2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/mutex.h>
84 #include <linux/string.h>
86 #include <linux/socket.h>
87 #include <linux/sockios.h>
88 #include <linux/errno.h>
89 #include <linux/interrupt.h>
90 #include <linux/if_ether.h>
91 #include <linux/netdevice.h>
92 #include <linux/etherdevice.h>
93 #include <linux/ethtool.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
96 #include <net/net_namespace.h>
98 #include <linux/rtnetlink.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/stat.h>
102 #include <linux/if_bridge.h>
103 #include <linux/if_macvlan.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/kallsyms.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
130 #include "net-sysfs.h"
133 * The list of packet types we will receive (as opposed to discard)
134 * and the routines to invoke.
136 * Why 16. Because with 16 the only overlap we get on a hash of the
137 * low nibble of the protocol value is RARP/SNAP/X.25.
139 * NOTE: That is no longer true with the addition of VLAN tags. Not
140 * sure which should go first, but I bet it won't make much
141 * difference if we are running VLANs. The good news is that
142 * this protocol won't be in the list unless compiled in, so
143 * the average user (w/out VLANs) will not be adversely affected.
160 #define PTYPE_HASH_SIZE (16)
161 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
163 static DEFINE_SPINLOCK(ptype_lock);
164 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
165 static struct list_head ptype_all __read_mostly; /* Taps */
167 #ifdef CONFIG_NET_DMA
169 struct dma_client client;
171 cpumask_t channel_mask;
172 struct dma_chan **channels;
175 static enum dma_state_client
176 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
177 enum dma_state state);
179 static struct net_dma net_dma = {
181 .event_callback = netdev_dma_event,
187 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
190 * Pure readers hold dev_base_lock for reading.
192 * Writers must hold the rtnl semaphore while they loop through the
193 * dev_base_head list, and hold dev_base_lock for writing when they do the
194 * actual updates. This allows pure readers to access the list even
195 * while a writer is preparing to update it.
197 * To put it another way, dev_base_lock is held for writing only to
198 * protect against pure readers; the rtnl semaphore provides the
199 * protection against other writers.
201 * See, for example usages, register_netdevice() and
202 * unregister_netdevice(), which must be called with the rtnl
205 DEFINE_RWLOCK(dev_base_lock);
207 EXPORT_SYMBOL(dev_base_lock);
209 #define NETDEV_HASHBITS 8
210 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
212 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
214 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
215 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
218 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
220 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
223 /* Device list insertion */
224 static int list_netdevice(struct net_device *dev)
226 struct net *net = dev_net(dev);
230 write_lock_bh(&dev_base_lock);
231 list_add_tail(&dev->dev_list, &net->dev_base_head);
232 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
233 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
234 write_unlock_bh(&dev_base_lock);
238 /* Device list removal */
239 static void unlist_netdevice(struct net_device *dev)
243 /* Unlink dev from the device chain */
244 write_lock_bh(&dev_base_lock);
245 list_del(&dev->dev_list);
246 hlist_del(&dev->name_hlist);
247 hlist_del(&dev->index_hlist);
248 write_unlock_bh(&dev_base_lock);
255 static RAW_NOTIFIER_HEAD(netdev_chain);
258 * Device drivers call our routines to queue packets here. We empty the
259 * queue in the local softnet handler.
262 DEFINE_PER_CPU(struct softnet_data, softnet_data);
264 #ifdef CONFIG_LOCKDEP
266 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
267 * according to dev->type
269 static const unsigned short netdev_lock_type[] =
270 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
271 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
272 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
273 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
274 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
275 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
276 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
277 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
278 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
279 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
280 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
281 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
282 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
283 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
286 static const char *netdev_lock_name[] =
287 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
288 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
289 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
290 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
291 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
292 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
293 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
294 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
295 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
296 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
297 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
298 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
299 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
300 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
303 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
304 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
306 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
310 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
311 if (netdev_lock_type[i] == dev_type)
313 /* the last key is used by default */
314 return ARRAY_SIZE(netdev_lock_type) - 1;
317 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
318 unsigned short dev_type)
322 i = netdev_lock_pos(dev_type);
323 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
324 netdev_lock_name[i]);
327 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
331 i = netdev_lock_pos(dev->type);
332 lockdep_set_class_and_name(&dev->addr_list_lock,
333 &netdev_addr_lock_key[i],
334 netdev_lock_name[i]);
337 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
338 unsigned short dev_type)
341 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
346 /*******************************************************************************
348 Protocol management and registration routines
350 *******************************************************************************/
353 * Add a protocol ID to the list. Now that the input handler is
354 * smarter we can dispense with all the messy stuff that used to be
357 * BEWARE!!! Protocol handlers, mangling input packets,
358 * MUST BE last in hash buckets and checking protocol handlers
359 * MUST start from promiscuous ptype_all chain in net_bh.
360 * It is true now, do not change it.
361 * Explanation follows: if protocol handler, mangling packet, will
362 * be the first on list, it is not able to sense, that packet
363 * is cloned and should be copied-on-write, so that it will
364 * change it and subsequent readers will get broken packet.
369 * dev_add_pack - add packet handler
370 * @pt: packet type declaration
372 * Add a protocol handler to the networking stack. The passed &packet_type
373 * is linked into kernel lists and may not be freed until it has been
374 * removed from the kernel lists.
376 * This call does not sleep therefore it can not
377 * guarantee all CPU's that are in middle of receiving packets
378 * will see the new packet type (until the next received packet).
381 void dev_add_pack(struct packet_type *pt)
385 spin_lock_bh(&ptype_lock);
386 if (pt->type == htons(ETH_P_ALL))
387 list_add_rcu(&pt->list, &ptype_all);
389 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
390 list_add_rcu(&pt->list, &ptype_base[hash]);
392 spin_unlock_bh(&ptype_lock);
396 * __dev_remove_pack - remove packet handler
397 * @pt: packet type declaration
399 * Remove a protocol handler that was previously added to the kernel
400 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
401 * from the kernel lists and can be freed or reused once this function
404 * The packet type might still be in use by receivers
405 * and must not be freed until after all the CPU's have gone
406 * through a quiescent state.
408 void __dev_remove_pack(struct packet_type *pt)
410 struct list_head *head;
411 struct packet_type *pt1;
413 spin_lock_bh(&ptype_lock);
415 if (pt->type == htons(ETH_P_ALL))
418 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
420 list_for_each_entry(pt1, head, list) {
422 list_del_rcu(&pt->list);
427 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
429 spin_unlock_bh(&ptype_lock);
432 * dev_remove_pack - remove packet handler
433 * @pt: packet type declaration
435 * Remove a protocol handler that was previously added to the kernel
436 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
437 * from the kernel lists and can be freed or reused once this function
440 * This call sleeps to guarantee that no CPU is looking at the packet
443 void dev_remove_pack(struct packet_type *pt)
445 __dev_remove_pack(pt);
450 /******************************************************************************
452 Device Boot-time Settings Routines
454 *******************************************************************************/
456 /* Boot time configuration table */
457 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
460 * netdev_boot_setup_add - add new setup entry
461 * @name: name of the device
462 * @map: configured settings for the device
464 * Adds new setup entry to the dev_boot_setup list. The function
465 * returns 0 on error and 1 on success. This is a generic routine to
468 static int netdev_boot_setup_add(char *name, struct ifmap *map)
470 struct netdev_boot_setup *s;
474 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
475 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
476 memset(s[i].name, 0, sizeof(s[i].name));
477 strlcpy(s[i].name, name, IFNAMSIZ);
478 memcpy(&s[i].map, map, sizeof(s[i].map));
483 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
487 * netdev_boot_setup_check - check boot time settings
488 * @dev: the netdevice
490 * Check boot time settings for the device.
491 * The found settings are set for the device to be used
492 * later in the device probing.
493 * Returns 0 if no settings found, 1 if they are.
495 int netdev_boot_setup_check(struct net_device *dev)
497 struct netdev_boot_setup *s = dev_boot_setup;
500 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
501 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
502 !strcmp(dev->name, s[i].name)) {
503 dev->irq = s[i].map.irq;
504 dev->base_addr = s[i].map.base_addr;
505 dev->mem_start = s[i].map.mem_start;
506 dev->mem_end = s[i].map.mem_end;
515 * netdev_boot_base - get address from boot time settings
516 * @prefix: prefix for network device
517 * @unit: id for network device
519 * Check boot time settings for the base address of device.
520 * The found settings are set for the device to be used
521 * later in the device probing.
522 * Returns 0 if no settings found.
524 unsigned long netdev_boot_base(const char *prefix, int unit)
526 const struct netdev_boot_setup *s = dev_boot_setup;
530 sprintf(name, "%s%d", prefix, unit);
533 * If device already registered then return base of 1
534 * to indicate not to probe for this interface
536 if (__dev_get_by_name(&init_net, name))
539 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
540 if (!strcmp(name, s[i].name))
541 return s[i].map.base_addr;
546 * Saves at boot time configured settings for any netdevice.
548 int __init netdev_boot_setup(char *str)
553 str = get_options(str, ARRAY_SIZE(ints), ints);
558 memset(&map, 0, sizeof(map));
562 map.base_addr = ints[2];
564 map.mem_start = ints[3];
566 map.mem_end = ints[4];
568 /* Add new entry to the list */
569 return netdev_boot_setup_add(str, &map);
572 __setup("netdev=", netdev_boot_setup);
574 /*******************************************************************************
576 Device Interface Subroutines
578 *******************************************************************************/
581 * __dev_get_by_name - find a device by its name
582 * @net: the applicable net namespace
583 * @name: name to find
585 * Find an interface by name. Must be called under RTNL semaphore
586 * or @dev_base_lock. If the name is found a pointer to the device
587 * is returned. If the name is not found then %NULL is returned. The
588 * reference counters are not incremented so the caller must be
589 * careful with locks.
592 struct net_device *__dev_get_by_name(struct net *net, const char *name)
594 struct hlist_node *p;
596 hlist_for_each(p, dev_name_hash(net, name)) {
597 struct net_device *dev
598 = hlist_entry(p, struct net_device, name_hlist);
599 if (!strncmp(dev->name, name, IFNAMSIZ))
606 * dev_get_by_name - find a device by its name
607 * @net: the applicable net namespace
608 * @name: name to find
610 * Find an interface by name. This can be called from any
611 * context and does its own locking. The returned handle has
612 * the usage count incremented and the caller must use dev_put() to
613 * release it when it is no longer needed. %NULL is returned if no
614 * matching device is found.
617 struct net_device *dev_get_by_name(struct net *net, const char *name)
619 struct net_device *dev;
621 read_lock(&dev_base_lock);
622 dev = __dev_get_by_name(net, name);
625 read_unlock(&dev_base_lock);
630 * __dev_get_by_index - find a device by its ifindex
631 * @net: the applicable net namespace
632 * @ifindex: index of device
634 * Search for an interface by index. Returns %NULL if the device
635 * is not found or a pointer to the device. The device has not
636 * had its reference counter increased so the caller must be careful
637 * about locking. The caller must hold either the RTNL semaphore
641 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
643 struct hlist_node *p;
645 hlist_for_each(p, dev_index_hash(net, ifindex)) {
646 struct net_device *dev
647 = hlist_entry(p, struct net_device, index_hlist);
648 if (dev->ifindex == ifindex)
656 * dev_get_by_index - find a device by its ifindex
657 * @net: the applicable net namespace
658 * @ifindex: index of device
660 * Search for an interface by index. Returns NULL if the device
661 * is not found or a pointer to the device. The device returned has
662 * had a reference added and the pointer is safe until the user calls
663 * dev_put to indicate they have finished with it.
666 struct net_device *dev_get_by_index(struct net *net, int ifindex)
668 struct net_device *dev;
670 read_lock(&dev_base_lock);
671 dev = __dev_get_by_index(net, ifindex);
674 read_unlock(&dev_base_lock);
679 * dev_getbyhwaddr - find a device by its hardware address
680 * @net: the applicable net namespace
681 * @type: media type of device
682 * @ha: hardware address
684 * Search for an interface by MAC address. Returns NULL if the device
685 * is not found or a pointer to the device. The caller must hold the
686 * rtnl semaphore. The returned device has not had its ref count increased
687 * and the caller must therefore be careful about locking
690 * If the API was consistent this would be __dev_get_by_hwaddr
693 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
695 struct net_device *dev;
699 for_each_netdev(net, dev)
700 if (dev->type == type &&
701 !memcmp(dev->dev_addr, ha, dev->addr_len))
707 EXPORT_SYMBOL(dev_getbyhwaddr);
709 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
711 struct net_device *dev;
714 for_each_netdev(net, dev)
715 if (dev->type == type)
721 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
723 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
725 struct net_device *dev;
728 dev = __dev_getfirstbyhwtype(net, type);
735 EXPORT_SYMBOL(dev_getfirstbyhwtype);
738 * dev_get_by_flags - find any device with given flags
739 * @net: the applicable net namespace
740 * @if_flags: IFF_* values
741 * @mask: bitmask of bits in if_flags to check
743 * Search for any interface with the given flags. Returns NULL if a device
744 * is not found or a pointer to the device. The device returned has
745 * had a reference added and the pointer is safe until the user calls
746 * dev_put to indicate they have finished with it.
749 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
751 struct net_device *dev, *ret;
754 read_lock(&dev_base_lock);
755 for_each_netdev(net, dev) {
756 if (((dev->flags ^ if_flags) & mask) == 0) {
762 read_unlock(&dev_base_lock);
767 * dev_valid_name - check if name is okay for network device
770 * Network device names need to be valid file names to
771 * to allow sysfs to work. We also disallow any kind of
774 int dev_valid_name(const char *name)
778 if (strlen(name) >= IFNAMSIZ)
780 if (!strcmp(name, ".") || !strcmp(name, ".."))
784 if (*name == '/' || isspace(*name))
792 * __dev_alloc_name - allocate a name for a device
793 * @net: network namespace to allocate the device name in
794 * @name: name format string
795 * @buf: scratch buffer and result name string
797 * Passed a format string - eg "lt%d" it will try and find a suitable
798 * id. It scans list of devices to build up a free map, then chooses
799 * the first empty slot. The caller must hold the dev_base or rtnl lock
800 * while allocating the name and adding the device in order to avoid
802 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
803 * Returns the number of the unit assigned or a negative errno code.
806 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
810 const int max_netdevices = 8*PAGE_SIZE;
811 unsigned long *inuse;
812 struct net_device *d;
814 p = strnchr(name, IFNAMSIZ-1, '%');
817 * Verify the string as this thing may have come from
818 * the user. There must be either one "%d" and no other "%"
821 if (p[1] != 'd' || strchr(p + 2, '%'))
824 /* Use one page as a bit array of possible slots */
825 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
829 for_each_netdev(net, d) {
830 if (!sscanf(d->name, name, &i))
832 if (i < 0 || i >= max_netdevices)
835 /* avoid cases where sscanf is not exact inverse of printf */
836 snprintf(buf, IFNAMSIZ, name, i);
837 if (!strncmp(buf, d->name, IFNAMSIZ))
841 i = find_first_zero_bit(inuse, max_netdevices);
842 free_page((unsigned long) inuse);
845 snprintf(buf, IFNAMSIZ, name, i);
846 if (!__dev_get_by_name(net, buf))
849 /* It is possible to run out of possible slots
850 * when the name is long and there isn't enough space left
851 * for the digits, or if all bits are used.
857 * dev_alloc_name - allocate a name for a device
859 * @name: name format string
861 * Passed a format string - eg "lt%d" it will try and find a suitable
862 * id. It scans list of devices to build up a free map, then chooses
863 * the first empty slot. The caller must hold the dev_base or rtnl lock
864 * while allocating the name and adding the device in order to avoid
866 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
867 * Returns the number of the unit assigned or a negative errno code.
870 int dev_alloc_name(struct net_device *dev, const char *name)
876 BUG_ON(!dev_net(dev));
878 ret = __dev_alloc_name(net, name, buf);
880 strlcpy(dev->name, buf, IFNAMSIZ);
886 * dev_change_name - change name of a device
888 * @newname: name (or format string) must be at least IFNAMSIZ
890 * Change name of a device, can pass format strings "eth%d".
893 int dev_change_name(struct net_device *dev, char *newname)
895 char oldname[IFNAMSIZ];
901 BUG_ON(!dev_net(dev));
904 if (dev->flags & IFF_UP)
907 if (!dev_valid_name(newname))
910 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
913 memcpy(oldname, dev->name, IFNAMSIZ);
915 if (strchr(newname, '%')) {
916 err = dev_alloc_name(dev, newname);
919 strcpy(newname, dev->name);
921 else if (__dev_get_by_name(net, newname))
924 strlcpy(dev->name, newname, IFNAMSIZ);
927 err = device_rename(&dev->dev, dev->name);
929 memcpy(dev->name, oldname, IFNAMSIZ);
933 write_lock_bh(&dev_base_lock);
934 hlist_del(&dev->name_hlist);
935 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
936 write_unlock_bh(&dev_base_lock);
938 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
939 ret = notifier_to_errno(ret);
944 "%s: name change rollback failed: %d.\n",
948 memcpy(dev->name, oldname, IFNAMSIZ);
957 * dev_set_alias - change ifalias of a device
959 * @alias: name up to IFALIASZ
961 * Set ifalias for a device,
963 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
970 dev->ifalias = krealloc(dev->ifalias, len+1, GFP_KERNEL);
974 strlcpy(dev->ifalias, alias, len+1);
980 * netdev_features_change - device changes features
981 * @dev: device to cause notification
983 * Called to indicate a device has changed features.
985 void netdev_features_change(struct net_device *dev)
987 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
989 EXPORT_SYMBOL(netdev_features_change);
992 * netdev_state_change - device changes state
993 * @dev: device to cause notification
995 * Called to indicate a device has changed state. This function calls
996 * the notifier chains for netdev_chain and sends a NEWLINK message
997 * to the routing socket.
999 void netdev_state_change(struct net_device *dev)
1001 if (dev->flags & IFF_UP) {
1002 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1003 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1007 void netdev_bonding_change(struct net_device *dev)
1009 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev);
1011 EXPORT_SYMBOL(netdev_bonding_change);
1014 * dev_load - load a network module
1015 * @net: the applicable net namespace
1016 * @name: name of interface
1018 * If a network interface is not present and the process has suitable
1019 * privileges this function loads the module. If module loading is not
1020 * available in this kernel then it becomes a nop.
1023 void dev_load(struct net *net, const char *name)
1025 struct net_device *dev;
1027 read_lock(&dev_base_lock);
1028 dev = __dev_get_by_name(net, name);
1029 read_unlock(&dev_base_lock);
1031 if (!dev && capable(CAP_SYS_MODULE))
1032 request_module("%s", name);
1036 * dev_open - prepare an interface for use.
1037 * @dev: device to open
1039 * Takes a device from down to up state. The device's private open
1040 * function is invoked and then the multicast lists are loaded. Finally
1041 * the device is moved into the up state and a %NETDEV_UP message is
1042 * sent to the netdev notifier chain.
1044 * Calling this function on an active interface is a nop. On a failure
1045 * a negative errno code is returned.
1047 int dev_open(struct net_device *dev)
1057 if (dev->flags & IFF_UP)
1061 * Is it even present?
1063 if (!netif_device_present(dev))
1067 * Call device private open method
1069 set_bit(__LINK_STATE_START, &dev->state);
1071 if (dev->validate_addr)
1072 ret = dev->validate_addr(dev);
1074 if (!ret && dev->open)
1075 ret = dev->open(dev);
1078 * If it went open OK then:
1082 clear_bit(__LINK_STATE_START, &dev->state);
1087 dev->flags |= IFF_UP;
1090 * Initialize multicasting status
1092 dev_set_rx_mode(dev);
1095 * Wakeup transmit queue engine
1100 * ... and announce new interface.
1102 call_netdevice_notifiers(NETDEV_UP, dev);
1109 * dev_close - shutdown an interface.
1110 * @dev: device to shutdown
1112 * This function moves an active device into down state. A
1113 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1114 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1117 int dev_close(struct net_device *dev)
1123 if (!(dev->flags & IFF_UP))
1127 * Tell people we are going down, so that they can
1128 * prepare to death, when device is still operating.
1130 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1132 clear_bit(__LINK_STATE_START, &dev->state);
1134 /* Synchronize to scheduled poll. We cannot touch poll list,
1135 * it can be even on different cpu. So just clear netif_running().
1137 * dev->stop() will invoke napi_disable() on all of it's
1138 * napi_struct instances on this device.
1140 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1142 dev_deactivate(dev);
1145 * Call the device specific close. This cannot fail.
1146 * Only if device is UP
1148 * We allow it to be called even after a DETACH hot-plug
1155 * Device is now down.
1158 dev->flags &= ~IFF_UP;
1161 * Tell people we are down
1163 call_netdevice_notifiers(NETDEV_DOWN, dev);
1170 * dev_disable_lro - disable Large Receive Offload on a device
1173 * Disable Large Receive Offload (LRO) on a net device. Must be
1174 * called under RTNL. This is needed if received packets may be
1175 * forwarded to another interface.
1177 void dev_disable_lro(struct net_device *dev)
1179 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1180 dev->ethtool_ops->set_flags) {
1181 u32 flags = dev->ethtool_ops->get_flags(dev);
1182 if (flags & ETH_FLAG_LRO) {
1183 flags &= ~ETH_FLAG_LRO;
1184 dev->ethtool_ops->set_flags(dev, flags);
1187 WARN_ON(dev->features & NETIF_F_LRO);
1189 EXPORT_SYMBOL(dev_disable_lro);
1192 static int dev_boot_phase = 1;
1195 * Device change register/unregister. These are not inline or static
1196 * as we export them to the world.
1200 * register_netdevice_notifier - register a network notifier block
1203 * Register a notifier to be called when network device events occur.
1204 * The notifier passed is linked into the kernel structures and must
1205 * not be reused until it has been unregistered. A negative errno code
1206 * is returned on a failure.
1208 * When registered all registration and up events are replayed
1209 * to the new notifier to allow device to have a race free
1210 * view of the network device list.
1213 int register_netdevice_notifier(struct notifier_block *nb)
1215 struct net_device *dev;
1216 struct net_device *last;
1221 err = raw_notifier_chain_register(&netdev_chain, nb);
1227 for_each_netdev(net, dev) {
1228 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1229 err = notifier_to_errno(err);
1233 if (!(dev->flags & IFF_UP))
1236 nb->notifier_call(nb, NETDEV_UP, dev);
1247 for_each_netdev(net, dev) {
1251 if (dev->flags & IFF_UP) {
1252 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1253 nb->notifier_call(nb, NETDEV_DOWN, dev);
1255 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1259 raw_notifier_chain_unregister(&netdev_chain, nb);
1264 * unregister_netdevice_notifier - unregister a network notifier block
1267 * Unregister a notifier previously registered by
1268 * register_netdevice_notifier(). The notifier is unlinked into the
1269 * kernel structures and may then be reused. A negative errno code
1270 * is returned on a failure.
1273 int unregister_netdevice_notifier(struct notifier_block *nb)
1278 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1284 * call_netdevice_notifiers - call all network notifier blocks
1285 * @val: value passed unmodified to notifier function
1286 * @dev: net_device pointer passed unmodified to notifier function
1288 * Call all network notifier blocks. Parameters and return value
1289 * are as for raw_notifier_call_chain().
1292 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1294 return raw_notifier_call_chain(&netdev_chain, val, dev);
1297 /* When > 0 there are consumers of rx skb time stamps */
1298 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1300 void net_enable_timestamp(void)
1302 atomic_inc(&netstamp_needed);
1305 void net_disable_timestamp(void)
1307 atomic_dec(&netstamp_needed);
1310 static inline void net_timestamp(struct sk_buff *skb)
1312 if (atomic_read(&netstamp_needed))
1313 __net_timestamp(skb);
1315 skb->tstamp.tv64 = 0;
1319 * Support routine. Sends outgoing frames to any network
1320 * taps currently in use.
1323 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1325 struct packet_type *ptype;
1330 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1331 /* Never send packets back to the socket
1332 * they originated from - MvS (miquels@drinkel.ow.org)
1334 if ((ptype->dev == dev || !ptype->dev) &&
1335 (ptype->af_packet_priv == NULL ||
1336 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1337 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1341 /* skb->nh should be correctly
1342 set by sender, so that the second statement is
1343 just protection against buggy protocols.
1345 skb_reset_mac_header(skb2);
1347 if (skb_network_header(skb2) < skb2->data ||
1348 skb2->network_header > skb2->tail) {
1349 if (net_ratelimit())
1350 printk(KERN_CRIT "protocol %04x is "
1352 skb2->protocol, dev->name);
1353 skb_reset_network_header(skb2);
1356 skb2->transport_header = skb2->network_header;
1357 skb2->pkt_type = PACKET_OUTGOING;
1358 ptype->func(skb2, skb->dev, ptype, skb->dev);
1365 static inline void __netif_reschedule(struct Qdisc *q)
1367 struct softnet_data *sd;
1368 unsigned long flags;
1370 local_irq_save(flags);
1371 sd = &__get_cpu_var(softnet_data);
1372 q->next_sched = sd->output_queue;
1373 sd->output_queue = q;
1374 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1375 local_irq_restore(flags);
1378 void __netif_schedule(struct Qdisc *q)
1380 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1381 __netif_reschedule(q);
1383 EXPORT_SYMBOL(__netif_schedule);
1385 void dev_kfree_skb_irq(struct sk_buff *skb)
1387 if (atomic_dec_and_test(&skb->users)) {
1388 struct softnet_data *sd;
1389 unsigned long flags;
1391 local_irq_save(flags);
1392 sd = &__get_cpu_var(softnet_data);
1393 skb->next = sd->completion_queue;
1394 sd->completion_queue = skb;
1395 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1396 local_irq_restore(flags);
1399 EXPORT_SYMBOL(dev_kfree_skb_irq);
1401 void dev_kfree_skb_any(struct sk_buff *skb)
1403 if (in_irq() || irqs_disabled())
1404 dev_kfree_skb_irq(skb);
1408 EXPORT_SYMBOL(dev_kfree_skb_any);
1412 * netif_device_detach - mark device as removed
1413 * @dev: network device
1415 * Mark device as removed from system and therefore no longer available.
1417 void netif_device_detach(struct net_device *dev)
1419 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1420 netif_running(dev)) {
1421 netif_stop_queue(dev);
1424 EXPORT_SYMBOL(netif_device_detach);
1427 * netif_device_attach - mark device as attached
1428 * @dev: network device
1430 * Mark device as attached from system and restart if needed.
1432 void netif_device_attach(struct net_device *dev)
1434 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1435 netif_running(dev)) {
1436 netif_wake_queue(dev);
1437 __netdev_watchdog_up(dev);
1440 EXPORT_SYMBOL(netif_device_attach);
1442 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1444 return ((features & NETIF_F_GEN_CSUM) ||
1445 ((features & NETIF_F_IP_CSUM) &&
1446 protocol == htons(ETH_P_IP)) ||
1447 ((features & NETIF_F_IPV6_CSUM) &&
1448 protocol == htons(ETH_P_IPV6)));
1451 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1453 if (can_checksum_protocol(dev->features, skb->protocol))
1456 if (skb->protocol == htons(ETH_P_8021Q)) {
1457 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1458 if (can_checksum_protocol(dev->features & dev->vlan_features,
1459 veh->h_vlan_encapsulated_proto))
1467 * Invalidate hardware checksum when packet is to be mangled, and
1468 * complete checksum manually on outgoing path.
1470 int skb_checksum_help(struct sk_buff *skb)
1473 int ret = 0, offset;
1475 if (skb->ip_summed == CHECKSUM_COMPLETE)
1476 goto out_set_summed;
1478 if (unlikely(skb_shinfo(skb)->gso_size)) {
1479 /* Let GSO fix up the checksum. */
1480 goto out_set_summed;
1483 offset = skb->csum_start - skb_headroom(skb);
1484 BUG_ON(offset >= skb_headlen(skb));
1485 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1487 offset += skb->csum_offset;
1488 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1490 if (skb_cloned(skb) &&
1491 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1492 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1497 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1499 skb->ip_summed = CHECKSUM_NONE;
1505 * skb_gso_segment - Perform segmentation on skb.
1506 * @skb: buffer to segment
1507 * @features: features for the output path (see dev->features)
1509 * This function segments the given skb and returns a list of segments.
1511 * It may return NULL if the skb requires no segmentation. This is
1512 * only possible when GSO is used for verifying header integrity.
1514 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1516 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1517 struct packet_type *ptype;
1518 __be16 type = skb->protocol;
1521 BUG_ON(skb_shinfo(skb)->frag_list);
1523 skb_reset_mac_header(skb);
1524 skb->mac_len = skb->network_header - skb->mac_header;
1525 __skb_pull(skb, skb->mac_len);
1527 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1528 if (skb_header_cloned(skb) &&
1529 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1530 return ERR_PTR(err);
1534 list_for_each_entry_rcu(ptype,
1535 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1536 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1537 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1538 err = ptype->gso_send_check(skb);
1539 segs = ERR_PTR(err);
1540 if (err || skb_gso_ok(skb, features))
1542 __skb_push(skb, (skb->data -
1543 skb_network_header(skb)));
1545 segs = ptype->gso_segment(skb, features);
1551 __skb_push(skb, skb->data - skb_mac_header(skb));
1556 EXPORT_SYMBOL(skb_gso_segment);
1558 /* Take action when hardware reception checksum errors are detected. */
1560 void netdev_rx_csum_fault(struct net_device *dev)
1562 if (net_ratelimit()) {
1563 printk(KERN_ERR "%s: hw csum failure.\n",
1564 dev ? dev->name : "<unknown>");
1568 EXPORT_SYMBOL(netdev_rx_csum_fault);
1571 /* Actually, we should eliminate this check as soon as we know, that:
1572 * 1. IOMMU is present and allows to map all the memory.
1573 * 2. No high memory really exists on this machine.
1576 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1578 #ifdef CONFIG_HIGHMEM
1581 if (dev->features & NETIF_F_HIGHDMA)
1584 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1585 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1593 void (*destructor)(struct sk_buff *skb);
1596 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1598 static void dev_gso_skb_destructor(struct sk_buff *skb)
1600 struct dev_gso_cb *cb;
1603 struct sk_buff *nskb = skb->next;
1605 skb->next = nskb->next;
1608 } while (skb->next);
1610 cb = DEV_GSO_CB(skb);
1612 cb->destructor(skb);
1616 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1617 * @skb: buffer to segment
1619 * This function segments the given skb and stores the list of segments
1622 static int dev_gso_segment(struct sk_buff *skb)
1624 struct net_device *dev = skb->dev;
1625 struct sk_buff *segs;
1626 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1629 segs = skb_gso_segment(skb, features);
1631 /* Verifying header integrity only. */
1636 return PTR_ERR(segs);
1639 DEV_GSO_CB(skb)->destructor = skb->destructor;
1640 skb->destructor = dev_gso_skb_destructor;
1645 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1646 struct netdev_queue *txq)
1648 if (likely(!skb->next)) {
1649 if (!list_empty(&ptype_all))
1650 dev_queue_xmit_nit(skb, dev);
1652 if (netif_needs_gso(dev, skb)) {
1653 if (unlikely(dev_gso_segment(skb)))
1659 return dev->hard_start_xmit(skb, dev);
1664 struct sk_buff *nskb = skb->next;
1667 skb->next = nskb->next;
1669 rc = dev->hard_start_xmit(nskb, dev);
1671 nskb->next = skb->next;
1675 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1676 return NETDEV_TX_BUSY;
1677 } while (skb->next);
1679 skb->destructor = DEV_GSO_CB(skb)->destructor;
1686 static u32 simple_tx_hashrnd;
1687 static int simple_tx_hashrnd_initialized = 0;
1689 static u16 simple_tx_hash(struct net_device *dev, struct sk_buff *skb)
1691 u32 addr1, addr2, ports;
1695 if (unlikely(!simple_tx_hashrnd_initialized)) {
1696 get_random_bytes(&simple_tx_hashrnd, 4);
1697 simple_tx_hashrnd_initialized = 1;
1700 switch (skb->protocol) {
1701 case htons(ETH_P_IP):
1702 ip_proto = ip_hdr(skb)->protocol;
1703 addr1 = ip_hdr(skb)->saddr;
1704 addr2 = ip_hdr(skb)->daddr;
1705 ihl = ip_hdr(skb)->ihl;
1707 case htons(ETH_P_IPV6):
1708 ip_proto = ipv6_hdr(skb)->nexthdr;
1709 addr1 = ipv6_hdr(skb)->saddr.s6_addr32[3];
1710 addr2 = ipv6_hdr(skb)->daddr.s6_addr32[3];
1725 case IPPROTO_UDPLITE:
1726 ports = *((u32 *) (skb_network_header(skb) + (ihl * 4)));
1734 hash = jhash_3words(addr1, addr2, ports, simple_tx_hashrnd);
1736 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1739 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1740 struct sk_buff *skb)
1742 u16 queue_index = 0;
1744 if (dev->select_queue)
1745 queue_index = dev->select_queue(dev, skb);
1746 else if (dev->real_num_tx_queues > 1)
1747 queue_index = simple_tx_hash(dev, skb);
1749 skb_set_queue_mapping(skb, queue_index);
1750 return netdev_get_tx_queue(dev, queue_index);
1754 * dev_queue_xmit - transmit a buffer
1755 * @skb: buffer to transmit
1757 * Queue a buffer for transmission to a network device. The caller must
1758 * have set the device and priority and built the buffer before calling
1759 * this function. The function can be called from an interrupt.
1761 * A negative errno code is returned on a failure. A success does not
1762 * guarantee the frame will be transmitted as it may be dropped due
1763 * to congestion or traffic shaping.
1765 * -----------------------------------------------------------------------------------
1766 * I notice this method can also return errors from the queue disciplines,
1767 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1770 * Regardless of the return value, the skb is consumed, so it is currently
1771 * difficult to retry a send to this method. (You can bump the ref count
1772 * before sending to hold a reference for retry if you are careful.)
1774 * When calling this method, interrupts MUST be enabled. This is because
1775 * the BH enable code must have IRQs enabled so that it will not deadlock.
1778 int dev_queue_xmit(struct sk_buff *skb)
1780 struct net_device *dev = skb->dev;
1781 struct netdev_queue *txq;
1785 /* GSO will handle the following emulations directly. */
1786 if (netif_needs_gso(dev, skb))
1789 if (skb_shinfo(skb)->frag_list &&
1790 !(dev->features & NETIF_F_FRAGLIST) &&
1791 __skb_linearize(skb))
1794 /* Fragmented skb is linearized if device does not support SG,
1795 * or if at least one of fragments is in highmem and device
1796 * does not support DMA from it.
1798 if (skb_shinfo(skb)->nr_frags &&
1799 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1800 __skb_linearize(skb))
1803 /* If packet is not checksummed and device does not support
1804 * checksumming for this protocol, complete checksumming here.
1806 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1807 skb_set_transport_header(skb, skb->csum_start -
1809 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1814 /* Disable soft irqs for various locks below. Also
1815 * stops preemption for RCU.
1819 txq = dev_pick_tx(dev, skb);
1820 q = rcu_dereference(txq->qdisc);
1822 #ifdef CONFIG_NET_CLS_ACT
1823 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1826 spinlock_t *root_lock = qdisc_lock(q);
1828 spin_lock(root_lock);
1830 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1834 rc = qdisc_enqueue_root(skb, q);
1837 spin_unlock(root_lock);
1842 /* The device has no queue. Common case for software devices:
1843 loopback, all the sorts of tunnels...
1845 Really, it is unlikely that netif_tx_lock protection is necessary
1846 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1848 However, it is possible, that they rely on protection
1851 Check this and shot the lock. It is not prone from deadlocks.
1852 Either shot noqueue qdisc, it is even simpler 8)
1854 if (dev->flags & IFF_UP) {
1855 int cpu = smp_processor_id(); /* ok because BHs are off */
1857 if (txq->xmit_lock_owner != cpu) {
1859 HARD_TX_LOCK(dev, txq, cpu);
1861 if (!netif_tx_queue_stopped(txq)) {
1863 if (!dev_hard_start_xmit(skb, dev, txq)) {
1864 HARD_TX_UNLOCK(dev, txq);
1868 HARD_TX_UNLOCK(dev, txq);
1869 if (net_ratelimit())
1870 printk(KERN_CRIT "Virtual device %s asks to "
1871 "queue packet!\n", dev->name);
1873 /* Recursion is detected! It is possible,
1875 if (net_ratelimit())
1876 printk(KERN_CRIT "Dead loop on virtual device "
1877 "%s, fix it urgently!\n", dev->name);
1882 rcu_read_unlock_bh();
1888 rcu_read_unlock_bh();
1893 /*=======================================================================
1895 =======================================================================*/
1897 int netdev_max_backlog __read_mostly = 1000;
1898 int netdev_budget __read_mostly = 300;
1899 int weight_p __read_mostly = 64; /* old backlog weight */
1901 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1905 * netif_rx - post buffer to the network code
1906 * @skb: buffer to post
1908 * This function receives a packet from a device driver and queues it for
1909 * the upper (protocol) levels to process. It always succeeds. The buffer
1910 * may be dropped during processing for congestion control or by the
1914 * NET_RX_SUCCESS (no congestion)
1915 * NET_RX_DROP (packet was dropped)
1919 int netif_rx(struct sk_buff *skb)
1921 struct softnet_data *queue;
1922 unsigned long flags;
1924 /* if netpoll wants it, pretend we never saw it */
1925 if (netpoll_rx(skb))
1928 if (!skb->tstamp.tv64)
1932 * The code is rearranged so that the path is the most
1933 * short when CPU is congested, but is still operating.
1935 local_irq_save(flags);
1936 queue = &__get_cpu_var(softnet_data);
1938 __get_cpu_var(netdev_rx_stat).total++;
1939 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1940 if (queue->input_pkt_queue.qlen) {
1942 __skb_queue_tail(&queue->input_pkt_queue, skb);
1943 local_irq_restore(flags);
1944 return NET_RX_SUCCESS;
1947 napi_schedule(&queue->backlog);
1951 __get_cpu_var(netdev_rx_stat).dropped++;
1952 local_irq_restore(flags);
1958 int netif_rx_ni(struct sk_buff *skb)
1963 err = netif_rx(skb);
1964 if (local_softirq_pending())
1971 EXPORT_SYMBOL(netif_rx_ni);
1973 static void net_tx_action(struct softirq_action *h)
1975 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1977 if (sd->completion_queue) {
1978 struct sk_buff *clist;
1980 local_irq_disable();
1981 clist = sd->completion_queue;
1982 sd->completion_queue = NULL;
1986 struct sk_buff *skb = clist;
1987 clist = clist->next;
1989 WARN_ON(atomic_read(&skb->users));
1994 if (sd->output_queue) {
1997 local_irq_disable();
1998 head = sd->output_queue;
1999 sd->output_queue = NULL;
2003 struct Qdisc *q = head;
2004 spinlock_t *root_lock;
2006 head = head->next_sched;
2008 root_lock = qdisc_lock(q);
2009 if (spin_trylock(root_lock)) {
2010 smp_mb__before_clear_bit();
2011 clear_bit(__QDISC_STATE_SCHED,
2014 spin_unlock(root_lock);
2016 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2018 __netif_reschedule(q);
2020 smp_mb__before_clear_bit();
2021 clear_bit(__QDISC_STATE_SCHED,
2029 static inline int deliver_skb(struct sk_buff *skb,
2030 struct packet_type *pt_prev,
2031 struct net_device *orig_dev)
2033 atomic_inc(&skb->users);
2034 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2037 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2038 /* These hooks defined here for ATM */
2040 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2041 unsigned char *addr);
2042 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
2045 * If bridge module is loaded call bridging hook.
2046 * returns NULL if packet was consumed.
2048 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2049 struct sk_buff *skb) __read_mostly;
2050 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2051 struct packet_type **pt_prev, int *ret,
2052 struct net_device *orig_dev)
2054 struct net_bridge_port *port;
2056 if (skb->pkt_type == PACKET_LOOPBACK ||
2057 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2061 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2065 return br_handle_frame_hook(port, skb);
2068 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2071 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2072 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2073 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2075 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2076 struct packet_type **pt_prev,
2078 struct net_device *orig_dev)
2080 if (skb->dev->macvlan_port == NULL)
2084 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2087 return macvlan_handle_frame_hook(skb);
2090 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2093 #ifdef CONFIG_NET_CLS_ACT
2094 /* TODO: Maybe we should just force sch_ingress to be compiled in
2095 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2096 * a compare and 2 stores extra right now if we dont have it on
2097 * but have CONFIG_NET_CLS_ACT
2098 * NOTE: This doesnt stop any functionality; if you dont have
2099 * the ingress scheduler, you just cant add policies on ingress.
2102 static int ing_filter(struct sk_buff *skb)
2104 struct net_device *dev = skb->dev;
2105 u32 ttl = G_TC_RTTL(skb->tc_verd);
2106 struct netdev_queue *rxq;
2107 int result = TC_ACT_OK;
2110 if (MAX_RED_LOOP < ttl++) {
2112 "Redir loop detected Dropping packet (%d->%d)\n",
2113 skb->iif, dev->ifindex);
2117 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2118 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2120 rxq = &dev->rx_queue;
2123 if (q != &noop_qdisc) {
2124 spin_lock(qdisc_lock(q));
2125 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2126 result = qdisc_enqueue_root(skb, q);
2127 spin_unlock(qdisc_lock(q));
2133 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2134 struct packet_type **pt_prev,
2135 int *ret, struct net_device *orig_dev)
2137 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2141 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2144 /* Huh? Why does turning on AF_PACKET affect this? */
2145 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2148 switch (ing_filter(skb)) {
2162 * netif_nit_deliver - deliver received packets to network taps
2165 * This function is used to deliver incoming packets to network
2166 * taps. It should be used when the normal netif_receive_skb path
2167 * is bypassed, for example because of VLAN acceleration.
2169 void netif_nit_deliver(struct sk_buff *skb)
2171 struct packet_type *ptype;
2173 if (list_empty(&ptype_all))
2176 skb_reset_network_header(skb);
2177 skb_reset_transport_header(skb);
2178 skb->mac_len = skb->network_header - skb->mac_header;
2181 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2182 if (!ptype->dev || ptype->dev == skb->dev)
2183 deliver_skb(skb, ptype, skb->dev);
2189 * netif_receive_skb - process receive buffer from network
2190 * @skb: buffer to process
2192 * netif_receive_skb() is the main receive data processing function.
2193 * It always succeeds. The buffer may be dropped during processing
2194 * for congestion control or by the protocol layers.
2196 * This function may only be called from softirq context and interrupts
2197 * should be enabled.
2199 * Return values (usually ignored):
2200 * NET_RX_SUCCESS: no congestion
2201 * NET_RX_DROP: packet was dropped
2203 int netif_receive_skb(struct sk_buff *skb)
2205 struct packet_type *ptype, *pt_prev;
2206 struct net_device *orig_dev;
2207 struct net_device *null_or_orig;
2208 int ret = NET_RX_DROP;
2211 /* if we've gotten here through NAPI, check netpoll */
2212 if (netpoll_receive_skb(skb))
2215 if (!skb->tstamp.tv64)
2219 skb->iif = skb->dev->ifindex;
2221 null_or_orig = NULL;
2222 orig_dev = skb->dev;
2223 if (orig_dev->master) {
2224 if (skb_bond_should_drop(skb))
2225 null_or_orig = orig_dev; /* deliver only exact match */
2227 skb->dev = orig_dev->master;
2230 __get_cpu_var(netdev_rx_stat).total++;
2232 skb_reset_network_header(skb);
2233 skb_reset_transport_header(skb);
2234 skb->mac_len = skb->network_header - skb->mac_header;
2240 /* Don't receive packets in an exiting network namespace */
2241 if (!net_alive(dev_net(skb->dev)))
2244 #ifdef CONFIG_NET_CLS_ACT
2245 if (skb->tc_verd & TC_NCLS) {
2246 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2251 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2252 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2253 ptype->dev == orig_dev) {
2255 ret = deliver_skb(skb, pt_prev, orig_dev);
2260 #ifdef CONFIG_NET_CLS_ACT
2261 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2267 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2270 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2274 type = skb->protocol;
2275 list_for_each_entry_rcu(ptype,
2276 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2277 if (ptype->type == type &&
2278 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2279 ptype->dev == orig_dev)) {
2281 ret = deliver_skb(skb, pt_prev, orig_dev);
2287 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2290 /* Jamal, now you will not able to escape explaining
2291 * me how you were going to use this. :-)
2301 /* Network device is going away, flush any packets still pending */
2302 static void flush_backlog(void *arg)
2304 struct net_device *dev = arg;
2305 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2306 struct sk_buff *skb, *tmp;
2308 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2309 if (skb->dev == dev) {
2310 __skb_unlink(skb, &queue->input_pkt_queue);
2315 static int process_backlog(struct napi_struct *napi, int quota)
2318 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2319 unsigned long start_time = jiffies;
2321 napi->weight = weight_p;
2323 struct sk_buff *skb;
2325 local_irq_disable();
2326 skb = __skb_dequeue(&queue->input_pkt_queue);
2328 __napi_complete(napi);
2334 netif_receive_skb(skb);
2335 } while (++work < quota && jiffies == start_time);
2341 * __napi_schedule - schedule for receive
2342 * @n: entry to schedule
2344 * The entry's receive function will be scheduled to run
2346 void __napi_schedule(struct napi_struct *n)
2348 unsigned long flags;
2350 local_irq_save(flags);
2351 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2352 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2353 local_irq_restore(flags);
2355 EXPORT_SYMBOL(__napi_schedule);
2358 static void net_rx_action(struct softirq_action *h)
2360 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2361 unsigned long start_time = jiffies;
2362 int budget = netdev_budget;
2365 local_irq_disable();
2367 while (!list_empty(list)) {
2368 struct napi_struct *n;
2371 /* If softirq window is exhuasted then punt.
2373 * Note that this is a slight policy change from the
2374 * previous NAPI code, which would allow up to 2
2375 * jiffies to pass before breaking out. The test
2376 * used to be "jiffies - start_time > 1".
2378 if (unlikely(budget <= 0 || jiffies != start_time))
2383 /* Even though interrupts have been re-enabled, this
2384 * access is safe because interrupts can only add new
2385 * entries to the tail of this list, and only ->poll()
2386 * calls can remove this head entry from the list.
2388 n = list_entry(list->next, struct napi_struct, poll_list);
2390 have = netpoll_poll_lock(n);
2394 /* This NAPI_STATE_SCHED test is for avoiding a race
2395 * with netpoll's poll_napi(). Only the entity which
2396 * obtains the lock and sees NAPI_STATE_SCHED set will
2397 * actually make the ->poll() call. Therefore we avoid
2398 * accidently calling ->poll() when NAPI is not scheduled.
2401 if (test_bit(NAPI_STATE_SCHED, &n->state))
2402 work = n->poll(n, weight);
2404 WARN_ON_ONCE(work > weight);
2408 local_irq_disable();
2410 /* Drivers must not modify the NAPI state if they
2411 * consume the entire weight. In such cases this code
2412 * still "owns" the NAPI instance and therefore can
2413 * move the instance around on the list at-will.
2415 if (unlikely(work == weight)) {
2416 if (unlikely(napi_disable_pending(n)))
2419 list_move_tail(&n->poll_list, list);
2422 netpoll_poll_unlock(have);
2427 #ifdef CONFIG_NET_DMA
2429 * There may not be any more sk_buffs coming right now, so push
2430 * any pending DMA copies to hardware
2432 if (!cpus_empty(net_dma.channel_mask)) {
2434 for_each_cpu_mask_nr(chan_idx, net_dma.channel_mask) {
2435 struct dma_chan *chan = net_dma.channels[chan_idx];
2437 dma_async_memcpy_issue_pending(chan);
2445 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2446 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2450 static gifconf_func_t * gifconf_list [NPROTO];
2453 * register_gifconf - register a SIOCGIF handler
2454 * @family: Address family
2455 * @gifconf: Function handler
2457 * Register protocol dependent address dumping routines. The handler
2458 * that is passed must not be freed or reused until it has been replaced
2459 * by another handler.
2461 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2463 if (family >= NPROTO)
2465 gifconf_list[family] = gifconf;
2471 * Map an interface index to its name (SIOCGIFNAME)
2475 * We need this ioctl for efficient implementation of the
2476 * if_indextoname() function required by the IPv6 API. Without
2477 * it, we would have to search all the interfaces to find a
2481 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2483 struct net_device *dev;
2487 * Fetch the caller's info block.
2490 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2493 read_lock(&dev_base_lock);
2494 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2496 read_unlock(&dev_base_lock);
2500 strcpy(ifr.ifr_name, dev->name);
2501 read_unlock(&dev_base_lock);
2503 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2509 * Perform a SIOCGIFCONF call. This structure will change
2510 * size eventually, and there is nothing I can do about it.
2511 * Thus we will need a 'compatibility mode'.
2514 static int dev_ifconf(struct net *net, char __user *arg)
2517 struct net_device *dev;
2524 * Fetch the caller's info block.
2527 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2534 * Loop over the interfaces, and write an info block for each.
2538 for_each_netdev(net, dev) {
2539 for (i = 0; i < NPROTO; i++) {
2540 if (gifconf_list[i]) {
2543 done = gifconf_list[i](dev, NULL, 0);
2545 done = gifconf_list[i](dev, pos + total,
2555 * All done. Write the updated control block back to the caller.
2557 ifc.ifc_len = total;
2560 * Both BSD and Solaris return 0 here, so we do too.
2562 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2565 #ifdef CONFIG_PROC_FS
2567 * This is invoked by the /proc filesystem handler to display a device
2570 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2571 __acquires(dev_base_lock)
2573 struct net *net = seq_file_net(seq);
2575 struct net_device *dev;
2577 read_lock(&dev_base_lock);
2579 return SEQ_START_TOKEN;
2582 for_each_netdev(net, dev)
2589 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2591 struct net *net = seq_file_net(seq);
2593 return v == SEQ_START_TOKEN ?
2594 first_net_device(net) : next_net_device((struct net_device *)v);
2597 void dev_seq_stop(struct seq_file *seq, void *v)
2598 __releases(dev_base_lock)
2600 read_unlock(&dev_base_lock);
2603 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2605 struct net_device_stats *stats = dev->get_stats(dev);
2607 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2608 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2609 dev->name, stats->rx_bytes, stats->rx_packets,
2611 stats->rx_dropped + stats->rx_missed_errors,
2612 stats->rx_fifo_errors,
2613 stats->rx_length_errors + stats->rx_over_errors +
2614 stats->rx_crc_errors + stats->rx_frame_errors,
2615 stats->rx_compressed, stats->multicast,
2616 stats->tx_bytes, stats->tx_packets,
2617 stats->tx_errors, stats->tx_dropped,
2618 stats->tx_fifo_errors, stats->collisions,
2619 stats->tx_carrier_errors +
2620 stats->tx_aborted_errors +
2621 stats->tx_window_errors +
2622 stats->tx_heartbeat_errors,
2623 stats->tx_compressed);
2627 * Called from the PROCfs module. This now uses the new arbitrary sized
2628 * /proc/net interface to create /proc/net/dev
2630 static int dev_seq_show(struct seq_file *seq, void *v)
2632 if (v == SEQ_START_TOKEN)
2633 seq_puts(seq, "Inter-| Receive "
2635 " face |bytes packets errs drop fifo frame "
2636 "compressed multicast|bytes packets errs "
2637 "drop fifo colls carrier compressed\n");
2639 dev_seq_printf_stats(seq, v);
2643 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2645 struct netif_rx_stats *rc = NULL;
2647 while (*pos < nr_cpu_ids)
2648 if (cpu_online(*pos)) {
2649 rc = &per_cpu(netdev_rx_stat, *pos);
2656 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2658 return softnet_get_online(pos);
2661 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2664 return softnet_get_online(pos);
2667 static void softnet_seq_stop(struct seq_file *seq, void *v)
2671 static int softnet_seq_show(struct seq_file *seq, void *v)
2673 struct netif_rx_stats *s = v;
2675 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2676 s->total, s->dropped, s->time_squeeze, 0,
2677 0, 0, 0, 0, /* was fastroute */
2682 static const struct seq_operations dev_seq_ops = {
2683 .start = dev_seq_start,
2684 .next = dev_seq_next,
2685 .stop = dev_seq_stop,
2686 .show = dev_seq_show,
2689 static int dev_seq_open(struct inode *inode, struct file *file)
2691 return seq_open_net(inode, file, &dev_seq_ops,
2692 sizeof(struct seq_net_private));
2695 static const struct file_operations dev_seq_fops = {
2696 .owner = THIS_MODULE,
2697 .open = dev_seq_open,
2699 .llseek = seq_lseek,
2700 .release = seq_release_net,
2703 static const struct seq_operations softnet_seq_ops = {
2704 .start = softnet_seq_start,
2705 .next = softnet_seq_next,
2706 .stop = softnet_seq_stop,
2707 .show = softnet_seq_show,
2710 static int softnet_seq_open(struct inode *inode, struct file *file)
2712 return seq_open(file, &softnet_seq_ops);
2715 static const struct file_operations softnet_seq_fops = {
2716 .owner = THIS_MODULE,
2717 .open = softnet_seq_open,
2719 .llseek = seq_lseek,
2720 .release = seq_release,
2723 static void *ptype_get_idx(loff_t pos)
2725 struct packet_type *pt = NULL;
2729 list_for_each_entry_rcu(pt, &ptype_all, list) {
2735 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
2736 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2745 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2749 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2752 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2754 struct packet_type *pt;
2755 struct list_head *nxt;
2759 if (v == SEQ_START_TOKEN)
2760 return ptype_get_idx(0);
2763 nxt = pt->list.next;
2764 if (pt->type == htons(ETH_P_ALL)) {
2765 if (nxt != &ptype_all)
2768 nxt = ptype_base[0].next;
2770 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
2772 while (nxt == &ptype_base[hash]) {
2773 if (++hash >= PTYPE_HASH_SIZE)
2775 nxt = ptype_base[hash].next;
2778 return list_entry(nxt, struct packet_type, list);
2781 static void ptype_seq_stop(struct seq_file *seq, void *v)
2787 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2789 #ifdef CONFIG_KALLSYMS
2790 unsigned long offset = 0, symsize;
2791 const char *symname;
2795 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2802 modname = delim = "";
2803 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2809 seq_printf(seq, "[%p]", sym);
2812 static int ptype_seq_show(struct seq_file *seq, void *v)
2814 struct packet_type *pt = v;
2816 if (v == SEQ_START_TOKEN)
2817 seq_puts(seq, "Type Device Function\n");
2818 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
2819 if (pt->type == htons(ETH_P_ALL))
2820 seq_puts(seq, "ALL ");
2822 seq_printf(seq, "%04x", ntohs(pt->type));
2824 seq_printf(seq, " %-8s ",
2825 pt->dev ? pt->dev->name : "");
2826 ptype_seq_decode(seq, pt->func);
2827 seq_putc(seq, '\n');
2833 static const struct seq_operations ptype_seq_ops = {
2834 .start = ptype_seq_start,
2835 .next = ptype_seq_next,
2836 .stop = ptype_seq_stop,
2837 .show = ptype_seq_show,
2840 static int ptype_seq_open(struct inode *inode, struct file *file)
2842 return seq_open_net(inode, file, &ptype_seq_ops,
2843 sizeof(struct seq_net_private));
2846 static const struct file_operations ptype_seq_fops = {
2847 .owner = THIS_MODULE,
2848 .open = ptype_seq_open,
2850 .llseek = seq_lseek,
2851 .release = seq_release_net,
2855 static int __net_init dev_proc_net_init(struct net *net)
2859 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2861 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2863 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2866 if (wext_proc_init(net))
2872 proc_net_remove(net, "ptype");
2874 proc_net_remove(net, "softnet_stat");
2876 proc_net_remove(net, "dev");
2880 static void __net_exit dev_proc_net_exit(struct net *net)
2882 wext_proc_exit(net);
2884 proc_net_remove(net, "ptype");
2885 proc_net_remove(net, "softnet_stat");
2886 proc_net_remove(net, "dev");
2889 static struct pernet_operations __net_initdata dev_proc_ops = {
2890 .init = dev_proc_net_init,
2891 .exit = dev_proc_net_exit,
2894 static int __init dev_proc_init(void)
2896 return register_pernet_subsys(&dev_proc_ops);
2899 #define dev_proc_init() 0
2900 #endif /* CONFIG_PROC_FS */
2904 * netdev_set_master - set up master/slave pair
2905 * @slave: slave device
2906 * @master: new master device
2908 * Changes the master device of the slave. Pass %NULL to break the
2909 * bonding. The caller must hold the RTNL semaphore. On a failure
2910 * a negative errno code is returned. On success the reference counts
2911 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2912 * function returns zero.
2914 int netdev_set_master(struct net_device *slave, struct net_device *master)
2916 struct net_device *old = slave->master;
2926 slave->master = master;
2934 slave->flags |= IFF_SLAVE;
2936 slave->flags &= ~IFF_SLAVE;
2938 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2942 static int __dev_set_promiscuity(struct net_device *dev, int inc)
2944 unsigned short old_flags = dev->flags;
2948 dev->flags |= IFF_PROMISC;
2949 dev->promiscuity += inc;
2950 if (dev->promiscuity == 0) {
2953 * If inc causes overflow, untouch promisc and return error.
2956 dev->flags &= ~IFF_PROMISC;
2958 dev->promiscuity -= inc;
2959 printk(KERN_WARNING "%s: promiscuity touches roof, "
2960 "set promiscuity failed, promiscuity feature "
2961 "of device might be broken.\n", dev->name);
2965 if (dev->flags != old_flags) {
2966 printk(KERN_INFO "device %s %s promiscuous mode\n",
2967 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2970 audit_log(current->audit_context, GFP_ATOMIC,
2971 AUDIT_ANOM_PROMISCUOUS,
2972 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2973 dev->name, (dev->flags & IFF_PROMISC),
2974 (old_flags & IFF_PROMISC),
2975 audit_get_loginuid(current),
2976 current->uid, current->gid,
2977 audit_get_sessionid(current));
2979 if (dev->change_rx_flags)
2980 dev->change_rx_flags(dev, IFF_PROMISC);
2986 * dev_set_promiscuity - update promiscuity count on a device
2990 * Add or remove promiscuity from a device. While the count in the device
2991 * remains above zero the interface remains promiscuous. Once it hits zero
2992 * the device reverts back to normal filtering operation. A negative inc
2993 * value is used to drop promiscuity on the device.
2994 * Return 0 if successful or a negative errno code on error.
2996 int dev_set_promiscuity(struct net_device *dev, int inc)
2998 unsigned short old_flags = dev->flags;
3001 err = __dev_set_promiscuity(dev, inc);
3004 if (dev->flags != old_flags)
3005 dev_set_rx_mode(dev);
3010 * dev_set_allmulti - update allmulti count on a device
3014 * Add or remove reception of all multicast frames to a device. While the
3015 * count in the device remains above zero the interface remains listening
3016 * to all interfaces. Once it hits zero the device reverts back to normal
3017 * filtering operation. A negative @inc value is used to drop the counter
3018 * when releasing a resource needing all multicasts.
3019 * Return 0 if successful or a negative errno code on error.
3022 int dev_set_allmulti(struct net_device *dev, int inc)
3024 unsigned short old_flags = dev->flags;
3028 dev->flags |= IFF_ALLMULTI;
3029 dev->allmulti += inc;
3030 if (dev->allmulti == 0) {
3033 * If inc causes overflow, untouch allmulti and return error.
3036 dev->flags &= ~IFF_ALLMULTI;
3038 dev->allmulti -= inc;
3039 printk(KERN_WARNING "%s: allmulti touches roof, "
3040 "set allmulti failed, allmulti feature of "
3041 "device might be broken.\n", dev->name);
3045 if (dev->flags ^ old_flags) {
3046 if (dev->change_rx_flags)
3047 dev->change_rx_flags(dev, IFF_ALLMULTI);
3048 dev_set_rx_mode(dev);
3054 * Upload unicast and multicast address lists to device and
3055 * configure RX filtering. When the device doesn't support unicast
3056 * filtering it is put in promiscuous mode while unicast addresses
3059 void __dev_set_rx_mode(struct net_device *dev)
3061 /* dev_open will call this function so the list will stay sane. */
3062 if (!(dev->flags&IFF_UP))
3065 if (!netif_device_present(dev))
3068 if (dev->set_rx_mode)
3069 dev->set_rx_mode(dev);
3071 /* Unicast addresses changes may only happen under the rtnl,
3072 * therefore calling __dev_set_promiscuity here is safe.
3074 if (dev->uc_count > 0 && !dev->uc_promisc) {
3075 __dev_set_promiscuity(dev, 1);
3076 dev->uc_promisc = 1;
3077 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3078 __dev_set_promiscuity(dev, -1);
3079 dev->uc_promisc = 0;
3082 if (dev->set_multicast_list)
3083 dev->set_multicast_list(dev);
3087 void dev_set_rx_mode(struct net_device *dev)
3089 netif_addr_lock_bh(dev);
3090 __dev_set_rx_mode(dev);
3091 netif_addr_unlock_bh(dev);
3094 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3095 void *addr, int alen, int glbl)
3097 struct dev_addr_list *da;
3099 for (; (da = *list) != NULL; list = &da->next) {
3100 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3101 alen == da->da_addrlen) {
3103 int old_glbl = da->da_gusers;
3120 int __dev_addr_add(struct dev_addr_list **list, int *count,
3121 void *addr, int alen, int glbl)
3123 struct dev_addr_list *da;
3125 for (da = *list; da != NULL; da = da->next) {
3126 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3127 da->da_addrlen == alen) {
3129 int old_glbl = da->da_gusers;
3139 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3142 memcpy(da->da_addr, addr, alen);
3143 da->da_addrlen = alen;
3145 da->da_gusers = glbl ? 1 : 0;
3153 * dev_unicast_delete - Release secondary unicast address.
3155 * @addr: address to delete
3156 * @alen: length of @addr
3158 * Release reference to a secondary unicast address and remove it
3159 * from the device if the reference count drops to zero.
3161 * The caller must hold the rtnl_mutex.
3163 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3169 netif_addr_lock_bh(dev);
3170 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3172 __dev_set_rx_mode(dev);
3173 netif_addr_unlock_bh(dev);
3176 EXPORT_SYMBOL(dev_unicast_delete);
3179 * dev_unicast_add - add a secondary unicast address
3181 * @addr: address to add
3182 * @alen: length of @addr
3184 * Add a secondary unicast address to the device or increase
3185 * the reference count if it already exists.
3187 * The caller must hold the rtnl_mutex.
3189 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3195 netif_addr_lock_bh(dev);
3196 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3198 __dev_set_rx_mode(dev);
3199 netif_addr_unlock_bh(dev);
3202 EXPORT_SYMBOL(dev_unicast_add);
3204 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3205 struct dev_addr_list **from, int *from_count)
3207 struct dev_addr_list *da, *next;
3211 while (da != NULL) {
3213 if (!da->da_synced) {
3214 err = __dev_addr_add(to, to_count,
3215 da->da_addr, da->da_addrlen, 0);
3220 } else if (da->da_users == 1) {
3221 __dev_addr_delete(to, to_count,
3222 da->da_addr, da->da_addrlen, 0);
3223 __dev_addr_delete(from, from_count,
3224 da->da_addr, da->da_addrlen, 0);
3231 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3232 struct dev_addr_list **from, int *from_count)
3234 struct dev_addr_list *da, *next;
3237 while (da != NULL) {
3239 if (da->da_synced) {
3240 __dev_addr_delete(to, to_count,
3241 da->da_addr, da->da_addrlen, 0);
3243 __dev_addr_delete(from, from_count,
3244 da->da_addr, da->da_addrlen, 0);
3251 * dev_unicast_sync - Synchronize device's unicast list to another device
3252 * @to: destination device
3253 * @from: source device
3255 * Add newly added addresses to the destination device and release
3256 * addresses that have no users left. The source device must be
3257 * locked by netif_tx_lock_bh.
3259 * This function is intended to be called from the dev->set_rx_mode
3260 * function of layered software devices.
3262 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3266 netif_addr_lock_bh(to);
3267 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3268 &from->uc_list, &from->uc_count);
3270 __dev_set_rx_mode(to);
3271 netif_addr_unlock_bh(to);
3274 EXPORT_SYMBOL(dev_unicast_sync);
3277 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3278 * @to: destination device
3279 * @from: source device
3281 * Remove all addresses that were added to the destination device by
3282 * dev_unicast_sync(). This function is intended to be called from the
3283 * dev->stop function of layered software devices.
3285 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3287 netif_addr_lock_bh(from);
3288 netif_addr_lock(to);
3290 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3291 &from->uc_list, &from->uc_count);
3292 __dev_set_rx_mode(to);
3294 netif_addr_unlock(to);
3295 netif_addr_unlock_bh(from);
3297 EXPORT_SYMBOL(dev_unicast_unsync);
3299 static void __dev_addr_discard(struct dev_addr_list **list)
3301 struct dev_addr_list *tmp;
3303 while (*list != NULL) {
3306 if (tmp->da_users > tmp->da_gusers)
3307 printk("__dev_addr_discard: address leakage! "
3308 "da_users=%d\n", tmp->da_users);
3313 static void dev_addr_discard(struct net_device *dev)
3315 netif_addr_lock_bh(dev);
3317 __dev_addr_discard(&dev->uc_list);
3320 __dev_addr_discard(&dev->mc_list);
3323 netif_addr_unlock_bh(dev);
3326 unsigned dev_get_flags(const struct net_device *dev)
3330 flags = (dev->flags & ~(IFF_PROMISC |
3335 (dev->gflags & (IFF_PROMISC |
3338 if (netif_running(dev)) {
3339 if (netif_oper_up(dev))
3340 flags |= IFF_RUNNING;
3341 if (netif_carrier_ok(dev))
3342 flags |= IFF_LOWER_UP;
3343 if (netif_dormant(dev))
3344 flags |= IFF_DORMANT;
3350 int dev_change_flags(struct net_device *dev, unsigned flags)
3353 int old_flags = dev->flags;
3358 * Set the flags on our device.
3361 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3362 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3364 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3368 * Load in the correct multicast list now the flags have changed.
3371 if (dev->change_rx_flags && (old_flags ^ flags) & IFF_MULTICAST)
3372 dev->change_rx_flags(dev, IFF_MULTICAST);
3374 dev_set_rx_mode(dev);
3377 * Have we downed the interface. We handle IFF_UP ourselves
3378 * according to user attempts to set it, rather than blindly
3383 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3384 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3387 dev_set_rx_mode(dev);
3390 if (dev->flags & IFF_UP &&
3391 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3393 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3395 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3396 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3397 dev->gflags ^= IFF_PROMISC;
3398 dev_set_promiscuity(dev, inc);
3401 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3402 is important. Some (broken) drivers set IFF_PROMISC, when
3403 IFF_ALLMULTI is requested not asking us and not reporting.
3405 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3406 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3407 dev->gflags ^= IFF_ALLMULTI;
3408 dev_set_allmulti(dev, inc);
3411 /* Exclude state transition flags, already notified */
3412 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3414 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3419 int dev_set_mtu(struct net_device *dev, int new_mtu)
3423 if (new_mtu == dev->mtu)
3426 /* MTU must be positive. */
3430 if (!netif_device_present(dev))
3434 if (dev->change_mtu)
3435 err = dev->change_mtu(dev, new_mtu);
3438 if (!err && dev->flags & IFF_UP)
3439 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3443 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3447 if (!dev->set_mac_address)
3449 if (sa->sa_family != dev->type)
3451 if (!netif_device_present(dev))
3453 err = dev->set_mac_address(dev, sa);
3455 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3460 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3462 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3465 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3471 case SIOCGIFFLAGS: /* Get interface flags */
3472 ifr->ifr_flags = dev_get_flags(dev);
3475 case SIOCGIFMETRIC: /* Get the metric on the interface
3476 (currently unused) */
3477 ifr->ifr_metric = 0;
3480 case SIOCGIFMTU: /* Get the MTU of a device */
3481 ifr->ifr_mtu = dev->mtu;
3486 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3488 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3489 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3490 ifr->ifr_hwaddr.sa_family = dev->type;
3498 ifr->ifr_map.mem_start = dev->mem_start;
3499 ifr->ifr_map.mem_end = dev->mem_end;
3500 ifr->ifr_map.base_addr = dev->base_addr;
3501 ifr->ifr_map.irq = dev->irq;
3502 ifr->ifr_map.dma = dev->dma;
3503 ifr->ifr_map.port = dev->if_port;
3507 ifr->ifr_ifindex = dev->ifindex;
3511 ifr->ifr_qlen = dev->tx_queue_len;
3515 /* dev_ioctl() should ensure this case
3527 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3529 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3532 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3538 case SIOCSIFFLAGS: /* Set interface flags */
3539 return dev_change_flags(dev, ifr->ifr_flags);
3541 case SIOCSIFMETRIC: /* Set the metric on the interface
3542 (currently unused) */
3545 case SIOCSIFMTU: /* Set the MTU of a device */
3546 return dev_set_mtu(dev, ifr->ifr_mtu);
3549 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3551 case SIOCSIFHWBROADCAST:
3552 if (ifr->ifr_hwaddr.sa_family != dev->type)
3554 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3555 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3556 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3560 if (dev->set_config) {
3561 if (!netif_device_present(dev))
3563 return dev->set_config(dev, &ifr->ifr_map);
3568 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3569 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3571 if (!netif_device_present(dev))
3573 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3577 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3578 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3580 if (!netif_device_present(dev))
3582 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3586 if (ifr->ifr_qlen < 0)
3588 dev->tx_queue_len = ifr->ifr_qlen;
3592 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3593 return dev_change_name(dev, ifr->ifr_newname);
3596 * Unknown or private ioctl
3600 if ((cmd >= SIOCDEVPRIVATE &&
3601 cmd <= SIOCDEVPRIVATE + 15) ||
3602 cmd == SIOCBONDENSLAVE ||
3603 cmd == SIOCBONDRELEASE ||
3604 cmd == SIOCBONDSETHWADDR ||
3605 cmd == SIOCBONDSLAVEINFOQUERY ||
3606 cmd == SIOCBONDINFOQUERY ||
3607 cmd == SIOCBONDCHANGEACTIVE ||
3608 cmd == SIOCGMIIPHY ||
3609 cmd == SIOCGMIIREG ||
3610 cmd == SIOCSMIIREG ||
3611 cmd == SIOCBRADDIF ||
3612 cmd == SIOCBRDELIF ||
3613 cmd == SIOCWANDEV) {
3615 if (dev->do_ioctl) {
3616 if (netif_device_present(dev))
3617 err = dev->do_ioctl(dev, ifr,
3630 * This function handles all "interface"-type I/O control requests. The actual
3631 * 'doing' part of this is dev_ifsioc above.
3635 * dev_ioctl - network device ioctl
3636 * @net: the applicable net namespace
3637 * @cmd: command to issue
3638 * @arg: pointer to a struct ifreq in user space
3640 * Issue ioctl functions to devices. This is normally called by the
3641 * user space syscall interfaces but can sometimes be useful for
3642 * other purposes. The return value is the return from the syscall if
3643 * positive or a negative errno code on error.
3646 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3652 /* One special case: SIOCGIFCONF takes ifconf argument
3653 and requires shared lock, because it sleeps writing
3657 if (cmd == SIOCGIFCONF) {
3659 ret = dev_ifconf(net, (char __user *) arg);
3663 if (cmd == SIOCGIFNAME)
3664 return dev_ifname(net, (struct ifreq __user *)arg);
3666 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3669 ifr.ifr_name[IFNAMSIZ-1] = 0;
3671 colon = strchr(ifr.ifr_name, ':');
3676 * See which interface the caller is talking about.
3681 * These ioctl calls:
3682 * - can be done by all.
3683 * - atomic and do not require locking.
3694 dev_load(net, ifr.ifr_name);
3695 read_lock(&dev_base_lock);
3696 ret = dev_ifsioc_locked(net, &ifr, cmd);
3697 read_unlock(&dev_base_lock);
3701 if (copy_to_user(arg, &ifr,
3702 sizeof(struct ifreq)))
3708 dev_load(net, ifr.ifr_name);
3710 ret = dev_ethtool(net, &ifr);
3715 if (copy_to_user(arg, &ifr,
3716 sizeof(struct ifreq)))
3722 * These ioctl calls:
3723 * - require superuser power.
3724 * - require strict serialization.
3730 if (!capable(CAP_NET_ADMIN))
3732 dev_load(net, ifr.ifr_name);
3734 ret = dev_ifsioc(net, &ifr, cmd);
3739 if (copy_to_user(arg, &ifr,
3740 sizeof(struct ifreq)))
3746 * These ioctl calls:
3747 * - require superuser power.
3748 * - require strict serialization.
3749 * - do not return a value
3759 case SIOCSIFHWBROADCAST:
3762 case SIOCBONDENSLAVE:
3763 case SIOCBONDRELEASE:
3764 case SIOCBONDSETHWADDR:
3765 case SIOCBONDCHANGEACTIVE:
3768 if (!capable(CAP_NET_ADMIN))
3771 case SIOCBONDSLAVEINFOQUERY:
3772 case SIOCBONDINFOQUERY:
3773 dev_load(net, ifr.ifr_name);
3775 ret = dev_ifsioc(net, &ifr, cmd);
3780 /* Get the per device memory space. We can add this but
3781 * currently do not support it */
3783 /* Set the per device memory buffer space.
3784 * Not applicable in our case */
3789 * Unknown or private ioctl.
3792 if (cmd == SIOCWANDEV ||
3793 (cmd >= SIOCDEVPRIVATE &&
3794 cmd <= SIOCDEVPRIVATE + 15)) {
3795 dev_load(net, ifr.ifr_name);
3797 ret = dev_ifsioc(net, &ifr, cmd);
3799 if (!ret && copy_to_user(arg, &ifr,
3800 sizeof(struct ifreq)))
3804 /* Take care of Wireless Extensions */
3805 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3806 return wext_handle_ioctl(net, &ifr, cmd, arg);
3813 * dev_new_index - allocate an ifindex
3814 * @net: the applicable net namespace
3816 * Returns a suitable unique value for a new device interface
3817 * number. The caller must hold the rtnl semaphore or the
3818 * dev_base_lock to be sure it remains unique.
3820 static int dev_new_index(struct net *net)
3826 if (!__dev_get_by_index(net, ifindex))
3831 /* Delayed registration/unregisteration */
3832 static DEFINE_SPINLOCK(net_todo_list_lock);
3833 static LIST_HEAD(net_todo_list);
3835 static void net_set_todo(struct net_device *dev)
3837 spin_lock(&net_todo_list_lock);
3838 list_add_tail(&dev->todo_list, &net_todo_list);
3839 spin_unlock(&net_todo_list_lock);
3842 static void rollback_registered(struct net_device *dev)
3844 BUG_ON(dev_boot_phase);
3847 /* Some devices call without registering for initialization unwind. */
3848 if (dev->reg_state == NETREG_UNINITIALIZED) {
3849 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3850 "was registered\n", dev->name, dev);
3856 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3858 /* If device is running, close it first. */
3861 /* And unlink it from device chain. */
3862 unlist_netdevice(dev);
3864 dev->reg_state = NETREG_UNREGISTERING;
3868 /* Shutdown queueing discipline. */
3872 /* Notify protocols, that we are about to destroy
3873 this device. They should clean all the things.
3875 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3878 * Flush the unicast and multicast chains
3880 dev_addr_discard(dev);
3885 /* Notifier chain MUST detach us from master device. */
3886 WARN_ON(dev->master);
3888 /* Remove entries from kobject tree */
3889 netdev_unregister_kobject(dev);
3896 static void __netdev_init_queue_locks_one(struct net_device *dev,
3897 struct netdev_queue *dev_queue,
3900 spin_lock_init(&dev_queue->_xmit_lock);
3901 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
3902 dev_queue->xmit_lock_owner = -1;
3905 static void netdev_init_queue_locks(struct net_device *dev)
3907 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
3908 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
3912 * register_netdevice - register a network device
3913 * @dev: device to register
3915 * Take a completed network device structure and add it to the kernel
3916 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3917 * chain. 0 is returned on success. A negative errno code is returned
3918 * on a failure to set up the device, or if the name is a duplicate.
3920 * Callers must hold the rtnl semaphore. You may want
3921 * register_netdev() instead of this.
3924 * The locking appears insufficient to guarantee two parallel registers
3925 * will not get the same name.
3928 int register_netdevice(struct net_device *dev)
3930 struct hlist_head *head;
3931 struct hlist_node *p;
3935 BUG_ON(dev_boot_phase);
3940 /* When net_device's are persistent, this will be fatal. */
3941 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3942 BUG_ON(!dev_net(dev));
3945 spin_lock_init(&dev->addr_list_lock);
3946 netdev_set_addr_lockdep_class(dev);
3947 netdev_init_queue_locks(dev);
3951 /* Init, if this function is available */
3953 ret = dev->init(dev);
3961 if (!dev_valid_name(dev->name)) {
3966 dev->ifindex = dev_new_index(net);
3967 if (dev->iflink == -1)
3968 dev->iflink = dev->ifindex;
3970 /* Check for existence of name */
3971 head = dev_name_hash(net, dev->name);
3972 hlist_for_each(p, head) {
3973 struct net_device *d
3974 = hlist_entry(p, struct net_device, name_hlist);
3975 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3981 /* Fix illegal checksum combinations */
3982 if ((dev->features & NETIF_F_HW_CSUM) &&
3983 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3984 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3986 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3989 if ((dev->features & NETIF_F_NO_CSUM) &&
3990 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3991 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3993 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3997 /* Fix illegal SG+CSUM combinations. */
3998 if ((dev->features & NETIF_F_SG) &&
3999 !(dev->features & NETIF_F_ALL_CSUM)) {
4000 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
4002 dev->features &= ~NETIF_F_SG;
4005 /* TSO requires that SG is present as well. */
4006 if ((dev->features & NETIF_F_TSO) &&
4007 !(dev->features & NETIF_F_SG)) {
4008 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
4010 dev->features &= ~NETIF_F_TSO;
4012 if (dev->features & NETIF_F_UFO) {
4013 if (!(dev->features & NETIF_F_HW_CSUM)) {
4014 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
4015 "NETIF_F_HW_CSUM feature.\n",
4017 dev->features &= ~NETIF_F_UFO;
4019 if (!(dev->features & NETIF_F_SG)) {
4020 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
4021 "NETIF_F_SG feature.\n",
4023 dev->features &= ~NETIF_F_UFO;
4027 /* Enable software GSO if SG is supported. */
4028 if (dev->features & NETIF_F_SG)
4029 dev->features |= NETIF_F_GSO;
4031 netdev_initialize_kobject(dev);
4032 ret = netdev_register_kobject(dev);
4035 dev->reg_state = NETREG_REGISTERED;
4038 * Default initial state at registry is that the
4039 * device is present.
4042 set_bit(__LINK_STATE_PRESENT, &dev->state);
4044 dev_init_scheduler(dev);
4046 list_netdevice(dev);
4048 /* Notify protocols, that a new device appeared. */
4049 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4050 ret = notifier_to_errno(ret);
4052 rollback_registered(dev);
4053 dev->reg_state = NETREG_UNREGISTERED;
4066 * register_netdev - register a network device
4067 * @dev: device to register
4069 * Take a completed network device structure and add it to the kernel
4070 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4071 * chain. 0 is returned on success. A negative errno code is returned
4072 * on a failure to set up the device, or if the name is a duplicate.
4074 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4075 * and expands the device name if you passed a format string to
4078 int register_netdev(struct net_device *dev)
4085 * If the name is a format string the caller wants us to do a
4088 if (strchr(dev->name, '%')) {
4089 err = dev_alloc_name(dev, dev->name);
4094 err = register_netdevice(dev);
4099 EXPORT_SYMBOL(register_netdev);
4102 * netdev_wait_allrefs - wait until all references are gone.
4104 * This is called when unregistering network devices.
4106 * Any protocol or device that holds a reference should register
4107 * for netdevice notification, and cleanup and put back the
4108 * reference if they receive an UNREGISTER event.
4109 * We can get stuck here if buggy protocols don't correctly
4112 static void netdev_wait_allrefs(struct net_device *dev)
4114 unsigned long rebroadcast_time, warning_time;
4116 rebroadcast_time = warning_time = jiffies;
4117 while (atomic_read(&dev->refcnt) != 0) {
4118 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4121 /* Rebroadcast unregister notification */
4122 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4124 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4126 /* We must not have linkwatch events
4127 * pending on unregister. If this
4128 * happens, we simply run the queue
4129 * unscheduled, resulting in a noop
4132 linkwatch_run_queue();
4137 rebroadcast_time = jiffies;
4142 if (time_after(jiffies, warning_time + 10 * HZ)) {
4143 printk(KERN_EMERG "unregister_netdevice: "
4144 "waiting for %s to become free. Usage "
4146 dev->name, atomic_read(&dev->refcnt));
4147 warning_time = jiffies;
4156 * register_netdevice(x1);
4157 * register_netdevice(x2);
4159 * unregister_netdevice(y1);
4160 * unregister_netdevice(y2);
4166 * We are invoked by rtnl_unlock() after it drops the semaphore.
4167 * This allows us to deal with problems:
4168 * 1) We can delete sysfs objects which invoke hotplug
4169 * without deadlocking with linkwatch via keventd.
4170 * 2) Since we run with the RTNL semaphore not held, we can sleep
4171 * safely in order to wait for the netdev refcnt to drop to zero.
4173 static DEFINE_MUTEX(net_todo_run_mutex);
4174 void netdev_run_todo(void)
4176 struct list_head list;
4178 /* Need to guard against multiple cpu's getting out of order. */
4179 mutex_lock(&net_todo_run_mutex);
4181 /* Not safe to do outside the semaphore. We must not return
4182 * until all unregister events invoked by the local processor
4183 * have been completed (either by this todo run, or one on
4186 if (list_empty(&net_todo_list))
4189 /* Snapshot list, allow later requests */
4190 spin_lock(&net_todo_list_lock);
4191 list_replace_init(&net_todo_list, &list);
4192 spin_unlock(&net_todo_list_lock);
4194 while (!list_empty(&list)) {
4195 struct net_device *dev
4196 = list_entry(list.next, struct net_device, todo_list);
4197 list_del(&dev->todo_list);
4199 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4200 printk(KERN_ERR "network todo '%s' but state %d\n",
4201 dev->name, dev->reg_state);
4206 dev->reg_state = NETREG_UNREGISTERED;
4208 on_each_cpu(flush_backlog, dev, 1);
4210 netdev_wait_allrefs(dev);
4213 BUG_ON(atomic_read(&dev->refcnt));
4214 WARN_ON(dev->ip_ptr);
4215 WARN_ON(dev->ip6_ptr);
4216 WARN_ON(dev->dn_ptr);
4218 if (dev->destructor)
4219 dev->destructor(dev);
4221 /* Free network device */
4222 kobject_put(&dev->dev.kobj);
4226 mutex_unlock(&net_todo_run_mutex);
4229 static struct net_device_stats *internal_stats(struct net_device *dev)
4234 static void netdev_init_one_queue(struct net_device *dev,
4235 struct netdev_queue *queue,
4241 static void netdev_init_queues(struct net_device *dev)
4243 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4244 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4245 spin_lock_init(&dev->tx_global_lock);
4249 * alloc_netdev_mq - allocate network device
4250 * @sizeof_priv: size of private data to allocate space for
4251 * @name: device name format string
4252 * @setup: callback to initialize device
4253 * @queue_count: the number of subqueues to allocate
4255 * Allocates a struct net_device with private data area for driver use
4256 * and performs basic initialization. Also allocates subquue structs
4257 * for each queue on the device at the end of the netdevice.
4259 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4260 void (*setup)(struct net_device *), unsigned int queue_count)
4262 struct netdev_queue *tx;
4263 struct net_device *dev;
4267 BUG_ON(strlen(name) >= sizeof(dev->name));
4269 alloc_size = sizeof(struct net_device);
4271 /* ensure 32-byte alignment of private area */
4272 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4273 alloc_size += sizeof_priv;
4275 /* ensure 32-byte alignment of whole construct */
4276 alloc_size += NETDEV_ALIGN_CONST;
4278 p = kzalloc(alloc_size, GFP_KERNEL);
4280 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4284 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
4286 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4292 dev = (struct net_device *)
4293 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4294 dev->padded = (char *)dev - (char *)p;
4295 dev_net_set(dev, &init_net);
4298 dev->num_tx_queues = queue_count;
4299 dev->real_num_tx_queues = queue_count;
4302 dev->priv = ((char *)dev +
4303 ((sizeof(struct net_device) + NETDEV_ALIGN_CONST)
4304 & ~NETDEV_ALIGN_CONST));
4307 dev->gso_max_size = GSO_MAX_SIZE;
4309 netdev_init_queues(dev);
4311 dev->get_stats = internal_stats;
4312 netpoll_netdev_init(dev);
4314 strcpy(dev->name, name);
4317 EXPORT_SYMBOL(alloc_netdev_mq);
4320 * free_netdev - free network device
4323 * This function does the last stage of destroying an allocated device
4324 * interface. The reference to the device object is released.
4325 * If this is the last reference then it will be freed.
4327 void free_netdev(struct net_device *dev)
4329 release_net(dev_net(dev));
4333 /* Compatibility with error handling in drivers */
4334 if (dev->reg_state == NETREG_UNINITIALIZED) {
4335 kfree((char *)dev - dev->padded);
4339 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4340 dev->reg_state = NETREG_RELEASED;
4342 /* will free via device release */
4343 put_device(&dev->dev);
4346 /* Synchronize with packet receive processing. */
4347 void synchronize_net(void)
4354 * unregister_netdevice - remove device from the kernel
4357 * This function shuts down a device interface and removes it
4358 * from the kernel tables.
4360 * Callers must hold the rtnl semaphore. You may want
4361 * unregister_netdev() instead of this.
4364 void unregister_netdevice(struct net_device *dev)
4368 rollback_registered(dev);
4369 /* Finish processing unregister after unlock */
4374 * unregister_netdev - remove device from the kernel
4377 * This function shuts down a device interface and removes it
4378 * from the kernel tables.
4380 * This is just a wrapper for unregister_netdevice that takes
4381 * the rtnl semaphore. In general you want to use this and not
4382 * unregister_netdevice.
4384 void unregister_netdev(struct net_device *dev)
4387 unregister_netdevice(dev);
4391 EXPORT_SYMBOL(unregister_netdev);
4394 * dev_change_net_namespace - move device to different nethost namespace
4396 * @net: network namespace
4397 * @pat: If not NULL name pattern to try if the current device name
4398 * is already taken in the destination network namespace.
4400 * This function shuts down a device interface and moves it
4401 * to a new network namespace. On success 0 is returned, on
4402 * a failure a netagive errno code is returned.
4404 * Callers must hold the rtnl semaphore.
4407 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4410 const char *destname;
4415 /* Don't allow namespace local devices to be moved. */
4417 if (dev->features & NETIF_F_NETNS_LOCAL)
4420 /* Ensure the device has been registrered */
4422 if (dev->reg_state != NETREG_REGISTERED)
4425 /* Get out if there is nothing todo */
4427 if (net_eq(dev_net(dev), net))
4430 /* Pick the destination device name, and ensure
4431 * we can use it in the destination network namespace.
4434 destname = dev->name;
4435 if (__dev_get_by_name(net, destname)) {
4436 /* We get here if we can't use the current device name */
4439 if (!dev_valid_name(pat))
4441 if (strchr(pat, '%')) {
4442 if (__dev_alloc_name(net, pat, buf) < 0)
4447 if (__dev_get_by_name(net, destname))
4452 * And now a mini version of register_netdevice unregister_netdevice.
4455 /* If device is running close it first. */
4458 /* And unlink it from device chain */
4460 unlist_netdevice(dev);
4464 /* Shutdown queueing discipline. */
4467 /* Notify protocols, that we are about to destroy
4468 this device. They should clean all the things.
4470 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4473 * Flush the unicast and multicast chains
4475 dev_addr_discard(dev);
4477 /* Actually switch the network namespace */
4478 dev_net_set(dev, net);
4480 /* Assign the new device name */
4481 if (destname != dev->name)
4482 strcpy(dev->name, destname);
4484 /* If there is an ifindex conflict assign a new one */
4485 if (__dev_get_by_index(net, dev->ifindex)) {
4486 int iflink = (dev->iflink == dev->ifindex);
4487 dev->ifindex = dev_new_index(net);
4489 dev->iflink = dev->ifindex;
4492 /* Fixup kobjects */
4493 netdev_unregister_kobject(dev);
4494 err = netdev_register_kobject(dev);
4497 /* Add the device back in the hashes */
4498 list_netdevice(dev);
4500 /* Notify protocols, that a new device appeared. */
4501 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4509 static int dev_cpu_callback(struct notifier_block *nfb,
4510 unsigned long action,
4513 struct sk_buff **list_skb;
4514 struct Qdisc **list_net;
4515 struct sk_buff *skb;
4516 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4517 struct softnet_data *sd, *oldsd;
4519 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4522 local_irq_disable();
4523 cpu = smp_processor_id();
4524 sd = &per_cpu(softnet_data, cpu);
4525 oldsd = &per_cpu(softnet_data, oldcpu);
4527 /* Find end of our completion_queue. */
4528 list_skb = &sd->completion_queue;
4530 list_skb = &(*list_skb)->next;
4531 /* Append completion queue from offline CPU. */
4532 *list_skb = oldsd->completion_queue;
4533 oldsd->completion_queue = NULL;
4535 /* Find end of our output_queue. */
4536 list_net = &sd->output_queue;
4538 list_net = &(*list_net)->next_sched;
4539 /* Append output queue from offline CPU. */
4540 *list_net = oldsd->output_queue;
4541 oldsd->output_queue = NULL;
4543 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4546 /* Process offline CPU's input_pkt_queue */
4547 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4553 #ifdef CONFIG_NET_DMA
4555 * net_dma_rebalance - try to maintain one DMA channel per CPU
4556 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4558 * This is called when the number of channels allocated to the net_dma client
4559 * changes. The net_dma client tries to have one DMA channel per CPU.
4562 static void net_dma_rebalance(struct net_dma *net_dma)
4564 unsigned int cpu, i, n, chan_idx;
4565 struct dma_chan *chan;
4567 if (cpus_empty(net_dma->channel_mask)) {
4568 for_each_online_cpu(cpu)
4569 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4574 cpu = first_cpu(cpu_online_map);
4576 for_each_cpu_mask_nr(chan_idx, net_dma->channel_mask) {
4577 chan = net_dma->channels[chan_idx];
4579 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4580 + (i < (num_online_cpus() %
4581 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4584 per_cpu(softnet_data, cpu).net_dma = chan;
4585 cpu = next_cpu(cpu, cpu_online_map);
4593 * netdev_dma_event - event callback for the net_dma_client
4594 * @client: should always be net_dma_client
4595 * @chan: DMA channel for the event
4596 * @state: DMA state to be handled
4598 static enum dma_state_client
4599 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4600 enum dma_state state)
4602 int i, found = 0, pos = -1;
4603 struct net_dma *net_dma =
4604 container_of(client, struct net_dma, client);
4605 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4607 spin_lock(&net_dma->lock);
4609 case DMA_RESOURCE_AVAILABLE:
4610 for (i = 0; i < nr_cpu_ids; i++)
4611 if (net_dma->channels[i] == chan) {
4614 } else if (net_dma->channels[i] == NULL && pos < 0)
4617 if (!found && pos >= 0) {
4619 net_dma->channels[pos] = chan;
4620 cpu_set(pos, net_dma->channel_mask);
4621 net_dma_rebalance(net_dma);
4624 case DMA_RESOURCE_REMOVED:
4625 for (i = 0; i < nr_cpu_ids; i++)
4626 if (net_dma->channels[i] == chan) {
4634 cpu_clear(pos, net_dma->channel_mask);
4635 net_dma->channels[i] = NULL;
4636 net_dma_rebalance(net_dma);
4642 spin_unlock(&net_dma->lock);
4648 * netdev_dma_regiser - register the networking subsystem as a DMA client
4650 static int __init netdev_dma_register(void)
4652 net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma),
4654 if (unlikely(!net_dma.channels)) {
4656 "netdev_dma: no memory for net_dma.channels\n");
4659 spin_lock_init(&net_dma.lock);
4660 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4661 dma_async_client_register(&net_dma.client);
4662 dma_async_client_chan_request(&net_dma.client);
4667 static int __init netdev_dma_register(void) { return -ENODEV; }
4668 #endif /* CONFIG_NET_DMA */
4671 * netdev_compute_feature - compute conjunction of two feature sets
4672 * @all: first feature set
4673 * @one: second feature set
4675 * Computes a new feature set after adding a device with feature set
4676 * @one to the master device with current feature set @all. Returns
4677 * the new feature set.
4679 int netdev_compute_features(unsigned long all, unsigned long one)
4681 /* if device needs checksumming, downgrade to hw checksumming */
4682 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4683 all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
4685 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4686 if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
4687 all ^= NETIF_F_HW_CSUM
4688 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4690 if (one & NETIF_F_GSO)
4691 one |= NETIF_F_GSO_SOFTWARE;
4695 * If even one device supports a GSO protocol with software fallback,
4696 * enable it for all.
4698 all |= one & NETIF_F_GSO_SOFTWARE;
4700 /* If even one device supports robust GSO, enable it for all. */
4701 if (one & NETIF_F_GSO_ROBUST)
4702 all |= NETIF_F_GSO_ROBUST;
4704 all &= one | NETIF_F_LLTX;
4706 if (!(all & NETIF_F_ALL_CSUM))
4708 if (!(all & NETIF_F_SG))
4709 all &= ~NETIF_F_GSO_MASK;
4713 EXPORT_SYMBOL(netdev_compute_features);
4715 static struct hlist_head *netdev_create_hash(void)
4718 struct hlist_head *hash;
4720 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4722 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4723 INIT_HLIST_HEAD(&hash[i]);
4728 /* Initialize per network namespace state */
4729 static int __net_init netdev_init(struct net *net)
4731 INIT_LIST_HEAD(&net->dev_base_head);
4733 net->dev_name_head = netdev_create_hash();
4734 if (net->dev_name_head == NULL)
4737 net->dev_index_head = netdev_create_hash();
4738 if (net->dev_index_head == NULL)
4744 kfree(net->dev_name_head);
4749 char *netdev_drivername(struct net_device *dev, char *buffer, int len)
4751 struct device_driver *driver;
4752 struct device *parent;
4754 if (len <= 0 || !buffer)
4758 parent = dev->dev.parent;
4763 driver = parent->driver;
4764 if (driver && driver->name)
4765 strlcpy(buffer, driver->name, len);
4769 static void __net_exit netdev_exit(struct net *net)
4771 kfree(net->dev_name_head);
4772 kfree(net->dev_index_head);
4775 static struct pernet_operations __net_initdata netdev_net_ops = {
4776 .init = netdev_init,
4777 .exit = netdev_exit,
4780 static void __net_exit default_device_exit(struct net *net)
4782 struct net_device *dev, *next;
4784 * Push all migratable of the network devices back to the
4785 * initial network namespace
4788 for_each_netdev_safe(net, dev, next) {
4790 char fb_name[IFNAMSIZ];
4792 /* Ignore unmoveable devices (i.e. loopback) */
4793 if (dev->features & NETIF_F_NETNS_LOCAL)
4796 /* Push remaing network devices to init_net */
4797 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
4798 err = dev_change_net_namespace(dev, &init_net, fb_name);
4800 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
4801 __func__, dev->name, err);
4808 static struct pernet_operations __net_initdata default_device_ops = {
4809 .exit = default_device_exit,
4813 * Initialize the DEV module. At boot time this walks the device list and
4814 * unhooks any devices that fail to initialise (normally hardware not
4815 * present) and leaves us with a valid list of present and active devices.
4820 * This is called single threaded during boot, so no need
4821 * to take the rtnl semaphore.
4823 static int __init net_dev_init(void)
4825 int i, rc = -ENOMEM;
4827 BUG_ON(!dev_boot_phase);
4829 if (dev_proc_init())
4832 if (netdev_kobject_init())
4835 INIT_LIST_HEAD(&ptype_all);
4836 for (i = 0; i < PTYPE_HASH_SIZE; i++)
4837 INIT_LIST_HEAD(&ptype_base[i]);
4839 if (register_pernet_subsys(&netdev_net_ops))
4842 if (register_pernet_device(&default_device_ops))
4846 * Initialise the packet receive queues.
4849 for_each_possible_cpu(i) {
4850 struct softnet_data *queue;
4852 queue = &per_cpu(softnet_data, i);
4853 skb_queue_head_init(&queue->input_pkt_queue);
4854 queue->completion_queue = NULL;
4855 INIT_LIST_HEAD(&queue->poll_list);
4857 queue->backlog.poll = process_backlog;
4858 queue->backlog.weight = weight_p;
4861 netdev_dma_register();
4865 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
4866 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
4868 hotcpu_notifier(dev_cpu_callback, 0);
4876 subsys_initcall(net_dev_init);
4878 EXPORT_SYMBOL(__dev_get_by_index);
4879 EXPORT_SYMBOL(__dev_get_by_name);
4880 EXPORT_SYMBOL(__dev_remove_pack);
4881 EXPORT_SYMBOL(dev_valid_name);
4882 EXPORT_SYMBOL(dev_add_pack);
4883 EXPORT_SYMBOL(dev_alloc_name);
4884 EXPORT_SYMBOL(dev_close);
4885 EXPORT_SYMBOL(dev_get_by_flags);
4886 EXPORT_SYMBOL(dev_get_by_index);
4887 EXPORT_SYMBOL(dev_get_by_name);
4888 EXPORT_SYMBOL(dev_open);
4889 EXPORT_SYMBOL(dev_queue_xmit);
4890 EXPORT_SYMBOL(dev_remove_pack);
4891 EXPORT_SYMBOL(dev_set_allmulti);
4892 EXPORT_SYMBOL(dev_set_promiscuity);
4893 EXPORT_SYMBOL(dev_change_flags);
4894 EXPORT_SYMBOL(dev_set_mtu);
4895 EXPORT_SYMBOL(dev_set_mac_address);
4896 EXPORT_SYMBOL(free_netdev);
4897 EXPORT_SYMBOL(netdev_boot_setup_check);
4898 EXPORT_SYMBOL(netdev_set_master);
4899 EXPORT_SYMBOL(netdev_state_change);
4900 EXPORT_SYMBOL(netif_receive_skb);
4901 EXPORT_SYMBOL(netif_rx);
4902 EXPORT_SYMBOL(register_gifconf);
4903 EXPORT_SYMBOL(register_netdevice);
4904 EXPORT_SYMBOL(register_netdevice_notifier);
4905 EXPORT_SYMBOL(skb_checksum_help);
4906 EXPORT_SYMBOL(synchronize_net);
4907 EXPORT_SYMBOL(unregister_netdevice);
4908 EXPORT_SYMBOL(unregister_netdevice_notifier);
4909 EXPORT_SYMBOL(net_enable_timestamp);
4910 EXPORT_SYMBOL(net_disable_timestamp);
4911 EXPORT_SYMBOL(dev_get_flags);
4913 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4914 EXPORT_SYMBOL(br_handle_frame_hook);
4915 EXPORT_SYMBOL(br_fdb_get_hook);
4916 EXPORT_SYMBOL(br_fdb_put_hook);
4920 EXPORT_SYMBOL(dev_load);
4923 EXPORT_PER_CPU_SYMBOL(softnet_data);