. */
4021 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4022 ret = notifier_to_errno(ret);
4024 rollback_registered(dev);
4025 dev->reg_state = NETREG_UNREGISTERED;
4038 * register_netdev - register a network device
4039 * @dev: device to register
4041 * Take a completed network device structure and add it to the kernel
4042 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4043 * chain. 0 is returned on success. A negative errno code is returned
4044 * on a failure to set up the device, or if the name is a duplicate.
4046 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4047 * and expands the device name if you passed a format string to
4050 int register_netdev(struct net_device *dev)
4057 * If the name is a format string the caller wants us to do a
4060 if (strchr(dev->name, '%')) {
4061 err = dev_alloc_name(dev, dev->name);
4066 err = register_netdevice(dev);
4071 EXPORT_SYMBOL(register_netdev);
4074 * netdev_wait_allrefs - wait until all references are gone.
4076 * This is called when unregistering network devices.
4078 * Any protocol or device that holds a reference should register
4079 * for netdevice notification, and cleanup and put back the
4080 * reference if they receive an UNREGISTER event.
4081 * We can get stuck here if buggy protocols don't correctly
4084 static void netdev_wait_allrefs(struct net_device *dev)
4086 unsigned long rebroadcast_time, warning_time;
4088 rebroadcast_time = warning_time = jiffies;
4089 while (atomic_read(&dev->refcnt) != 0) {
4090 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4093 /* Rebroadcast unregister notification */
4094 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4096 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4098 /* We must not have linkwatch events
4099 * pending on unregister. If this
4100 * happens, we simply run the queue
4101 * unscheduled, resulting in a noop
4104 linkwatch_run_queue();
4109 rebroadcast_time = jiffies;
4114 if (time_after(jiffies, warning_time + 10 * HZ)) {
4115 printk(KERN_EMERG "unregister_netdevice: "
4116 "waiting for %s to become free. Usage "
4118 dev->name, atomic_read(&dev->refcnt));
4119 warning_time = jiffies;
4128 * register_netdevice(x1);
4129 * register_netdevice(x2);
4131 * unregister_netdevice(y1);
4132 * unregister_netdevice(y2);
4138 * We are invoked by rtnl_unlock() after it drops the semaphore.
4139 * This allows us to deal with problems:
4140 * 1) We can delete sysfs objects which invoke hotplug
4141 * without deadlocking with linkwatch via keventd.
4142 * 2) Since we run with the RTNL semaphore not held, we can sleep
4143 * safely in order to wait for the netdev refcnt to drop to zero.
4145 static DEFINE_MUTEX(net_todo_run_mutex);
4146 void netdev_run_todo(void)
4148 struct list_head list;
4150 /* Need to guard against multiple cpu's getting out of order. */
4151 mutex_lock(&net_todo_run_mutex);
4153 /* Not safe to do outside the semaphore. We must not return
4154 * until all unregister events invoked by the local processor
4155 * have been completed (either by this todo run, or one on
4158 if (list_empty(&net_todo_list))
4161 /* Snapshot list, allow later requests */
4162 spin_lock(&net_todo_list_lock);
4163 list_replace_init(&net_todo_list, &list);
4164 spin_unlock(&net_todo_list_lock);
4166 while (!list_empty(&list)) {
4167 struct net_device *dev
4168 = list_entry(list.next, struct net_device, todo_list);
4169 list_del(&dev->todo_list);
4171 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4172 printk(KERN_ERR "network todo '%s' but state %d\n",
4173 dev->name, dev->reg_state);
4178 dev->reg_state = NETREG_UNREGISTERED;
4180 on_each_cpu(flush_backlog, dev, 1);
4182 netdev_wait_allrefs(dev);
4185 BUG_ON(atomic_read(&dev->refcnt));
4186 WARN_ON(dev->ip_ptr);
4187 WARN_ON(dev->ip6_ptr);
4188 WARN_ON(dev->dn_ptr);
4190 if (dev->destructor)
4191 dev->destructor(dev);
4193 /* Free network device */
4194 kobject_put(&dev->dev.kobj);
4198 mutex_unlock(&net_todo_run_mutex);
4201 static struct net_device_stats *internal_stats(struct net_device *dev)
4206 static void netdev_init_one_queue(struct net_device *dev,
4207 struct netdev_queue *queue,
4213 static void netdev_init_queues(struct net_device *dev)
4215 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4216 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4217 spin_lock_init(&dev->tx_global_lock);
4221 * alloc_netdev_mq - allocate network device
4222 * @sizeof_priv: size of private data to allocate space for
4223 * @name: device name format string
4224 * @setup: callback to initialize device
4225 * @queue_count: the number of subqueues to allocate
4227 * Allocates a struct net_device with private data area for driver use
4228 * and performs basic initialization. Also allocates subquue structs
4229 * for each queue on the device at the end of the netdevice.
4231 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4232 void (*setup)(struct net_device *), unsigned int queue_count)
4234 struct netdev_queue *tx;
4235 struct net_device *dev;
4239 BUG_ON(strlen(name) >= sizeof(dev->name));
4241 alloc_size = sizeof(struct net_device);
4243 /* ensure 32-byte alignment of private area */
4244 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4245 alloc_size += sizeof_priv;
4247 /* ensure 32-byte alignment of whole construct */
4248 alloc_size += NETDEV_ALIGN_CONST;
4250 p = kzalloc(alloc_size, GFP_KERNEL);
4252 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4256 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
4258 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4264 dev = (struct net_device *)
4265 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4266 dev->padded = (char *)dev - (char *)p;
4267 dev_net_set(dev, &init_net);
4270 dev->num_tx_queues = queue_count;
4271 dev->real_num_tx_queues = queue_count;
4274 dev->priv = ((char *)dev +
4275 ((sizeof(struct net_device) + NETDEV_ALIGN_CONST)
4276 & ~NETDEV_ALIGN_CONST));
4279 dev->gso_max_size = GSO_MAX_SIZE;
4281 netdev_init_queues(dev);
4283 dev->get_stats = internal_stats;
4284 netpoll_netdev_init(dev);
4286 strcpy(dev->name, name);
4289 EXPORT_SYMBOL(alloc_netdev_mq);
4292 * free_netdev - free network device
4295 * This function does the last stage of destroying an allocated device
4296 * interface. The reference to the device object is released.
4297 * If this is the last reference then it will be freed.
4299 void free_netdev(struct net_device *dev)
4301 release_net(dev_net(dev));
4305 /* Compatibility with error handling in drivers */
4306 if (dev->reg_state == NETREG_UNINITIALIZED) {
4307 kfree((char *)dev - dev->padded);
4311 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4312 dev->reg_state = NETREG_RELEASED;
4314 /* will free via device release */
4315 put_device(&dev->dev);
4318 /* Synchronize with packet receive processing. */
4319 void synchronize_net(void)
4326 * unregister_netdevice - remove device from the kernel
4329 * This function shuts down a device interface and removes it
4330 * from the kernel tables.
4332 * Callers must hold the rtnl semaphore. You may want
4333 * unregister_netdev() instead of this.
4336 void unregister_netdevice(struct net_device *dev)
4340 rollback_registered(dev);
4341 /* Finish processing unregister after unlock */
4346 * unregister_netdev - remove device from the kernel
4349 * This function shuts down a device interface and removes it
4350 * from the kernel tables.
4352 * This is just a wrapper for unregister_netdevice that takes
4353 * the rtnl semaphore. In general you want to use this and not
4354 * unregister_netdevice.
4356 void unregister_netdev(struct net_device *dev)
4359 unregister_netdevice(dev);
4363 EXPORT_SYMBOL(unregister_netdev);
4366 * dev_change_net_namespace - move device to different nethost namespace
4368 * @net: network namespace
4369 * @pat: If not NULL name pattern to try if the current device name
4370 * is already taken in the destination network namespace.
4372 * This function shuts down a device interface and moves it
4373 * to a new network namespace. On success 0 is returned, on
4374 * a failure a netagive errno code is returned.
4376 * Callers must hold the rtnl semaphore.
4379 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4382 const char *destname;
4387 /* Don't allow namespace local devices to be moved. */
4389 if (dev->features & NETIF_F_NETNS_LOCAL)
4392 /* Ensure the device has been registrered */
4394 if (dev->reg_state != NETREG_REGISTERED)
4397 /* Get out if there is nothing todo */
4399 if (net_eq(dev_net(dev), net))
4402 /* Pick the destination device name, and ensure
4403 * we can use it in the destination network namespace.
4406 destname = dev->name;
4407 if (__dev_get_by_name(net, destname)) {
4408 /* We get here if we can't use the current device name */
4411 if (!dev_valid_name(pat))
4413 if (strchr(pat, '%')) {
4414 if (__dev_alloc_name(net, pat, buf) < 0)
4419 if (__dev_get_by_name(net, destname))
4424 * And now a mini version of register_netdevice unregister_netdevice.
4427 /* If device is running close it first. */
4430 /* And unlink it from device chain */
4432 unlist_netdevice(dev);
4436 /* Shutdown queueing discipline. */
4439 /* Notify protocols, that we are about to destroy
4440 this device. They should clean all the things.
4442 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4445 * Flush the unicast and multicast chains
4447 dev_addr_discard(dev);
4449 /* Actually switch the network namespace */
4450 dev_net_set(dev, net);
4452 /* Assign the new device name */
4453 if (destname != dev->name)
4454 strcpy(dev->name, destname);
4456 /* If there is an ifindex conflict assign a new one */
4457 if (__dev_get_by_index(net, dev->ifindex)) {
4458 int iflink = (dev->iflink == dev->ifindex);
4459 dev->ifindex = dev_new_index(net);
4461 dev->iflink = dev->ifindex;
4464 /* Fixup kobjects */
4465 netdev_unregister_kobject(dev);
4466 err = netdev_register_kobject(dev);
4469 /* Add the device back in the hashes */
4470 list_netdevice(dev);
4472 /* Notify protocols, that a new device appeared. */
4473 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4481 static int dev_cpu_callback(struct notifier_block *nfb,
4482 unsigned long action,
4485 struct sk_buff **list_skb;
4486 struct Qdisc **list_net;
4487 struct sk_buff *skb;
4488 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4489 struct softnet_data *sd, *oldsd;
4491 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4494 local_irq_disable();
4495 cpu = smp_processor_id();
4496 sd = &per_cpu(softnet_data, cpu);
4497 oldsd = &per_cpu(softnet_data, oldcpu);
4499 /* Find end of our completion_queue. */
4500 list_skb = &sd->completion_queue;
4502 list_skb = &(*list_skb)->next;
4503 /* Append completion queue from offline CPU. */
4504 *list_skb = oldsd->completion_queue;
4505 oldsd->completion_queue = NULL;
4507 /* Find end of our output_queue. */
4508 list_net = &sd->output_queue;
4510 list_net = &(*list_net)->next_sched;
4511 /* Append output queue from offline CPU. */
4512 *list_net = oldsd->output_queue;
4513 oldsd->output_queue = NULL;
4515 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4518 /* Process offline CPU's input_pkt_queue */
4519 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4525 #ifdef CONFIG_NET_DMA
4527 * net_dma_rebalance - try to maintain one DMA channel per CPU
4528 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4530 * This is called when the number of channels allocated to the net_dma client
4531 * changes. The net_dma client tries to have one DMA channel per CPU.
4534 static void net_dma_rebalance(struct net_dma *net_dma)
4536 unsigned int cpu, i, n, chan_idx;
4537 struct dma_chan *chan;
4539 if (cpus_empty(net_dma->channel_mask)) {
4540 for_each_online_cpu(cpu)
4541 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4546 cpu = first_cpu(cpu_online_map);
4548 for_each_cpu_mask_nr(chan_idx, net_dma->channel_mask) {
4549 chan = net_dma->channels[chan_idx];
4551 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4552 + (i < (num_online_cpus() %
4553 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4556 per_cpu(softnet_data, cpu).net_dma = chan;
4557 cpu = next_cpu(cpu, cpu_online_map);
4565 * netdev_dma_event - event callback for the net_dma_client
4566 * @client: should always be net_dma_client
4567 * @chan: DMA channel for the event
4568 * @state: DMA state to be handled
4570 static enum dma_state_client
4571 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4572 enum dma_state state)
4574 int i, found = 0, pos = -1;
4575 struct net_dma *net_dma =
4576 container_of(client, struct net_dma, client);
4577 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4579 spin_lock(&net_dma->lock);
4581 case DMA_RESOURCE_AVAILABLE:
4582 for (i = 0; i < nr_cpu_ids; i++)
4583 if (net_dma->channels[i] == chan) {
4586 } else if (net_dma->channels[i] == NULL && pos < 0)
4589 if (!found && pos >= 0) {
4591 net_dma->channels[pos] = chan;
4592 cpu_set(pos, net_dma->channel_mask);
4593 net_dma_rebalance(net_dma);
4596 case DMA_RESOURCE_REMOVED:
4597 for (i = 0; i < nr_cpu_ids; i++)
4598 if (net_dma->channels[i] == chan) {
4606 cpu_clear(pos, net_dma->channel_mask);
4607 net_dma->channels[i] = NULL;
4608 net_dma_rebalance(net_dma);
4614 spin_unlock(&net_dma->lock);
4620 * netdev_dma_regiser - register the networking subsystem as a DMA client
4622 static int __init netdev_dma_register(void)
4624 net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma),
4626 if (unlikely(!net_dma.channels)) {
4628 "netdev_dma: no memory for net_dma.channels\n");
4631 spin_lock_init(&net_dma.lock);
4632 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4633 dma_async_client_register(&net_dma.client);
4634 dma_async_client_chan_request(&net_dma.client);
4639 static int __init netdev_dma_register(void) { return -ENODEV; }
4640 #endif /* CONFIG_NET_DMA */
4643 * netdev_compute_feature - compute conjunction of two feature sets
4644 * @all: first feature set
4645 * @one: second feature set
4647 * Computes a new feature set after adding a device with feature set
4648 * @one to the master device with current feature set @all. Returns
4649 * the new feature set.
4651 int netdev_compute_features(unsigned long all, unsigned long one)
4653 /* if device needs checksumming, downgrade to hw checksumming */
4654 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4655 all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
4657 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4658 if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
4659 all ^= NETIF_F_HW_CSUM
4660 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4662 if (one & NETIF_F_GSO)
4663 one |= NETIF_F_GSO_SOFTWARE;
4666 /* If even one device supports robust GSO, enable it for all. */
4667 if (one & NETIF_F_GSO_ROBUST)
4668 all |= NETIF_F_GSO_ROBUST;
4670 all &= one | NETIF_F_LLTX;
4672 if (!(all & NETIF_F_ALL_CSUM))
4674 if (!(all & NETIF_F_SG))
4675 all &= ~NETIF_F_GSO_MASK;
4679 EXPORT_SYMBOL(netdev_compute_features);
4681 static struct hlist_head *netdev_create_hash(void)
4684 struct hlist_head *hash;
4686 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4688 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4689 INIT_HLIST_HEAD(&hash[i]);
4694 /* Initialize per network namespace state */
4695 static int __net_init netdev_init(struct net *net)
4697 INIT_LIST_HEAD(&net->dev_base_head);
4699 net->dev_name_head = netdev_create_hash();
4700 if (net->dev_name_head == NULL)
4703 net->dev_index_head = netdev_create_hash();
4704 if (net->dev_index_head == NULL)
4710 kfree(net->dev_name_head);
4715 char *netdev_drivername(struct net_device *dev, char *buffer, int len)
4717 struct device_driver *driver;
4718 struct device *parent;
4720 if (len <= 0 || !buffer)
4724 parent = dev->dev.parent;
4729 driver = parent->driver;
4730 if (driver && driver->name)
4731 strlcpy(buffer, driver->name, len);
4735 static void __net_exit netdev_exit(struct net *net)
4737 kfree(net->dev_name_head);
4738 kfree(net->dev_index_head);
4741 static struct pernet_operations __net_initdata netdev_net_ops = {
4742 .init = netdev_init,
4743 .exit = netdev_exit,
4746 static void __net_exit default_device_exit(struct net *net)
4748 struct net_device *dev, *next;
4750 * Push all migratable of the network devices back to the
4751 * initial network namespace
4754 for_each_netdev_safe(net, dev, next) {
4756 char fb_name[IFNAMSIZ];
4758 /* Ignore unmoveable devices (i.e. loopback) */
4759 if (dev->features & NETIF_F_NETNS_LOCAL)
4762 /* Push remaing network devices to init_net */
4763 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
4764 err = dev_change_net_namespace(dev, &init_net, fb_name);
4766 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
4767 __func__, dev->name, err);
4774 static struct pernet_operations __net_initdata default_device_ops = {
4775 .exit = default_device_exit,
4779 * Initialize the DEV module. At boot time this walks the device list and
4780 * unhooks any devices that fail to initialise (normally hardware not
4781 * present) and leaves us with a valid list of present and active devices.
4786 * This is called single threaded during boot, so no need
4787 * to take the rtnl semaphore.
4789 static int __init net_dev_init(void)
4791 int i, rc = -ENOMEM;
4793 BUG_ON(!dev_boot_phase);
4795 if (dev_proc_init())
4798 if (netdev_kobject_init())
4801 INIT_LIST_HEAD(&ptype_all);
4802 for (i = 0; i < PTYPE_HASH_SIZE; i++)
4803 INIT_LIST_HEAD(&ptype_base[i]);
4805 if (register_pernet_subsys(&netdev_net_ops))
4808 if (register_pernet_device(&default_device_ops))
4812 * Initialise the packet receive queues.
4815 for_each_possible_cpu(i) {
4816 struct softnet_data *queue;
4818 queue = &per_cpu(softnet_data, i);
4819 skb_queue_head_init(&queue->input_pkt_queue);
4820 queue->completion_queue = NULL;
4821 INIT_LIST_HEAD(&queue->poll_list);
4823 queue->backlog.poll = process_backlog;
4824 queue->backlog.weight = weight_p;
4827 netdev_dma_register();
4831 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
4832 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
4834 hotcpu_notifier(dev_cpu_callback, 0);
4842 subsys_initcall(net_dev_init);
4844 EXPORT_SYMBOL(__dev_get_by_index);
4845 EXPORT_SYMBOL(__dev_get_by_name);
4846 EXPORT_SYMBOL(__dev_remove_pack);
4847 EXPORT_SYMBOL(dev_valid_name);
4848 EXPORT_SYMBOL(dev_add_pack);
4849 EXPORT_SYMBOL(dev_alloc_name);
4850 EXPORT_SYMBOL(dev_close);
4851 EXPORT_SYMBOL(dev_get_by_flags);
4852 EXPORT_SYMBOL(dev_get_by_index);
4853 EXPORT_SYMBOL(dev_get_by_name);
4854 EXPORT_SYMBOL(dev_open);
4855 EXPORT_SYMBOL(dev_queue_xmit);
4856 EXPORT_SYMBOL(dev_remove_pack);
4857 EXPORT_SYMBOL(dev_set_allmulti);
4858 EXPORT_SYMBOL(dev_set_promiscuity);
4859 EXPORT_SYMBOL(dev_change_flags);
4860 EXPORT_SYMBOL(dev_set_mtu);
4861 EXPORT_SYMBOL(dev_set_mac_address);
4862 EXPORT_SYMBOL(free_netdev);
4863 EXPORT_SYMBOL(netdev_boot_setup_check);
4864 EXPORT_SYMBOL(netdev_set_master);
4865 EXPORT_SYMBOL(netdev_state_change);
4866 EXPORT_SYMBOL(netif_receive_skb);
4867 EXPORT_SYMBOL(netif_rx);
4868 EXPORT_SYMBOL(register_gifconf);
4869 EXPORT_SYMBOL(register_netdevice);
4870 EXPORT_SYMBOL(register_netdevice_notifier);
4871 EXPORT_SYMBOL(skb_checksum_help);
4872 EXPORT_SYMBOL(synchronize_net);
4873 EXPORT_SYMBOL(unregister_netdevice);
4874 EXPORT_SYMBOL(unregister_netdevice_notifier);
4875 EXPORT_SYMBOL(net_enable_timestamp);
4876 EXPORT_SYMBOL(net_disable_timestamp);
4877 EXPORT_SYMBOL(dev_get_flags);
4879 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4880 EXPORT_SYMBOL(br_handle_frame_hook);
4881 EXPORT_SYMBOL(br_fdb_get_hook);
4882 EXPORT_SYMBOL(br_fdb_put_hook);
4886 EXPORT_SYMBOL(dev_load);
4889 EXPORT_PER_CPU_SYMBOL(softnet_data);