2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/mutex.h>
84 #include <linux/string.h>
86 #include <linux/socket.h>
87 #include <linux/sockios.h>
88 #include <linux/errno.h>
89 #include <linux/interrupt.h>
90 #include <linux/if_ether.h>
91 #include <linux/netdevice.h>
92 #include <linux/etherdevice.h>
93 #include <linux/ethtool.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
96 #include <net/net_namespace.h>
98 #include <linux/rtnetlink.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/stat.h>
102 #include <linux/if_bridge.h>
103 #include <linux/if_macvlan.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
130 #include "net-sysfs.h"
132 /* Instead of increasing this, you should create a hash table. */
133 #define MAX_GRO_SKBS 8
135 /* This should be increased if a protocol with a bigger head is added. */
136 #define GRO_MAX_HEAD (MAX_HEADER + 128)
139 * The list of packet types we will receive (as opposed to discard)
140 * and the routines to invoke.
142 * Why 16. Because with 16 the only overlap we get on a hash of the
143 * low nibble of the protocol value is RARP/SNAP/X.25.
145 * NOTE: That is no longer true with the addition of VLAN tags. Not
146 * sure which should go first, but I bet it won't make much
147 * difference if we are running VLANs. The good news is that
148 * this protocol won't be in the list unless compiled in, so
149 * the average user (w/out VLANs) will not be adversely affected.
166 #define PTYPE_HASH_SIZE (16)
167 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
169 static DEFINE_SPINLOCK(ptype_lock);
170 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
171 static struct list_head ptype_all __read_mostly; /* Taps */
174 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
177 * Pure readers hold dev_base_lock for reading.
179 * Writers must hold the rtnl semaphore while they loop through the
180 * dev_base_head list, and hold dev_base_lock for writing when they do the
181 * actual updates. This allows pure readers to access the list even
182 * while a writer is preparing to update it.
184 * To put it another way, dev_base_lock is held for writing only to
185 * protect against pure readers; the rtnl semaphore provides the
186 * protection against other writers.
188 * See, for example usages, register_netdevice() and
189 * unregister_netdevice(), which must be called with the rtnl
192 DEFINE_RWLOCK(dev_base_lock);
194 EXPORT_SYMBOL(dev_base_lock);
196 #define NETDEV_HASHBITS 8
197 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
199 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
201 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
202 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
205 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
207 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
210 /* Device list insertion */
211 static int list_netdevice(struct net_device *dev)
213 struct net *net = dev_net(dev);
217 write_lock_bh(&dev_base_lock);
218 list_add_tail(&dev->dev_list, &net->dev_base_head);
219 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
220 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
221 write_unlock_bh(&dev_base_lock);
225 /* Device list removal */
226 static void unlist_netdevice(struct net_device *dev)
230 /* Unlink dev from the device chain */
231 write_lock_bh(&dev_base_lock);
232 list_del(&dev->dev_list);
233 hlist_del(&dev->name_hlist);
234 hlist_del(&dev->index_hlist);
235 write_unlock_bh(&dev_base_lock);
242 static RAW_NOTIFIER_HEAD(netdev_chain);
245 * Device drivers call our routines to queue packets here. We empty the
246 * queue in the local softnet handler.
249 DEFINE_PER_CPU(struct softnet_data, softnet_data);
251 #ifdef CONFIG_LOCKDEP
253 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
254 * according to dev->type
256 static const unsigned short netdev_lock_type[] =
257 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
258 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
259 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
260 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
261 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
262 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
263 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
264 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
265 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
266 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
267 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
268 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
269 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
270 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
271 ARPHRD_PHONET_PIPE, ARPHRD_VOID, ARPHRD_NONE};
273 static const char *netdev_lock_name[] =
274 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
275 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
276 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
277 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
278 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
279 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
280 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
281 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
282 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
283 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
284 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
285 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
286 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
287 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
288 "_xmit_PHONET_PIPE", "_xmit_VOID", "_xmit_NONE"};
290 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
291 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
293 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
297 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
298 if (netdev_lock_type[i] == dev_type)
300 /* the last key is used by default */
301 return ARRAY_SIZE(netdev_lock_type) - 1;
304 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
305 unsigned short dev_type)
309 i = netdev_lock_pos(dev_type);
310 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
311 netdev_lock_name[i]);
314 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
318 i = netdev_lock_pos(dev->type);
319 lockdep_set_class_and_name(&dev->addr_list_lock,
320 &netdev_addr_lock_key[i],
321 netdev_lock_name[i]);
324 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
325 unsigned short dev_type)
328 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
333 /*******************************************************************************
335 Protocol management and registration routines
337 *******************************************************************************/
340 * Add a protocol ID to the list. Now that the input handler is
341 * smarter we can dispense with all the messy stuff that used to be
344 * BEWARE!!! Protocol handlers, mangling input packets,
345 * MUST BE last in hash buckets and checking protocol handlers
346 * MUST start from promiscuous ptype_all chain in net_bh.
347 * It is true now, do not change it.
348 * Explanation follows: if protocol handler, mangling packet, will
349 * be the first on list, it is not able to sense, that packet
350 * is cloned and should be copied-on-write, so that it will
351 * change it and subsequent readers will get broken packet.
356 * dev_add_pack - add packet handler
357 * @pt: packet type declaration
359 * Add a protocol handler to the networking stack. The passed &packet_type
360 * is linked into kernel lists and may not be freed until it has been
361 * removed from the kernel lists.
363 * This call does not sleep therefore it can not
364 * guarantee all CPU's that are in middle of receiving packets
365 * will see the new packet type (until the next received packet).
368 void dev_add_pack(struct packet_type *pt)
372 spin_lock_bh(&ptype_lock);
373 if (pt->type == htons(ETH_P_ALL))
374 list_add_rcu(&pt->list, &ptype_all);
376 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
377 list_add_rcu(&pt->list, &ptype_base[hash]);
379 spin_unlock_bh(&ptype_lock);
383 * __dev_remove_pack - remove packet handler
384 * @pt: packet type declaration
386 * Remove a protocol handler that was previously added to the kernel
387 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
388 * from the kernel lists and can be freed or reused once this function
391 * The packet type might still be in use by receivers
392 * and must not be freed until after all the CPU's have gone
393 * through a quiescent state.
395 void __dev_remove_pack(struct packet_type *pt)
397 struct list_head *head;
398 struct packet_type *pt1;
400 spin_lock_bh(&ptype_lock);
402 if (pt->type == htons(ETH_P_ALL))
405 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
407 list_for_each_entry(pt1, head, list) {
409 list_del_rcu(&pt->list);
414 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
416 spin_unlock_bh(&ptype_lock);
419 * dev_remove_pack - remove packet handler
420 * @pt: packet type declaration
422 * Remove a protocol handler that was previously added to the kernel
423 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
424 * from the kernel lists and can be freed or reused once this function
427 * This call sleeps to guarantee that no CPU is looking at the packet
430 void dev_remove_pack(struct packet_type *pt)
432 __dev_remove_pack(pt);
437 /******************************************************************************
439 Device Boot-time Settings Routines
441 *******************************************************************************/
443 /* Boot time configuration table */
444 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
447 * netdev_boot_setup_add - add new setup entry
448 * @name: name of the device
449 * @map: configured settings for the device
451 * Adds new setup entry to the dev_boot_setup list. The function
452 * returns 0 on error and 1 on success. This is a generic routine to
455 static int netdev_boot_setup_add(char *name, struct ifmap *map)
457 struct netdev_boot_setup *s;
461 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
462 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
463 memset(s[i].name, 0, sizeof(s[i].name));
464 strlcpy(s[i].name, name, IFNAMSIZ);
465 memcpy(&s[i].map, map, sizeof(s[i].map));
470 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
474 * netdev_boot_setup_check - check boot time settings
475 * @dev: the netdevice
477 * Check boot time settings for the device.
478 * The found settings are set for the device to be used
479 * later in the device probing.
480 * Returns 0 if no settings found, 1 if they are.
482 int netdev_boot_setup_check(struct net_device *dev)
484 struct netdev_boot_setup *s = dev_boot_setup;
487 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
488 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
489 !strcmp(dev->name, s[i].name)) {
490 dev->irq = s[i].map.irq;
491 dev->base_addr = s[i].map.base_addr;
492 dev->mem_start = s[i].map.mem_start;
493 dev->mem_end = s[i].map.mem_end;
502 * netdev_boot_base - get address from boot time settings
503 * @prefix: prefix for network device
504 * @unit: id for network device
506 * Check boot time settings for the base address of device.
507 * The found settings are set for the device to be used
508 * later in the device probing.
509 * Returns 0 if no settings found.
511 unsigned long netdev_boot_base(const char *prefix, int unit)
513 const struct netdev_boot_setup *s = dev_boot_setup;
517 sprintf(name, "%s%d", prefix, unit);
520 * If device already registered then return base of 1
521 * to indicate not to probe for this interface
523 if (__dev_get_by_name(&init_net, name))
526 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
527 if (!strcmp(name, s[i].name))
528 return s[i].map.base_addr;
533 * Saves at boot time configured settings for any netdevice.
535 int __init netdev_boot_setup(char *str)
540 str = get_options(str, ARRAY_SIZE(ints), ints);
545 memset(&map, 0, sizeof(map));
549 map.base_addr = ints[2];
551 map.mem_start = ints[3];
553 map.mem_end = ints[4];
555 /* Add new entry to the list */
556 return netdev_boot_setup_add(str, &map);
559 __setup("netdev=", netdev_boot_setup);
561 /*******************************************************************************
563 Device Interface Subroutines
565 *******************************************************************************/
568 * __dev_get_by_name - find a device by its name
569 * @net: the applicable net namespace
570 * @name: name to find
572 * Find an interface by name. Must be called under RTNL semaphore
573 * or @dev_base_lock. If the name is found a pointer to the device
574 * is returned. If the name is not found then %NULL is returned. The
575 * reference counters are not incremented so the caller must be
576 * careful with locks.
579 struct net_device *__dev_get_by_name(struct net *net, const char *name)
581 struct hlist_node *p;
583 hlist_for_each(p, dev_name_hash(net, name)) {
584 struct net_device *dev
585 = hlist_entry(p, struct net_device, name_hlist);
586 if (!strncmp(dev->name, name, IFNAMSIZ))
593 * dev_get_by_name - find a device by its name
594 * @net: the applicable net namespace
595 * @name: name to find
597 * Find an interface by name. This can be called from any
598 * context and does its own locking. The returned handle has
599 * the usage count incremented and the caller must use dev_put() to
600 * release it when it is no longer needed. %NULL is returned if no
601 * matching device is found.
604 struct net_device *dev_get_by_name(struct net *net, const char *name)
606 struct net_device *dev;
608 read_lock(&dev_base_lock);
609 dev = __dev_get_by_name(net, name);
612 read_unlock(&dev_base_lock);
617 * __dev_get_by_index - find a device by its ifindex
618 * @net: the applicable net namespace
619 * @ifindex: index of device
621 * Search for an interface by index. Returns %NULL if the device
622 * is not found or a pointer to the device. The device has not
623 * had its reference counter increased so the caller must be careful
624 * about locking. The caller must hold either the RTNL semaphore
628 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
630 struct hlist_node *p;
632 hlist_for_each(p, dev_index_hash(net, ifindex)) {
633 struct net_device *dev
634 = hlist_entry(p, struct net_device, index_hlist);
635 if (dev->ifindex == ifindex)
643 * dev_get_by_index - find a device by its ifindex
644 * @net: the applicable net namespace
645 * @ifindex: index of device
647 * Search for an interface by index. Returns NULL if the device
648 * is not found or a pointer to the device. The device returned has
649 * had a reference added and the pointer is safe until the user calls
650 * dev_put to indicate they have finished with it.
653 struct net_device *dev_get_by_index(struct net *net, int ifindex)
655 struct net_device *dev;
657 read_lock(&dev_base_lock);
658 dev = __dev_get_by_index(net, ifindex);
661 read_unlock(&dev_base_lock);
666 * dev_getbyhwaddr - find a device by its hardware address
667 * @net: the applicable net namespace
668 * @type: media type of device
669 * @ha: hardware address
671 * Search for an interface by MAC address. Returns NULL if the device
672 * is not found or a pointer to the device. The caller must hold the
673 * rtnl semaphore. The returned device has not had its ref count increased
674 * and the caller must therefore be careful about locking
677 * If the API was consistent this would be __dev_get_by_hwaddr
680 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
682 struct net_device *dev;
686 for_each_netdev(net, dev)
687 if (dev->type == type &&
688 !memcmp(dev->dev_addr, ha, dev->addr_len))
694 EXPORT_SYMBOL(dev_getbyhwaddr);
696 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
698 struct net_device *dev;
701 for_each_netdev(net, dev)
702 if (dev->type == type)
708 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
710 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
712 struct net_device *dev;
715 dev = __dev_getfirstbyhwtype(net, type);
722 EXPORT_SYMBOL(dev_getfirstbyhwtype);
725 * dev_get_by_flags - find any device with given flags
726 * @net: the applicable net namespace
727 * @if_flags: IFF_* values
728 * @mask: bitmask of bits in if_flags to check
730 * Search for any interface with the given flags. Returns NULL if a device
731 * is not found or a pointer to the device. The device returned has
732 * had a reference added and the pointer is safe until the user calls
733 * dev_put to indicate they have finished with it.
736 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
738 struct net_device *dev, *ret;
741 read_lock(&dev_base_lock);
742 for_each_netdev(net, dev) {
743 if (((dev->flags ^ if_flags) & mask) == 0) {
749 read_unlock(&dev_base_lock);
754 * dev_valid_name - check if name is okay for network device
757 * Network device names need to be valid file names to
758 * to allow sysfs to work. We also disallow any kind of
761 int dev_valid_name(const char *name)
765 if (strlen(name) >= IFNAMSIZ)
767 if (!strcmp(name, ".") || !strcmp(name, ".."))
771 if (*name == '/' || isspace(*name))
779 * __dev_alloc_name - allocate a name for a device
780 * @net: network namespace to allocate the device name in
781 * @name: name format string
782 * @buf: scratch buffer and result name string
784 * Passed a format string - eg "lt%d" it will try and find a suitable
785 * id. It scans list of devices to build up a free map, then chooses
786 * the first empty slot. The caller must hold the dev_base or rtnl lock
787 * while allocating the name and adding the device in order to avoid
789 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
790 * Returns the number of the unit assigned or a negative errno code.
793 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
797 const int max_netdevices = 8*PAGE_SIZE;
798 unsigned long *inuse;
799 struct net_device *d;
801 p = strnchr(name, IFNAMSIZ-1, '%');
804 * Verify the string as this thing may have come from
805 * the user. There must be either one "%d" and no other "%"
808 if (p[1] != 'd' || strchr(p + 2, '%'))
811 /* Use one page as a bit array of possible slots */
812 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
816 for_each_netdev(net, d) {
817 if (!sscanf(d->name, name, &i))
819 if (i < 0 || i >= max_netdevices)
822 /* avoid cases where sscanf is not exact inverse of printf */
823 snprintf(buf, IFNAMSIZ, name, i);
824 if (!strncmp(buf, d->name, IFNAMSIZ))
828 i = find_first_zero_bit(inuse, max_netdevices);
829 free_page((unsigned long) inuse);
832 snprintf(buf, IFNAMSIZ, name, i);
833 if (!__dev_get_by_name(net, buf))
836 /* It is possible to run out of possible slots
837 * when the name is long and there isn't enough space left
838 * for the digits, or if all bits are used.
844 * dev_alloc_name - allocate a name for a device
846 * @name: name format string
848 * Passed a format string - eg "lt%d" it will try and find a suitable
849 * id. It scans list of devices to build up a free map, then chooses
850 * the first empty slot. The caller must hold the dev_base or rtnl lock
851 * while allocating the name and adding the device in order to avoid
853 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
854 * Returns the number of the unit assigned or a negative errno code.
857 int dev_alloc_name(struct net_device *dev, const char *name)
863 BUG_ON(!dev_net(dev));
865 ret = __dev_alloc_name(net, name, buf);
867 strlcpy(dev->name, buf, IFNAMSIZ);
873 * dev_change_name - change name of a device
875 * @newname: name (or format string) must be at least IFNAMSIZ
877 * Change name of a device, can pass format strings "eth%d".
880 int dev_change_name(struct net_device *dev, const char *newname)
882 char oldname[IFNAMSIZ];
888 BUG_ON(!dev_net(dev));
891 if (dev->flags & IFF_UP)
894 if (!dev_valid_name(newname))
897 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
900 memcpy(oldname, dev->name, IFNAMSIZ);
902 if (strchr(newname, '%')) {
903 err = dev_alloc_name(dev, newname);
907 else if (__dev_get_by_name(net, newname))
910 strlcpy(dev->name, newname, IFNAMSIZ);
913 /* For now only devices in the initial network namespace
916 if (net == &init_net) {
917 ret = device_rename(&dev->dev, dev->name);
919 memcpy(dev->name, oldname, IFNAMSIZ);
924 write_lock_bh(&dev_base_lock);
925 hlist_del(&dev->name_hlist);
926 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
927 write_unlock_bh(&dev_base_lock);
929 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
930 ret = notifier_to_errno(ret);
935 "%s: name change rollback failed: %d.\n",
939 memcpy(dev->name, oldname, IFNAMSIZ);
948 * dev_set_alias - change ifalias of a device
950 * @alias: name up to IFALIASZ
951 * @len: limit of bytes to copy from info
953 * Set ifalias for a device,
955 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)
1049 const struct net_device_ops *ops = dev->netdev_ops;
1058 if (dev->flags & IFF_UP)
1062 * Is it even present?
1064 if (!netif_device_present(dev))
1068 * Call device private open method
1070 set_bit(__LINK_STATE_START, &dev->state);
1072 if (ops->ndo_validate_addr)
1073 ret = ops->ndo_validate_addr(dev);
1075 if (!ret && ops->ndo_open)
1076 ret = ops->ndo_open(dev);
1079 * If it went open OK then:
1083 clear_bit(__LINK_STATE_START, &dev->state);
1088 dev->flags |= IFF_UP;
1091 * Initialize multicasting status
1093 dev_set_rx_mode(dev);
1096 * Wakeup transmit queue engine
1101 * ... and announce new interface.
1103 call_netdevice_notifiers(NETDEV_UP, dev);
1110 * dev_close - shutdown an interface.
1111 * @dev: device to shutdown
1113 * This function moves an active device into down state. A
1114 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1115 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1118 int dev_close(struct net_device *dev)
1120 const struct net_device_ops *ops = dev->netdev_ops;
1125 if (!(dev->flags & IFF_UP))
1129 * Tell people we are going down, so that they can
1130 * prepare to death, when device is still operating.
1132 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1134 clear_bit(__LINK_STATE_START, &dev->state);
1136 /* Synchronize to scheduled poll. We cannot touch poll list,
1137 * it can be even on different cpu. So just clear netif_running().
1139 * dev->stop() will invoke napi_disable() on all of it's
1140 * napi_struct instances on this device.
1142 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1144 dev_deactivate(dev);
1147 * Call the device specific close. This cannot fail.
1148 * Only if device is UP
1150 * We allow it to be called even after a DETACH hot-plug
1157 * Device is now down.
1160 dev->flags &= ~IFF_UP;
1163 * Tell people we are down
1165 call_netdevice_notifiers(NETDEV_DOWN, dev);
1172 * dev_disable_lro - disable Large Receive Offload on a device
1175 * Disable Large Receive Offload (LRO) on a net device. Must be
1176 * called under RTNL. This is needed if received packets may be
1177 * forwarded to another interface.
1179 void dev_disable_lro(struct net_device *dev)
1181 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1182 dev->ethtool_ops->set_flags) {
1183 u32 flags = dev->ethtool_ops->get_flags(dev);
1184 if (flags & ETH_FLAG_LRO) {
1185 flags &= ~ETH_FLAG_LRO;
1186 dev->ethtool_ops->set_flags(dev, flags);
1189 WARN_ON(dev->features & NETIF_F_LRO);
1191 EXPORT_SYMBOL(dev_disable_lro);
1194 static int dev_boot_phase = 1;
1197 * Device change register/unregister. These are not inline or static
1198 * as we export them to the world.
1202 * register_netdevice_notifier - register a network notifier block
1205 * Register a notifier to be called when network device events occur.
1206 * The notifier passed is linked into the kernel structures and must
1207 * not be reused until it has been unregistered. A negative errno code
1208 * is returned on a failure.
1210 * When registered all registration and up events are replayed
1211 * to the new notifier to allow device to have a race free
1212 * view of the network device list.
1215 int register_netdevice_notifier(struct notifier_block *nb)
1217 struct net_device *dev;
1218 struct net_device *last;
1223 err = raw_notifier_chain_register(&netdev_chain, nb);
1229 for_each_netdev(net, dev) {
1230 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1231 err = notifier_to_errno(err);
1235 if (!(dev->flags & IFF_UP))
1238 nb->notifier_call(nb, NETDEV_UP, dev);
1249 for_each_netdev(net, dev) {
1253 if (dev->flags & IFF_UP) {
1254 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1255 nb->notifier_call(nb, NETDEV_DOWN, dev);
1257 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1261 raw_notifier_chain_unregister(&netdev_chain, nb);
1266 * unregister_netdevice_notifier - unregister a network notifier block
1269 * Unregister a notifier previously registered by
1270 * register_netdevice_notifier(). The notifier is unlinked into the
1271 * kernel structures and may then be reused. A negative errno code
1272 * is returned on a failure.
1275 int unregister_netdevice_notifier(struct notifier_block *nb)
1280 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1286 * call_netdevice_notifiers - call all network notifier blocks
1287 * @val: value passed unmodified to notifier function
1288 * @dev: net_device pointer passed unmodified to notifier function
1290 * Call all network notifier blocks. Parameters and return value
1291 * are as for raw_notifier_call_chain().
1294 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1296 return raw_notifier_call_chain(&netdev_chain, val, dev);
1299 /* When > 0 there are consumers of rx skb time stamps */
1300 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1302 void net_enable_timestamp(void)
1304 atomic_inc(&netstamp_needed);
1307 void net_disable_timestamp(void)
1309 atomic_dec(&netstamp_needed);
1312 static inline void net_timestamp(struct sk_buff *skb)
1314 if (atomic_read(&netstamp_needed))
1315 __net_timestamp(skb);
1317 skb->tstamp.tv64 = 0;
1321 * Support routine. Sends outgoing frames to any network
1322 * taps currently in use.
1325 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1327 struct packet_type *ptype;
1332 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1333 /* Never send packets back to the socket
1334 * they originated from - MvS (miquels@drinkel.ow.org)
1336 if ((ptype->dev == dev || !ptype->dev) &&
1337 (ptype->af_packet_priv == NULL ||
1338 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1339 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1343 /* skb->nh should be correctly
1344 set by sender, so that the second statement is
1345 just protection against buggy protocols.
1347 skb_reset_mac_header(skb2);
1349 if (skb_network_header(skb2) < skb2->data ||
1350 skb2->network_header > skb2->tail) {
1351 if (net_ratelimit())
1352 printk(KERN_CRIT "protocol %04x is "
1354 skb2->protocol, dev->name);
1355 skb_reset_network_header(skb2);
1358 skb2->transport_header = skb2->network_header;
1359 skb2->pkt_type = PACKET_OUTGOING;
1360 ptype->func(skb2, skb->dev, ptype, skb->dev);
1367 static inline void __netif_reschedule(struct Qdisc *q)
1369 struct softnet_data *sd;
1370 unsigned long flags;
1372 local_irq_save(flags);
1373 sd = &__get_cpu_var(softnet_data);
1374 q->next_sched = sd->output_queue;
1375 sd->output_queue = q;
1376 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1377 local_irq_restore(flags);
1380 void __netif_schedule(struct Qdisc *q)
1382 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1383 __netif_reschedule(q);
1385 EXPORT_SYMBOL(__netif_schedule);
1387 void dev_kfree_skb_irq(struct sk_buff *skb)
1389 if (atomic_dec_and_test(&skb->users)) {
1390 struct softnet_data *sd;
1391 unsigned long flags;
1393 local_irq_save(flags);
1394 sd = &__get_cpu_var(softnet_data);
1395 skb->next = sd->completion_queue;
1396 sd->completion_queue = skb;
1397 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1398 local_irq_restore(flags);
1401 EXPORT_SYMBOL(dev_kfree_skb_irq);
1403 void dev_kfree_skb_any(struct sk_buff *skb)
1405 if (in_irq() || irqs_disabled())
1406 dev_kfree_skb_irq(skb);
1410 EXPORT_SYMBOL(dev_kfree_skb_any);
1414 * netif_device_detach - mark device as removed
1415 * @dev: network device
1417 * Mark device as removed from system and therefore no longer available.
1419 void netif_device_detach(struct net_device *dev)
1421 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1422 netif_running(dev)) {
1423 netif_stop_queue(dev);
1426 EXPORT_SYMBOL(netif_device_detach);
1429 * netif_device_attach - mark device as attached
1430 * @dev: network device
1432 * Mark device as attached from system and restart if needed.
1434 void netif_device_attach(struct net_device *dev)
1436 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1437 netif_running(dev)) {
1438 netif_wake_queue(dev);
1439 __netdev_watchdog_up(dev);
1442 EXPORT_SYMBOL(netif_device_attach);
1444 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1446 return ((features & NETIF_F_GEN_CSUM) ||
1447 ((features & NETIF_F_IP_CSUM) &&
1448 protocol == htons(ETH_P_IP)) ||
1449 ((features & NETIF_F_IPV6_CSUM) &&
1450 protocol == htons(ETH_P_IPV6)));
1453 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1455 if (can_checksum_protocol(dev->features, skb->protocol))
1458 if (skb->protocol == htons(ETH_P_8021Q)) {
1459 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1460 if (can_checksum_protocol(dev->features & dev->vlan_features,
1461 veh->h_vlan_encapsulated_proto))
1469 * Invalidate hardware checksum when packet is to be mangled, and
1470 * complete checksum manually on outgoing path.
1472 int skb_checksum_help(struct sk_buff *skb)
1475 int ret = 0, offset;
1477 if (skb->ip_summed == CHECKSUM_COMPLETE)
1478 goto out_set_summed;
1480 if (unlikely(skb_shinfo(skb)->gso_size)) {
1481 /* Let GSO fix up the checksum. */
1482 goto out_set_summed;
1485 offset = skb->csum_start - skb_headroom(skb);
1486 BUG_ON(offset >= skb_headlen(skb));
1487 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1489 offset += skb->csum_offset;
1490 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1492 if (skb_cloned(skb) &&
1493 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1494 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1499 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1501 skb->ip_summed = CHECKSUM_NONE;
1507 * skb_gso_segment - Perform segmentation on skb.
1508 * @skb: buffer to segment
1509 * @features: features for the output path (see dev->features)
1511 * This function segments the given skb and returns a list of segments.
1513 * It may return NULL if the skb requires no segmentation. This is
1514 * only possible when GSO is used for verifying header integrity.
1516 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1518 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1519 struct packet_type *ptype;
1520 __be16 type = skb->protocol;
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 const struct net_device_ops *ops = dev->netdev_ops;
1650 prefetch(&dev->netdev_ops->ndo_start_xmit);
1651 if (likely(!skb->next)) {
1652 if (!list_empty(&ptype_all))
1653 dev_queue_xmit_nit(skb, dev);
1655 if (netif_needs_gso(dev, skb)) {
1656 if (unlikely(dev_gso_segment(skb)))
1662 return ops->ndo_start_xmit(skb, dev);
1667 struct sk_buff *nskb = skb->next;
1670 skb->next = nskb->next;
1672 rc = ops->ndo_start_xmit(nskb, dev);
1674 nskb->next = skb->next;
1678 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1679 return NETDEV_TX_BUSY;
1680 } while (skb->next);
1682 skb->destructor = DEV_GSO_CB(skb)->destructor;
1689 static u32 simple_tx_hashrnd;
1690 static int simple_tx_hashrnd_initialized = 0;
1692 static u16 simple_tx_hash(struct net_device *dev, struct sk_buff *skb)
1694 u32 addr1, addr2, ports;
1698 if (unlikely(!simple_tx_hashrnd_initialized)) {
1699 get_random_bytes(&simple_tx_hashrnd, 4);
1700 simple_tx_hashrnd_initialized = 1;
1703 switch (skb->protocol) {
1704 case htons(ETH_P_IP):
1705 if (!(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)))
1706 ip_proto = ip_hdr(skb)->protocol;
1707 addr1 = ip_hdr(skb)->saddr;
1708 addr2 = ip_hdr(skb)->daddr;
1709 ihl = ip_hdr(skb)->ihl;
1711 case htons(ETH_P_IPV6):
1712 ip_proto = ipv6_hdr(skb)->nexthdr;
1713 addr1 = ipv6_hdr(skb)->saddr.s6_addr32[3];
1714 addr2 = ipv6_hdr(skb)->daddr.s6_addr32[3];
1729 case IPPROTO_UDPLITE:
1730 ports = *((u32 *) (skb_network_header(skb) + (ihl * 4)));
1738 hash = jhash_3words(addr1, addr2, ports, simple_tx_hashrnd);
1740 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1743 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1744 struct sk_buff *skb)
1746 const struct net_device_ops *ops = dev->netdev_ops;
1747 u16 queue_index = 0;
1749 if (ops->ndo_select_queue)
1750 queue_index = ops->ndo_select_queue(dev, skb);
1751 else if (dev->real_num_tx_queues > 1)
1752 queue_index = simple_tx_hash(dev, skb);
1754 skb_set_queue_mapping(skb, queue_index);
1755 return netdev_get_tx_queue(dev, queue_index);
1759 * dev_queue_xmit - transmit a buffer
1760 * @skb: buffer to transmit
1762 * Queue a buffer for transmission to a network device. The caller must
1763 * have set the device and priority and built the buffer before calling
1764 * this function. The function can be called from an interrupt.
1766 * A negative errno code is returned on a failure. A success does not
1767 * guarantee the frame will be transmitted as it may be dropped due
1768 * to congestion or traffic shaping.
1770 * -----------------------------------------------------------------------------------
1771 * I notice this method can also return errors from the queue disciplines,
1772 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1775 * Regardless of the return value, the skb is consumed, so it is currently
1776 * difficult to retry a send to this method. (You can bump the ref count
1777 * before sending to hold a reference for retry if you are careful.)
1779 * When calling this method, interrupts MUST be enabled. This is because
1780 * the BH enable code must have IRQs enabled so that it will not deadlock.
1783 int dev_queue_xmit(struct sk_buff *skb)
1785 struct net_device *dev = skb->dev;
1786 struct netdev_queue *txq;
1790 /* GSO will handle the following emulations directly. */
1791 if (netif_needs_gso(dev, skb))
1794 if (skb_shinfo(skb)->frag_list &&
1795 !(dev->features & NETIF_F_FRAGLIST) &&
1796 __skb_linearize(skb))
1799 /* Fragmented skb is linearized if device does not support SG,
1800 * or if at least one of fragments is in highmem and device
1801 * does not support DMA from it.
1803 if (skb_shinfo(skb)->nr_frags &&
1804 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1805 __skb_linearize(skb))
1808 /* If packet is not checksummed and device does not support
1809 * checksumming for this protocol, complete checksumming here.
1811 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1812 skb_set_transport_header(skb, skb->csum_start -
1814 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1819 /* Disable soft irqs for various locks below. Also
1820 * stops preemption for RCU.
1824 txq = dev_pick_tx(dev, skb);
1825 q = rcu_dereference(txq->qdisc);
1827 #ifdef CONFIG_NET_CLS_ACT
1828 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1831 spinlock_t *root_lock = qdisc_lock(q);
1833 spin_lock(root_lock);
1835 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1839 rc = qdisc_enqueue_root(skb, q);
1842 spin_unlock(root_lock);
1847 /* The device has no queue. Common case for software devices:
1848 loopback, all the sorts of tunnels...
1850 Really, it is unlikely that netif_tx_lock protection is necessary
1851 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1853 However, it is possible, that they rely on protection
1856 Check this and shot the lock. It is not prone from deadlocks.
1857 Either shot noqueue qdisc, it is even simpler 8)
1859 if (dev->flags & IFF_UP) {
1860 int cpu = smp_processor_id(); /* ok because BHs are off */
1862 if (txq->xmit_lock_owner != cpu) {
1864 HARD_TX_LOCK(dev, txq, cpu);
1866 if (!netif_tx_queue_stopped(txq)) {
1868 if (!dev_hard_start_xmit(skb, dev, txq)) {
1869 HARD_TX_UNLOCK(dev, txq);
1873 HARD_TX_UNLOCK(dev, txq);
1874 if (net_ratelimit())
1875 printk(KERN_CRIT "Virtual device %s asks to "
1876 "queue packet!\n", dev->name);
1878 /* Recursion is detected! It is possible,
1880 if (net_ratelimit())
1881 printk(KERN_CRIT "Dead loop on virtual device "
1882 "%s, fix it urgently!\n", dev->name);
1887 rcu_read_unlock_bh();
1893 rcu_read_unlock_bh();
1898 /*=======================================================================
1900 =======================================================================*/
1902 int netdev_max_backlog __read_mostly = 1000;
1903 int netdev_budget __read_mostly = 300;
1904 int weight_p __read_mostly = 64; /* old backlog weight */
1906 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1910 * netif_rx - post buffer to the network code
1911 * @skb: buffer to post
1913 * This function receives a packet from a device driver and queues it for
1914 * the upper (protocol) levels to process. It always succeeds. The buffer
1915 * may be dropped during processing for congestion control or by the
1919 * NET_RX_SUCCESS (no congestion)
1920 * NET_RX_DROP (packet was dropped)
1924 int netif_rx(struct sk_buff *skb)
1926 struct softnet_data *queue;
1927 unsigned long flags;
1929 /* if netpoll wants it, pretend we never saw it */
1930 if (netpoll_rx(skb))
1933 if (!skb->tstamp.tv64)
1937 * The code is rearranged so that the path is the most
1938 * short when CPU is congested, but is still operating.
1940 local_irq_save(flags);
1941 queue = &__get_cpu_var(softnet_data);
1943 __get_cpu_var(netdev_rx_stat).total++;
1944 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1945 if (queue->input_pkt_queue.qlen) {
1947 __skb_queue_tail(&queue->input_pkt_queue, skb);
1948 local_irq_restore(flags);
1949 return NET_RX_SUCCESS;
1952 napi_schedule(&queue->backlog);
1956 __get_cpu_var(netdev_rx_stat).dropped++;
1957 local_irq_restore(flags);
1963 int netif_rx_ni(struct sk_buff *skb)
1968 err = netif_rx(skb);
1969 if (local_softirq_pending())
1976 EXPORT_SYMBOL(netif_rx_ni);
1978 static void net_tx_action(struct softirq_action *h)
1980 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1982 if (sd->completion_queue) {
1983 struct sk_buff *clist;
1985 local_irq_disable();
1986 clist = sd->completion_queue;
1987 sd->completion_queue = NULL;
1991 struct sk_buff *skb = clist;
1992 clist = clist->next;
1994 WARN_ON(atomic_read(&skb->users));
1999 if (sd->output_queue) {
2002 local_irq_disable();
2003 head = sd->output_queue;
2004 sd->output_queue = NULL;
2008 struct Qdisc *q = head;
2009 spinlock_t *root_lock;
2011 head = head->next_sched;
2013 root_lock = qdisc_lock(q);
2014 if (spin_trylock(root_lock)) {
2015 smp_mb__before_clear_bit();
2016 clear_bit(__QDISC_STATE_SCHED,
2019 spin_unlock(root_lock);
2021 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2023 __netif_reschedule(q);
2025 smp_mb__before_clear_bit();
2026 clear_bit(__QDISC_STATE_SCHED,
2034 static inline int deliver_skb(struct sk_buff *skb,
2035 struct packet_type *pt_prev,
2036 struct net_device *orig_dev)
2038 atomic_inc(&skb->users);
2039 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2042 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2043 /* These hooks defined here for ATM */
2045 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2046 unsigned char *addr);
2047 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
2050 * If bridge module is loaded call bridging hook.
2051 * returns NULL if packet was consumed.
2053 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2054 struct sk_buff *skb) __read_mostly;
2055 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2056 struct packet_type **pt_prev, int *ret,
2057 struct net_device *orig_dev)
2059 struct net_bridge_port *port;
2061 if (skb->pkt_type == PACKET_LOOPBACK ||
2062 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2066 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2070 return br_handle_frame_hook(port, skb);
2073 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2076 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2077 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2078 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2080 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2081 struct packet_type **pt_prev,
2083 struct net_device *orig_dev)
2085 if (skb->dev->macvlan_port == NULL)
2089 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2092 return macvlan_handle_frame_hook(skb);
2095 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2098 #ifdef CONFIG_NET_CLS_ACT
2099 /* TODO: Maybe we should just force sch_ingress to be compiled in
2100 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2101 * a compare and 2 stores extra right now if we dont have it on
2102 * but have CONFIG_NET_CLS_ACT
2103 * NOTE: This doesnt stop any functionality; if you dont have
2104 * the ingress scheduler, you just cant add policies on ingress.
2107 static int ing_filter(struct sk_buff *skb)
2109 struct net_device *dev = skb->dev;
2110 u32 ttl = G_TC_RTTL(skb->tc_verd);
2111 struct netdev_queue *rxq;
2112 int result = TC_ACT_OK;
2115 if (MAX_RED_LOOP < ttl++) {
2117 "Redir loop detected Dropping packet (%d->%d)\n",
2118 skb->iif, dev->ifindex);
2122 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2123 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2125 rxq = &dev->rx_queue;
2128 if (q != &noop_qdisc) {
2129 spin_lock(qdisc_lock(q));
2130 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2131 result = qdisc_enqueue_root(skb, q);
2132 spin_unlock(qdisc_lock(q));
2138 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2139 struct packet_type **pt_prev,
2140 int *ret, struct net_device *orig_dev)
2142 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2146 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2149 /* Huh? Why does turning on AF_PACKET affect this? */
2150 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2153 switch (ing_filter(skb)) {
2167 * netif_nit_deliver - deliver received packets to network taps
2170 * This function is used to deliver incoming packets to network
2171 * taps. It should be used when the normal netif_receive_skb path
2172 * is bypassed, for example because of VLAN acceleration.
2174 void netif_nit_deliver(struct sk_buff *skb)
2176 struct packet_type *ptype;
2178 if (list_empty(&ptype_all))
2181 skb_reset_network_header(skb);
2182 skb_reset_transport_header(skb);
2183 skb->mac_len = skb->network_header - skb->mac_header;
2186 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2187 if (!ptype->dev || ptype->dev == skb->dev)
2188 deliver_skb(skb, ptype, skb->dev);
2194 * netif_receive_skb - process receive buffer from network
2195 * @skb: buffer to process
2197 * netif_receive_skb() is the main receive data processing function.
2198 * It always succeeds. The buffer may be dropped during processing
2199 * for congestion control or by the protocol layers.
2201 * This function may only be called from softirq context and interrupts
2202 * should be enabled.
2204 * Return values (usually ignored):
2205 * NET_RX_SUCCESS: no congestion
2206 * NET_RX_DROP: packet was dropped
2208 int netif_receive_skb(struct sk_buff *skb)
2210 struct packet_type *ptype, *pt_prev;
2211 struct net_device *orig_dev;
2212 struct net_device *null_or_orig;
2213 int ret = NET_RX_DROP;
2216 if (skb->vlan_tci && vlan_hwaccel_do_receive(skb))
2217 return NET_RX_SUCCESS;
2219 /* if we've gotten here through NAPI, check netpoll */
2220 if (netpoll_receive_skb(skb))
2223 if (!skb->tstamp.tv64)
2227 skb->iif = skb->dev->ifindex;
2229 null_or_orig = NULL;
2230 orig_dev = skb->dev;
2231 if (orig_dev->master) {
2232 if (skb_bond_should_drop(skb))
2233 null_or_orig = orig_dev; /* deliver only exact match */
2235 skb->dev = orig_dev->master;
2238 __get_cpu_var(netdev_rx_stat).total++;
2240 skb_reset_network_header(skb);
2241 skb_reset_transport_header(skb);
2242 skb->mac_len = skb->network_header - skb->mac_header;
2248 /* Don't receive packets in an exiting network namespace */
2249 if (!net_alive(dev_net(skb->dev))) {
2254 #ifdef CONFIG_NET_CLS_ACT
2255 if (skb->tc_verd & TC_NCLS) {
2256 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2261 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2262 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2263 ptype->dev == orig_dev) {
2265 ret = deliver_skb(skb, pt_prev, orig_dev);
2270 #ifdef CONFIG_NET_CLS_ACT
2271 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2277 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2280 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2284 type = skb->protocol;
2285 list_for_each_entry_rcu(ptype,
2286 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2287 if (ptype->type == type &&
2288 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2289 ptype->dev == orig_dev)) {
2291 ret = deliver_skb(skb, pt_prev, orig_dev);
2297 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2300 /* Jamal, now you will not able to escape explaining
2301 * me how you were going to use this. :-)
2311 /* Network device is going away, flush any packets still pending */
2312 static void flush_backlog(void *arg)
2314 struct net_device *dev = arg;
2315 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2316 struct sk_buff *skb, *tmp;
2318 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2319 if (skb->dev == dev) {
2320 __skb_unlink(skb, &queue->input_pkt_queue);
2325 static int napi_gro_complete(struct sk_buff *skb)
2327 struct packet_type *ptype;
2328 __be16 type = skb->protocol;
2329 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2332 if (NAPI_GRO_CB(skb)->count == 1)
2336 list_for_each_entry_rcu(ptype, head, list) {
2337 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2340 err = ptype->gro_complete(skb);
2346 WARN_ON(&ptype->list == head);
2348 return NET_RX_SUCCESS;
2352 skb_shinfo(skb)->gso_size = 0;
2353 __skb_push(skb, -skb_network_offset(skb));
2354 return netif_receive_skb(skb);
2357 void napi_gro_flush(struct napi_struct *napi)
2359 struct sk_buff *skb, *next;
2361 for (skb = napi->gro_list; skb; skb = next) {
2364 napi_gro_complete(skb);
2367 napi->gro_list = NULL;
2369 EXPORT_SYMBOL(napi_gro_flush);
2371 int dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2373 struct sk_buff **pp = NULL;
2374 struct packet_type *ptype;
2375 __be16 type = skb->protocol;
2376 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2382 if (!(skb->dev->features & NETIF_F_GRO))
2386 list_for_each_entry_rcu(ptype, head, list) {
2389 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2392 skb_reset_network_header(skb);
2393 mac_len = skb->network_header - skb->mac_header;
2394 skb->mac_len = mac_len;
2395 NAPI_GRO_CB(skb)->same_flow = 0;
2396 NAPI_GRO_CB(skb)->flush = 0;
2397 NAPI_GRO_CB(skb)->free = 0;
2399 for (p = napi->gro_list; p; p = p->next) {
2402 if (!NAPI_GRO_CB(p)->same_flow)
2405 if (p->mac_len != mac_len ||
2406 memcmp(skb_mac_header(p), skb_mac_header(skb),
2408 NAPI_GRO_CB(p)->same_flow = 0;
2411 pp = ptype->gro_receive(&napi->gro_list, skb);
2416 if (&ptype->list == head)
2419 same_flow = NAPI_GRO_CB(skb)->same_flow;
2420 free = NAPI_GRO_CB(skb)->free;
2423 struct sk_buff *nskb = *pp;
2427 napi_gro_complete(nskb);
2434 if (NAPI_GRO_CB(skb)->flush || count >= MAX_GRO_SKBS) {
2435 __skb_push(skb, -skb_network_offset(skb));
2439 NAPI_GRO_CB(skb)->count = 1;
2440 skb_shinfo(skb)->gso_size = skb->len;
2441 skb->next = napi->gro_list;
2442 napi->gro_list = skb;
2450 EXPORT_SYMBOL(dev_gro_receive);
2452 static int __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2456 for (p = napi->gro_list; p; p = p->next) {
2457 NAPI_GRO_CB(p)->same_flow = 1;
2458 NAPI_GRO_CB(p)->flush = 0;
2461 return dev_gro_receive(napi, skb);
2464 int napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2466 switch (__napi_gro_receive(napi, skb)) {
2468 return netif_receive_skb(skb);
2475 return NET_RX_SUCCESS;
2477 EXPORT_SYMBOL(napi_gro_receive);
2479 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2481 skb_shinfo(skb)->nr_frags = 0;
2483 skb->len -= skb->data_len;
2484 skb->truesize -= skb->data_len;
2487 __skb_pull(skb, skb_headlen(skb));
2488 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2492 EXPORT_SYMBOL(napi_reuse_skb);
2494 struct sk_buff *napi_fraginfo_skb(struct napi_struct *napi,
2495 struct napi_gro_fraginfo *info)
2497 struct net_device *dev = napi->dev;
2498 struct sk_buff *skb = napi->skb;
2503 skb = netdev_alloc_skb(dev, GRO_MAX_HEAD + NET_IP_ALIGN);
2507 skb_reserve(skb, NET_IP_ALIGN);
2510 BUG_ON(info->nr_frags > MAX_SKB_FRAGS);
2511 skb_shinfo(skb)->nr_frags = info->nr_frags;
2512 memcpy(skb_shinfo(skb)->frags, info->frags, sizeof(info->frags));
2514 skb->data_len = info->len;
2515 skb->len += info->len;
2516 skb->truesize += info->len;
2518 if (!pskb_may_pull(skb, ETH_HLEN)) {
2519 napi_reuse_skb(napi, skb);
2523 skb->protocol = eth_type_trans(skb, dev);
2525 skb->ip_summed = info->ip_summed;
2526 skb->csum = info->csum;
2531 EXPORT_SYMBOL(napi_fraginfo_skb);
2533 int napi_gro_frags(struct napi_struct *napi, struct napi_gro_fraginfo *info)
2535 struct sk_buff *skb = napi_fraginfo_skb(napi, info);
2536 int err = NET_RX_DROP;
2541 err = NET_RX_SUCCESS;
2543 switch (__napi_gro_receive(napi, skb)) {
2545 return netif_receive_skb(skb);
2551 napi_reuse_skb(napi, skb);
2556 EXPORT_SYMBOL(napi_gro_frags);
2558 static int process_backlog(struct napi_struct *napi, int quota)
2561 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2562 unsigned long start_time = jiffies;
2564 napi->weight = weight_p;
2566 struct sk_buff *skb;
2568 local_irq_disable();
2569 skb = __skb_dequeue(&queue->input_pkt_queue);
2571 __napi_complete(napi);
2577 napi_gro_receive(napi, skb);
2578 } while (++work < quota && jiffies == start_time);
2580 napi_gro_flush(napi);
2586 * __napi_schedule - schedule for receive
2587 * @n: entry to schedule
2589 * The entry's receive function will be scheduled to run
2591 void __napi_schedule(struct napi_struct *n)
2593 unsigned long flags;
2595 local_irq_save(flags);
2596 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2597 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2598 local_irq_restore(flags);
2600 EXPORT_SYMBOL(__napi_schedule);
2602 void __napi_complete(struct napi_struct *n)
2604 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2605 BUG_ON(n->gro_list);
2607 list_del(&n->poll_list);
2608 smp_mb__before_clear_bit();
2609 clear_bit(NAPI_STATE_SCHED, &n->state);
2611 EXPORT_SYMBOL(__napi_complete);
2613 void napi_complete(struct napi_struct *n)
2615 unsigned long flags;
2618 * don't let napi dequeue from the cpu poll list
2619 * just in case its running on a different cpu
2621 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2625 local_irq_save(flags);
2627 local_irq_restore(flags);
2629 EXPORT_SYMBOL(napi_complete);
2631 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2632 int (*poll)(struct napi_struct *, int), int weight)
2634 INIT_LIST_HEAD(&napi->poll_list);
2635 napi->gro_list = NULL;
2638 napi->weight = weight;
2639 list_add(&napi->dev_list, &dev->napi_list);
2641 #ifdef CONFIG_NETPOLL
2642 spin_lock_init(&napi->poll_lock);
2643 napi->poll_owner = -1;
2645 set_bit(NAPI_STATE_SCHED, &napi->state);
2647 EXPORT_SYMBOL(netif_napi_add);
2649 void netif_napi_del(struct napi_struct *napi)
2651 struct sk_buff *skb, *next;
2653 list_del_init(&napi->dev_list);
2656 for (skb = napi->gro_list; skb; skb = next) {
2662 napi->gro_list = NULL;
2664 EXPORT_SYMBOL(netif_napi_del);
2667 static void net_rx_action(struct softirq_action *h)
2669 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2670 unsigned long time_limit = jiffies + 2;
2671 int budget = netdev_budget;
2674 local_irq_disable();
2676 while (!list_empty(list)) {
2677 struct napi_struct *n;
2680 /* If softirq window is exhuasted then punt.
2681 * Allow this to run for 2 jiffies since which will allow
2682 * an average latency of 1.5/HZ.
2684 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2689 /* Even though interrupts have been re-enabled, this
2690 * access is safe because interrupts can only add new
2691 * entries to the tail of this list, and only ->poll()
2692 * calls can remove this head entry from the list.
2694 n = list_entry(list->next, struct napi_struct, poll_list);
2696 have = netpoll_poll_lock(n);
2700 /* This NAPI_STATE_SCHED test is for avoiding a race
2701 * with netpoll's poll_napi(). Only the entity which
2702 * obtains the lock and sees NAPI_STATE_SCHED set will
2703 * actually make the ->poll() call. Therefore we avoid
2704 * accidently calling ->poll() when NAPI is not scheduled.
2707 if (test_bit(NAPI_STATE_SCHED, &n->state))
2708 work = n->poll(n, weight);
2710 WARN_ON_ONCE(work > weight);
2714 local_irq_disable();
2716 /* Drivers must not modify the NAPI state if they
2717 * consume the entire weight. In such cases this code
2718 * still "owns" the NAPI instance and therefore can
2719 * move the instance around on the list at-will.
2721 if (unlikely(work == weight)) {
2722 if (unlikely(napi_disable_pending(n)))
2725 list_move_tail(&n->poll_list, list);
2728 netpoll_poll_unlock(have);
2733 #ifdef CONFIG_NET_DMA
2735 * There may not be any more sk_buffs coming right now, so push
2736 * any pending DMA copies to hardware
2738 dma_issue_pending_all();
2744 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2745 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2749 static gifconf_func_t * gifconf_list [NPROTO];
2752 * register_gifconf - register a SIOCGIF handler
2753 * @family: Address family
2754 * @gifconf: Function handler
2756 * Register protocol dependent address dumping routines. The handler
2757 * that is passed must not be freed or reused until it has been replaced
2758 * by another handler.
2760 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2762 if (family >= NPROTO)
2764 gifconf_list[family] = gifconf;
2770 * Map an interface index to its name (SIOCGIFNAME)
2774 * We need this ioctl for efficient implementation of the
2775 * if_indextoname() function required by the IPv6 API. Without
2776 * it, we would have to search all the interfaces to find a
2780 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2782 struct net_device *dev;
2786 * Fetch the caller's info block.
2789 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2792 read_lock(&dev_base_lock);
2793 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2795 read_unlock(&dev_base_lock);
2799 strcpy(ifr.ifr_name, dev->name);
2800 read_unlock(&dev_base_lock);
2802 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2808 * Perform a SIOCGIFCONF call. This structure will change
2809 * size eventually, and there is nothing I can do about it.
2810 * Thus we will need a 'compatibility mode'.
2813 static int dev_ifconf(struct net *net, char __user *arg)
2816 struct net_device *dev;
2823 * Fetch the caller's info block.
2826 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2833 * Loop over the interfaces, and write an info block for each.
2837 for_each_netdev(net, dev) {
2838 for (i = 0; i < NPROTO; i++) {
2839 if (gifconf_list[i]) {
2842 done = gifconf_list[i](dev, NULL, 0);
2844 done = gifconf_list[i](dev, pos + total,
2854 * All done. Write the updated control block back to the caller.
2856 ifc.ifc_len = total;
2859 * Both BSD and Solaris return 0 here, so we do too.
2861 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2864 #ifdef CONFIG_PROC_FS
2866 * This is invoked by the /proc filesystem handler to display a device
2869 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2870 __acquires(dev_base_lock)
2872 struct net *net = seq_file_net(seq);
2874 struct net_device *dev;
2876 read_lock(&dev_base_lock);
2878 return SEQ_START_TOKEN;
2881 for_each_netdev(net, dev)
2888 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2890 struct net *net = seq_file_net(seq);
2892 return v == SEQ_START_TOKEN ?
2893 first_net_device(net) : next_net_device((struct net_device *)v);
2896 void dev_seq_stop(struct seq_file *seq, void *v)
2897 __releases(dev_base_lock)
2899 read_unlock(&dev_base_lock);
2902 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2904 const struct net_device_stats *stats = dev_get_stats(dev);
2906 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2907 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2908 dev->name, stats->rx_bytes, stats->rx_packets,
2910 stats->rx_dropped + stats->rx_missed_errors,
2911 stats->rx_fifo_errors,
2912 stats->rx_length_errors + stats->rx_over_errors +
2913 stats->rx_crc_errors + stats->rx_frame_errors,
2914 stats->rx_compressed, stats->multicast,
2915 stats->tx_bytes, stats->tx_packets,
2916 stats->tx_errors, stats->tx_dropped,
2917 stats->tx_fifo_errors, stats->collisions,
2918 stats->tx_carrier_errors +
2919 stats->tx_aborted_errors +
2920 stats->tx_window_errors +
2921 stats->tx_heartbeat_errors,
2922 stats->tx_compressed);
2926 * Called from the PROCfs module. This now uses the new arbitrary sized
2927 * /proc/net interface to create /proc/net/dev
2929 static int dev_seq_show(struct seq_file *seq, void *v)
2931 if (v == SEQ_START_TOKEN)
2932 seq_puts(seq, "Inter-| Receive "
2934 " face |bytes packets errs drop fifo frame "
2935 "compressed multicast|bytes packets errs "
2936 "drop fifo colls carrier compressed\n");
2938 dev_seq_printf_stats(seq, v);
2942 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2944 struct netif_rx_stats *rc = NULL;
2946 while (*pos < nr_cpu_ids)
2947 if (cpu_online(*pos)) {
2948 rc = &per_cpu(netdev_rx_stat, *pos);
2955 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2957 return softnet_get_online(pos);
2960 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2963 return softnet_get_online(pos);
2966 static void softnet_seq_stop(struct seq_file *seq, void *v)
2970 static int softnet_seq_show(struct seq_file *seq, void *v)
2972 struct netif_rx_stats *s = v;
2974 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2975 s->total, s->dropped, s->time_squeeze, 0,
2976 0, 0, 0, 0, /* was fastroute */
2981 static const struct seq_operations dev_seq_ops = {
2982 .start = dev_seq_start,
2983 .next = dev_seq_next,
2984 .stop = dev_seq_stop,
2985 .show = dev_seq_show,
2988 static int dev_seq_open(struct inode *inode, struct file *file)
2990 return seq_open_net(inode, file, &dev_seq_ops,
2991 sizeof(struct seq_net_private));
2994 static const struct file_operations dev_seq_fops = {
2995 .owner = THIS_MODULE,
2996 .open = dev_seq_open,
2998 .llseek = seq_lseek,
2999 .release = seq_release_net,
3002 static const struct seq_operations softnet_seq_ops = {
3003 .start = softnet_seq_start,
3004 .next = softnet_seq_next,
3005 .stop = softnet_seq_stop,
3006 .show = softnet_seq_show,
3009 static int softnet_seq_open(struct inode *inode, struct file *file)
3011 return seq_open(file, &softnet_seq_ops);
3014 static const struct file_operations softnet_seq_fops = {
3015 .owner = THIS_MODULE,
3016 .open = softnet_seq_open,
3018 .llseek = seq_lseek,
3019 .release = seq_release,
3022 static void *ptype_get_idx(loff_t pos)
3024 struct packet_type *pt = NULL;
3028 list_for_each_entry_rcu(pt, &ptype_all, list) {
3034 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3035 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3044 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3048 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3051 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3053 struct packet_type *pt;
3054 struct list_head *nxt;
3058 if (v == SEQ_START_TOKEN)
3059 return ptype_get_idx(0);
3062 nxt = pt->list.next;
3063 if (pt->type == htons(ETH_P_ALL)) {
3064 if (nxt != &ptype_all)
3067 nxt = ptype_base[0].next;
3069 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3071 while (nxt == &ptype_base[hash]) {
3072 if (++hash >= PTYPE_HASH_SIZE)
3074 nxt = ptype_base[hash].next;
3077 return list_entry(nxt, struct packet_type, list);
3080 static void ptype_seq_stop(struct seq_file *seq, void *v)
3086 static int ptype_seq_show(struct seq_file *seq, void *v)
3088 struct packet_type *pt = v;
3090 if (v == SEQ_START_TOKEN)
3091 seq_puts(seq, "Type Device Function\n");
3092 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3093 if (pt->type == htons(ETH_P_ALL))
3094 seq_puts(seq, "ALL ");
3096 seq_printf(seq, "%04x", ntohs(pt->type));
3098 seq_printf(seq, " %-8s %pF\n",
3099 pt->dev ? pt->dev->name : "", pt->func);
3105 static const struct seq_operations ptype_seq_ops = {
3106 .start = ptype_seq_start,
3107 .next = ptype_seq_next,
3108 .stop = ptype_seq_stop,
3109 .show = ptype_seq_show,
3112 static int ptype_seq_open(struct inode *inode, struct file *file)
3114 return seq_open_net(inode, file, &ptype_seq_ops,
3115 sizeof(struct seq_net_private));
3118 static const struct file_operations ptype_seq_fops = {
3119 .owner = THIS_MODULE,
3120 .open = ptype_seq_open,
3122 .llseek = seq_lseek,
3123 .release = seq_release_net,
3127 static int __net_init dev_proc_net_init(struct net *net)
3131 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3133 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3135 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3138 if (wext_proc_init(net))
3144 proc_net_remove(net, "ptype");
3146 proc_net_remove(net, "softnet_stat");
3148 proc_net_remove(net, "dev");
3152 static void __net_exit dev_proc_net_exit(struct net *net)
3154 wext_proc_exit(net);
3156 proc_net_remove(net, "ptype");
3157 proc_net_remove(net, "softnet_stat");
3158 proc_net_remove(net, "dev");
3161 static struct pernet_operations __net_initdata dev_proc_ops = {
3162 .init = dev_proc_net_init,
3163 .exit = dev_proc_net_exit,
3166 static int __init dev_proc_init(void)
3168 return register_pernet_subsys(&dev_proc_ops);
3171 #define dev_proc_init() 0
3172 #endif /* CONFIG_PROC_FS */
3176 * netdev_set_master - set up master/slave pair
3177 * @slave: slave device
3178 * @master: new master device
3180 * Changes the master device of the slave. Pass %NULL to break the
3181 * bonding. The caller must hold the RTNL semaphore. On a failure
3182 * a negative errno code is returned. On success the reference counts
3183 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3184 * function returns zero.
3186 int netdev_set_master(struct net_device *slave, struct net_device *master)
3188 struct net_device *old = slave->master;
3198 slave->master = master;
3206 slave->flags |= IFF_SLAVE;
3208 slave->flags &= ~IFF_SLAVE;
3210 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3214 static void dev_change_rx_flags(struct net_device *dev, int flags)
3216 const struct net_device_ops *ops = dev->netdev_ops;
3218 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3219 ops->ndo_change_rx_flags(dev, flags);
3222 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3224 unsigned short old_flags = dev->flags;
3230 dev->flags |= IFF_PROMISC;
3231 dev->promiscuity += inc;
3232 if (dev->promiscuity == 0) {
3235 * If inc causes overflow, untouch promisc and return error.
3238 dev->flags &= ~IFF_PROMISC;
3240 dev->promiscuity -= inc;
3241 printk(KERN_WARNING "%s: promiscuity touches roof, "
3242 "set promiscuity failed, promiscuity feature "
3243 "of device might be broken.\n", dev->name);
3247 if (dev->flags != old_flags) {
3248 printk(KERN_INFO "device %s %s promiscuous mode\n",
3249 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3251 if (audit_enabled) {
3252 current_uid_gid(&uid, &gid);
3253 audit_log(current->audit_context, GFP_ATOMIC,
3254 AUDIT_ANOM_PROMISCUOUS,
3255 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3256 dev->name, (dev->flags & IFF_PROMISC),
3257 (old_flags & IFF_PROMISC),
3258 audit_get_loginuid(current),
3260 audit_get_sessionid(current));
3263 dev_change_rx_flags(dev, IFF_PROMISC);
3269 * dev_set_promiscuity - update promiscuity count on a device
3273 * Add or remove promiscuity from a device. While the count in the device
3274 * remains above zero the interface remains promiscuous. Once it hits zero
3275 * the device reverts back to normal filtering operation. A negative inc
3276 * value is used to drop promiscuity on the device.
3277 * Return 0 if successful or a negative errno code on error.
3279 int dev_set_promiscuity(struct net_device *dev, int inc)
3281 unsigned short old_flags = dev->flags;
3284 err = __dev_set_promiscuity(dev, inc);
3287 if (dev->flags != old_flags)
3288 dev_set_rx_mode(dev);
3293 * dev_set_allmulti - update allmulti count on a device
3297 * Add or remove reception of all multicast frames to a device. While the
3298 * count in the device remains above zero the interface remains listening
3299 * to all interfaces. Once it hits zero the device reverts back to normal
3300 * filtering operation. A negative @inc value is used to drop the counter
3301 * when releasing a resource needing all multicasts.
3302 * Return 0 if successful or a negative errno code on error.
3305 int dev_set_allmulti(struct net_device *dev, int inc)
3307 unsigned short old_flags = dev->flags;
3311 dev->flags |= IFF_ALLMULTI;
3312 dev->allmulti += inc;
3313 if (dev->allmulti == 0) {
3316 * If inc causes overflow, untouch allmulti and return error.
3319 dev->flags &= ~IFF_ALLMULTI;
3321 dev->allmulti -= inc;
3322 printk(KERN_WARNING "%s: allmulti touches roof, "
3323 "set allmulti failed, allmulti feature of "
3324 "device might be broken.\n", dev->name);
3328 if (dev->flags ^ old_flags) {
3329 dev_change_rx_flags(dev, IFF_ALLMULTI);
3330 dev_set_rx_mode(dev);
3336 * Upload unicast and multicast address lists to device and
3337 * configure RX filtering. When the device doesn't support unicast
3338 * filtering it is put in promiscuous mode while unicast addresses
3341 void __dev_set_rx_mode(struct net_device *dev)
3343 const struct net_device_ops *ops = dev->netdev_ops;
3345 /* dev_open will call this function so the list will stay sane. */
3346 if (!(dev->flags&IFF_UP))
3349 if (!netif_device_present(dev))
3352 if (ops->ndo_set_rx_mode)
3353 ops->ndo_set_rx_mode(dev);
3355 /* Unicast addresses changes may only happen under the rtnl,
3356 * therefore calling __dev_set_promiscuity here is safe.
3358 if (dev->uc_count > 0 && !dev->uc_promisc) {
3359 __dev_set_promiscuity(dev, 1);
3360 dev->uc_promisc = 1;
3361 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3362 __dev_set_promiscuity(dev, -1);
3363 dev->uc_promisc = 0;
3366 if (ops->ndo_set_multicast_list)
3367 ops->ndo_set_multicast_list(dev);
3371 void dev_set_rx_mode(struct net_device *dev)
3373 netif_addr_lock_bh(dev);
3374 __dev_set_rx_mode(dev);
3375 netif_addr_unlock_bh(dev);
3378 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3379 void *addr, int alen, int glbl)
3381 struct dev_addr_list *da;
3383 for (; (da = *list) != NULL; list = &da->next) {
3384 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3385 alen == da->da_addrlen) {
3387 int old_glbl = da->da_gusers;
3404 int __dev_addr_add(struct dev_addr_list **list, int *count,
3405 void *addr, int alen, int glbl)
3407 struct dev_addr_list *da;
3409 for (da = *list; da != NULL; da = da->next) {
3410 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3411 da->da_addrlen == alen) {
3413 int old_glbl = da->da_gusers;
3423 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3426 memcpy(da->da_addr, addr, alen);
3427 da->da_addrlen = alen;
3429 da->da_gusers = glbl ? 1 : 0;
3437 * dev_unicast_delete - Release secondary unicast address.
3439 * @addr: address to delete
3440 * @alen: length of @addr
3442 * Release reference to a secondary unicast address and remove it
3443 * from the device if the reference count drops to zero.
3445 * The caller must hold the rtnl_mutex.
3447 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3453 netif_addr_lock_bh(dev);
3454 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3456 __dev_set_rx_mode(dev);
3457 netif_addr_unlock_bh(dev);
3460 EXPORT_SYMBOL(dev_unicast_delete);
3463 * dev_unicast_add - add a secondary unicast address
3465 * @addr: address to add
3466 * @alen: length of @addr
3468 * Add a secondary unicast address to the device or increase
3469 * the reference count if it already exists.
3471 * The caller must hold the rtnl_mutex.
3473 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3479 netif_addr_lock_bh(dev);
3480 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3482 __dev_set_rx_mode(dev);
3483 netif_addr_unlock_bh(dev);
3486 EXPORT_SYMBOL(dev_unicast_add);
3488 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3489 struct dev_addr_list **from, int *from_count)
3491 struct dev_addr_list *da, *next;
3495 while (da != NULL) {
3497 if (!da->da_synced) {
3498 err = __dev_addr_add(to, to_count,
3499 da->da_addr, da->da_addrlen, 0);
3504 } else if (da->da_users == 1) {
3505 __dev_addr_delete(to, to_count,
3506 da->da_addr, da->da_addrlen, 0);
3507 __dev_addr_delete(from, from_count,
3508 da->da_addr, da->da_addrlen, 0);
3515 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3516 struct dev_addr_list **from, int *from_count)
3518 struct dev_addr_list *da, *next;
3521 while (da != NULL) {
3523 if (da->da_synced) {
3524 __dev_addr_delete(to, to_count,
3525 da->da_addr, da->da_addrlen, 0);
3527 __dev_addr_delete(from, from_count,
3528 da->da_addr, da->da_addrlen, 0);
3535 * dev_unicast_sync - Synchronize device's unicast list to another device
3536 * @to: destination device
3537 * @from: source device
3539 * Add newly added addresses to the destination device and release
3540 * addresses that have no users left. The source device must be
3541 * locked by netif_tx_lock_bh.
3543 * This function is intended to be called from the dev->set_rx_mode
3544 * function of layered software devices.
3546 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3550 netif_addr_lock_bh(to);
3551 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3552 &from->uc_list, &from->uc_count);
3554 __dev_set_rx_mode(to);
3555 netif_addr_unlock_bh(to);
3558 EXPORT_SYMBOL(dev_unicast_sync);
3561 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3562 * @to: destination device
3563 * @from: source device
3565 * Remove all addresses that were added to the destination device by
3566 * dev_unicast_sync(). This function is intended to be called from the
3567 * dev->stop function of layered software devices.
3569 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3571 netif_addr_lock_bh(from);
3572 netif_addr_lock(to);
3574 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3575 &from->uc_list, &from->uc_count);
3576 __dev_set_rx_mode(to);
3578 netif_addr_unlock(to);
3579 netif_addr_unlock_bh(from);
3581 EXPORT_SYMBOL(dev_unicast_unsync);
3583 static void __dev_addr_discard(struct dev_addr_list **list)
3585 struct dev_addr_list *tmp;
3587 while (*list != NULL) {
3590 if (tmp->da_users > tmp->da_gusers)
3591 printk("__dev_addr_discard: address leakage! "
3592 "da_users=%d\n", tmp->da_users);
3597 static void dev_addr_discard(struct net_device *dev)
3599 netif_addr_lock_bh(dev);
3601 __dev_addr_discard(&dev->uc_list);
3604 __dev_addr_discard(&dev->mc_list);
3607 netif_addr_unlock_bh(dev);
3611 * dev_get_flags - get flags reported to userspace
3614 * Get the combination of flag bits exported through APIs to userspace.
3616 unsigned dev_get_flags(const struct net_device *dev)
3620 flags = (dev->flags & ~(IFF_PROMISC |
3625 (dev->gflags & (IFF_PROMISC |
3628 if (netif_running(dev)) {
3629 if (netif_oper_up(dev))
3630 flags |= IFF_RUNNING;
3631 if (netif_carrier_ok(dev))
3632 flags |= IFF_LOWER_UP;
3633 if (netif_dormant(dev))
3634 flags |= IFF_DORMANT;
3641 * dev_change_flags - change device settings
3643 * @flags: device state flags
3645 * Change settings on device based state flags. The flags are
3646 * in the userspace exported format.
3648 int dev_change_flags(struct net_device *dev, unsigned flags)
3651 int old_flags = dev->flags;
3656 * Set the flags on our device.
3659 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3660 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3662 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3666 * Load in the correct multicast list now the flags have changed.
3669 if ((old_flags ^ flags) & IFF_MULTICAST)
3670 dev_change_rx_flags(dev, IFF_MULTICAST);
3672 dev_set_rx_mode(dev);
3675 * Have we downed the interface. We handle IFF_UP ourselves
3676 * according to user attempts to set it, rather than blindly
3681 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3682 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3685 dev_set_rx_mode(dev);
3688 if (dev->flags & IFF_UP &&
3689 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3691 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3693 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3694 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3695 dev->gflags ^= IFF_PROMISC;
3696 dev_set_promiscuity(dev, inc);
3699 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3700 is important. Some (broken) drivers set IFF_PROMISC, when
3701 IFF_ALLMULTI is requested not asking us and not reporting.
3703 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3704 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3705 dev->gflags ^= IFF_ALLMULTI;
3706 dev_set_allmulti(dev, inc);
3709 /* Exclude state transition flags, already notified */
3710 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3712 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3718 * dev_set_mtu - Change maximum transfer unit
3720 * @new_mtu: new transfer unit
3722 * Change the maximum transfer size of the network device.
3724 int dev_set_mtu(struct net_device *dev, int new_mtu)
3726 const struct net_device_ops *ops = dev->netdev_ops;
3729 if (new_mtu == dev->mtu)
3732 /* MTU must be positive. */
3736 if (!netif_device_present(dev))
3740 if (ops->ndo_change_mtu)
3741 err = ops->ndo_change_mtu(dev, new_mtu);
3745 if (!err && dev->flags & IFF_UP)
3746 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3751 * dev_set_mac_address - Change Media Access Control Address
3755 * Change the hardware (MAC) address of the device
3757 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3759 const struct net_device_ops *ops = dev->netdev_ops;
3762 if (!ops->ndo_set_mac_address)
3764 if (sa->sa_family != dev->type)
3766 if (!netif_device_present(dev))
3768 err = ops->ndo_set_mac_address(dev, sa);
3770 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3775 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3777 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3780 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3786 case SIOCGIFFLAGS: /* Get interface flags */
3787 ifr->ifr_flags = dev_get_flags(dev);
3790 case SIOCGIFMETRIC: /* Get the metric on the interface
3791 (currently unused) */
3792 ifr->ifr_metric = 0;
3795 case SIOCGIFMTU: /* Get the MTU of a device */
3796 ifr->ifr_mtu = dev->mtu;
3801 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3803 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3804 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3805 ifr->ifr_hwaddr.sa_family = dev->type;
3813 ifr->ifr_map.mem_start = dev->mem_start;
3814 ifr->ifr_map.mem_end = dev->mem_end;
3815 ifr->ifr_map.base_addr = dev->base_addr;
3816 ifr->ifr_map.irq = dev->irq;
3817 ifr->ifr_map.dma = dev->dma;
3818 ifr->ifr_map.port = dev->if_port;
3822 ifr->ifr_ifindex = dev->ifindex;
3826 ifr->ifr_qlen = dev->tx_queue_len;
3830 /* dev_ioctl() should ensure this case
3842 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3844 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3847 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3848 const struct net_device_ops *ops;
3853 ops = dev->netdev_ops;
3856 case SIOCSIFFLAGS: /* Set interface flags */
3857 return dev_change_flags(dev, ifr->ifr_flags);
3859 case SIOCSIFMETRIC: /* Set the metric on the interface
3860 (currently unused) */
3863 case SIOCSIFMTU: /* Set the MTU of a device */
3864 return dev_set_mtu(dev, ifr->ifr_mtu);
3867 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3869 case SIOCSIFHWBROADCAST:
3870 if (ifr->ifr_hwaddr.sa_family != dev->type)
3872 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3873 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3874 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3878 if (ops->ndo_set_config) {
3879 if (!netif_device_present(dev))
3881 return ops->ndo_set_config(dev, &ifr->ifr_map);
3886 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
3887 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3889 if (!netif_device_present(dev))
3891 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3895 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
3896 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3898 if (!netif_device_present(dev))
3900 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3904 if (ifr->ifr_qlen < 0)
3906 dev->tx_queue_len = ifr->ifr_qlen;
3910 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3911 return dev_change_name(dev, ifr->ifr_newname);
3914 * Unknown or private ioctl
3918 if ((cmd >= SIOCDEVPRIVATE &&
3919 cmd <= SIOCDEVPRIVATE + 15) ||
3920 cmd == SIOCBONDENSLAVE ||
3921 cmd == SIOCBONDRELEASE ||
3922 cmd == SIOCBONDSETHWADDR ||
3923 cmd == SIOCBONDSLAVEINFOQUERY ||
3924 cmd == SIOCBONDINFOQUERY ||
3925 cmd == SIOCBONDCHANGEACTIVE ||
3926 cmd == SIOCGMIIPHY ||
3927 cmd == SIOCGMIIREG ||
3928 cmd == SIOCSMIIREG ||
3929 cmd == SIOCBRADDIF ||
3930 cmd == SIOCBRDELIF ||
3931 cmd == SIOCWANDEV) {
3933 if (ops->ndo_do_ioctl) {
3934 if (netif_device_present(dev))
3935 err = ops->ndo_do_ioctl(dev, ifr, cmd);
3947 * This function handles all "interface"-type I/O control requests. The actual
3948 * 'doing' part of this is dev_ifsioc above.
3952 * dev_ioctl - network device ioctl
3953 * @net: the applicable net namespace
3954 * @cmd: command to issue
3955 * @arg: pointer to a struct ifreq in user space
3957 * Issue ioctl functions to devices. This is normally called by the
3958 * user space syscall interfaces but can sometimes be useful for
3959 * other purposes. The return value is the return from the syscall if
3960 * positive or a negative errno code on error.
3963 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3969 /* One special case: SIOCGIFCONF takes ifconf argument
3970 and requires shared lock, because it sleeps writing
3974 if (cmd == SIOCGIFCONF) {
3976 ret = dev_ifconf(net, (char __user *) arg);
3980 if (cmd == SIOCGIFNAME)
3981 return dev_ifname(net, (struct ifreq __user *)arg);
3983 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3986 ifr.ifr_name[IFNAMSIZ-1] = 0;
3988 colon = strchr(ifr.ifr_name, ':');
3993 * See which interface the caller is talking about.
3998 * These ioctl calls:
3999 * - can be done by all.
4000 * - atomic and do not require locking.
4011 dev_load(net, ifr.ifr_name);
4012 read_lock(&dev_base_lock);
4013 ret = dev_ifsioc_locked(net, &ifr, cmd);
4014 read_unlock(&dev_base_lock);
4018 if (copy_to_user(arg, &ifr,
4019 sizeof(struct ifreq)))
4025 dev_load(net, ifr.ifr_name);
4027 ret = dev_ethtool(net, &ifr);
4032 if (copy_to_user(arg, &ifr,
4033 sizeof(struct ifreq)))
4039 * These ioctl calls:
4040 * - require superuser power.
4041 * - require strict serialization.
4047 if (!capable(CAP_NET_ADMIN))
4049 dev_load(net, ifr.ifr_name);
4051 ret = dev_ifsioc(net, &ifr, cmd);
4056 if (copy_to_user(arg, &ifr,
4057 sizeof(struct ifreq)))
4063 * These ioctl calls:
4064 * - require superuser power.
4065 * - require strict serialization.
4066 * - do not return a value
4076 case SIOCSIFHWBROADCAST:
4079 case SIOCBONDENSLAVE:
4080 case SIOCBONDRELEASE:
4081 case SIOCBONDSETHWADDR:
4082 case SIOCBONDCHANGEACTIVE:
4085 if (!capable(CAP_NET_ADMIN))
4088 case SIOCBONDSLAVEINFOQUERY:
4089 case SIOCBONDINFOQUERY:
4090 dev_load(net, ifr.ifr_name);
4092 ret = dev_ifsioc(net, &ifr, cmd);
4097 /* Get the per device memory space. We can add this but
4098 * currently do not support it */
4100 /* Set the per device memory buffer space.
4101 * Not applicable in our case */
4106 * Unknown or private ioctl.
4109 if (cmd == SIOCWANDEV ||
4110 (cmd >= SIOCDEVPRIVATE &&
4111 cmd <= SIOCDEVPRIVATE + 15)) {
4112 dev_load(net, ifr.ifr_name);
4114 ret = dev_ifsioc(net, &ifr, cmd);
4116 if (!ret && copy_to_user(arg, &ifr,
4117 sizeof(struct ifreq)))
4121 /* Take care of Wireless Extensions */
4122 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4123 return wext_handle_ioctl(net, &ifr, cmd, arg);
4130 * dev_new_index - allocate an ifindex
4131 * @net: the applicable net namespace
4133 * Returns a suitable unique value for a new device interface
4134 * number. The caller must hold the rtnl semaphore or the
4135 * dev_base_lock to be sure it remains unique.
4137 static int dev_new_index(struct net *net)
4143 if (!__dev_get_by_index(net, ifindex))
4148 /* Delayed registration/unregisteration */
4149 static LIST_HEAD(net_todo_list);
4151 static void net_set_todo(struct net_device *dev)
4153 list_add_tail(&dev->todo_list, &net_todo_list);
4156 static void rollback_registered(struct net_device *dev)
4158 BUG_ON(dev_boot_phase);
4161 /* Some devices call without registering for initialization unwind. */
4162 if (dev->reg_state == NETREG_UNINITIALIZED) {
4163 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
4164 "was registered\n", dev->name, dev);
4170 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4172 /* If device is running, close it first. */
4175 /* And unlink it from device chain. */
4176 unlist_netdevice(dev);
4178 dev->reg_state = NETREG_UNREGISTERING;
4182 /* Shutdown queueing discipline. */
4186 /* Notify protocols, that we are about to destroy
4187 this device. They should clean all the things.
4189 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4192 * Flush the unicast and multicast chains
4194 dev_addr_discard(dev);
4196 if (dev->netdev_ops->ndo_uninit)
4197 dev->netdev_ops->ndo_uninit(dev);
4199 /* Notifier chain MUST detach us from master device. */
4200 WARN_ON(dev->master);
4202 /* Remove entries from kobject tree */
4203 netdev_unregister_kobject(dev);
4210 static void __netdev_init_queue_locks_one(struct net_device *dev,
4211 struct netdev_queue *dev_queue,
4214 spin_lock_init(&dev_queue->_xmit_lock);
4215 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4216 dev_queue->xmit_lock_owner = -1;
4219 static void netdev_init_queue_locks(struct net_device *dev)
4221 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4222 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4225 unsigned long netdev_fix_features(unsigned long features, const char *name)
4227 /* Fix illegal SG+CSUM combinations. */
4228 if ((features & NETIF_F_SG) &&
4229 !(features & NETIF_F_ALL_CSUM)) {
4231 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4232 "checksum feature.\n", name);
4233 features &= ~NETIF_F_SG;
4236 /* TSO requires that SG is present as well. */
4237 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4239 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4240 "SG feature.\n", name);
4241 features &= ~NETIF_F_TSO;
4244 if (features & NETIF_F_UFO) {
4245 if (!(features & NETIF_F_GEN_CSUM)) {
4247 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4248 "since no NETIF_F_HW_CSUM feature.\n",
4250 features &= ~NETIF_F_UFO;
4253 if (!(features & NETIF_F_SG)) {
4255 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4256 "since no NETIF_F_SG feature.\n", name);
4257 features &= ~NETIF_F_UFO;
4263 EXPORT_SYMBOL(netdev_fix_features);
4266 * register_netdevice - register a network device
4267 * @dev: device to register
4269 * Take a completed network device structure and add it to the kernel
4270 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4271 * chain. 0 is returned on success. A negative errno code is returned
4272 * on a failure to set up the device, or if the name is a duplicate.
4274 * Callers must hold the rtnl semaphore. You may want
4275 * register_netdev() instead of this.
4278 * The locking appears insufficient to guarantee two parallel registers
4279 * will not get the same name.
4282 int register_netdevice(struct net_device *dev)
4284 struct hlist_head *head;
4285 struct hlist_node *p;
4287 struct net *net = dev_net(dev);
4289 BUG_ON(dev_boot_phase);
4294 /* When net_device's are persistent, this will be fatal. */
4295 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4298 spin_lock_init(&dev->addr_list_lock);
4299 netdev_set_addr_lockdep_class(dev);
4300 netdev_init_queue_locks(dev);
4304 #ifdef CONFIG_COMPAT_NET_DEV_OPS
4305 /* Netdevice_ops API compatiability support.
4306 * This is temporary until all network devices are converted.
4308 if (dev->netdev_ops) {
4309 const struct net_device_ops *ops = dev->netdev_ops;
4311 dev->init = ops->ndo_init;
4312 dev->uninit = ops->ndo_uninit;
4313 dev->open = ops->ndo_open;
4314 dev->change_rx_flags = ops->ndo_change_rx_flags;
4315 dev->set_rx_mode = ops->ndo_set_rx_mode;
4316 dev->set_multicast_list = ops->ndo_set_multicast_list;
4317 dev->set_mac_address = ops->ndo_set_mac_address;
4318 dev->validate_addr = ops->ndo_validate_addr;
4319 dev->do_ioctl = ops->ndo_do_ioctl;
4320 dev->set_config = ops->ndo_set_config;
4321 dev->change_mtu = ops->ndo_change_mtu;
4322 dev->tx_timeout = ops->ndo_tx_timeout;
4323 dev->get_stats = ops->ndo_get_stats;
4324 dev->vlan_rx_register = ops->ndo_vlan_rx_register;
4325 dev->vlan_rx_add_vid = ops->ndo_vlan_rx_add_vid;
4326 dev->vlan_rx_kill_vid = ops->ndo_vlan_rx_kill_vid;
4327 #ifdef CONFIG_NET_POLL_CONTROLLER
4328 dev->poll_controller = ops->ndo_poll_controller;
4331 char drivername[64];
4332 pr_info("%s (%s): not using net_device_ops yet\n",
4333 dev->name, netdev_drivername(dev, drivername, 64));
4335 /* This works only because net_device_ops and the
4336 compatiablity structure are the same. */
4337 dev->netdev_ops = (void *) &(dev->init);
4341 /* Init, if this function is available */
4342 if (dev->netdev_ops->ndo_init) {
4343 ret = dev->netdev_ops->ndo_init(dev);
4351 if (!dev_valid_name(dev->name)) {
4356 dev->ifindex = dev_new_index(net);
4357 if (dev->iflink == -1)
4358 dev->iflink = dev->ifindex;
4360 /* Check for existence of name */
4361 head = dev_name_hash(net, dev->name);
4362 hlist_for_each(p, head) {
4363 struct net_device *d
4364 = hlist_entry(p, struct net_device, name_hlist);
4365 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4371 /* Fix illegal checksum combinations */
4372 if ((dev->features & NETIF_F_HW_CSUM) &&
4373 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4374 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4376 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4379 if ((dev->features & NETIF_F_NO_CSUM) &&
4380 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4381 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4383 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4386 dev->features = netdev_fix_features(dev->features, dev->name);
4388 /* Enable software GSO if SG is supported. */
4389 if (dev->features & NETIF_F_SG)
4390 dev->features |= NETIF_F_GSO;
4392 netdev_initialize_kobject(dev);
4393 ret = netdev_register_kobject(dev);
4396 dev->reg_state = NETREG_REGISTERED;
4399 * Default initial state at registry is that the
4400 * device is present.
4403 set_bit(__LINK_STATE_PRESENT, &dev->state);
4405 dev_init_scheduler(dev);
4407 list_netdevice(dev);
4409 /* Notify protocols, that a new device appeared. */
4410 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4411 ret = notifier_to_errno(ret);
4413 rollback_registered(dev);
4414 dev->reg_state = NETREG_UNREGISTERED;
4421 if (dev->netdev_ops->ndo_uninit)
4422 dev->netdev_ops->ndo_uninit(dev);
4427 * register_netdev - register a network device
4428 * @dev: device to register
4430 * Take a completed network device structure and add it to the kernel
4431 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4432 * chain. 0 is returned on success. A negative errno code is returned
4433 * on a failure to set up the device, or if the name is a duplicate.
4435 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4436 * and expands the device name if you passed a format string to
4439 int register_netdev(struct net_device *dev)
4446 * If the name is a format string the caller wants us to do a
4449 if (strchr(dev->name, '%')) {
4450 err = dev_alloc_name(dev, dev->name);
4455 err = register_netdevice(dev);
4460 EXPORT_SYMBOL(register_netdev);
4463 * netdev_wait_allrefs - wait until all references are gone.
4465 * This is called when unregistering network devices.
4467 * Any protocol or device that holds a reference should register
4468 * for netdevice notification, and cleanup and put back the
4469 * reference if they receive an UNREGISTER event.
4470 * We can get stuck here if buggy protocols don't correctly
4473 static void netdev_wait_allrefs(struct net_device *dev)
4475 unsigned long rebroadcast_time, warning_time;
4477 rebroadcast_time = warning_time = jiffies;
4478 while (atomic_read(&dev->refcnt) != 0) {
4479 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4482 /* Rebroadcast unregister notification */
4483 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4485 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4487 /* We must not have linkwatch events
4488 * pending on unregister. If this
4489 * happens, we simply run the queue
4490 * unscheduled, resulting in a noop
4493 linkwatch_run_queue();
4498 rebroadcast_time = jiffies;
4503 if (time_after(jiffies, warning_time + 10 * HZ)) {
4504 printk(KERN_EMERG "unregister_netdevice: "
4505 "waiting for %s to become free. Usage "
4507 dev->name, atomic_read(&dev->refcnt));
4508 warning_time = jiffies;
4517 * register_netdevice(x1);
4518 * register_netdevice(x2);
4520 * unregister_netdevice(y1);
4521 * unregister_netdevice(y2);
4527 * We are invoked by rtnl_unlock().
4528 * This allows us to deal with problems:
4529 * 1) We can delete sysfs objects which invoke hotplug
4530 * without deadlocking with linkwatch via keventd.
4531 * 2) Since we run with the RTNL semaphore not held, we can sleep
4532 * safely in order to wait for the netdev refcnt to drop to zero.
4534 * We must not return until all unregister events added during
4535 * the interval the lock was held have been completed.
4537 void netdev_run_todo(void)
4539 struct list_head list;
4541 /* Snapshot list, allow later requests */
4542 list_replace_init(&net_todo_list, &list);
4546 while (!list_empty(&list)) {
4547 struct net_device *dev
4548 = list_entry(list.next, struct net_device, todo_list);
4549 list_del(&dev->todo_list);
4551 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4552 printk(KERN_ERR "network todo '%s' but state %d\n",
4553 dev->name, dev->reg_state);
4558 dev->reg_state = NETREG_UNREGISTERED;
4560 on_each_cpu(flush_backlog, dev, 1);
4562 netdev_wait_allrefs(dev);
4565 BUG_ON(atomic_read(&dev->refcnt));
4566 WARN_ON(dev->ip_ptr);
4567 WARN_ON(dev->ip6_ptr);
4568 WARN_ON(dev->dn_ptr);
4570 if (dev->destructor)
4571 dev->destructor(dev);
4573 /* Free network device */
4574 kobject_put(&dev->dev.kobj);
4579 * dev_get_stats - get network device statistics
4580 * @dev: device to get statistics from
4582 * Get network statistics from device. The device driver may provide
4583 * its own method by setting dev->netdev_ops->get_stats; otherwise
4584 * the internal statistics structure is used.
4586 const struct net_device_stats *dev_get_stats(struct net_device *dev)
4588 const struct net_device_ops *ops = dev->netdev_ops;
4590 if (ops->ndo_get_stats)
4591 return ops->ndo_get_stats(dev);
4595 EXPORT_SYMBOL(dev_get_stats);
4597 static void netdev_init_one_queue(struct net_device *dev,
4598 struct netdev_queue *queue,
4604 static void netdev_init_queues(struct net_device *dev)
4606 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4607 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4608 spin_lock_init(&dev->tx_global_lock);
4612 * alloc_netdev_mq - allocate network device
4613 * @sizeof_priv: size of private data to allocate space for
4614 * @name: device name format string
4615 * @setup: callback to initialize device
4616 * @queue_count: the number of subqueues to allocate
4618 * Allocates a struct net_device with private data area for driver use
4619 * and performs basic initialization. Also allocates subquue structs
4620 * for each queue on the device at the end of the netdevice.
4622 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4623 void (*setup)(struct net_device *), unsigned int queue_count)
4625 struct netdev_queue *tx;
4626 struct net_device *dev;
4630 BUG_ON(strlen(name) >= sizeof(dev->name));
4632 alloc_size = sizeof(struct net_device);
4634 /* ensure 32-byte alignment of private area */
4635 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4636 alloc_size += sizeof_priv;
4638 /* ensure 32-byte alignment of whole construct */
4639 alloc_size += NETDEV_ALIGN_CONST;
4641 p = kzalloc(alloc_size, GFP_KERNEL);
4643 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4647 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
4649 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4655 dev = (struct net_device *)
4656 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4657 dev->padded = (char *)dev - (char *)p;
4658 dev_net_set(dev, &init_net);
4661 dev->num_tx_queues = queue_count;
4662 dev->real_num_tx_queues = queue_count;
4664 dev->gso_max_size = GSO_MAX_SIZE;
4666 netdev_init_queues(dev);
4668 INIT_LIST_HEAD(&dev->napi_list);
4670 strcpy(dev->name, name);
4673 EXPORT_SYMBOL(alloc_netdev_mq);
4676 * free_netdev - free network device
4679 * This function does the last stage of destroying an allocated device
4680 * interface. The reference to the device object is released.
4681 * If this is the last reference then it will be freed.
4683 void free_netdev(struct net_device *dev)
4685 struct napi_struct *p, *n;
4687 release_net(dev_net(dev));
4691 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
4694 /* Compatibility with error handling in drivers */
4695 if (dev->reg_state == NETREG_UNINITIALIZED) {
4696 kfree((char *)dev - dev->padded);
4700 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4701 dev->reg_state = NETREG_RELEASED;
4703 /* will free via device release */
4704 put_device(&dev->dev);
4708 * synchronize_net - Synchronize with packet receive processing
4710 * Wait for packets currently being received to be done.
4711 * Does not block later packets from starting.
4713 void synchronize_net(void)
4720 * unregister_netdevice - remove device from the kernel
4723 * This function shuts down a device interface and removes it
4724 * from the kernel tables.
4726 * Callers must hold the rtnl semaphore. You may want
4727 * unregister_netdev() instead of this.
4730 void unregister_netdevice(struct net_device *dev)
4734 rollback_registered(dev);
4735 /* Finish processing unregister after unlock */
4740 * unregister_netdev - remove device from the kernel
4743 * This function shuts down a device interface and removes it
4744 * from the kernel tables.
4746 * This is just a wrapper for unregister_netdevice that takes
4747 * the rtnl semaphore. In general you want to use this and not
4748 * unregister_netdevice.
4750 void unregister_netdev(struct net_device *dev)
4753 unregister_netdevice(dev);
4757 EXPORT_SYMBOL(unregister_netdev);
4760 * dev_change_net_namespace - move device to different nethost namespace
4762 * @net: network namespace
4763 * @pat: If not NULL name pattern to try if the current device name
4764 * is already taken in the destination network namespace.
4766 * This function shuts down a device interface and moves it
4767 * to a new network namespace. On success 0 is returned, on
4768 * a failure a netagive errno code is returned.
4770 * Callers must hold the rtnl semaphore.
4773 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4776 const char *destname;
4781 /* Don't allow namespace local devices to be moved. */
4783 if (dev->features & NETIF_F_NETNS_LOCAL)
4787 /* Don't allow real devices to be moved when sysfs
4791 if (dev->dev.parent)
4795 /* Ensure the device has been registrered */
4797 if (dev->reg_state != NETREG_REGISTERED)
4800 /* Get out if there is nothing todo */
4802 if (net_eq(dev_net(dev), net))
4805 /* Pick the destination device name, and ensure
4806 * we can use it in the destination network namespace.
4809 destname = dev->name;
4810 if (__dev_get_by_name(net, destname)) {
4811 /* We get here if we can't use the current device name */
4814 if (!dev_valid_name(pat))
4816 if (strchr(pat, '%')) {
4817 if (__dev_alloc_name(net, pat, buf) < 0)
4822 if (__dev_get_by_name(net, destname))
4827 * And now a mini version of register_netdevice unregister_netdevice.
4830 /* If device is running close it first. */
4833 /* And unlink it from device chain */
4835 unlist_netdevice(dev);
4839 /* Shutdown queueing discipline. */
4842 /* Notify protocols, that we are about to destroy
4843 this device. They should clean all the things.
4845 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4848 * Flush the unicast and multicast chains
4850 dev_addr_discard(dev);
4852 netdev_unregister_kobject(dev);
4854 /* Actually switch the network namespace */
4855 dev_net_set(dev, net);
4857 /* Assign the new device name */
4858 if (destname != dev->name)
4859 strcpy(dev->name, destname);
4861 /* If there is an ifindex conflict assign a new one */
4862 if (__dev_get_by_index(net, dev->ifindex)) {
4863 int iflink = (dev->iflink == dev->ifindex);
4864 dev->ifindex = dev_new_index(net);
4866 dev->iflink = dev->ifindex;
4869 /* Fixup kobjects */
4870 err = netdev_register_kobject(dev);
4873 /* Add the device back in the hashes */
4874 list_netdevice(dev);
4876 /* Notify protocols, that a new device appeared. */
4877 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4885 static int dev_cpu_callback(struct notifier_block *nfb,
4886 unsigned long action,
4889 struct sk_buff **list_skb;
4890 struct Qdisc **list_net;
4891 struct sk_buff *skb;
4892 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4893 struct softnet_data *sd, *oldsd;
4895 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4898 local_irq_disable();
4899 cpu = smp_processor_id();
4900 sd = &per_cpu(softnet_data, cpu);
4901 oldsd = &per_cpu(softnet_data, oldcpu);
4903 /* Find end of our completion_queue. */
4904 list_skb = &sd->completion_queue;
4906 list_skb = &(*list_skb)->next;
4907 /* Append completion queue from offline CPU. */
4908 *list_skb = oldsd->completion_queue;
4909 oldsd->completion_queue = NULL;
4911 /* Find end of our output_queue. */
4912 list_net = &sd->output_queue;
4914 list_net = &(*list_net)->next_sched;
4915 /* Append output queue from offline CPU. */
4916 *list_net = oldsd->output_queue;
4917 oldsd->output_queue = NULL;
4919 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4922 /* Process offline CPU's input_pkt_queue */
4923 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4931 * netdev_increment_features - increment feature set by one
4932 * @all: current feature set
4933 * @one: new feature set
4934 * @mask: mask feature set
4936 * Computes a new feature set after adding a device with feature set
4937 * @one to the master device with current feature set @all. Will not
4938 * enable anything that is off in @mask. Returns the new feature set.
4940 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
4943 /* If device needs checksumming, downgrade to it. */
4944 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4945 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
4946 else if (mask & NETIF_F_ALL_CSUM) {
4947 /* If one device supports v4/v6 checksumming, set for all. */
4948 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
4949 !(all & NETIF_F_GEN_CSUM)) {
4950 all &= ~NETIF_F_ALL_CSUM;
4951 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
4954 /* If one device supports hw checksumming, set for all. */
4955 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
4956 all &= ~NETIF_F_ALL_CSUM;
4957 all |= NETIF_F_HW_CSUM;
4961 one |= NETIF_F_ALL_CSUM;
4963 one |= all & NETIF_F_ONE_FOR_ALL;
4964 all &= one | NETIF_F_LLTX | NETIF_F_GSO;
4965 all |= one & mask & NETIF_F_ONE_FOR_ALL;
4969 EXPORT_SYMBOL(netdev_increment_features);
4971 static struct hlist_head *netdev_create_hash(void)
4974 struct hlist_head *hash;
4976 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4978 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4979 INIT_HLIST_HEAD(&hash[i]);
4984 /* Initialize per network namespace state */
4985 static int __net_init netdev_init(struct net *net)
4987 INIT_LIST_HEAD(&net->dev_base_head);
4989 net->dev_name_head = netdev_create_hash();
4990 if (net->dev_name_head == NULL)
4993 net->dev_index_head = netdev_create_hash();
4994 if (net->dev_index_head == NULL)
5000 kfree(net->dev_name_head);
5006 * netdev_drivername - network driver for the device
5007 * @dev: network device
5008 * @buffer: buffer for resulting name
5009 * @len: size of buffer
5011 * Determine network driver for device.
5013 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5015 const struct device_driver *driver;
5016 const struct device *parent;
5018 if (len <= 0 || !buffer)
5022 parent = dev->dev.parent;
5027 driver = parent->driver;
5028 if (driver && driver->name)
5029 strlcpy(buffer, driver->name, len);
5033 static void __net_exit netdev_exit(struct net *net)
5035 kfree(net->dev_name_head);
5036 kfree(net->dev_index_head);
5039 static struct pernet_operations __net_initdata netdev_net_ops = {
5040 .init = netdev_init,
5041 .exit = netdev_exit,
5044 static void __net_exit default_device_exit(struct net *net)
5046 struct net_device *dev;
5048 * Push all migratable of the network devices back to the
5049 * initial network namespace
5053 for_each_netdev(net, dev) {
5055 char fb_name[IFNAMSIZ];
5057 /* Ignore unmoveable devices (i.e. loopback) */
5058 if (dev->features & NETIF_F_NETNS_LOCAL)
5061 /* Delete virtual devices */
5062 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5063 dev->rtnl_link_ops->dellink(dev);
5067 /* Push remaing network devices to init_net */
5068 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5069 err = dev_change_net_namespace(dev, &init_net, fb_name);
5071 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5072 __func__, dev->name, err);
5080 static struct pernet_operations __net_initdata default_device_ops = {
5081 .exit = default_device_exit,
5085 * Initialize the DEV module. At boot time this walks the device list and
5086 * unhooks any devices that fail to initialise (normally hardware not
5087 * present) and leaves us with a valid list of present and active devices.
5092 * This is called single threaded during boot, so no need
5093 * to take the rtnl semaphore.
5095 static int __init net_dev_init(void)
5097 int i, rc = -ENOMEM;
5099 BUG_ON(!dev_boot_phase);
5101 if (dev_proc_init())
5104 if (netdev_kobject_init())
5107 INIT_LIST_HEAD(&ptype_all);
5108 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5109 INIT_LIST_HEAD(&ptype_base[i]);
5111 if (register_pernet_subsys(&netdev_net_ops))
5115 * Initialise the packet receive queues.
5118 for_each_possible_cpu(i) {
5119 struct softnet_data *queue;
5121 queue = &per_cpu(softnet_data, i);
5122 skb_queue_head_init(&queue->input_pkt_queue);
5123 queue->completion_queue = NULL;
5124 INIT_LIST_HEAD(&queue->poll_list);
5126 queue->backlog.poll = process_backlog;
5127 queue->backlog.weight = weight_p;
5128 queue->backlog.gro_list = NULL;
5133 /* The loopback device is special if any other network devices
5134 * is present in a network namespace the loopback device must
5135 * be present. Since we now dynamically allocate and free the
5136 * loopback device ensure this invariant is maintained by
5137 * keeping the loopback device as the first device on the
5138 * list of network devices. Ensuring the loopback devices
5139 * is the first device that appears and the last network device
5142 if (register_pernet_device(&loopback_net_ops))
5145 if (register_pernet_device(&default_device_ops))
5148 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5149 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5151 hotcpu_notifier(dev_cpu_callback, 0);
5154 #ifdef CONFIG_NET_DMA
5162 subsys_initcall(net_dev_init);
5164 EXPORT_SYMBOL(__dev_get_by_index);
5165 EXPORT_SYMBOL(__dev_get_by_name);
5166 EXPORT_SYMBOL(__dev_remove_pack);
5167 EXPORT_SYMBOL(dev_valid_name);
5168 EXPORT_SYMBOL(dev_add_pack);
5169 EXPORT_SYMBOL(dev_alloc_name);
5170 EXPORT_SYMBOL(dev_close);
5171 EXPORT_SYMBOL(dev_get_by_flags);
5172 EXPORT_SYMBOL(dev_get_by_index);
5173 EXPORT_SYMBOL(dev_get_by_name);
5174 EXPORT_SYMBOL(dev_open);
5175 EXPORT_SYMBOL(dev_queue_xmit);
5176 EXPORT_SYMBOL(dev_remove_pack);
5177 EXPORT_SYMBOL(dev_set_allmulti);
5178 EXPORT_SYMBOL(dev_set_promiscuity);
5179 EXPORT_SYMBOL(dev_change_flags);
5180 EXPORT_SYMBOL(dev_set_mtu);
5181 EXPORT_SYMBOL(dev_set_mac_address);
5182 EXPORT_SYMBOL(free_netdev);
5183 EXPORT_SYMBOL(netdev_boot_setup_check);
5184 EXPORT_SYMBOL(netdev_set_master);
5185 EXPORT_SYMBOL(netdev_state_change);
5186 EXPORT_SYMBOL(netif_receive_skb);
5187 EXPORT_SYMBOL(netif_rx);
5188 EXPORT_SYMBOL(register_gifconf);
5189 EXPORT_SYMBOL(register_netdevice);
5190 EXPORT_SYMBOL(register_netdevice_notifier);
5191 EXPORT_SYMBOL(skb_checksum_help);
5192 EXPORT_SYMBOL(synchronize_net);
5193 EXPORT_SYMBOL(unregister_netdevice);
5194 EXPORT_SYMBOL(unregister_netdevice_notifier);
5195 EXPORT_SYMBOL(net_enable_timestamp);
5196 EXPORT_SYMBOL(net_disable_timestamp);
5197 EXPORT_SYMBOL(dev_get_flags);
5199 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
5200 EXPORT_SYMBOL(br_handle_frame_hook);
5201 EXPORT_SYMBOL(br_fdb_get_hook);
5202 EXPORT_SYMBOL(br_fdb_put_hook);
5205 EXPORT_SYMBOL(dev_load);
5207 EXPORT_PER_CPU_SYMBOL(softnet_data);