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/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #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, const 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);
920 else if (__dev_get_by_name(net, newname))
923 strlcpy(dev->name, newname, IFNAMSIZ);
926 /* For now only devices in the initial network namespace
929 if (net == &init_net) {
930 ret = device_rename(&dev->dev, dev->name);
932 memcpy(dev->name, oldname, IFNAMSIZ);
937 write_lock_bh(&dev_base_lock);
938 hlist_del(&dev->name_hlist);
939 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
940 write_unlock_bh(&dev_base_lock);
942 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
943 ret = notifier_to_errno(ret);
948 "%s: name change rollback failed: %d.\n",
952 memcpy(dev->name, oldname, IFNAMSIZ);
961 * dev_set_alias - change ifalias of a device
963 * @alias: name up to IFALIASZ
964 * @len: limit of bytes to copy from info
966 * Set ifalias for a device,
968 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
983 dev->ifalias = krealloc(dev->ifalias, len+1, GFP_KERNEL);
987 strlcpy(dev->ifalias, alias, len+1);
993 * netdev_features_change - device changes features
994 * @dev: device to cause notification
996 * Called to indicate a device has changed features.
998 void netdev_features_change(struct net_device *dev)
1000 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1002 EXPORT_SYMBOL(netdev_features_change);
1005 * netdev_state_change - device changes state
1006 * @dev: device to cause notification
1008 * Called to indicate a device has changed state. This function calls
1009 * the notifier chains for netdev_chain and sends a NEWLINK message
1010 * to the routing socket.
1012 void netdev_state_change(struct net_device *dev)
1014 if (dev->flags & IFF_UP) {
1015 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1016 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1020 void netdev_bonding_change(struct net_device *dev)
1022 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev);
1024 EXPORT_SYMBOL(netdev_bonding_change);
1027 * dev_load - load a network module
1028 * @net: the applicable net namespace
1029 * @name: name of interface
1031 * If a network interface is not present and the process has suitable
1032 * privileges this function loads the module. If module loading is not
1033 * available in this kernel then it becomes a nop.
1036 void dev_load(struct net *net, const char *name)
1038 struct net_device *dev;
1040 read_lock(&dev_base_lock);
1041 dev = __dev_get_by_name(net, name);
1042 read_unlock(&dev_base_lock);
1044 if (!dev && capable(CAP_SYS_MODULE))
1045 request_module("%s", name);
1049 * dev_open - prepare an interface for use.
1050 * @dev: device to open
1052 * Takes a device from down to up state. The device's private open
1053 * function is invoked and then the multicast lists are loaded. Finally
1054 * the device is moved into the up state and a %NETDEV_UP message is
1055 * sent to the netdev notifier chain.
1057 * Calling this function on an active interface is a nop. On a failure
1058 * a negative errno code is returned.
1060 int dev_open(struct net_device *dev)
1070 if (dev->flags & IFF_UP)
1074 * Is it even present?
1076 if (!netif_device_present(dev))
1080 * Call device private open method
1082 set_bit(__LINK_STATE_START, &dev->state);
1084 if (dev->validate_addr)
1085 ret = dev->validate_addr(dev);
1087 if (!ret && dev->open)
1088 ret = dev->open(dev);
1091 * If it went open OK then:
1095 clear_bit(__LINK_STATE_START, &dev->state);
1100 dev->flags |= IFF_UP;
1103 * Initialize multicasting status
1105 dev_set_rx_mode(dev);
1108 * Wakeup transmit queue engine
1113 * ... and announce new interface.
1115 call_netdevice_notifiers(NETDEV_UP, dev);
1122 * dev_close - shutdown an interface.
1123 * @dev: device to shutdown
1125 * This function moves an active device into down state. A
1126 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1127 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1130 int dev_close(struct net_device *dev)
1136 if (!(dev->flags & IFF_UP))
1140 * Tell people we are going down, so that they can
1141 * prepare to death, when device is still operating.
1143 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1145 clear_bit(__LINK_STATE_START, &dev->state);
1147 /* Synchronize to scheduled poll. We cannot touch poll list,
1148 * it can be even on different cpu. So just clear netif_running().
1150 * dev->stop() will invoke napi_disable() on all of it's
1151 * napi_struct instances on this device.
1153 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1155 dev_deactivate(dev);
1158 * Call the device specific close. This cannot fail.
1159 * Only if device is UP
1161 * We allow it to be called even after a DETACH hot-plug
1168 * Device is now down.
1171 dev->flags &= ~IFF_UP;
1174 * Tell people we are down
1176 call_netdevice_notifiers(NETDEV_DOWN, dev);
1183 * dev_disable_lro - disable Large Receive Offload on a device
1186 * Disable Large Receive Offload (LRO) on a net device. Must be
1187 * called under RTNL. This is needed if received packets may be
1188 * forwarded to another interface.
1190 void dev_disable_lro(struct net_device *dev)
1192 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1193 dev->ethtool_ops->set_flags) {
1194 u32 flags = dev->ethtool_ops->get_flags(dev);
1195 if (flags & ETH_FLAG_LRO) {
1196 flags &= ~ETH_FLAG_LRO;
1197 dev->ethtool_ops->set_flags(dev, flags);
1200 WARN_ON(dev->features & NETIF_F_LRO);
1202 EXPORT_SYMBOL(dev_disable_lro);
1205 static int dev_boot_phase = 1;
1208 * Device change register/unregister. These are not inline or static
1209 * as we export them to the world.
1213 * register_netdevice_notifier - register a network notifier block
1216 * Register a notifier to be called when network device events occur.
1217 * The notifier passed is linked into the kernel structures and must
1218 * not be reused until it has been unregistered. A negative errno code
1219 * is returned on a failure.
1221 * When registered all registration and up events are replayed
1222 * to the new notifier to allow device to have a race free
1223 * view of the network device list.
1226 int register_netdevice_notifier(struct notifier_block *nb)
1228 struct net_device *dev;
1229 struct net_device *last;
1234 err = raw_notifier_chain_register(&netdev_chain, nb);
1240 for_each_netdev(net, dev) {
1241 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1242 err = notifier_to_errno(err);
1246 if (!(dev->flags & IFF_UP))
1249 nb->notifier_call(nb, NETDEV_UP, dev);
1260 for_each_netdev(net, dev) {
1264 if (dev->flags & IFF_UP) {
1265 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1266 nb->notifier_call(nb, NETDEV_DOWN, dev);
1268 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1272 raw_notifier_chain_unregister(&netdev_chain, nb);
1277 * unregister_netdevice_notifier - unregister a network notifier block
1280 * Unregister a notifier previously registered by
1281 * register_netdevice_notifier(). The notifier is unlinked into the
1282 * kernel structures and may then be reused. A negative errno code
1283 * is returned on a failure.
1286 int unregister_netdevice_notifier(struct notifier_block *nb)
1291 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1297 * call_netdevice_notifiers - call all network notifier blocks
1298 * @val: value passed unmodified to notifier function
1299 * @dev: net_device pointer passed unmodified to notifier function
1301 * Call all network notifier blocks. Parameters and return value
1302 * are as for raw_notifier_call_chain().
1305 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1307 return raw_notifier_call_chain(&netdev_chain, val, dev);
1310 /* When > 0 there are consumers of rx skb time stamps */
1311 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1313 void net_enable_timestamp(void)
1315 atomic_inc(&netstamp_needed);
1318 void net_disable_timestamp(void)
1320 atomic_dec(&netstamp_needed);
1323 static inline void net_timestamp(struct sk_buff *skb)
1325 if (atomic_read(&netstamp_needed))
1326 __net_timestamp(skb);
1328 skb->tstamp.tv64 = 0;
1332 * Support routine. Sends outgoing frames to any network
1333 * taps currently in use.
1336 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1338 struct packet_type *ptype;
1343 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1344 /* Never send packets back to the socket
1345 * they originated from - MvS (miquels@drinkel.ow.org)
1347 if ((ptype->dev == dev || !ptype->dev) &&
1348 (ptype->af_packet_priv == NULL ||
1349 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1350 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1354 /* skb->nh should be correctly
1355 set by sender, so that the second statement is
1356 just protection against buggy protocols.
1358 skb_reset_mac_header(skb2);
1360 if (skb_network_header(skb2) < skb2->data ||
1361 skb2->network_header > skb2->tail) {
1362 if (net_ratelimit())
1363 printk(KERN_CRIT "protocol %04x is "
1365 skb2->protocol, dev->name);
1366 skb_reset_network_header(skb2);
1369 skb2->transport_header = skb2->network_header;
1370 skb2->pkt_type = PACKET_OUTGOING;
1371 ptype->func(skb2, skb->dev, ptype, skb->dev);
1378 static inline void __netif_reschedule(struct Qdisc *q)
1380 struct softnet_data *sd;
1381 unsigned long flags;
1383 local_irq_save(flags);
1384 sd = &__get_cpu_var(softnet_data);
1385 q->next_sched = sd->output_queue;
1386 sd->output_queue = q;
1387 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1388 local_irq_restore(flags);
1391 void __netif_schedule(struct Qdisc *q)
1393 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1394 __netif_reschedule(q);
1396 EXPORT_SYMBOL(__netif_schedule);
1398 void dev_kfree_skb_irq(struct sk_buff *skb)
1400 if (atomic_dec_and_test(&skb->users)) {
1401 struct softnet_data *sd;
1402 unsigned long flags;
1404 local_irq_save(flags);
1405 sd = &__get_cpu_var(softnet_data);
1406 skb->next = sd->completion_queue;
1407 sd->completion_queue = skb;
1408 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1409 local_irq_restore(flags);
1412 EXPORT_SYMBOL(dev_kfree_skb_irq);
1414 void dev_kfree_skb_any(struct sk_buff *skb)
1416 if (in_irq() || irqs_disabled())
1417 dev_kfree_skb_irq(skb);
1421 EXPORT_SYMBOL(dev_kfree_skb_any);
1425 * netif_device_detach - mark device as removed
1426 * @dev: network device
1428 * Mark device as removed from system and therefore no longer available.
1430 void netif_device_detach(struct net_device *dev)
1432 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1433 netif_running(dev)) {
1434 netif_stop_queue(dev);
1437 EXPORT_SYMBOL(netif_device_detach);
1440 * netif_device_attach - mark device as attached
1441 * @dev: network device
1443 * Mark device as attached from system and restart if needed.
1445 void netif_device_attach(struct net_device *dev)
1447 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1448 netif_running(dev)) {
1449 netif_wake_queue(dev);
1450 __netdev_watchdog_up(dev);
1453 EXPORT_SYMBOL(netif_device_attach);
1455 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1457 return ((features & NETIF_F_GEN_CSUM) ||
1458 ((features & NETIF_F_IP_CSUM) &&
1459 protocol == htons(ETH_P_IP)) ||
1460 ((features & NETIF_F_IPV6_CSUM) &&
1461 protocol == htons(ETH_P_IPV6)));
1464 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1466 if (can_checksum_protocol(dev->features, skb->protocol))
1469 if (skb->protocol == htons(ETH_P_8021Q)) {
1470 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1471 if (can_checksum_protocol(dev->features & dev->vlan_features,
1472 veh->h_vlan_encapsulated_proto))
1480 * Invalidate hardware checksum when packet is to be mangled, and
1481 * complete checksum manually on outgoing path.
1483 int skb_checksum_help(struct sk_buff *skb)
1486 int ret = 0, offset;
1488 if (skb->ip_summed == CHECKSUM_COMPLETE)
1489 goto out_set_summed;
1491 if (unlikely(skb_shinfo(skb)->gso_size)) {
1492 /* Let GSO fix up the checksum. */
1493 goto out_set_summed;
1496 offset = skb->csum_start - skb_headroom(skb);
1497 BUG_ON(offset >= skb_headlen(skb));
1498 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1500 offset += skb->csum_offset;
1501 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1503 if (skb_cloned(skb) &&
1504 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1505 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1510 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1512 skb->ip_summed = CHECKSUM_NONE;
1518 * skb_gso_segment - Perform segmentation on skb.
1519 * @skb: buffer to segment
1520 * @features: features for the output path (see dev->features)
1522 * This function segments the given skb and returns a list of segments.
1524 * It may return NULL if the skb requires no segmentation. This is
1525 * only possible when GSO is used for verifying header integrity.
1527 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1529 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1530 struct packet_type *ptype;
1531 __be16 type = skb->protocol;
1534 BUG_ON(skb_shinfo(skb)->frag_list);
1536 skb_reset_mac_header(skb);
1537 skb->mac_len = skb->network_header - skb->mac_header;
1538 __skb_pull(skb, skb->mac_len);
1540 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1541 if (skb_header_cloned(skb) &&
1542 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1543 return ERR_PTR(err);
1547 list_for_each_entry_rcu(ptype,
1548 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1549 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1550 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1551 err = ptype->gso_send_check(skb);
1552 segs = ERR_PTR(err);
1553 if (err || skb_gso_ok(skb, features))
1555 __skb_push(skb, (skb->data -
1556 skb_network_header(skb)));
1558 segs = ptype->gso_segment(skb, features);
1564 __skb_push(skb, skb->data - skb_mac_header(skb));
1569 EXPORT_SYMBOL(skb_gso_segment);
1571 /* Take action when hardware reception checksum errors are detected. */
1573 void netdev_rx_csum_fault(struct net_device *dev)
1575 if (net_ratelimit()) {
1576 printk(KERN_ERR "%s: hw csum failure.\n",
1577 dev ? dev->name : "<unknown>");
1581 EXPORT_SYMBOL(netdev_rx_csum_fault);
1584 /* Actually, we should eliminate this check as soon as we know, that:
1585 * 1. IOMMU is present and allows to map all the memory.
1586 * 2. No high memory really exists on this machine.
1589 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1591 #ifdef CONFIG_HIGHMEM
1594 if (dev->features & NETIF_F_HIGHDMA)
1597 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1598 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1606 void (*destructor)(struct sk_buff *skb);
1609 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1611 static void dev_gso_skb_destructor(struct sk_buff *skb)
1613 struct dev_gso_cb *cb;
1616 struct sk_buff *nskb = skb->next;
1618 skb->next = nskb->next;
1621 } while (skb->next);
1623 cb = DEV_GSO_CB(skb);
1625 cb->destructor(skb);
1629 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1630 * @skb: buffer to segment
1632 * This function segments the given skb and stores the list of segments
1635 static int dev_gso_segment(struct sk_buff *skb)
1637 struct net_device *dev = skb->dev;
1638 struct sk_buff *segs;
1639 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1642 segs = skb_gso_segment(skb, features);
1644 /* Verifying header integrity only. */
1649 return PTR_ERR(segs);
1652 DEV_GSO_CB(skb)->destructor = skb->destructor;
1653 skb->destructor = dev_gso_skb_destructor;
1658 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1659 struct netdev_queue *txq)
1661 if (likely(!skb->next)) {
1662 if (!list_empty(&ptype_all))
1663 dev_queue_xmit_nit(skb, dev);
1665 if (netif_needs_gso(dev, skb)) {
1666 if (unlikely(dev_gso_segment(skb)))
1672 return dev->hard_start_xmit(skb, dev);
1677 struct sk_buff *nskb = skb->next;
1680 skb->next = nskb->next;
1682 rc = dev->hard_start_xmit(nskb, dev);
1684 nskb->next = skb->next;
1688 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1689 return NETDEV_TX_BUSY;
1690 } while (skb->next);
1692 skb->destructor = DEV_GSO_CB(skb)->destructor;
1699 static u32 simple_tx_hashrnd;
1700 static int simple_tx_hashrnd_initialized = 0;
1702 static u16 simple_tx_hash(struct net_device *dev, struct sk_buff *skb)
1704 u32 addr1, addr2, ports;
1708 if (unlikely(!simple_tx_hashrnd_initialized)) {
1709 get_random_bytes(&simple_tx_hashrnd, 4);
1710 simple_tx_hashrnd_initialized = 1;
1713 switch (skb->protocol) {
1714 case htons(ETH_P_IP):
1715 if (!(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)))
1716 ip_proto = ip_hdr(skb)->protocol;
1717 addr1 = ip_hdr(skb)->saddr;
1718 addr2 = ip_hdr(skb)->daddr;
1719 ihl = ip_hdr(skb)->ihl;
1721 case htons(ETH_P_IPV6):
1722 ip_proto = ipv6_hdr(skb)->nexthdr;
1723 addr1 = ipv6_hdr(skb)->saddr.s6_addr32[3];
1724 addr2 = ipv6_hdr(skb)->daddr.s6_addr32[3];
1739 case IPPROTO_UDPLITE:
1740 ports = *((u32 *) (skb_network_header(skb) + (ihl * 4)));
1748 hash = jhash_3words(addr1, addr2, ports, simple_tx_hashrnd);
1750 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1753 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1754 struct sk_buff *skb)
1756 u16 queue_index = 0;
1758 if (dev->select_queue)
1759 queue_index = dev->select_queue(dev, skb);
1760 else if (dev->real_num_tx_queues > 1)
1761 queue_index = simple_tx_hash(dev, skb);
1763 skb_set_queue_mapping(skb, queue_index);
1764 return netdev_get_tx_queue(dev, queue_index);
1768 * dev_queue_xmit - transmit a buffer
1769 * @skb: buffer to transmit
1771 * Queue a buffer for transmission to a network device. The caller must
1772 * have set the device and priority and built the buffer before calling
1773 * this function. The function can be called from an interrupt.
1775 * A negative errno code is returned on a failure. A success does not
1776 * guarantee the frame will be transmitted as it may be dropped due
1777 * to congestion or traffic shaping.
1779 * -----------------------------------------------------------------------------------
1780 * I notice this method can also return errors from the queue disciplines,
1781 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1784 * Regardless of the return value, the skb is consumed, so it is currently
1785 * difficult to retry a send to this method. (You can bump the ref count
1786 * before sending to hold a reference for retry if you are careful.)
1788 * When calling this method, interrupts MUST be enabled. This is because
1789 * the BH enable code must have IRQs enabled so that it will not deadlock.
1792 int dev_queue_xmit(struct sk_buff *skb)
1794 struct net_device *dev = skb->dev;
1795 struct netdev_queue *txq;
1799 /* GSO will handle the following emulations directly. */
1800 if (netif_needs_gso(dev, skb))
1803 if (skb_shinfo(skb)->frag_list &&
1804 !(dev->features & NETIF_F_FRAGLIST) &&
1805 __skb_linearize(skb))
1808 /* Fragmented skb is linearized if device does not support SG,
1809 * or if at least one of fragments is in highmem and device
1810 * does not support DMA from it.
1812 if (skb_shinfo(skb)->nr_frags &&
1813 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1814 __skb_linearize(skb))
1817 /* If packet is not checksummed and device does not support
1818 * checksumming for this protocol, complete checksumming here.
1820 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1821 skb_set_transport_header(skb, skb->csum_start -
1823 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1828 /* Disable soft irqs for various locks below. Also
1829 * stops preemption for RCU.
1833 txq = dev_pick_tx(dev, skb);
1834 q = rcu_dereference(txq->qdisc);
1836 #ifdef CONFIG_NET_CLS_ACT
1837 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1840 spinlock_t *root_lock = qdisc_lock(q);
1842 spin_lock(root_lock);
1844 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1848 rc = qdisc_enqueue_root(skb, q);
1851 spin_unlock(root_lock);
1856 /* The device has no queue. Common case for software devices:
1857 loopback, all the sorts of tunnels...
1859 Really, it is unlikely that netif_tx_lock protection is necessary
1860 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1862 However, it is possible, that they rely on protection
1865 Check this and shot the lock. It is not prone from deadlocks.
1866 Either shot noqueue qdisc, it is even simpler 8)
1868 if (dev->flags & IFF_UP) {
1869 int cpu = smp_processor_id(); /* ok because BHs are off */
1871 if (txq->xmit_lock_owner != cpu) {
1873 HARD_TX_LOCK(dev, txq, cpu);
1875 if (!netif_tx_queue_stopped(txq)) {
1877 if (!dev_hard_start_xmit(skb, dev, txq)) {
1878 HARD_TX_UNLOCK(dev, txq);
1882 HARD_TX_UNLOCK(dev, txq);
1883 if (net_ratelimit())
1884 printk(KERN_CRIT "Virtual device %s asks to "
1885 "queue packet!\n", dev->name);
1887 /* Recursion is detected! It is possible,
1889 if (net_ratelimit())
1890 printk(KERN_CRIT "Dead loop on virtual device "
1891 "%s, fix it urgently!\n", dev->name);
1896 rcu_read_unlock_bh();
1902 rcu_read_unlock_bh();
1907 /*=======================================================================
1909 =======================================================================*/
1911 int netdev_max_backlog __read_mostly = 1000;
1912 int netdev_budget __read_mostly = 300;
1913 int weight_p __read_mostly = 64; /* old backlog weight */
1915 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1919 * netif_rx - post buffer to the network code
1920 * @skb: buffer to post
1922 * This function receives a packet from a device driver and queues it for
1923 * the upper (protocol) levels to process. It always succeeds. The buffer
1924 * may be dropped during processing for congestion control or by the
1928 * NET_RX_SUCCESS (no congestion)
1929 * NET_RX_DROP (packet was dropped)
1933 int netif_rx(struct sk_buff *skb)
1935 struct softnet_data *queue;
1936 unsigned long flags;
1938 /* if netpoll wants it, pretend we never saw it */
1939 if (netpoll_rx(skb))
1942 if (!skb->tstamp.tv64)
1946 * The code is rearranged so that the path is the most
1947 * short when CPU is congested, but is still operating.
1949 local_irq_save(flags);
1950 queue = &__get_cpu_var(softnet_data);
1952 __get_cpu_var(netdev_rx_stat).total++;
1953 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1954 if (queue->input_pkt_queue.qlen) {
1956 __skb_queue_tail(&queue->input_pkt_queue, skb);
1957 local_irq_restore(flags);
1958 return NET_RX_SUCCESS;
1961 napi_schedule(&queue->backlog);
1965 __get_cpu_var(netdev_rx_stat).dropped++;
1966 local_irq_restore(flags);
1972 int netif_rx_ni(struct sk_buff *skb)
1977 err = netif_rx(skb);
1978 if (local_softirq_pending())
1985 EXPORT_SYMBOL(netif_rx_ni);
1987 static void net_tx_action(struct softirq_action *h)
1989 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1991 if (sd->completion_queue) {
1992 struct sk_buff *clist;
1994 local_irq_disable();
1995 clist = sd->completion_queue;
1996 sd->completion_queue = NULL;
2000 struct sk_buff *skb = clist;
2001 clist = clist->next;
2003 WARN_ON(atomic_read(&skb->users));
2008 if (sd->output_queue) {
2011 local_irq_disable();
2012 head = sd->output_queue;
2013 sd->output_queue = NULL;
2017 struct Qdisc *q = head;
2018 spinlock_t *root_lock;
2020 head = head->next_sched;
2022 root_lock = qdisc_lock(q);
2023 if (spin_trylock(root_lock)) {
2024 smp_mb__before_clear_bit();
2025 clear_bit(__QDISC_STATE_SCHED,
2028 spin_unlock(root_lock);
2030 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2032 __netif_reschedule(q);
2034 smp_mb__before_clear_bit();
2035 clear_bit(__QDISC_STATE_SCHED,
2043 static inline int deliver_skb(struct sk_buff *skb,
2044 struct packet_type *pt_prev,
2045 struct net_device *orig_dev)
2047 atomic_inc(&skb->users);
2048 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2051 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2052 /* These hooks defined here for ATM */
2054 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2055 unsigned char *addr);
2056 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
2059 * If bridge module is loaded call bridging hook.
2060 * returns NULL if packet was consumed.
2062 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2063 struct sk_buff *skb) __read_mostly;
2064 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2065 struct packet_type **pt_prev, int *ret,
2066 struct net_device *orig_dev)
2068 struct net_bridge_port *port;
2070 if (skb->pkt_type == PACKET_LOOPBACK ||
2071 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2075 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2079 return br_handle_frame_hook(port, skb);
2082 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2085 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2086 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2087 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2089 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2090 struct packet_type **pt_prev,
2092 struct net_device *orig_dev)
2094 if (skb->dev->macvlan_port == NULL)
2098 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2101 return macvlan_handle_frame_hook(skb);
2104 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2107 #ifdef CONFIG_NET_CLS_ACT
2108 /* TODO: Maybe we should just force sch_ingress to be compiled in
2109 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2110 * a compare and 2 stores extra right now if we dont have it on
2111 * but have CONFIG_NET_CLS_ACT
2112 * NOTE: This doesnt stop any functionality; if you dont have
2113 * the ingress scheduler, you just cant add policies on ingress.
2116 static int ing_filter(struct sk_buff *skb)
2118 struct net_device *dev = skb->dev;
2119 u32 ttl = G_TC_RTTL(skb->tc_verd);
2120 struct netdev_queue *rxq;
2121 int result = TC_ACT_OK;
2124 if (MAX_RED_LOOP < ttl++) {
2126 "Redir loop detected Dropping packet (%d->%d)\n",
2127 skb->iif, dev->ifindex);
2131 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2132 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2134 rxq = &dev->rx_queue;
2137 if (q != &noop_qdisc) {
2138 spin_lock(qdisc_lock(q));
2139 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2140 result = qdisc_enqueue_root(skb, q);
2141 spin_unlock(qdisc_lock(q));
2147 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2148 struct packet_type **pt_prev,
2149 int *ret, struct net_device *orig_dev)
2151 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2155 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2158 /* Huh? Why does turning on AF_PACKET affect this? */
2159 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2162 switch (ing_filter(skb)) {
2176 * netif_nit_deliver - deliver received packets to network taps
2179 * This function is used to deliver incoming packets to network
2180 * taps. It should be used when the normal netif_receive_skb path
2181 * is bypassed, for example because of VLAN acceleration.
2183 void netif_nit_deliver(struct sk_buff *skb)
2185 struct packet_type *ptype;
2187 if (list_empty(&ptype_all))
2190 skb_reset_network_header(skb);
2191 skb_reset_transport_header(skb);
2192 skb->mac_len = skb->network_header - skb->mac_header;
2195 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2196 if (!ptype->dev || ptype->dev == skb->dev)
2197 deliver_skb(skb, ptype, skb->dev);
2203 * netif_receive_skb - process receive buffer from network
2204 * @skb: buffer to process
2206 * netif_receive_skb() is the main receive data processing function.
2207 * It always succeeds. The buffer may be dropped during processing
2208 * for congestion control or by the protocol layers.
2210 * This function may only be called from softirq context and interrupts
2211 * should be enabled.
2213 * Return values (usually ignored):
2214 * NET_RX_SUCCESS: no congestion
2215 * NET_RX_DROP: packet was dropped
2217 int netif_receive_skb(struct sk_buff *skb)
2219 struct packet_type *ptype, *pt_prev;
2220 struct net_device *orig_dev;
2221 struct net_device *null_or_orig;
2222 int ret = NET_RX_DROP;
2225 if (skb->vlan_tci && vlan_hwaccel_do_receive(skb))
2226 return NET_RX_SUCCESS;
2228 /* if we've gotten here through NAPI, check netpoll */
2229 if (netpoll_receive_skb(skb))
2232 if (!skb->tstamp.tv64)
2236 skb->iif = skb->dev->ifindex;
2238 null_or_orig = NULL;
2239 orig_dev = skb->dev;
2240 if (orig_dev->master) {
2241 if (skb_bond_should_drop(skb))
2242 null_or_orig = orig_dev; /* deliver only exact match */
2244 skb->dev = orig_dev->master;
2247 __get_cpu_var(netdev_rx_stat).total++;
2249 skb_reset_network_header(skb);
2250 skb_reset_transport_header(skb);
2251 skb->mac_len = skb->network_header - skb->mac_header;
2257 /* Don't receive packets in an exiting network namespace */
2258 if (!net_alive(dev_net(skb->dev))) {
2263 #ifdef CONFIG_NET_CLS_ACT
2264 if (skb->tc_verd & TC_NCLS) {
2265 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2270 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2271 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2272 ptype->dev == orig_dev) {
2274 ret = deliver_skb(skb, pt_prev, orig_dev);
2279 #ifdef CONFIG_NET_CLS_ACT
2280 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2286 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2289 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2293 type = skb->protocol;
2294 list_for_each_entry_rcu(ptype,
2295 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2296 if (ptype->type == type &&
2297 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2298 ptype->dev == orig_dev)) {
2300 ret = deliver_skb(skb, pt_prev, orig_dev);
2306 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2309 /* Jamal, now you will not able to escape explaining
2310 * me how you were going to use this. :-)
2320 /* Network device is going away, flush any packets still pending */
2321 static void flush_backlog(void *arg)
2323 struct net_device *dev = arg;
2324 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2325 struct sk_buff *skb, *tmp;
2327 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2328 if (skb->dev == dev) {
2329 __skb_unlink(skb, &queue->input_pkt_queue);
2334 static int process_backlog(struct napi_struct *napi, int quota)
2337 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2338 unsigned long start_time = jiffies;
2340 napi->weight = weight_p;
2342 struct sk_buff *skb;
2344 local_irq_disable();
2345 skb = __skb_dequeue(&queue->input_pkt_queue);
2347 __napi_complete(napi);
2353 netif_receive_skb(skb);
2354 } while (++work < quota && jiffies == start_time);
2360 * __napi_schedule - schedule for receive
2361 * @n: entry to schedule
2363 * The entry's receive function will be scheduled to run
2365 void __napi_schedule(struct napi_struct *n)
2367 unsigned long flags;
2369 local_irq_save(flags);
2370 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2371 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2372 local_irq_restore(flags);
2374 EXPORT_SYMBOL(__napi_schedule);
2377 static void net_rx_action(struct softirq_action *h)
2379 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2380 unsigned long time_limit = jiffies + 2;
2381 int budget = netdev_budget;
2384 local_irq_disable();
2386 while (!list_empty(list)) {
2387 struct napi_struct *n;
2390 /* If softirq window is exhuasted then punt.
2391 * Allow this to run for 2 jiffies since which will allow
2392 * an average latency of 1.5/HZ.
2394 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2399 /* Even though interrupts have been re-enabled, this
2400 * access is safe because interrupts can only add new
2401 * entries to the tail of this list, and only ->poll()
2402 * calls can remove this head entry from the list.
2404 n = list_entry(list->next, struct napi_struct, poll_list);
2406 have = netpoll_poll_lock(n);
2410 /* This NAPI_STATE_SCHED test is for avoiding a race
2411 * with netpoll's poll_napi(). Only the entity which
2412 * obtains the lock and sees NAPI_STATE_SCHED set will
2413 * actually make the ->poll() call. Therefore we avoid
2414 * accidently calling ->poll() when NAPI is not scheduled.
2417 if (test_bit(NAPI_STATE_SCHED, &n->state))
2418 work = n->poll(n, weight);
2420 WARN_ON_ONCE(work > weight);
2424 local_irq_disable();
2426 /* Drivers must not modify the NAPI state if they
2427 * consume the entire weight. In such cases this code
2428 * still "owns" the NAPI instance and therefore can
2429 * move the instance around on the list at-will.
2431 if (unlikely(work == weight)) {
2432 if (unlikely(napi_disable_pending(n)))
2435 list_move_tail(&n->poll_list, list);
2438 netpoll_poll_unlock(have);
2443 #ifdef CONFIG_NET_DMA
2445 * There may not be any more sk_buffs coming right now, so push
2446 * any pending DMA copies to hardware
2448 if (!cpus_empty(net_dma.channel_mask)) {
2450 for_each_cpu_mask_nr(chan_idx, net_dma.channel_mask) {
2451 struct dma_chan *chan = net_dma.channels[chan_idx];
2453 dma_async_memcpy_issue_pending(chan);
2461 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2462 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2466 static gifconf_func_t * gifconf_list [NPROTO];
2469 * register_gifconf - register a SIOCGIF handler
2470 * @family: Address family
2471 * @gifconf: Function handler
2473 * Register protocol dependent address dumping routines. The handler
2474 * that is passed must not be freed or reused until it has been replaced
2475 * by another handler.
2477 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2479 if (family >= NPROTO)
2481 gifconf_list[family] = gifconf;
2487 * Map an interface index to its name (SIOCGIFNAME)
2491 * We need this ioctl for efficient implementation of the
2492 * if_indextoname() function required by the IPv6 API. Without
2493 * it, we would have to search all the interfaces to find a
2497 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2499 struct net_device *dev;
2503 * Fetch the caller's info block.
2506 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2509 read_lock(&dev_base_lock);
2510 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2512 read_unlock(&dev_base_lock);
2516 strcpy(ifr.ifr_name, dev->name);
2517 read_unlock(&dev_base_lock);
2519 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2525 * Perform a SIOCGIFCONF call. This structure will change
2526 * size eventually, and there is nothing I can do about it.
2527 * Thus we will need a 'compatibility mode'.
2530 static int dev_ifconf(struct net *net, char __user *arg)
2533 struct net_device *dev;
2540 * Fetch the caller's info block.
2543 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2550 * Loop over the interfaces, and write an info block for each.
2554 for_each_netdev(net, dev) {
2555 for (i = 0; i < NPROTO; i++) {
2556 if (gifconf_list[i]) {
2559 done = gifconf_list[i](dev, NULL, 0);
2561 done = gifconf_list[i](dev, pos + total,
2571 * All done. Write the updated control block back to the caller.
2573 ifc.ifc_len = total;
2576 * Both BSD and Solaris return 0 here, so we do too.
2578 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2581 #ifdef CONFIG_PROC_FS
2583 * This is invoked by the /proc filesystem handler to display a device
2586 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2587 __acquires(dev_base_lock)
2589 struct net *net = seq_file_net(seq);
2591 struct net_device *dev;
2593 read_lock(&dev_base_lock);
2595 return SEQ_START_TOKEN;
2598 for_each_netdev(net, dev)
2605 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2607 struct net *net = seq_file_net(seq);
2609 return v == SEQ_START_TOKEN ?
2610 first_net_device(net) : next_net_device((struct net_device *)v);
2613 void dev_seq_stop(struct seq_file *seq, void *v)
2614 __releases(dev_base_lock)
2616 read_unlock(&dev_base_lock);
2619 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2621 struct net_device_stats *stats = dev->get_stats(dev);
2623 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2624 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2625 dev->name, stats->rx_bytes, stats->rx_packets,
2627 stats->rx_dropped + stats->rx_missed_errors,
2628 stats->rx_fifo_errors,
2629 stats->rx_length_errors + stats->rx_over_errors +
2630 stats->rx_crc_errors + stats->rx_frame_errors,
2631 stats->rx_compressed, stats->multicast,
2632 stats->tx_bytes, stats->tx_packets,
2633 stats->tx_errors, stats->tx_dropped,
2634 stats->tx_fifo_errors, stats->collisions,
2635 stats->tx_carrier_errors +
2636 stats->tx_aborted_errors +
2637 stats->tx_window_errors +
2638 stats->tx_heartbeat_errors,
2639 stats->tx_compressed);
2643 * Called from the PROCfs module. This now uses the new arbitrary sized
2644 * /proc/net interface to create /proc/net/dev
2646 static int dev_seq_show(struct seq_file *seq, void *v)
2648 if (v == SEQ_START_TOKEN)
2649 seq_puts(seq, "Inter-| Receive "
2651 " face |bytes packets errs drop fifo frame "
2652 "compressed multicast|bytes packets errs "
2653 "drop fifo colls carrier compressed\n");
2655 dev_seq_printf_stats(seq, v);
2659 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2661 struct netif_rx_stats *rc = NULL;
2663 while (*pos < nr_cpu_ids)
2664 if (cpu_online(*pos)) {
2665 rc = &per_cpu(netdev_rx_stat, *pos);
2672 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2674 return softnet_get_online(pos);
2677 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2680 return softnet_get_online(pos);
2683 static void softnet_seq_stop(struct seq_file *seq, void *v)
2687 static int softnet_seq_show(struct seq_file *seq, void *v)
2689 struct netif_rx_stats *s = v;
2691 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2692 s->total, s->dropped, s->time_squeeze, 0,
2693 0, 0, 0, 0, /* was fastroute */
2698 static const struct seq_operations dev_seq_ops = {
2699 .start = dev_seq_start,
2700 .next = dev_seq_next,
2701 .stop = dev_seq_stop,
2702 .show = dev_seq_show,
2705 static int dev_seq_open(struct inode *inode, struct file *file)
2707 return seq_open_net(inode, file, &dev_seq_ops,
2708 sizeof(struct seq_net_private));
2711 static const struct file_operations dev_seq_fops = {
2712 .owner = THIS_MODULE,
2713 .open = dev_seq_open,
2715 .llseek = seq_lseek,
2716 .release = seq_release_net,
2719 static const struct seq_operations softnet_seq_ops = {
2720 .start = softnet_seq_start,
2721 .next = softnet_seq_next,
2722 .stop = softnet_seq_stop,
2723 .show = softnet_seq_show,
2726 static int softnet_seq_open(struct inode *inode, struct file *file)
2728 return seq_open(file, &softnet_seq_ops);
2731 static const struct file_operations softnet_seq_fops = {
2732 .owner = THIS_MODULE,
2733 .open = softnet_seq_open,
2735 .llseek = seq_lseek,
2736 .release = seq_release,
2739 static void *ptype_get_idx(loff_t pos)
2741 struct packet_type *pt = NULL;
2745 list_for_each_entry_rcu(pt, &ptype_all, list) {
2751 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
2752 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2761 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2765 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2768 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2770 struct packet_type *pt;
2771 struct list_head *nxt;
2775 if (v == SEQ_START_TOKEN)
2776 return ptype_get_idx(0);
2779 nxt = pt->list.next;
2780 if (pt->type == htons(ETH_P_ALL)) {
2781 if (nxt != &ptype_all)
2784 nxt = ptype_base[0].next;
2786 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
2788 while (nxt == &ptype_base[hash]) {
2789 if (++hash >= PTYPE_HASH_SIZE)
2791 nxt = ptype_base[hash].next;
2794 return list_entry(nxt, struct packet_type, list);
2797 static void ptype_seq_stop(struct seq_file *seq, void *v)
2803 static int ptype_seq_show(struct seq_file *seq, void *v)
2805 struct packet_type *pt = v;
2807 if (v == SEQ_START_TOKEN)
2808 seq_puts(seq, "Type Device Function\n");
2809 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
2810 if (pt->type == htons(ETH_P_ALL))
2811 seq_puts(seq, "ALL ");
2813 seq_printf(seq, "%04x", ntohs(pt->type));
2815 seq_printf(seq, " %-8s %pF\n",
2816 pt->dev ? pt->dev->name : "", pt->func);
2822 static const struct seq_operations ptype_seq_ops = {
2823 .start = ptype_seq_start,
2824 .next = ptype_seq_next,
2825 .stop = ptype_seq_stop,
2826 .show = ptype_seq_show,
2829 static int ptype_seq_open(struct inode *inode, struct file *file)
2831 return seq_open_net(inode, file, &ptype_seq_ops,
2832 sizeof(struct seq_net_private));
2835 static const struct file_operations ptype_seq_fops = {
2836 .owner = THIS_MODULE,
2837 .open = ptype_seq_open,
2839 .llseek = seq_lseek,
2840 .release = seq_release_net,
2844 static int __net_init dev_proc_net_init(struct net *net)
2848 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2850 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2852 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2855 if (wext_proc_init(net))
2861 proc_net_remove(net, "ptype");
2863 proc_net_remove(net, "softnet_stat");
2865 proc_net_remove(net, "dev");
2869 static void __net_exit dev_proc_net_exit(struct net *net)
2871 wext_proc_exit(net);
2873 proc_net_remove(net, "ptype");
2874 proc_net_remove(net, "softnet_stat");
2875 proc_net_remove(net, "dev");
2878 static struct pernet_operations __net_initdata dev_proc_ops = {
2879 .init = dev_proc_net_init,
2880 .exit = dev_proc_net_exit,
2883 static int __init dev_proc_init(void)
2885 return register_pernet_subsys(&dev_proc_ops);
2888 #define dev_proc_init() 0
2889 #endif /* CONFIG_PROC_FS */
2893 * netdev_set_master - set up master/slave pair
2894 * @slave: slave device
2895 * @master: new master device
2897 * Changes the master device of the slave. Pass %NULL to break the
2898 * bonding. The caller must hold the RTNL semaphore. On a failure
2899 * a negative errno code is returned. On success the reference counts
2900 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2901 * function returns zero.
2903 int netdev_set_master(struct net_device *slave, struct net_device *master)
2905 struct net_device *old = slave->master;
2915 slave->master = master;
2923 slave->flags |= IFF_SLAVE;
2925 slave->flags &= ~IFF_SLAVE;
2927 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2931 static void dev_change_rx_flags(struct net_device *dev, int flags)
2933 if (dev->flags & IFF_UP && dev->change_rx_flags)
2934 dev->change_rx_flags(dev, flags);
2937 static int __dev_set_promiscuity(struct net_device *dev, int inc)
2939 unsigned short old_flags = dev->flags;
2943 dev->flags |= IFF_PROMISC;
2944 dev->promiscuity += inc;
2945 if (dev->promiscuity == 0) {
2948 * If inc causes overflow, untouch promisc and return error.
2951 dev->flags &= ~IFF_PROMISC;
2953 dev->promiscuity -= inc;
2954 printk(KERN_WARNING "%s: promiscuity touches roof, "
2955 "set promiscuity failed, promiscuity feature "
2956 "of device might be broken.\n", dev->name);
2960 if (dev->flags != old_flags) {
2961 printk(KERN_INFO "device %s %s promiscuous mode\n",
2962 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2965 audit_log(current->audit_context, GFP_ATOMIC,
2966 AUDIT_ANOM_PROMISCUOUS,
2967 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2968 dev->name, (dev->flags & IFF_PROMISC),
2969 (old_flags & IFF_PROMISC),
2970 audit_get_loginuid(current),
2971 current->uid, current->gid,
2972 audit_get_sessionid(current));
2974 dev_change_rx_flags(dev, IFF_PROMISC);
2980 * dev_set_promiscuity - update promiscuity count on a device
2984 * Add or remove promiscuity from a device. While the count in the device
2985 * remains above zero the interface remains promiscuous. Once it hits zero
2986 * the device reverts back to normal filtering operation. A negative inc
2987 * value is used to drop promiscuity on the device.
2988 * Return 0 if successful or a negative errno code on error.
2990 int dev_set_promiscuity(struct net_device *dev, int inc)
2992 unsigned short old_flags = dev->flags;
2995 err = __dev_set_promiscuity(dev, inc);
2998 if (dev->flags != old_flags)
2999 dev_set_rx_mode(dev);
3004 * dev_set_allmulti - update allmulti count on a device
3008 * Add or remove reception of all multicast frames to a device. While the
3009 * count in the device remains above zero the interface remains listening
3010 * to all interfaces. Once it hits zero the device reverts back to normal
3011 * filtering operation. A negative @inc value is used to drop the counter
3012 * when releasing a resource needing all multicasts.
3013 * Return 0 if successful or a negative errno code on error.
3016 int dev_set_allmulti(struct net_device *dev, int inc)
3018 unsigned short old_flags = dev->flags;
3022 dev->flags |= IFF_ALLMULTI;
3023 dev->allmulti += inc;
3024 if (dev->allmulti == 0) {
3027 * If inc causes overflow, untouch allmulti and return error.
3030 dev->flags &= ~IFF_ALLMULTI;
3032 dev->allmulti -= inc;
3033 printk(KERN_WARNING "%s: allmulti touches roof, "
3034 "set allmulti failed, allmulti feature of "
3035 "device might be broken.\n", dev->name);
3039 if (dev->flags ^ old_flags) {
3040 dev_change_rx_flags(dev, IFF_ALLMULTI);
3041 dev_set_rx_mode(dev);
3047 * Upload unicast and multicast address lists to device and
3048 * configure RX filtering. When the device doesn't support unicast
3049 * filtering it is put in promiscuous mode while unicast addresses
3052 void __dev_set_rx_mode(struct net_device *dev)
3054 /* dev_open will call this function so the list will stay sane. */
3055 if (!(dev->flags&IFF_UP))
3058 if (!netif_device_present(dev))
3061 if (dev->set_rx_mode)
3062 dev->set_rx_mode(dev);
3064 /* Unicast addresses changes may only happen under the rtnl,
3065 * therefore calling __dev_set_promiscuity here is safe.
3067 if (dev->uc_count > 0 && !dev->uc_promisc) {
3068 __dev_set_promiscuity(dev, 1);
3069 dev->uc_promisc = 1;
3070 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3071 __dev_set_promiscuity(dev, -1);
3072 dev->uc_promisc = 0;
3075 if (dev->set_multicast_list)
3076 dev->set_multicast_list(dev);
3080 void dev_set_rx_mode(struct net_device *dev)
3082 netif_addr_lock_bh(dev);
3083 __dev_set_rx_mode(dev);
3084 netif_addr_unlock_bh(dev);
3087 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3088 void *addr, int alen, int glbl)
3090 struct dev_addr_list *da;
3092 for (; (da = *list) != NULL; list = &da->next) {
3093 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3094 alen == da->da_addrlen) {
3096 int old_glbl = da->da_gusers;
3113 int __dev_addr_add(struct dev_addr_list **list, int *count,
3114 void *addr, int alen, int glbl)
3116 struct dev_addr_list *da;
3118 for (da = *list; da != NULL; da = da->next) {
3119 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3120 da->da_addrlen == alen) {
3122 int old_glbl = da->da_gusers;
3132 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3135 memcpy(da->da_addr, addr, alen);
3136 da->da_addrlen = alen;
3138 da->da_gusers = glbl ? 1 : 0;
3146 * dev_unicast_delete - Release secondary unicast address.
3148 * @addr: address to delete
3149 * @alen: length of @addr
3151 * Release reference to a secondary unicast address and remove it
3152 * from the device if the reference count drops to zero.
3154 * The caller must hold the rtnl_mutex.
3156 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3162 netif_addr_lock_bh(dev);
3163 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3165 __dev_set_rx_mode(dev);
3166 netif_addr_unlock_bh(dev);
3169 EXPORT_SYMBOL(dev_unicast_delete);
3172 * dev_unicast_add - add a secondary unicast address
3174 * @addr: address to add
3175 * @alen: length of @addr
3177 * Add a secondary unicast address to the device or increase
3178 * the reference count if it already exists.
3180 * The caller must hold the rtnl_mutex.
3182 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3188 netif_addr_lock_bh(dev);
3189 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3191 __dev_set_rx_mode(dev);
3192 netif_addr_unlock_bh(dev);
3195 EXPORT_SYMBOL(dev_unicast_add);
3197 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3198 struct dev_addr_list **from, int *from_count)
3200 struct dev_addr_list *da, *next;
3204 while (da != NULL) {
3206 if (!da->da_synced) {
3207 err = __dev_addr_add(to, to_count,
3208 da->da_addr, da->da_addrlen, 0);
3213 } else if (da->da_users == 1) {
3214 __dev_addr_delete(to, to_count,
3215 da->da_addr, da->da_addrlen, 0);
3216 __dev_addr_delete(from, from_count,
3217 da->da_addr, da->da_addrlen, 0);
3224 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3225 struct dev_addr_list **from, int *from_count)
3227 struct dev_addr_list *da, *next;
3230 while (da != NULL) {
3232 if (da->da_synced) {
3233 __dev_addr_delete(to, to_count,
3234 da->da_addr, da->da_addrlen, 0);
3236 __dev_addr_delete(from, from_count,
3237 da->da_addr, da->da_addrlen, 0);
3244 * dev_unicast_sync - Synchronize device's unicast list to another device
3245 * @to: destination device
3246 * @from: source device
3248 * Add newly added addresses to the destination device and release
3249 * addresses that have no users left. The source device must be
3250 * locked by netif_tx_lock_bh.
3252 * This function is intended to be called from the dev->set_rx_mode
3253 * function of layered software devices.
3255 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3259 netif_addr_lock_bh(to);
3260 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3261 &from->uc_list, &from->uc_count);
3263 __dev_set_rx_mode(to);
3264 netif_addr_unlock_bh(to);
3267 EXPORT_SYMBOL(dev_unicast_sync);
3270 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3271 * @to: destination device
3272 * @from: source device
3274 * Remove all addresses that were added to the destination device by
3275 * dev_unicast_sync(). This function is intended to be called from the
3276 * dev->stop function of layered software devices.
3278 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3280 netif_addr_lock_bh(from);
3281 netif_addr_lock(to);
3283 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3284 &from->uc_list, &from->uc_count);
3285 __dev_set_rx_mode(to);
3287 netif_addr_unlock(to);
3288 netif_addr_unlock_bh(from);
3290 EXPORT_SYMBOL(dev_unicast_unsync);
3292 static void __dev_addr_discard(struct dev_addr_list **list)
3294 struct dev_addr_list *tmp;
3296 while (*list != NULL) {
3299 if (tmp->da_users > tmp->da_gusers)
3300 printk("__dev_addr_discard: address leakage! "
3301 "da_users=%d\n", tmp->da_users);
3306 static void dev_addr_discard(struct net_device *dev)
3308 netif_addr_lock_bh(dev);
3310 __dev_addr_discard(&dev->uc_list);
3313 __dev_addr_discard(&dev->mc_list);
3316 netif_addr_unlock_bh(dev);
3320 * dev_get_flags - get flags reported to userspace
3323 * Get the combination of flag bits exported through APIs to userspace.
3325 unsigned dev_get_flags(const struct net_device *dev)
3329 flags = (dev->flags & ~(IFF_PROMISC |
3334 (dev->gflags & (IFF_PROMISC |
3337 if (netif_running(dev)) {
3338 if (netif_oper_up(dev))
3339 flags |= IFF_RUNNING;
3340 if (netif_carrier_ok(dev))
3341 flags |= IFF_LOWER_UP;
3342 if (netif_dormant(dev))
3343 flags |= IFF_DORMANT;
3350 * dev_change_flags - change device settings
3352 * @flags: device state flags
3354 * Change settings on device based state flags. The flags are
3355 * in the userspace exported format.
3357 int dev_change_flags(struct net_device *dev, unsigned flags)
3360 int old_flags = dev->flags;
3365 * Set the flags on our device.
3368 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3369 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3371 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3375 * Load in the correct multicast list now the flags have changed.
3378 if ((old_flags ^ flags) & IFF_MULTICAST)
3379 dev_change_rx_flags(dev, IFF_MULTICAST);
3381 dev_set_rx_mode(dev);
3384 * Have we downed the interface. We handle IFF_UP ourselves
3385 * according to user attempts to set it, rather than blindly
3390 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3391 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3394 dev_set_rx_mode(dev);
3397 if (dev->flags & IFF_UP &&
3398 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3400 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3402 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3403 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3404 dev->gflags ^= IFF_PROMISC;
3405 dev_set_promiscuity(dev, inc);
3408 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3409 is important. Some (broken) drivers set IFF_PROMISC, when
3410 IFF_ALLMULTI is requested not asking us and not reporting.
3412 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3413 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3414 dev->gflags ^= IFF_ALLMULTI;
3415 dev_set_allmulti(dev, inc);
3418 /* Exclude state transition flags, already notified */
3419 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3421 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3427 * dev_set_mtu - Change maximum transfer unit
3429 * @new_mtu: new transfer unit
3431 * Change the maximum transfer size of the network device.
3433 int dev_set_mtu(struct net_device *dev, int new_mtu)
3437 if (new_mtu == dev->mtu)
3440 /* MTU must be positive. */
3444 if (!netif_device_present(dev))
3448 if (dev->change_mtu)
3449 err = dev->change_mtu(dev, new_mtu);
3452 if (!err && dev->flags & IFF_UP)
3453 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3458 * dev_set_mac_address - Change Media Access Control Address
3462 * Change the hardware (MAC) address of the device
3464 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3468 if (!dev->set_mac_address)
3470 if (sa->sa_family != dev->type)
3472 if (!netif_device_present(dev))
3474 err = dev->set_mac_address(dev, sa);
3476 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3481 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3483 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3486 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3492 case SIOCGIFFLAGS: /* Get interface flags */
3493 ifr->ifr_flags = dev_get_flags(dev);
3496 case SIOCGIFMETRIC: /* Get the metric on the interface
3497 (currently unused) */
3498 ifr->ifr_metric = 0;
3501 case SIOCGIFMTU: /* Get the MTU of a device */
3502 ifr->ifr_mtu = dev->mtu;
3507 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3509 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3510 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3511 ifr->ifr_hwaddr.sa_family = dev->type;
3519 ifr->ifr_map.mem_start = dev->mem_start;
3520 ifr->ifr_map.mem_end = dev->mem_end;
3521 ifr->ifr_map.base_addr = dev->base_addr;
3522 ifr->ifr_map.irq = dev->irq;
3523 ifr->ifr_map.dma = dev->dma;
3524 ifr->ifr_map.port = dev->if_port;
3528 ifr->ifr_ifindex = dev->ifindex;
3532 ifr->ifr_qlen = dev->tx_queue_len;
3536 /* dev_ioctl() should ensure this case
3548 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3550 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3553 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3559 case SIOCSIFFLAGS: /* Set interface flags */
3560 return dev_change_flags(dev, ifr->ifr_flags);
3562 case SIOCSIFMETRIC: /* Set the metric on the interface
3563 (currently unused) */
3566 case SIOCSIFMTU: /* Set the MTU of a device */
3567 return dev_set_mtu(dev, ifr->ifr_mtu);
3570 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3572 case SIOCSIFHWBROADCAST:
3573 if (ifr->ifr_hwaddr.sa_family != dev->type)
3575 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3576 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3577 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3581 if (dev->set_config) {
3582 if (!netif_device_present(dev))
3584 return dev->set_config(dev, &ifr->ifr_map);
3589 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3590 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3592 if (!netif_device_present(dev))
3594 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3598 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3599 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3601 if (!netif_device_present(dev))
3603 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3607 if (ifr->ifr_qlen < 0)
3609 dev->tx_queue_len = ifr->ifr_qlen;
3613 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3614 return dev_change_name(dev, ifr->ifr_newname);
3617 * Unknown or private ioctl
3621 if ((cmd >= SIOCDEVPRIVATE &&
3622 cmd <= SIOCDEVPRIVATE + 15) ||
3623 cmd == SIOCBONDENSLAVE ||
3624 cmd == SIOCBONDRELEASE ||
3625 cmd == SIOCBONDSETHWADDR ||
3626 cmd == SIOCBONDSLAVEINFOQUERY ||
3627 cmd == SIOCBONDINFOQUERY ||
3628 cmd == SIOCBONDCHANGEACTIVE ||
3629 cmd == SIOCGMIIPHY ||
3630 cmd == SIOCGMIIREG ||
3631 cmd == SIOCSMIIREG ||
3632 cmd == SIOCBRADDIF ||
3633 cmd == SIOCBRDELIF ||
3634 cmd == SIOCWANDEV) {
3636 if (dev->do_ioctl) {
3637 if (netif_device_present(dev))
3638 err = dev->do_ioctl(dev, ifr,
3651 * This function handles all "interface"-type I/O control requests. The actual
3652 * 'doing' part of this is dev_ifsioc above.
3656 * dev_ioctl - network device ioctl
3657 * @net: the applicable net namespace
3658 * @cmd: command to issue
3659 * @arg: pointer to a struct ifreq in user space
3661 * Issue ioctl functions to devices. This is normally called by the
3662 * user space syscall interfaces but can sometimes be useful for
3663 * other purposes. The return value is the return from the syscall if
3664 * positive or a negative errno code on error.
3667 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3673 /* One special case: SIOCGIFCONF takes ifconf argument
3674 and requires shared lock, because it sleeps writing
3678 if (cmd == SIOCGIFCONF) {
3680 ret = dev_ifconf(net, (char __user *) arg);
3684 if (cmd == SIOCGIFNAME)
3685 return dev_ifname(net, (struct ifreq __user *)arg);
3687 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3690 ifr.ifr_name[IFNAMSIZ-1] = 0;
3692 colon = strchr(ifr.ifr_name, ':');
3697 * See which interface the caller is talking about.
3702 * These ioctl calls:
3703 * - can be done by all.
3704 * - atomic and do not require locking.
3715 dev_load(net, ifr.ifr_name);
3716 read_lock(&dev_base_lock);
3717 ret = dev_ifsioc_locked(net, &ifr, cmd);
3718 read_unlock(&dev_base_lock);
3722 if (copy_to_user(arg, &ifr,
3723 sizeof(struct ifreq)))
3729 dev_load(net, ifr.ifr_name);
3731 ret = dev_ethtool(net, &ifr);
3736 if (copy_to_user(arg, &ifr,
3737 sizeof(struct ifreq)))
3743 * These ioctl calls:
3744 * - require superuser power.
3745 * - require strict serialization.
3751 if (!capable(CAP_NET_ADMIN))
3753 dev_load(net, ifr.ifr_name);
3755 ret = dev_ifsioc(net, &ifr, cmd);
3760 if (copy_to_user(arg, &ifr,
3761 sizeof(struct ifreq)))
3767 * These ioctl calls:
3768 * - require superuser power.
3769 * - require strict serialization.
3770 * - do not return a value
3780 case SIOCSIFHWBROADCAST:
3783 case SIOCBONDENSLAVE:
3784 case SIOCBONDRELEASE:
3785 case SIOCBONDSETHWADDR:
3786 case SIOCBONDCHANGEACTIVE:
3789 if (!capable(CAP_NET_ADMIN))
3792 case SIOCBONDSLAVEINFOQUERY:
3793 case SIOCBONDINFOQUERY:
3794 dev_load(net, ifr.ifr_name);
3796 ret = dev_ifsioc(net, &ifr, cmd);
3801 /* Get the per device memory space. We can add this but
3802 * currently do not support it */
3804 /* Set the per device memory buffer space.
3805 * Not applicable in our case */
3810 * Unknown or private ioctl.
3813 if (cmd == SIOCWANDEV ||
3814 (cmd >= SIOCDEVPRIVATE &&
3815 cmd <= SIOCDEVPRIVATE + 15)) {
3816 dev_load(net, ifr.ifr_name);
3818 ret = dev_ifsioc(net, &ifr, cmd);
3820 if (!ret && copy_to_user(arg, &ifr,
3821 sizeof(struct ifreq)))
3825 /* Take care of Wireless Extensions */
3826 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3827 return wext_handle_ioctl(net, &ifr, cmd, arg);
3834 * dev_new_index - allocate an ifindex
3835 * @net: the applicable net namespace
3837 * Returns a suitable unique value for a new device interface
3838 * number. The caller must hold the rtnl semaphore or the
3839 * dev_base_lock to be sure it remains unique.
3841 static int dev_new_index(struct net *net)
3847 if (!__dev_get_by_index(net, ifindex))
3852 /* Delayed registration/unregisteration */
3853 static LIST_HEAD(net_todo_list);
3855 static void net_set_todo(struct net_device *dev)
3857 list_add_tail(&dev->todo_list, &net_todo_list);
3860 static void rollback_registered(struct net_device *dev)
3862 BUG_ON(dev_boot_phase);
3865 /* Some devices call without registering for initialization unwind. */
3866 if (dev->reg_state == NETREG_UNINITIALIZED) {
3867 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3868 "was registered\n", dev->name, dev);
3874 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3876 /* If device is running, close it first. */
3879 /* And unlink it from device chain. */
3880 unlist_netdevice(dev);
3882 dev->reg_state = NETREG_UNREGISTERING;
3886 /* Shutdown queueing discipline. */
3890 /* Notify protocols, that we are about to destroy
3891 this device. They should clean all the things.
3893 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3896 * Flush the unicast and multicast chains
3898 dev_addr_discard(dev);
3903 /* Notifier chain MUST detach us from master device. */
3904 WARN_ON(dev->master);
3906 /* Remove entries from kobject tree */
3907 netdev_unregister_kobject(dev);
3914 static void __netdev_init_queue_locks_one(struct net_device *dev,
3915 struct netdev_queue *dev_queue,
3918 spin_lock_init(&dev_queue->_xmit_lock);
3919 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
3920 dev_queue->xmit_lock_owner = -1;
3923 static void netdev_init_queue_locks(struct net_device *dev)
3925 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
3926 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
3929 unsigned long netdev_fix_features(unsigned long features, const char *name)
3931 /* Fix illegal SG+CSUM combinations. */
3932 if ((features & NETIF_F_SG) &&
3933 !(features & NETIF_F_ALL_CSUM)) {
3935 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
3936 "checksum feature.\n", name);
3937 features &= ~NETIF_F_SG;
3940 /* TSO requires that SG is present as well. */
3941 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
3943 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
3944 "SG feature.\n", name);
3945 features &= ~NETIF_F_TSO;
3948 if (features & NETIF_F_UFO) {
3949 if (!(features & NETIF_F_GEN_CSUM)) {
3951 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
3952 "since no NETIF_F_HW_CSUM feature.\n",
3954 features &= ~NETIF_F_UFO;
3957 if (!(features & NETIF_F_SG)) {
3959 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
3960 "since no NETIF_F_SG feature.\n", name);
3961 features &= ~NETIF_F_UFO;
3967 EXPORT_SYMBOL(netdev_fix_features);
3970 * register_netdevice - register a network device
3971 * @dev: device to register
3973 * Take a completed network device structure and add it to the kernel
3974 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3975 * chain. 0 is returned on success. A negative errno code is returned
3976 * on a failure to set up the device, or if the name is a duplicate.
3978 * Callers must hold the rtnl semaphore. You may want
3979 * register_netdev() instead of this.
3982 * The locking appears insufficient to guarantee two parallel registers
3983 * will not get the same name.
3986 int register_netdevice(struct net_device *dev)
3988 struct hlist_head *head;
3989 struct hlist_node *p;
3993 BUG_ON(dev_boot_phase);
3998 /* When net_device's are persistent, this will be fatal. */
3999 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4000 BUG_ON(!dev_net(dev));
4003 spin_lock_init(&dev->addr_list_lock);
4004 netdev_set_addr_lockdep_class(dev);
4005 netdev_init_queue_locks(dev);
4009 /* Init, if this function is available */
4011 ret = dev->init(dev);
4019 if (!dev_valid_name(dev->name)) {
4024 dev->ifindex = dev_new_index(net);
4025 if (dev->iflink == -1)
4026 dev->iflink = dev->ifindex;
4028 /* Check for existence of name */
4029 head = dev_name_hash(net, dev->name);
4030 hlist_for_each(p, head) {
4031 struct net_device *d
4032 = hlist_entry(p, struct net_device, name_hlist);
4033 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4039 /* Fix illegal checksum combinations */
4040 if ((dev->features & NETIF_F_HW_CSUM) &&
4041 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4042 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4044 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4047 if ((dev->features & NETIF_F_NO_CSUM) &&
4048 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4049 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4051 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4054 dev->features = netdev_fix_features(dev->features, dev->name);
4056 /* Enable software GSO if SG is supported. */
4057 if (dev->features & NETIF_F_SG)
4058 dev->features |= NETIF_F_GSO;
4060 netdev_initialize_kobject(dev);
4061 ret = netdev_register_kobject(dev);
4064 dev->reg_state = NETREG_REGISTERED;
4067 * Default initial state at registry is that the
4068 * device is present.
4071 set_bit(__LINK_STATE_PRESENT, &dev->state);
4073 dev_init_scheduler(dev);
4075 list_netdevice(dev);
4077 /* Notify protocols, that a new device appeared. */
4078 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4079 ret = notifier_to_errno(ret);
4081 rollback_registered(dev);
4082 dev->reg_state = NETREG_UNREGISTERED;
4095 * register_netdev - register a network device
4096 * @dev: device to register
4098 * Take a completed network device structure and add it to the kernel
4099 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4100 * chain. 0 is returned on success. A negative errno code is returned
4101 * on a failure to set up the device, or if the name is a duplicate.
4103 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4104 * and expands the device name if you passed a format string to
4107 int register_netdev(struct net_device *dev)
4114 * If the name is a format string the caller wants us to do a
4117 if (strchr(dev->name, '%')) {
4118 err = dev_alloc_name(dev, dev->name);
4123 err = register_netdevice(dev);
4128 EXPORT_SYMBOL(register_netdev);
4131 * netdev_wait_allrefs - wait until all references are gone.
4133 * This is called when unregistering network devices.
4135 * Any protocol or device that holds a reference should register
4136 * for netdevice notification, and cleanup and put back the
4137 * reference if they receive an UNREGISTER event.
4138 * We can get stuck here if buggy protocols don't correctly
4141 static void netdev_wait_allrefs(struct net_device *dev)
4143 unsigned long rebroadcast_time, warning_time;
4145 rebroadcast_time = warning_time = jiffies;
4146 while (atomic_read(&dev->refcnt) != 0) {
4147 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4150 /* Rebroadcast unregister notification */
4151 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4153 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4155 /* We must not have linkwatch events
4156 * pending on unregister. If this
4157 * happens, we simply run the queue
4158 * unscheduled, resulting in a noop
4161 linkwatch_run_queue();
4166 rebroadcast_time = jiffies;
4171 if (time_after(jiffies, warning_time + 10 * HZ)) {
4172 printk(KERN_EMERG "unregister_netdevice: "
4173 "waiting for %s to become free. Usage "
4175 dev->name, atomic_read(&dev->refcnt));
4176 warning_time = jiffies;
4185 * register_netdevice(x1);
4186 * register_netdevice(x2);
4188 * unregister_netdevice(y1);
4189 * unregister_netdevice(y2);
4195 * We are invoked by rtnl_unlock().
4196 * This allows us to deal with problems:
4197 * 1) We can delete sysfs objects which invoke hotplug
4198 * without deadlocking with linkwatch via keventd.
4199 * 2) Since we run with the RTNL semaphore not held, we can sleep
4200 * safely in order to wait for the netdev refcnt to drop to zero.
4202 * We must not return until all unregister events added during
4203 * the interval the lock was held have been completed.
4205 void netdev_run_todo(void)
4207 struct list_head list;
4209 /* Snapshot list, allow later requests */
4210 list_replace_init(&net_todo_list, &list);
4214 while (!list_empty(&list)) {
4215 struct net_device *dev
4216 = list_entry(list.next, struct net_device, todo_list);
4217 list_del(&dev->todo_list);
4219 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4220 printk(KERN_ERR "network todo '%s' but state %d\n",
4221 dev->name, dev->reg_state);
4226 dev->reg_state = NETREG_UNREGISTERED;
4228 on_each_cpu(flush_backlog, dev, 1);
4230 netdev_wait_allrefs(dev);
4233 BUG_ON(atomic_read(&dev->refcnt));
4234 WARN_ON(dev->ip_ptr);
4235 WARN_ON(dev->ip6_ptr);
4236 WARN_ON(dev->dn_ptr);
4238 if (dev->destructor)
4239 dev->destructor(dev);
4241 /* Free network device */
4242 kobject_put(&dev->dev.kobj);
4246 static struct net_device_stats *internal_stats(struct net_device *dev)
4251 static void netdev_init_one_queue(struct net_device *dev,
4252 struct netdev_queue *queue,
4258 static void netdev_init_queues(struct net_device *dev)
4260 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4261 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4262 spin_lock_init(&dev->tx_global_lock);
4266 * alloc_netdev_mq - allocate network device
4267 * @sizeof_priv: size of private data to allocate space for
4268 * @name: device name format string
4269 * @setup: callback to initialize device
4270 * @queue_count: the number of subqueues to allocate
4272 * Allocates a struct net_device with private data area for driver use
4273 * and performs basic initialization. Also allocates subquue structs
4274 * for each queue on the device at the end of the netdevice.
4276 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4277 void (*setup)(struct net_device *), unsigned int queue_count)
4279 struct netdev_queue *tx;
4280 struct net_device *dev;
4284 BUG_ON(strlen(name) >= sizeof(dev->name));
4286 alloc_size = sizeof(struct net_device);
4288 /* ensure 32-byte alignment of private area */
4289 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4290 alloc_size += sizeof_priv;
4292 /* ensure 32-byte alignment of whole construct */
4293 alloc_size += NETDEV_ALIGN_CONST;
4295 p = kzalloc(alloc_size, GFP_KERNEL);
4297 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4301 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
4303 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4309 dev = (struct net_device *)
4310 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4311 dev->padded = (char *)dev - (char *)p;
4312 dev_net_set(dev, &init_net);
4315 dev->num_tx_queues = queue_count;
4316 dev->real_num_tx_queues = queue_count;
4319 dev->priv = ((char *)dev +
4320 ((sizeof(struct net_device) + NETDEV_ALIGN_CONST)
4321 & ~NETDEV_ALIGN_CONST));
4324 dev->gso_max_size = GSO_MAX_SIZE;
4326 netdev_init_queues(dev);
4328 dev->get_stats = internal_stats;
4329 netpoll_netdev_init(dev);
4331 strcpy(dev->name, name);
4334 EXPORT_SYMBOL(alloc_netdev_mq);
4337 * free_netdev - free network device
4340 * This function does the last stage of destroying an allocated device
4341 * interface. The reference to the device object is released.
4342 * If this is the last reference then it will be freed.
4344 void free_netdev(struct net_device *dev)
4346 release_net(dev_net(dev));
4350 /* Compatibility with error handling in drivers */
4351 if (dev->reg_state == NETREG_UNINITIALIZED) {
4352 kfree((char *)dev - dev->padded);
4356 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4357 dev->reg_state = NETREG_RELEASED;
4359 /* will free via device release */
4360 put_device(&dev->dev);
4364 * synchronize_net - Synchronize with packet receive processing
4366 * Wait for packets currently being received to be done.
4367 * Does not block later packets from starting.
4369 void synchronize_net(void)
4376 * unregister_netdevice - remove device from the kernel
4379 * This function shuts down a device interface and removes it
4380 * from the kernel tables.
4382 * Callers must hold the rtnl semaphore. You may want
4383 * unregister_netdev() instead of this.
4386 void unregister_netdevice(struct net_device *dev)
4390 rollback_registered(dev);
4391 /* Finish processing unregister after unlock */
4396 * unregister_netdev - remove device from the kernel
4399 * This function shuts down a device interface and removes it
4400 * from the kernel tables.
4402 * This is just a wrapper for unregister_netdevice that takes
4403 * the rtnl semaphore. In general you want to use this and not
4404 * unregister_netdevice.
4406 void unregister_netdev(struct net_device *dev)
4409 unregister_netdevice(dev);
4413 EXPORT_SYMBOL(unregister_netdev);
4416 * dev_change_net_namespace - move device to different nethost namespace
4418 * @net: network namespace
4419 * @pat: If not NULL name pattern to try if the current device name
4420 * is already taken in the destination network namespace.
4422 * This function shuts down a device interface and moves it
4423 * to a new network namespace. On success 0 is returned, on
4424 * a failure a netagive errno code is returned.
4426 * Callers must hold the rtnl semaphore.
4429 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4432 const char *destname;
4437 /* Don't allow namespace local devices to be moved. */
4439 if (dev->features & NETIF_F_NETNS_LOCAL)
4443 /* Don't allow real devices to be moved when sysfs
4447 if (dev->dev.parent)
4451 /* Ensure the device has been registrered */
4453 if (dev->reg_state != NETREG_REGISTERED)
4456 /* Get out if there is nothing todo */
4458 if (net_eq(dev_net(dev), net))
4461 /* Pick the destination device name, and ensure
4462 * we can use it in the destination network namespace.
4465 destname = dev->name;
4466 if (__dev_get_by_name(net, destname)) {
4467 /* We get here if we can't use the current device name */
4470 if (!dev_valid_name(pat))
4472 if (strchr(pat, '%')) {
4473 if (__dev_alloc_name(net, pat, buf) < 0)
4478 if (__dev_get_by_name(net, destname))
4483 * And now a mini version of register_netdevice unregister_netdevice.
4486 /* If device is running close it first. */
4489 /* And unlink it from device chain */
4491 unlist_netdevice(dev);
4495 /* Shutdown queueing discipline. */
4498 /* Notify protocols, that we are about to destroy
4499 this device. They should clean all the things.
4501 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4504 * Flush the unicast and multicast chains
4506 dev_addr_discard(dev);
4508 netdev_unregister_kobject(dev);
4510 /* Actually switch the network namespace */
4511 dev_net_set(dev, net);
4513 /* Assign the new device name */
4514 if (destname != dev->name)
4515 strcpy(dev->name, destname);
4517 /* If there is an ifindex conflict assign a new one */
4518 if (__dev_get_by_index(net, dev->ifindex)) {
4519 int iflink = (dev->iflink == dev->ifindex);
4520 dev->ifindex = dev_new_index(net);
4522 dev->iflink = dev->ifindex;
4525 /* Fixup kobjects */
4526 err = netdev_register_kobject(dev);
4529 /* Add the device back in the hashes */
4530 list_netdevice(dev);
4532 /* Notify protocols, that a new device appeared. */
4533 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4541 static int dev_cpu_callback(struct notifier_block *nfb,
4542 unsigned long action,
4545 struct sk_buff **list_skb;
4546 struct Qdisc **list_net;
4547 struct sk_buff *skb;
4548 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4549 struct softnet_data *sd, *oldsd;
4551 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4554 local_irq_disable();
4555 cpu = smp_processor_id();
4556 sd = &per_cpu(softnet_data, cpu);
4557 oldsd = &per_cpu(softnet_data, oldcpu);
4559 /* Find end of our completion_queue. */
4560 list_skb = &sd->completion_queue;
4562 list_skb = &(*list_skb)->next;
4563 /* Append completion queue from offline CPU. */
4564 *list_skb = oldsd->completion_queue;
4565 oldsd->completion_queue = NULL;
4567 /* Find end of our output_queue. */
4568 list_net = &sd->output_queue;
4570 list_net = &(*list_net)->next_sched;
4571 /* Append output queue from offline CPU. */
4572 *list_net = oldsd->output_queue;
4573 oldsd->output_queue = NULL;
4575 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4578 /* Process offline CPU's input_pkt_queue */
4579 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4585 #ifdef CONFIG_NET_DMA
4587 * net_dma_rebalance - try to maintain one DMA channel per CPU
4588 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4590 * This is called when the number of channels allocated to the net_dma client
4591 * changes. The net_dma client tries to have one DMA channel per CPU.
4594 static void net_dma_rebalance(struct net_dma *net_dma)
4596 unsigned int cpu, i, n, chan_idx;
4597 struct dma_chan *chan;
4599 if (cpus_empty(net_dma->channel_mask)) {
4600 for_each_online_cpu(cpu)
4601 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4606 cpu = first_cpu(cpu_online_map);
4608 for_each_cpu_mask_nr(chan_idx, net_dma->channel_mask) {
4609 chan = net_dma->channels[chan_idx];
4611 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4612 + (i < (num_online_cpus() %
4613 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4616 per_cpu(softnet_data, cpu).net_dma = chan;
4617 cpu = next_cpu(cpu, cpu_online_map);
4625 * netdev_dma_event - event callback for the net_dma_client
4626 * @client: should always be net_dma_client
4627 * @chan: DMA channel for the event
4628 * @state: DMA state to be handled
4630 static enum dma_state_client
4631 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4632 enum dma_state state)
4634 int i, found = 0, pos = -1;
4635 struct net_dma *net_dma =
4636 container_of(client, struct net_dma, client);
4637 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4639 spin_lock(&net_dma->lock);
4641 case DMA_RESOURCE_AVAILABLE:
4642 for (i = 0; i < nr_cpu_ids; i++)
4643 if (net_dma->channels[i] == chan) {
4646 } else if (net_dma->channels[i] == NULL && pos < 0)
4649 if (!found && pos >= 0) {
4651 net_dma->channels[pos] = chan;
4652 cpu_set(pos, net_dma->channel_mask);
4653 net_dma_rebalance(net_dma);
4656 case DMA_RESOURCE_REMOVED:
4657 for (i = 0; i < nr_cpu_ids; i++)
4658 if (net_dma->channels[i] == chan) {
4666 cpu_clear(pos, net_dma->channel_mask);
4667 net_dma->channels[i] = NULL;
4668 net_dma_rebalance(net_dma);
4674 spin_unlock(&net_dma->lock);
4680 * netdev_dma_register - register the networking subsystem as a DMA client
4682 static int __init netdev_dma_register(void)
4684 net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma),
4686 if (unlikely(!net_dma.channels)) {
4688 "netdev_dma: no memory for net_dma.channels\n");
4691 spin_lock_init(&net_dma.lock);
4692 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4693 dma_async_client_register(&net_dma.client);
4694 dma_async_client_chan_request(&net_dma.client);
4699 static int __init netdev_dma_register(void) { return -ENODEV; }
4700 #endif /* CONFIG_NET_DMA */
4703 * netdev_increment_features - increment feature set by one
4704 * @all: current feature set
4705 * @one: new feature set
4706 * @mask: mask feature set
4708 * Computes a new feature set after adding a device with feature set
4709 * @one to the master device with current feature set @all. Will not
4710 * enable anything that is off in @mask. Returns the new feature set.
4712 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
4715 /* If device needs checksumming, downgrade to it. */
4716 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4717 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
4718 else if (mask & NETIF_F_ALL_CSUM) {
4719 /* If one device supports v4/v6 checksumming, set for all. */
4720 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
4721 !(all & NETIF_F_GEN_CSUM)) {
4722 all &= ~NETIF_F_ALL_CSUM;
4723 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
4726 /* If one device supports hw checksumming, set for all. */
4727 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
4728 all &= ~NETIF_F_ALL_CSUM;
4729 all |= NETIF_F_HW_CSUM;
4733 one |= NETIF_F_ALL_CSUM;
4735 one |= all & NETIF_F_ONE_FOR_ALL;
4736 all &= one | NETIF_F_LLTX | NETIF_F_GSO;
4737 all |= one & mask & NETIF_F_ONE_FOR_ALL;
4741 EXPORT_SYMBOL(netdev_increment_features);
4743 static struct hlist_head *netdev_create_hash(void)
4746 struct hlist_head *hash;
4748 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4750 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4751 INIT_HLIST_HEAD(&hash[i]);
4756 /* Initialize per network namespace state */
4757 static int __net_init netdev_init(struct net *net)
4759 INIT_LIST_HEAD(&net->dev_base_head);
4761 net->dev_name_head = netdev_create_hash();
4762 if (net->dev_name_head == NULL)
4765 net->dev_index_head = netdev_create_hash();
4766 if (net->dev_index_head == NULL)
4772 kfree(net->dev_name_head);
4778 * netdev_drivername - network driver for the device
4779 * @dev: network device
4780 * @buffer: buffer for resulting name
4781 * @len: size of buffer
4783 * Determine network driver for device.
4785 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
4787 const struct device_driver *driver;
4788 const struct device *parent;
4790 if (len <= 0 || !buffer)
4794 parent = dev->dev.parent;
4799 driver = parent->driver;
4800 if (driver && driver->name)
4801 strlcpy(buffer, driver->name, len);
4805 static void __net_exit netdev_exit(struct net *net)
4807 kfree(net->dev_name_head);
4808 kfree(net->dev_index_head);
4811 static struct pernet_operations __net_initdata netdev_net_ops = {
4812 .init = netdev_init,
4813 .exit = netdev_exit,
4816 static void __net_exit default_device_exit(struct net *net)
4818 struct net_device *dev, *next;
4820 * Push all migratable of the network devices back to the
4821 * initial network namespace
4824 for_each_netdev_safe(net, dev, next) {
4826 char fb_name[IFNAMSIZ];
4828 /* Ignore unmoveable devices (i.e. loopback) */
4829 if (dev->features & NETIF_F_NETNS_LOCAL)
4832 /* Delete virtual devices */
4833 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
4834 dev->rtnl_link_ops->dellink(dev);
4838 /* Push remaing network devices to init_net */
4839 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
4840 err = dev_change_net_namespace(dev, &init_net, fb_name);
4842 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
4843 __func__, dev->name, err);
4850 static struct pernet_operations __net_initdata default_device_ops = {
4851 .exit = default_device_exit,
4855 * Initialize the DEV module. At boot time this walks the device list and
4856 * unhooks any devices that fail to initialise (normally hardware not
4857 * present) and leaves us with a valid list of present and active devices.
4862 * This is called single threaded during boot, so no need
4863 * to take the rtnl semaphore.
4865 static int __init net_dev_init(void)
4867 int i, rc = -ENOMEM;
4869 BUG_ON(!dev_boot_phase);
4871 if (dev_proc_init())
4874 if (netdev_kobject_init())
4877 INIT_LIST_HEAD(&ptype_all);
4878 for (i = 0; i < PTYPE_HASH_SIZE; i++)
4879 INIT_LIST_HEAD(&ptype_base[i]);
4881 if (register_pernet_subsys(&netdev_net_ops))
4885 * Initialise the packet receive queues.
4888 for_each_possible_cpu(i) {
4889 struct softnet_data *queue;
4891 queue = &per_cpu(softnet_data, i);
4892 skb_queue_head_init(&queue->input_pkt_queue);
4893 queue->completion_queue = NULL;
4894 INIT_LIST_HEAD(&queue->poll_list);
4896 queue->backlog.poll = process_backlog;
4897 queue->backlog.weight = weight_p;
4902 /* The loopback device is special if any other network devices
4903 * is present in a network namespace the loopback device must
4904 * be present. Since we now dynamically allocate and free the
4905 * loopback device ensure this invariant is maintained by
4906 * keeping the loopback device as the first device on the
4907 * list of network devices. Ensuring the loopback devices
4908 * is the first device that appears and the last network device
4911 if (register_pernet_device(&loopback_net_ops))
4914 if (register_pernet_device(&default_device_ops))
4917 netdev_dma_register();
4919 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
4920 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
4922 hotcpu_notifier(dev_cpu_callback, 0);
4930 subsys_initcall(net_dev_init);
4932 EXPORT_SYMBOL(__dev_get_by_index);
4933 EXPORT_SYMBOL(__dev_get_by_name);
4934 EXPORT_SYMBOL(__dev_remove_pack);
4935 EXPORT_SYMBOL(dev_valid_name);
4936 EXPORT_SYMBOL(dev_add_pack);
4937 EXPORT_SYMBOL(dev_alloc_name);
4938 EXPORT_SYMBOL(dev_close);
4939 EXPORT_SYMBOL(dev_get_by_flags);
4940 EXPORT_SYMBOL(dev_get_by_index);
4941 EXPORT_SYMBOL(dev_get_by_name);
4942 EXPORT_SYMBOL(dev_open);
4943 EXPORT_SYMBOL(dev_queue_xmit);
4944 EXPORT_SYMBOL(dev_remove_pack);
4945 EXPORT_SYMBOL(dev_set_allmulti);
4946 EXPORT_SYMBOL(dev_set_promiscuity);
4947 EXPORT_SYMBOL(dev_change_flags);
4948 EXPORT_SYMBOL(dev_set_mtu);
4949 EXPORT_SYMBOL(dev_set_mac_address);
4950 EXPORT_SYMBOL(free_netdev);
4951 EXPORT_SYMBOL(netdev_boot_setup_check);
4952 EXPORT_SYMBOL(netdev_set_master);
4953 EXPORT_SYMBOL(netdev_state_change);
4954 EXPORT_SYMBOL(netif_receive_skb);
4955 EXPORT_SYMBOL(netif_rx);
4956 EXPORT_SYMBOL(register_gifconf);
4957 EXPORT_SYMBOL(register_netdevice);
4958 EXPORT_SYMBOL(register_netdevice_notifier);
4959 EXPORT_SYMBOL(skb_checksum_help);
4960 EXPORT_SYMBOL(synchronize_net);
4961 EXPORT_SYMBOL(unregister_netdevice);
4962 EXPORT_SYMBOL(unregister_netdevice_notifier);
4963 EXPORT_SYMBOL(net_enable_timestamp);
4964 EXPORT_SYMBOL(net_disable_timestamp);
4965 EXPORT_SYMBOL(dev_get_flags);
4967 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4968 EXPORT_SYMBOL(br_handle_frame_hook);
4969 EXPORT_SYMBOL(br_fdb_get_hook);
4970 EXPORT_SYMBOL(br_fdb_put_hook);
4973 EXPORT_SYMBOL(dev_load);
4975 EXPORT_PER_CPU_SYMBOL(softnet_data);