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;
1093 net_dmaengine_get();
1096 * Initialize multicasting status
1098 dev_set_rx_mode(dev);
1101 * Wakeup transmit queue engine
1106 * ... and announce new interface.
1108 call_netdevice_notifiers(NETDEV_UP, dev);
1115 * dev_close - shutdown an interface.
1116 * @dev: device to shutdown
1118 * This function moves an active device into down state. A
1119 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1120 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1123 int dev_close(struct net_device *dev)
1125 const struct net_device_ops *ops = dev->netdev_ops;
1130 if (!(dev->flags & IFF_UP))
1134 * Tell people we are going down, so that they can
1135 * prepare to death, when device is still operating.
1137 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1139 clear_bit(__LINK_STATE_START, &dev->state);
1141 /* Synchronize to scheduled poll. We cannot touch poll list,
1142 * it can be even on different cpu. So just clear netif_running().
1144 * dev->stop() will invoke napi_disable() on all of it's
1145 * napi_struct instances on this device.
1147 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1149 dev_deactivate(dev);
1152 * Call the device specific close. This cannot fail.
1153 * Only if device is UP
1155 * We allow it to be called even after a DETACH hot-plug
1162 * Device is now down.
1165 dev->flags &= ~IFF_UP;
1168 * Tell people we are down
1170 call_netdevice_notifiers(NETDEV_DOWN, dev);
1175 net_dmaengine_put();
1182 * dev_disable_lro - disable Large Receive Offload on a device
1185 * Disable Large Receive Offload (LRO) on a net device. Must be
1186 * called under RTNL. This is needed if received packets may be
1187 * forwarded to another interface.
1189 void dev_disable_lro(struct net_device *dev)
1191 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1192 dev->ethtool_ops->set_flags) {
1193 u32 flags = dev->ethtool_ops->get_flags(dev);
1194 if (flags & ETH_FLAG_LRO) {
1195 flags &= ~ETH_FLAG_LRO;
1196 dev->ethtool_ops->set_flags(dev, flags);
1199 WARN_ON(dev->features & NETIF_F_LRO);
1201 EXPORT_SYMBOL(dev_disable_lro);
1204 static int dev_boot_phase = 1;
1207 * Device change register/unregister. These are not inline or static
1208 * as we export them to the world.
1212 * register_netdevice_notifier - register a network notifier block
1215 * Register a notifier to be called when network device events occur.
1216 * The notifier passed is linked into the kernel structures and must
1217 * not be reused until it has been unregistered. A negative errno code
1218 * is returned on a failure.
1220 * When registered all registration and up events are replayed
1221 * to the new notifier to allow device to have a race free
1222 * view of the network device list.
1225 int register_netdevice_notifier(struct notifier_block *nb)
1227 struct net_device *dev;
1228 struct net_device *last;
1233 err = raw_notifier_chain_register(&netdev_chain, nb);
1239 for_each_netdev(net, dev) {
1240 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1241 err = notifier_to_errno(err);
1245 if (!(dev->flags & IFF_UP))
1248 nb->notifier_call(nb, NETDEV_UP, dev);
1259 for_each_netdev(net, dev) {
1263 if (dev->flags & IFF_UP) {
1264 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1265 nb->notifier_call(nb, NETDEV_DOWN, dev);
1267 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1271 raw_notifier_chain_unregister(&netdev_chain, nb);
1276 * unregister_netdevice_notifier - unregister a network notifier block
1279 * Unregister a notifier previously registered by
1280 * register_netdevice_notifier(). The notifier is unlinked into the
1281 * kernel structures and may then be reused. A negative errno code
1282 * is returned on a failure.
1285 int unregister_netdevice_notifier(struct notifier_block *nb)
1290 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1296 * call_netdevice_notifiers - call all network notifier blocks
1297 * @val: value passed unmodified to notifier function
1298 * @dev: net_device pointer passed unmodified to notifier function
1300 * Call all network notifier blocks. Parameters and return value
1301 * are as for raw_notifier_call_chain().
1304 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1306 return raw_notifier_call_chain(&netdev_chain, val, dev);
1309 /* When > 0 there are consumers of rx skb time stamps */
1310 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1312 void net_enable_timestamp(void)
1314 atomic_inc(&netstamp_needed);
1317 void net_disable_timestamp(void)
1319 atomic_dec(&netstamp_needed);
1322 static inline void net_timestamp(struct sk_buff *skb)
1324 if (atomic_read(&netstamp_needed))
1325 __net_timestamp(skb);
1327 skb->tstamp.tv64 = 0;
1331 * Support routine. Sends outgoing frames to any network
1332 * taps currently in use.
1335 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1337 struct packet_type *ptype;
1339 #ifdef CONFIG_NET_CLS_ACT
1340 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1347 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1348 /* Never send packets back to the socket
1349 * they originated from - MvS (miquels@drinkel.ow.org)
1351 if ((ptype->dev == dev || !ptype->dev) &&
1352 (ptype->af_packet_priv == NULL ||
1353 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1354 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1358 /* skb->nh should be correctly
1359 set by sender, so that the second statement is
1360 just protection against buggy protocols.
1362 skb_reset_mac_header(skb2);
1364 if (skb_network_header(skb2) < skb2->data ||
1365 skb2->network_header > skb2->tail) {
1366 if (net_ratelimit())
1367 printk(KERN_CRIT "protocol %04x is "
1369 skb2->protocol, dev->name);
1370 skb_reset_network_header(skb2);
1373 skb2->transport_header = skb2->network_header;
1374 skb2->pkt_type = PACKET_OUTGOING;
1375 ptype->func(skb2, skb->dev, ptype, skb->dev);
1382 static inline void __netif_reschedule(struct Qdisc *q)
1384 struct softnet_data *sd;
1385 unsigned long flags;
1387 local_irq_save(flags);
1388 sd = &__get_cpu_var(softnet_data);
1389 q->next_sched = sd->output_queue;
1390 sd->output_queue = q;
1391 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1392 local_irq_restore(flags);
1395 void __netif_schedule(struct Qdisc *q)
1397 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1398 __netif_reschedule(q);
1400 EXPORT_SYMBOL(__netif_schedule);
1402 void dev_kfree_skb_irq(struct sk_buff *skb)
1404 if (atomic_dec_and_test(&skb->users)) {
1405 struct softnet_data *sd;
1406 unsigned long flags;
1408 local_irq_save(flags);
1409 sd = &__get_cpu_var(softnet_data);
1410 skb->next = sd->completion_queue;
1411 sd->completion_queue = skb;
1412 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1413 local_irq_restore(flags);
1416 EXPORT_SYMBOL(dev_kfree_skb_irq);
1418 void dev_kfree_skb_any(struct sk_buff *skb)
1420 if (in_irq() || irqs_disabled())
1421 dev_kfree_skb_irq(skb);
1425 EXPORT_SYMBOL(dev_kfree_skb_any);
1429 * netif_device_detach - mark device as removed
1430 * @dev: network device
1432 * Mark device as removed from system and therefore no longer available.
1434 void netif_device_detach(struct net_device *dev)
1436 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1437 netif_running(dev)) {
1438 netif_tx_stop_all_queues(dev);
1441 EXPORT_SYMBOL(netif_device_detach);
1444 * netif_device_attach - mark device as attached
1445 * @dev: network device
1447 * Mark device as attached from system and restart if needed.
1449 void netif_device_attach(struct net_device *dev)
1451 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1452 netif_running(dev)) {
1453 netif_tx_wake_all_queues(dev);
1454 __netdev_watchdog_up(dev);
1457 EXPORT_SYMBOL(netif_device_attach);
1459 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1461 return ((features & NETIF_F_GEN_CSUM) ||
1462 ((features & NETIF_F_IP_CSUM) &&
1463 protocol == htons(ETH_P_IP)) ||
1464 ((features & NETIF_F_IPV6_CSUM) &&
1465 protocol == htons(ETH_P_IPV6)) ||
1466 ((features & NETIF_F_FCOE_CRC) &&
1467 protocol == htons(ETH_P_FCOE)));
1470 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1472 if (can_checksum_protocol(dev->features, skb->protocol))
1475 if (skb->protocol == htons(ETH_P_8021Q)) {
1476 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1477 if (can_checksum_protocol(dev->features & dev->vlan_features,
1478 veh->h_vlan_encapsulated_proto))
1486 * Invalidate hardware checksum when packet is to be mangled, and
1487 * complete checksum manually on outgoing path.
1489 int skb_checksum_help(struct sk_buff *skb)
1492 int ret = 0, offset;
1494 if (skb->ip_summed == CHECKSUM_COMPLETE)
1495 goto out_set_summed;
1497 if (unlikely(skb_shinfo(skb)->gso_size)) {
1498 /* Let GSO fix up the checksum. */
1499 goto out_set_summed;
1502 offset = skb->csum_start - skb_headroom(skb);
1503 BUG_ON(offset >= skb_headlen(skb));
1504 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1506 offset += skb->csum_offset;
1507 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1509 if (skb_cloned(skb) &&
1510 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1511 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1516 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1518 skb->ip_summed = CHECKSUM_NONE;
1524 * skb_gso_segment - Perform segmentation on skb.
1525 * @skb: buffer to segment
1526 * @features: features for the output path (see dev->features)
1528 * This function segments the given skb and returns a list of segments.
1530 * It may return NULL if the skb requires no segmentation. This is
1531 * only possible when GSO is used for verifying header integrity.
1533 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1535 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1536 struct packet_type *ptype;
1537 __be16 type = skb->protocol;
1540 skb_reset_mac_header(skb);
1541 skb->mac_len = skb->network_header - skb->mac_header;
1542 __skb_pull(skb, skb->mac_len);
1544 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1545 struct net_device *dev = skb->dev;
1546 struct ethtool_drvinfo info = {};
1548 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1549 dev->ethtool_ops->get_drvinfo(dev, &info);
1551 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1553 info.driver, dev ? dev->features : 0L,
1554 skb->sk ? skb->sk->sk_route_caps : 0L,
1555 skb->len, skb->data_len, skb->ip_summed);
1557 if (skb_header_cloned(skb) &&
1558 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1559 return ERR_PTR(err);
1563 list_for_each_entry_rcu(ptype,
1564 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1565 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1566 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1567 err = ptype->gso_send_check(skb);
1568 segs = ERR_PTR(err);
1569 if (err || skb_gso_ok(skb, features))
1571 __skb_push(skb, (skb->data -
1572 skb_network_header(skb)));
1574 segs = ptype->gso_segment(skb, features);
1580 __skb_push(skb, skb->data - skb_mac_header(skb));
1585 EXPORT_SYMBOL(skb_gso_segment);
1587 /* Take action when hardware reception checksum errors are detected. */
1589 void netdev_rx_csum_fault(struct net_device *dev)
1591 if (net_ratelimit()) {
1592 printk(KERN_ERR "%s: hw csum failure.\n",
1593 dev ? dev->name : "<unknown>");
1597 EXPORT_SYMBOL(netdev_rx_csum_fault);
1600 /* Actually, we should eliminate this check as soon as we know, that:
1601 * 1. IOMMU is present and allows to map all the memory.
1602 * 2. No high memory really exists on this machine.
1605 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1607 #ifdef CONFIG_HIGHMEM
1610 if (dev->features & NETIF_F_HIGHDMA)
1613 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1614 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1622 void (*destructor)(struct sk_buff *skb);
1625 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1627 static void dev_gso_skb_destructor(struct sk_buff *skb)
1629 struct dev_gso_cb *cb;
1632 struct sk_buff *nskb = skb->next;
1634 skb->next = nskb->next;
1637 } while (skb->next);
1639 cb = DEV_GSO_CB(skb);
1641 cb->destructor(skb);
1645 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1646 * @skb: buffer to segment
1648 * This function segments the given skb and stores the list of segments
1651 static int dev_gso_segment(struct sk_buff *skb)
1653 struct net_device *dev = skb->dev;
1654 struct sk_buff *segs;
1655 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1658 segs = skb_gso_segment(skb, features);
1660 /* Verifying header integrity only. */
1665 return PTR_ERR(segs);
1668 DEV_GSO_CB(skb)->destructor = skb->destructor;
1669 skb->destructor = dev_gso_skb_destructor;
1674 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1675 struct netdev_queue *txq)
1677 const struct net_device_ops *ops = dev->netdev_ops;
1680 if (likely(!skb->next)) {
1681 if (!list_empty(&ptype_all))
1682 dev_queue_xmit_nit(skb, dev);
1684 if (netif_needs_gso(dev, skb)) {
1685 if (unlikely(dev_gso_segment(skb)))
1691 rc = ops->ndo_start_xmit(skb, dev);
1693 * TODO: if skb_orphan() was called by
1694 * dev->hard_start_xmit() (for example, the unmodified
1695 * igb driver does that; bnx2 doesn't), then
1696 * skb_tx_software_timestamp() will be unable to send
1697 * back the time stamp.
1699 * How can this be prevented? Always create another
1700 * reference to the socket before calling
1701 * dev->hard_start_xmit()? Prevent that skb_orphan()
1702 * does anything in dev->hard_start_xmit() by clearing
1703 * the skb destructor before the call and restoring it
1704 * afterwards, then doing the skb_orphan() ourselves?
1711 struct sk_buff *nskb = skb->next;
1713 skb->next = nskb->next;
1715 rc = ops->ndo_start_xmit(nskb, dev);
1717 nskb->next = skb->next;
1721 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1722 return NETDEV_TX_BUSY;
1723 } while (skb->next);
1725 skb->destructor = DEV_GSO_CB(skb)->destructor;
1732 static u32 skb_tx_hashrnd;
1734 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1738 if (skb_rx_queue_recorded(skb)) {
1739 hash = skb_get_rx_queue(skb);
1740 } else if (skb->sk && skb->sk->sk_hash) {
1741 hash = skb->sk->sk_hash;
1743 hash = skb->protocol;
1745 hash = jhash_1word(hash, skb_tx_hashrnd);
1747 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1749 EXPORT_SYMBOL(skb_tx_hash);
1751 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1752 struct sk_buff *skb)
1754 const struct net_device_ops *ops = dev->netdev_ops;
1755 u16 queue_index = 0;
1757 if (ops->ndo_select_queue)
1758 queue_index = ops->ndo_select_queue(dev, skb);
1759 else if (dev->real_num_tx_queues > 1)
1760 queue_index = skb_tx_hash(dev, skb);
1762 skb_set_queue_mapping(skb, queue_index);
1763 return netdev_get_tx_queue(dev, queue_index);
1767 * dev_queue_xmit - transmit a buffer
1768 * @skb: buffer to transmit
1770 * Queue a buffer for transmission to a network device. The caller must
1771 * have set the device and priority and built the buffer before calling
1772 * this function. The function can be called from an interrupt.
1774 * A negative errno code is returned on a failure. A success does not
1775 * guarantee the frame will be transmitted as it may be dropped due
1776 * to congestion or traffic shaping.
1778 * -----------------------------------------------------------------------------------
1779 * I notice this method can also return errors from the queue disciplines,
1780 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1783 * Regardless of the return value, the skb is consumed, so it is currently
1784 * difficult to retry a send to this method. (You can bump the ref count
1785 * before sending to hold a reference for retry if you are careful.)
1787 * When calling this method, interrupts MUST be enabled. This is because
1788 * the BH enable code must have IRQs enabled so that it will not deadlock.
1791 int dev_queue_xmit(struct sk_buff *skb)
1793 struct net_device *dev = skb->dev;
1794 struct netdev_queue *txq;
1798 /* GSO will handle the following emulations directly. */
1799 if (netif_needs_gso(dev, skb))
1802 if (skb_shinfo(skb)->frag_list &&
1803 !(dev->features & NETIF_F_FRAGLIST) &&
1804 __skb_linearize(skb))
1807 /* Fragmented skb is linearized if device does not support SG,
1808 * or if at least one of fragments is in highmem and device
1809 * does not support DMA from it.
1811 if (skb_shinfo(skb)->nr_frags &&
1812 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1813 __skb_linearize(skb))
1816 /* If packet is not checksummed and device does not support
1817 * checksumming for this protocol, complete checksumming here.
1819 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1820 skb_set_transport_header(skb, skb->csum_start -
1822 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1827 /* Disable soft irqs for various locks below. Also
1828 * stops preemption for RCU.
1832 txq = dev_pick_tx(dev, skb);
1833 q = rcu_dereference(txq->qdisc);
1835 #ifdef CONFIG_NET_CLS_ACT
1836 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1839 spinlock_t *root_lock = qdisc_lock(q);
1841 spin_lock(root_lock);
1843 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1847 rc = qdisc_enqueue_root(skb, q);
1850 spin_unlock(root_lock);
1855 /* The device has no queue. Common case for software devices:
1856 loopback, all the sorts of tunnels...
1858 Really, it is unlikely that netif_tx_lock protection is necessary
1859 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1861 However, it is possible, that they rely on protection
1864 Check this and shot the lock. It is not prone from deadlocks.
1865 Either shot noqueue qdisc, it is even simpler 8)
1867 if (dev->flags & IFF_UP) {
1868 int cpu = smp_processor_id(); /* ok because BHs are off */
1870 if (txq->xmit_lock_owner != cpu) {
1872 HARD_TX_LOCK(dev, txq, cpu);
1874 if (!netif_tx_queue_stopped(txq)) {
1876 if (!dev_hard_start_xmit(skb, dev, txq)) {
1877 HARD_TX_UNLOCK(dev, txq);
1881 HARD_TX_UNLOCK(dev, txq);
1882 if (net_ratelimit())
1883 printk(KERN_CRIT "Virtual device %s asks to "
1884 "queue packet!\n", dev->name);
1886 /* Recursion is detected! It is possible,
1888 if (net_ratelimit())
1889 printk(KERN_CRIT "Dead loop on virtual device "
1890 "%s, fix it urgently!\n", dev->name);
1895 rcu_read_unlock_bh();
1901 rcu_read_unlock_bh();
1906 /*=======================================================================
1908 =======================================================================*/
1910 int netdev_max_backlog __read_mostly = 1000;
1911 int netdev_budget __read_mostly = 300;
1912 int weight_p __read_mostly = 64; /* old backlog weight */
1914 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1918 * netif_rx - post buffer to the network code
1919 * @skb: buffer to post
1921 * This function receives a packet from a device driver and queues it for
1922 * the upper (protocol) levels to process. It always succeeds. The buffer
1923 * may be dropped during processing for congestion control or by the
1927 * NET_RX_SUCCESS (no congestion)
1928 * NET_RX_DROP (packet was dropped)
1932 int netif_rx(struct sk_buff *skb)
1934 struct softnet_data *queue;
1935 unsigned long flags;
1937 /* if netpoll wants it, pretend we never saw it */
1938 if (netpoll_rx(skb))
1941 if (!skb->tstamp.tv64)
1945 * The code is rearranged so that the path is the most
1946 * short when CPU is congested, but is still operating.
1948 local_irq_save(flags);
1949 queue = &__get_cpu_var(softnet_data);
1951 __get_cpu_var(netdev_rx_stat).total++;
1952 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1953 if (queue->input_pkt_queue.qlen) {
1955 __skb_queue_tail(&queue->input_pkt_queue, skb);
1956 local_irq_restore(flags);
1957 return NET_RX_SUCCESS;
1960 napi_schedule(&queue->backlog);
1964 __get_cpu_var(netdev_rx_stat).dropped++;
1965 local_irq_restore(flags);
1971 int netif_rx_ni(struct sk_buff *skb)
1976 err = netif_rx(skb);
1977 if (local_softirq_pending())
1984 EXPORT_SYMBOL(netif_rx_ni);
1986 static void net_tx_action(struct softirq_action *h)
1988 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1990 if (sd->completion_queue) {
1991 struct sk_buff *clist;
1993 local_irq_disable();
1994 clist = sd->completion_queue;
1995 sd->completion_queue = NULL;
1999 struct sk_buff *skb = clist;
2000 clist = clist->next;
2002 WARN_ON(atomic_read(&skb->users));
2007 if (sd->output_queue) {
2010 local_irq_disable();
2011 head = sd->output_queue;
2012 sd->output_queue = NULL;
2016 struct Qdisc *q = head;
2017 spinlock_t *root_lock;
2019 head = head->next_sched;
2021 root_lock = qdisc_lock(q);
2022 if (spin_trylock(root_lock)) {
2023 smp_mb__before_clear_bit();
2024 clear_bit(__QDISC_STATE_SCHED,
2027 spin_unlock(root_lock);
2029 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2031 __netif_reschedule(q);
2033 smp_mb__before_clear_bit();
2034 clear_bit(__QDISC_STATE_SCHED,
2042 static inline int deliver_skb(struct sk_buff *skb,
2043 struct packet_type *pt_prev,
2044 struct net_device *orig_dev)
2046 atomic_inc(&skb->users);
2047 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2050 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2051 /* These hooks defined here for ATM */
2053 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2054 unsigned char *addr);
2055 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
2058 * If bridge module is loaded call bridging hook.
2059 * returns NULL if packet was consumed.
2061 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2062 struct sk_buff *skb) __read_mostly;
2063 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2064 struct packet_type **pt_prev, int *ret,
2065 struct net_device *orig_dev)
2067 struct net_bridge_port *port;
2069 if (skb->pkt_type == PACKET_LOOPBACK ||
2070 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2074 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2078 return br_handle_frame_hook(port, skb);
2081 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2084 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2085 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2086 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2088 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2089 struct packet_type **pt_prev,
2091 struct net_device *orig_dev)
2093 if (skb->dev->macvlan_port == NULL)
2097 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2100 return macvlan_handle_frame_hook(skb);
2103 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2106 #ifdef CONFIG_NET_CLS_ACT
2107 /* TODO: Maybe we should just force sch_ingress to be compiled in
2108 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2109 * a compare and 2 stores extra right now if we dont have it on
2110 * but have CONFIG_NET_CLS_ACT
2111 * NOTE: This doesnt stop any functionality; if you dont have
2112 * the ingress scheduler, you just cant add policies on ingress.
2115 static int ing_filter(struct sk_buff *skb)
2117 struct net_device *dev = skb->dev;
2118 u32 ttl = G_TC_RTTL(skb->tc_verd);
2119 struct netdev_queue *rxq;
2120 int result = TC_ACT_OK;
2123 if (MAX_RED_LOOP < ttl++) {
2125 "Redir loop detected Dropping packet (%d->%d)\n",
2126 skb->iif, dev->ifindex);
2130 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2131 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2133 rxq = &dev->rx_queue;
2136 if (q != &noop_qdisc) {
2137 spin_lock(qdisc_lock(q));
2138 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2139 result = qdisc_enqueue_root(skb, q);
2140 spin_unlock(qdisc_lock(q));
2146 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2147 struct packet_type **pt_prev,
2148 int *ret, struct net_device *orig_dev)
2150 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2154 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2157 /* Huh? Why does turning on AF_PACKET affect this? */
2158 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2161 switch (ing_filter(skb)) {
2175 * netif_nit_deliver - deliver received packets to network taps
2178 * This function is used to deliver incoming packets to network
2179 * taps. It should be used when the normal netif_receive_skb path
2180 * is bypassed, for example because of VLAN acceleration.
2182 void netif_nit_deliver(struct sk_buff *skb)
2184 struct packet_type *ptype;
2186 if (list_empty(&ptype_all))
2189 skb_reset_network_header(skb);
2190 skb_reset_transport_header(skb);
2191 skb->mac_len = skb->network_header - skb->mac_header;
2194 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2195 if (!ptype->dev || ptype->dev == skb->dev)
2196 deliver_skb(skb, ptype, skb->dev);
2202 * netif_receive_skb - process receive buffer from network
2203 * @skb: buffer to process
2205 * netif_receive_skb() is the main receive data processing function.
2206 * It always succeeds. The buffer may be dropped during processing
2207 * for congestion control or by the protocol layers.
2209 * This function may only be called from softirq context and interrupts
2210 * should be enabled.
2212 * Return values (usually ignored):
2213 * NET_RX_SUCCESS: no congestion
2214 * NET_RX_DROP: packet was dropped
2216 int netif_receive_skb(struct sk_buff *skb)
2218 struct packet_type *ptype, *pt_prev;
2219 struct net_device *orig_dev;
2220 struct net_device *null_or_orig;
2221 int ret = NET_RX_DROP;
2224 if (skb->vlan_tci && vlan_hwaccel_do_receive(skb))
2225 return NET_RX_SUCCESS;
2227 /* if we've gotten here through NAPI, check netpoll */
2228 if (netpoll_receive_skb(skb))
2231 if (!skb->tstamp.tv64)
2235 skb->iif = skb->dev->ifindex;
2237 null_or_orig = NULL;
2238 orig_dev = skb->dev;
2239 if (orig_dev->master) {
2240 if (skb_bond_should_drop(skb))
2241 null_or_orig = orig_dev; /* deliver only exact match */
2243 skb->dev = orig_dev->master;
2246 __get_cpu_var(netdev_rx_stat).total++;
2248 skb_reset_network_header(skb);
2249 skb_reset_transport_header(skb);
2250 skb->mac_len = skb->network_header - skb->mac_header;
2256 #ifdef CONFIG_NET_CLS_ACT
2257 if (skb->tc_verd & TC_NCLS) {
2258 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2263 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2264 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2265 ptype->dev == orig_dev) {
2267 ret = deliver_skb(skb, pt_prev, orig_dev);
2272 #ifdef CONFIG_NET_CLS_ACT
2273 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2279 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2282 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2288 type = skb->protocol;
2289 list_for_each_entry_rcu(ptype,
2290 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2291 if (ptype->type == type &&
2292 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2293 ptype->dev == orig_dev)) {
2295 ret = deliver_skb(skb, pt_prev, orig_dev);
2301 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2304 /* Jamal, now you will not able to escape explaining
2305 * me how you were going to use this. :-)
2315 /* Network device is going away, flush any packets still pending */
2316 static void flush_backlog(void *arg)
2318 struct net_device *dev = arg;
2319 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2320 struct sk_buff *skb, *tmp;
2322 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2323 if (skb->dev == dev) {
2324 __skb_unlink(skb, &queue->input_pkt_queue);
2329 static int napi_gro_complete(struct sk_buff *skb)
2331 struct packet_type *ptype;
2332 __be16 type = skb->protocol;
2333 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2336 if (NAPI_GRO_CB(skb)->count == 1) {
2337 skb_shinfo(skb)->gso_size = 0;
2342 list_for_each_entry_rcu(ptype, head, list) {
2343 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2346 err = ptype->gro_complete(skb);
2352 WARN_ON(&ptype->list == head);
2354 return NET_RX_SUCCESS;
2358 return netif_receive_skb(skb);
2361 void napi_gro_flush(struct napi_struct *napi)
2363 struct sk_buff *skb, *next;
2365 for (skb = napi->gro_list; skb; skb = next) {
2368 napi_gro_complete(skb);
2371 napi->gro_count = 0;
2372 napi->gro_list = NULL;
2374 EXPORT_SYMBOL(napi_gro_flush);
2376 void *skb_gro_header(struct sk_buff *skb, unsigned int hlen)
2378 unsigned int offset = skb_gro_offset(skb);
2381 if (unlikely(skb_headlen(skb) ||
2382 skb_shinfo(skb)->frags[0].size < hlen ||
2383 PageHighMem(skb_shinfo(skb)->frags[0].page)))
2384 return pskb_may_pull(skb, hlen) ? skb->data + offset : NULL;
2386 return page_address(skb_shinfo(skb)->frags[0].page) +
2387 skb_shinfo(skb)->frags[0].page_offset + offset;
2389 EXPORT_SYMBOL(skb_gro_header);
2391 int dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2393 struct sk_buff **pp = NULL;
2394 struct packet_type *ptype;
2395 __be16 type = skb->protocol;
2396 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2401 if (!(skb->dev->features & NETIF_F_GRO))
2404 if (skb_is_gso(skb) || skb_shinfo(skb)->frag_list)
2408 list_for_each_entry_rcu(ptype, head, list) {
2409 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2412 skb_set_network_header(skb, skb_gro_offset(skb));
2413 mac_len = skb->network_header - skb->mac_header;
2414 skb->mac_len = mac_len;
2415 NAPI_GRO_CB(skb)->same_flow = 0;
2416 NAPI_GRO_CB(skb)->flush = 0;
2417 NAPI_GRO_CB(skb)->free = 0;
2419 pp = ptype->gro_receive(&napi->gro_list, skb);
2424 if (&ptype->list == head)
2427 same_flow = NAPI_GRO_CB(skb)->same_flow;
2428 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2431 struct sk_buff *nskb = *pp;
2435 napi_gro_complete(nskb);
2442 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2446 NAPI_GRO_CB(skb)->count = 1;
2447 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2448 skb->next = napi->gro_list;
2449 napi->gro_list = skb;
2453 if (unlikely(!pskb_may_pull(skb, skb_gro_offset(skb)))) {
2454 if (napi->gro_list == skb)
2455 napi->gro_list = skb->next;
2466 EXPORT_SYMBOL(dev_gro_receive);
2468 static int __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2472 if (netpoll_rx_on(skb))
2475 for (p = napi->gro_list; p; p = p->next) {
2476 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2477 && !compare_ether_header(skb_mac_header(p),
2478 skb_gro_mac_header(skb));
2479 NAPI_GRO_CB(p)->flush = 0;
2482 return dev_gro_receive(napi, skb);
2485 int napi_skb_finish(int ret, struct sk_buff *skb)
2487 int err = NET_RX_SUCCESS;
2491 return netif_receive_skb(skb);
2497 case GRO_MERGED_FREE:
2504 EXPORT_SYMBOL(napi_skb_finish);
2506 int napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2508 skb_gro_reset_offset(skb);
2510 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2512 EXPORT_SYMBOL(napi_gro_receive);
2514 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2516 __skb_pull(skb, skb_headlen(skb));
2517 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2521 EXPORT_SYMBOL(napi_reuse_skb);
2523 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2525 struct net_device *dev = napi->dev;
2526 struct sk_buff *skb = napi->skb;
2529 skb = netdev_alloc_skb(dev, GRO_MAX_HEAD + NET_IP_ALIGN);
2533 skb_reserve(skb, NET_IP_ALIGN);
2541 EXPORT_SYMBOL(napi_get_frags);
2543 int napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, int ret)
2545 int err = NET_RX_SUCCESS;
2550 skb->protocol = eth_type_trans(skb, napi->dev);
2552 if (ret == GRO_NORMAL)
2553 return netif_receive_skb(skb);
2555 skb_gro_pull(skb, -ETH_HLEN);
2562 case GRO_MERGED_FREE:
2563 napi_reuse_skb(napi, skb);
2569 EXPORT_SYMBOL(napi_frags_finish);
2571 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2573 struct sk_buff *skb = napi->skb;
2578 skb_reset_mac_header(skb);
2579 skb_gro_reset_offset(skb);
2581 eth = skb_gro_header(skb, sizeof(*eth));
2583 napi_reuse_skb(napi, skb);
2588 skb_gro_pull(skb, sizeof(*eth));
2591 * This works because the only protocols we care about don't require
2592 * special handling. We'll fix it up properly at the end.
2594 skb->protocol = eth->h_proto;
2599 EXPORT_SYMBOL(napi_frags_skb);
2601 int napi_gro_frags(struct napi_struct *napi)
2603 struct sk_buff *skb = napi_frags_skb(napi);
2608 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2610 EXPORT_SYMBOL(napi_gro_frags);
2612 static int process_backlog(struct napi_struct *napi, int quota)
2615 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2616 unsigned long start_time = jiffies;
2618 napi->weight = weight_p;
2620 struct sk_buff *skb;
2622 local_irq_disable();
2623 skb = __skb_dequeue(&queue->input_pkt_queue);
2625 __napi_complete(napi);
2631 netif_receive_skb(skb);
2632 } while (++work < quota && jiffies == start_time);
2638 * __napi_schedule - schedule for receive
2639 * @n: entry to schedule
2641 * The entry's receive function will be scheduled to run
2643 void __napi_schedule(struct napi_struct *n)
2645 unsigned long flags;
2647 local_irq_save(flags);
2648 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2649 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2650 local_irq_restore(flags);
2652 EXPORT_SYMBOL(__napi_schedule);
2654 void __napi_complete(struct napi_struct *n)
2656 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2657 BUG_ON(n->gro_list);
2659 list_del(&n->poll_list);
2660 smp_mb__before_clear_bit();
2661 clear_bit(NAPI_STATE_SCHED, &n->state);
2663 EXPORT_SYMBOL(__napi_complete);
2665 void napi_complete(struct napi_struct *n)
2667 unsigned long flags;
2670 * don't let napi dequeue from the cpu poll list
2671 * just in case its running on a different cpu
2673 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2677 local_irq_save(flags);
2679 local_irq_restore(flags);
2681 EXPORT_SYMBOL(napi_complete);
2683 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2684 int (*poll)(struct napi_struct *, int), int weight)
2686 INIT_LIST_HEAD(&napi->poll_list);
2687 napi->gro_count = 0;
2688 napi->gro_list = NULL;
2691 napi->weight = weight;
2692 list_add(&napi->dev_list, &dev->napi_list);
2694 #ifdef CONFIG_NETPOLL
2695 spin_lock_init(&napi->poll_lock);
2696 napi->poll_owner = -1;
2698 set_bit(NAPI_STATE_SCHED, &napi->state);
2700 EXPORT_SYMBOL(netif_napi_add);
2702 void netif_napi_del(struct napi_struct *napi)
2704 struct sk_buff *skb, *next;
2706 list_del_init(&napi->dev_list);
2707 napi_free_frags(napi);
2709 for (skb = napi->gro_list; skb; skb = next) {
2715 napi->gro_list = NULL;
2716 napi->gro_count = 0;
2718 EXPORT_SYMBOL(netif_napi_del);
2721 static void net_rx_action(struct softirq_action *h)
2723 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2724 unsigned long time_limit = jiffies + 2;
2725 int budget = netdev_budget;
2728 local_irq_disable();
2730 while (!list_empty(list)) {
2731 struct napi_struct *n;
2734 /* If softirq window is exhuasted then punt.
2735 * Allow this to run for 2 jiffies since which will allow
2736 * an average latency of 1.5/HZ.
2738 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2743 /* Even though interrupts have been re-enabled, this
2744 * access is safe because interrupts can only add new
2745 * entries to the tail of this list, and only ->poll()
2746 * calls can remove this head entry from the list.
2748 n = list_entry(list->next, struct napi_struct, poll_list);
2750 have = netpoll_poll_lock(n);
2754 /* This NAPI_STATE_SCHED test is for avoiding a race
2755 * with netpoll's poll_napi(). Only the entity which
2756 * obtains the lock and sees NAPI_STATE_SCHED set will
2757 * actually make the ->poll() call. Therefore we avoid
2758 * accidently calling ->poll() when NAPI is not scheduled.
2761 if (test_bit(NAPI_STATE_SCHED, &n->state))
2762 work = n->poll(n, weight);
2764 WARN_ON_ONCE(work > weight);
2768 local_irq_disable();
2770 /* Drivers must not modify the NAPI state if they
2771 * consume the entire weight. In such cases this code
2772 * still "owns" the NAPI instance and therefore can
2773 * move the instance around on the list at-will.
2775 if (unlikely(work == weight)) {
2776 if (unlikely(napi_disable_pending(n)))
2779 list_move_tail(&n->poll_list, list);
2782 netpoll_poll_unlock(have);
2787 #ifdef CONFIG_NET_DMA
2789 * There may not be any more sk_buffs coming right now, so push
2790 * any pending DMA copies to hardware
2792 dma_issue_pending_all();
2798 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2799 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2803 static gifconf_func_t * gifconf_list [NPROTO];
2806 * register_gifconf - register a SIOCGIF handler
2807 * @family: Address family
2808 * @gifconf: Function handler
2810 * Register protocol dependent address dumping routines. The handler
2811 * that is passed must not be freed or reused until it has been replaced
2812 * by another handler.
2814 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2816 if (family >= NPROTO)
2818 gifconf_list[family] = gifconf;
2824 * Map an interface index to its name (SIOCGIFNAME)
2828 * We need this ioctl for efficient implementation of the
2829 * if_indextoname() function required by the IPv6 API. Without
2830 * it, we would have to search all the interfaces to find a
2834 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2836 struct net_device *dev;
2840 * Fetch the caller's info block.
2843 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2846 read_lock(&dev_base_lock);
2847 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2849 read_unlock(&dev_base_lock);
2853 strcpy(ifr.ifr_name, dev->name);
2854 read_unlock(&dev_base_lock);
2856 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2862 * Perform a SIOCGIFCONF call. This structure will change
2863 * size eventually, and there is nothing I can do about it.
2864 * Thus we will need a 'compatibility mode'.
2867 static int dev_ifconf(struct net *net, char __user *arg)
2870 struct net_device *dev;
2877 * Fetch the caller's info block.
2880 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2887 * Loop over the interfaces, and write an info block for each.
2891 for_each_netdev(net, dev) {
2892 for (i = 0; i < NPROTO; i++) {
2893 if (gifconf_list[i]) {
2896 done = gifconf_list[i](dev, NULL, 0);
2898 done = gifconf_list[i](dev, pos + total,
2908 * All done. Write the updated control block back to the caller.
2910 ifc.ifc_len = total;
2913 * Both BSD and Solaris return 0 here, so we do too.
2915 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2918 #ifdef CONFIG_PROC_FS
2920 * This is invoked by the /proc filesystem handler to display a device
2923 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2924 __acquires(dev_base_lock)
2926 struct net *net = seq_file_net(seq);
2928 struct net_device *dev;
2930 read_lock(&dev_base_lock);
2932 return SEQ_START_TOKEN;
2935 for_each_netdev(net, dev)
2942 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2944 struct net *net = seq_file_net(seq);
2946 return v == SEQ_START_TOKEN ?
2947 first_net_device(net) : next_net_device((struct net_device *)v);
2950 void dev_seq_stop(struct seq_file *seq, void *v)
2951 __releases(dev_base_lock)
2953 read_unlock(&dev_base_lock);
2956 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2958 const struct net_device_stats *stats = dev_get_stats(dev);
2960 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2961 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2962 dev->name, stats->rx_bytes, stats->rx_packets,
2964 stats->rx_dropped + stats->rx_missed_errors,
2965 stats->rx_fifo_errors,
2966 stats->rx_length_errors + stats->rx_over_errors +
2967 stats->rx_crc_errors + stats->rx_frame_errors,
2968 stats->rx_compressed, stats->multicast,
2969 stats->tx_bytes, stats->tx_packets,
2970 stats->tx_errors, stats->tx_dropped,
2971 stats->tx_fifo_errors, stats->collisions,
2972 stats->tx_carrier_errors +
2973 stats->tx_aborted_errors +
2974 stats->tx_window_errors +
2975 stats->tx_heartbeat_errors,
2976 stats->tx_compressed);
2980 * Called from the PROCfs module. This now uses the new arbitrary sized
2981 * /proc/net interface to create /proc/net/dev
2983 static int dev_seq_show(struct seq_file *seq, void *v)
2985 if (v == SEQ_START_TOKEN)
2986 seq_puts(seq, "Inter-| Receive "
2988 " face |bytes packets errs drop fifo frame "
2989 "compressed multicast|bytes packets errs "
2990 "drop fifo colls carrier compressed\n");
2992 dev_seq_printf_stats(seq, v);
2996 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2998 struct netif_rx_stats *rc = NULL;
3000 while (*pos < nr_cpu_ids)
3001 if (cpu_online(*pos)) {
3002 rc = &per_cpu(netdev_rx_stat, *pos);
3009 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3011 return softnet_get_online(pos);
3014 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3017 return softnet_get_online(pos);
3020 static void softnet_seq_stop(struct seq_file *seq, void *v)
3024 static int softnet_seq_show(struct seq_file *seq, void *v)
3026 struct netif_rx_stats *s = v;
3028 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3029 s->total, s->dropped, s->time_squeeze, 0,
3030 0, 0, 0, 0, /* was fastroute */
3035 static const struct seq_operations dev_seq_ops = {
3036 .start = dev_seq_start,
3037 .next = dev_seq_next,
3038 .stop = dev_seq_stop,
3039 .show = dev_seq_show,
3042 static int dev_seq_open(struct inode *inode, struct file *file)
3044 return seq_open_net(inode, file, &dev_seq_ops,
3045 sizeof(struct seq_net_private));
3048 static const struct file_operations dev_seq_fops = {
3049 .owner = THIS_MODULE,
3050 .open = dev_seq_open,
3052 .llseek = seq_lseek,
3053 .release = seq_release_net,
3056 static const struct seq_operations softnet_seq_ops = {
3057 .start = softnet_seq_start,
3058 .next = softnet_seq_next,
3059 .stop = softnet_seq_stop,
3060 .show = softnet_seq_show,
3063 static int softnet_seq_open(struct inode *inode, struct file *file)
3065 return seq_open(file, &softnet_seq_ops);
3068 static const struct file_operations softnet_seq_fops = {
3069 .owner = THIS_MODULE,
3070 .open = softnet_seq_open,
3072 .llseek = seq_lseek,
3073 .release = seq_release,
3076 static void *ptype_get_idx(loff_t pos)
3078 struct packet_type *pt = NULL;
3082 list_for_each_entry_rcu(pt, &ptype_all, list) {
3088 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3089 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3098 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3102 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3105 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3107 struct packet_type *pt;
3108 struct list_head *nxt;
3112 if (v == SEQ_START_TOKEN)
3113 return ptype_get_idx(0);
3116 nxt = pt->list.next;
3117 if (pt->type == htons(ETH_P_ALL)) {
3118 if (nxt != &ptype_all)
3121 nxt = ptype_base[0].next;
3123 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3125 while (nxt == &ptype_base[hash]) {
3126 if (++hash >= PTYPE_HASH_SIZE)
3128 nxt = ptype_base[hash].next;
3131 return list_entry(nxt, struct packet_type, list);
3134 static void ptype_seq_stop(struct seq_file *seq, void *v)
3140 static int ptype_seq_show(struct seq_file *seq, void *v)
3142 struct packet_type *pt = v;
3144 if (v == SEQ_START_TOKEN)
3145 seq_puts(seq, "Type Device Function\n");
3146 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3147 if (pt->type == htons(ETH_P_ALL))
3148 seq_puts(seq, "ALL ");
3150 seq_printf(seq, "%04x", ntohs(pt->type));
3152 seq_printf(seq, " %-8s %pF\n",
3153 pt->dev ? pt->dev->name : "", pt->func);
3159 static const struct seq_operations ptype_seq_ops = {
3160 .start = ptype_seq_start,
3161 .next = ptype_seq_next,
3162 .stop = ptype_seq_stop,
3163 .show = ptype_seq_show,
3166 static int ptype_seq_open(struct inode *inode, struct file *file)
3168 return seq_open_net(inode, file, &ptype_seq_ops,
3169 sizeof(struct seq_net_private));
3172 static const struct file_operations ptype_seq_fops = {
3173 .owner = THIS_MODULE,
3174 .open = ptype_seq_open,
3176 .llseek = seq_lseek,
3177 .release = seq_release_net,
3181 static int __net_init dev_proc_net_init(struct net *net)
3185 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3187 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3189 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3192 if (wext_proc_init(net))
3198 proc_net_remove(net, "ptype");
3200 proc_net_remove(net, "softnet_stat");
3202 proc_net_remove(net, "dev");
3206 static void __net_exit dev_proc_net_exit(struct net *net)
3208 wext_proc_exit(net);
3210 proc_net_remove(net, "ptype");
3211 proc_net_remove(net, "softnet_stat");
3212 proc_net_remove(net, "dev");
3215 static struct pernet_operations __net_initdata dev_proc_ops = {
3216 .init = dev_proc_net_init,
3217 .exit = dev_proc_net_exit,
3220 static int __init dev_proc_init(void)
3222 return register_pernet_subsys(&dev_proc_ops);
3225 #define dev_proc_init() 0
3226 #endif /* CONFIG_PROC_FS */
3230 * netdev_set_master - set up master/slave pair
3231 * @slave: slave device
3232 * @master: new master device
3234 * Changes the master device of the slave. Pass %NULL to break the
3235 * bonding. The caller must hold the RTNL semaphore. On a failure
3236 * a negative errno code is returned. On success the reference counts
3237 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3238 * function returns zero.
3240 int netdev_set_master(struct net_device *slave, struct net_device *master)
3242 struct net_device *old = slave->master;
3252 slave->master = master;
3260 slave->flags |= IFF_SLAVE;
3262 slave->flags &= ~IFF_SLAVE;
3264 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3268 static void dev_change_rx_flags(struct net_device *dev, int flags)
3270 const struct net_device_ops *ops = dev->netdev_ops;
3272 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3273 ops->ndo_change_rx_flags(dev, flags);
3276 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3278 unsigned short old_flags = dev->flags;
3284 dev->flags |= IFF_PROMISC;
3285 dev->promiscuity += inc;
3286 if (dev->promiscuity == 0) {
3289 * If inc causes overflow, untouch promisc and return error.
3292 dev->flags &= ~IFF_PROMISC;
3294 dev->promiscuity -= inc;
3295 printk(KERN_WARNING "%s: promiscuity touches roof, "
3296 "set promiscuity failed, promiscuity feature "
3297 "of device might be broken.\n", dev->name);
3301 if (dev->flags != old_flags) {
3302 printk(KERN_INFO "device %s %s promiscuous mode\n",
3303 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3305 if (audit_enabled) {
3306 current_uid_gid(&uid, &gid);
3307 audit_log(current->audit_context, GFP_ATOMIC,
3308 AUDIT_ANOM_PROMISCUOUS,
3309 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3310 dev->name, (dev->flags & IFF_PROMISC),
3311 (old_flags & IFF_PROMISC),
3312 audit_get_loginuid(current),
3314 audit_get_sessionid(current));
3317 dev_change_rx_flags(dev, IFF_PROMISC);
3323 * dev_set_promiscuity - update promiscuity count on a device
3327 * Add or remove promiscuity from a device. While the count in the device
3328 * remains above zero the interface remains promiscuous. Once it hits zero
3329 * the device reverts back to normal filtering operation. A negative inc
3330 * value is used to drop promiscuity on the device.
3331 * Return 0 if successful or a negative errno code on error.
3333 int dev_set_promiscuity(struct net_device *dev, int inc)
3335 unsigned short old_flags = dev->flags;
3338 err = __dev_set_promiscuity(dev, inc);
3341 if (dev->flags != old_flags)
3342 dev_set_rx_mode(dev);
3347 * dev_set_allmulti - update allmulti count on a device
3351 * Add or remove reception of all multicast frames to a device. While the
3352 * count in the device remains above zero the interface remains listening
3353 * to all interfaces. Once it hits zero the device reverts back to normal
3354 * filtering operation. A negative @inc value is used to drop the counter
3355 * when releasing a resource needing all multicasts.
3356 * Return 0 if successful or a negative errno code on error.
3359 int dev_set_allmulti(struct net_device *dev, int inc)
3361 unsigned short old_flags = dev->flags;
3365 dev->flags |= IFF_ALLMULTI;
3366 dev->allmulti += inc;
3367 if (dev->allmulti == 0) {
3370 * If inc causes overflow, untouch allmulti and return error.
3373 dev->flags &= ~IFF_ALLMULTI;
3375 dev->allmulti -= inc;
3376 printk(KERN_WARNING "%s: allmulti touches roof, "
3377 "set allmulti failed, allmulti feature of "
3378 "device might be broken.\n", dev->name);
3382 if (dev->flags ^ old_flags) {
3383 dev_change_rx_flags(dev, IFF_ALLMULTI);
3384 dev_set_rx_mode(dev);
3390 * Upload unicast and multicast address lists to device and
3391 * configure RX filtering. When the device doesn't support unicast
3392 * filtering it is put in promiscuous mode while unicast addresses
3395 void __dev_set_rx_mode(struct net_device *dev)
3397 const struct net_device_ops *ops = dev->netdev_ops;
3399 /* dev_open will call this function so the list will stay sane. */
3400 if (!(dev->flags&IFF_UP))
3403 if (!netif_device_present(dev))
3406 if (ops->ndo_set_rx_mode)
3407 ops->ndo_set_rx_mode(dev);
3409 /* Unicast addresses changes may only happen under the rtnl,
3410 * therefore calling __dev_set_promiscuity here is safe.
3412 if (dev->uc_count > 0 && !dev->uc_promisc) {
3413 __dev_set_promiscuity(dev, 1);
3414 dev->uc_promisc = 1;
3415 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3416 __dev_set_promiscuity(dev, -1);
3417 dev->uc_promisc = 0;
3420 if (ops->ndo_set_multicast_list)
3421 ops->ndo_set_multicast_list(dev);
3425 void dev_set_rx_mode(struct net_device *dev)
3427 netif_addr_lock_bh(dev);
3428 __dev_set_rx_mode(dev);
3429 netif_addr_unlock_bh(dev);
3432 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3433 void *addr, int alen, int glbl)
3435 struct dev_addr_list *da;
3437 for (; (da = *list) != NULL; list = &da->next) {
3438 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3439 alen == da->da_addrlen) {
3441 int old_glbl = da->da_gusers;
3458 int __dev_addr_add(struct dev_addr_list **list, int *count,
3459 void *addr, int alen, int glbl)
3461 struct dev_addr_list *da;
3463 for (da = *list; da != NULL; da = da->next) {
3464 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3465 da->da_addrlen == alen) {
3467 int old_glbl = da->da_gusers;
3477 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3480 memcpy(da->da_addr, addr, alen);
3481 da->da_addrlen = alen;
3483 da->da_gusers = glbl ? 1 : 0;
3491 * dev_unicast_delete - Release secondary unicast address.
3493 * @addr: address to delete
3494 * @alen: length of @addr
3496 * Release reference to a secondary unicast address and remove it
3497 * from the device if the reference count drops to zero.
3499 * The caller must hold the rtnl_mutex.
3501 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3507 netif_addr_lock_bh(dev);
3508 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3510 __dev_set_rx_mode(dev);
3511 netif_addr_unlock_bh(dev);
3514 EXPORT_SYMBOL(dev_unicast_delete);
3517 * dev_unicast_add - add a secondary unicast address
3519 * @addr: address to add
3520 * @alen: length of @addr
3522 * Add a secondary unicast address to the device or increase
3523 * the reference count if it already exists.
3525 * The caller must hold the rtnl_mutex.
3527 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3533 netif_addr_lock_bh(dev);
3534 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3536 __dev_set_rx_mode(dev);
3537 netif_addr_unlock_bh(dev);
3540 EXPORT_SYMBOL(dev_unicast_add);
3542 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3543 struct dev_addr_list **from, int *from_count)
3545 struct dev_addr_list *da, *next;
3549 while (da != NULL) {
3551 if (!da->da_synced) {
3552 err = __dev_addr_add(to, to_count,
3553 da->da_addr, da->da_addrlen, 0);
3558 } else if (da->da_users == 1) {
3559 __dev_addr_delete(to, to_count,
3560 da->da_addr, da->da_addrlen, 0);
3561 __dev_addr_delete(from, from_count,
3562 da->da_addr, da->da_addrlen, 0);
3569 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3570 struct dev_addr_list **from, int *from_count)
3572 struct dev_addr_list *da, *next;
3575 while (da != NULL) {
3577 if (da->da_synced) {
3578 __dev_addr_delete(to, to_count,
3579 da->da_addr, da->da_addrlen, 0);
3581 __dev_addr_delete(from, from_count,
3582 da->da_addr, da->da_addrlen, 0);
3589 * dev_unicast_sync - Synchronize device's unicast list to another device
3590 * @to: destination device
3591 * @from: source device
3593 * Add newly added addresses to the destination device and release
3594 * addresses that have no users left. The source device must be
3595 * locked by netif_tx_lock_bh.
3597 * This function is intended to be called from the dev->set_rx_mode
3598 * function of layered software devices.
3600 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3604 netif_addr_lock_bh(to);
3605 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3606 &from->uc_list, &from->uc_count);
3608 __dev_set_rx_mode(to);
3609 netif_addr_unlock_bh(to);
3612 EXPORT_SYMBOL(dev_unicast_sync);
3615 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3616 * @to: destination device
3617 * @from: source device
3619 * Remove all addresses that were added to the destination device by
3620 * dev_unicast_sync(). This function is intended to be called from the
3621 * dev->stop function of layered software devices.
3623 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3625 netif_addr_lock_bh(from);
3626 netif_addr_lock(to);
3628 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3629 &from->uc_list, &from->uc_count);
3630 __dev_set_rx_mode(to);
3632 netif_addr_unlock(to);
3633 netif_addr_unlock_bh(from);
3635 EXPORT_SYMBOL(dev_unicast_unsync);
3637 static void __dev_addr_discard(struct dev_addr_list **list)
3639 struct dev_addr_list *tmp;
3641 while (*list != NULL) {
3644 if (tmp->da_users > tmp->da_gusers)
3645 printk("__dev_addr_discard: address leakage! "
3646 "da_users=%d\n", tmp->da_users);
3651 static void dev_addr_discard(struct net_device *dev)
3653 netif_addr_lock_bh(dev);
3655 __dev_addr_discard(&dev->uc_list);
3658 __dev_addr_discard(&dev->mc_list);
3661 netif_addr_unlock_bh(dev);
3665 * dev_get_flags - get flags reported to userspace
3668 * Get the combination of flag bits exported through APIs to userspace.
3670 unsigned dev_get_flags(const struct net_device *dev)
3674 flags = (dev->flags & ~(IFF_PROMISC |
3679 (dev->gflags & (IFF_PROMISC |
3682 if (netif_running(dev)) {
3683 if (netif_oper_up(dev))
3684 flags |= IFF_RUNNING;
3685 if (netif_carrier_ok(dev))
3686 flags |= IFF_LOWER_UP;
3687 if (netif_dormant(dev))
3688 flags |= IFF_DORMANT;
3695 * dev_change_flags - change device settings
3697 * @flags: device state flags
3699 * Change settings on device based state flags. The flags are
3700 * in the userspace exported format.
3702 int dev_change_flags(struct net_device *dev, unsigned flags)
3705 int old_flags = dev->flags;
3710 * Set the flags on our device.
3713 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3714 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3716 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3720 * Load in the correct multicast list now the flags have changed.
3723 if ((old_flags ^ flags) & IFF_MULTICAST)
3724 dev_change_rx_flags(dev, IFF_MULTICAST);
3726 dev_set_rx_mode(dev);
3729 * Have we downed the interface. We handle IFF_UP ourselves
3730 * according to user attempts to set it, rather than blindly
3735 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3736 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3739 dev_set_rx_mode(dev);
3742 if (dev->flags & IFF_UP &&
3743 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3745 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3747 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3748 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3749 dev->gflags ^= IFF_PROMISC;
3750 dev_set_promiscuity(dev, inc);
3753 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3754 is important. Some (broken) drivers set IFF_PROMISC, when
3755 IFF_ALLMULTI is requested not asking us and not reporting.
3757 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3758 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3759 dev->gflags ^= IFF_ALLMULTI;
3760 dev_set_allmulti(dev, inc);
3763 /* Exclude state transition flags, already notified */
3764 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3766 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3772 * dev_set_mtu - Change maximum transfer unit
3774 * @new_mtu: new transfer unit
3776 * Change the maximum transfer size of the network device.
3778 int dev_set_mtu(struct net_device *dev, int new_mtu)
3780 const struct net_device_ops *ops = dev->netdev_ops;
3783 if (new_mtu == dev->mtu)
3786 /* MTU must be positive. */
3790 if (!netif_device_present(dev))
3794 if (ops->ndo_change_mtu)
3795 err = ops->ndo_change_mtu(dev, new_mtu);
3799 if (!err && dev->flags & IFF_UP)
3800 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3805 * dev_set_mac_address - Change Media Access Control Address
3809 * Change the hardware (MAC) address of the device
3811 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3813 const struct net_device_ops *ops = dev->netdev_ops;
3816 if (!ops->ndo_set_mac_address)
3818 if (sa->sa_family != dev->type)
3820 if (!netif_device_present(dev))
3822 err = ops->ndo_set_mac_address(dev, sa);
3824 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3829 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3831 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3834 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3840 case SIOCGIFFLAGS: /* Get interface flags */
3841 ifr->ifr_flags = dev_get_flags(dev);
3844 case SIOCGIFMETRIC: /* Get the metric on the interface
3845 (currently unused) */
3846 ifr->ifr_metric = 0;
3849 case SIOCGIFMTU: /* Get the MTU of a device */
3850 ifr->ifr_mtu = dev->mtu;
3855 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3857 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3858 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3859 ifr->ifr_hwaddr.sa_family = dev->type;
3867 ifr->ifr_map.mem_start = dev->mem_start;
3868 ifr->ifr_map.mem_end = dev->mem_end;
3869 ifr->ifr_map.base_addr = dev->base_addr;
3870 ifr->ifr_map.irq = dev->irq;
3871 ifr->ifr_map.dma = dev->dma;
3872 ifr->ifr_map.port = dev->if_port;
3876 ifr->ifr_ifindex = dev->ifindex;
3880 ifr->ifr_qlen = dev->tx_queue_len;
3884 /* dev_ioctl() should ensure this case
3896 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3898 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3901 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3902 const struct net_device_ops *ops;
3907 ops = dev->netdev_ops;
3910 case SIOCSIFFLAGS: /* Set interface flags */
3911 return dev_change_flags(dev, ifr->ifr_flags);
3913 case SIOCSIFMETRIC: /* Set the metric on the interface
3914 (currently unused) */
3917 case SIOCSIFMTU: /* Set the MTU of a device */
3918 return dev_set_mtu(dev, ifr->ifr_mtu);
3921 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3923 case SIOCSIFHWBROADCAST:
3924 if (ifr->ifr_hwaddr.sa_family != dev->type)
3926 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3927 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3928 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3932 if (ops->ndo_set_config) {
3933 if (!netif_device_present(dev))
3935 return ops->ndo_set_config(dev, &ifr->ifr_map);
3940 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
3941 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3943 if (!netif_device_present(dev))
3945 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3949 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
3950 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3952 if (!netif_device_present(dev))
3954 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3958 if (ifr->ifr_qlen < 0)
3960 dev->tx_queue_len = ifr->ifr_qlen;
3964 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3965 return dev_change_name(dev, ifr->ifr_newname);
3968 * Unknown or private ioctl
3972 if ((cmd >= SIOCDEVPRIVATE &&
3973 cmd <= SIOCDEVPRIVATE + 15) ||
3974 cmd == SIOCBONDENSLAVE ||
3975 cmd == SIOCBONDRELEASE ||
3976 cmd == SIOCBONDSETHWADDR ||
3977 cmd == SIOCBONDSLAVEINFOQUERY ||
3978 cmd == SIOCBONDINFOQUERY ||
3979 cmd == SIOCBONDCHANGEACTIVE ||
3980 cmd == SIOCGMIIPHY ||
3981 cmd == SIOCGMIIREG ||
3982 cmd == SIOCSMIIREG ||
3983 cmd == SIOCBRADDIF ||
3984 cmd == SIOCBRDELIF ||
3985 cmd == SIOCSHWTSTAMP ||
3986 cmd == SIOCWANDEV) {
3988 if (ops->ndo_do_ioctl) {
3989 if (netif_device_present(dev))
3990 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4002 * This function handles all "interface"-type I/O control requests. The actual
4003 * 'doing' part of this is dev_ifsioc above.
4007 * dev_ioctl - network device ioctl
4008 * @net: the applicable net namespace
4009 * @cmd: command to issue
4010 * @arg: pointer to a struct ifreq in user space
4012 * Issue ioctl functions to devices. This is normally called by the
4013 * user space syscall interfaces but can sometimes be useful for
4014 * other purposes. The return value is the return from the syscall if
4015 * positive or a negative errno code on error.
4018 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4024 /* One special case: SIOCGIFCONF takes ifconf argument
4025 and requires shared lock, because it sleeps writing
4029 if (cmd == SIOCGIFCONF) {
4031 ret = dev_ifconf(net, (char __user *) arg);
4035 if (cmd == SIOCGIFNAME)
4036 return dev_ifname(net, (struct ifreq __user *)arg);
4038 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4041 ifr.ifr_name[IFNAMSIZ-1] = 0;
4043 colon = strchr(ifr.ifr_name, ':');
4048 * See which interface the caller is talking about.
4053 * These ioctl calls:
4054 * - can be done by all.
4055 * - atomic and do not require locking.
4066 dev_load(net, ifr.ifr_name);
4067 read_lock(&dev_base_lock);
4068 ret = dev_ifsioc_locked(net, &ifr, cmd);
4069 read_unlock(&dev_base_lock);
4073 if (copy_to_user(arg, &ifr,
4074 sizeof(struct ifreq)))
4080 dev_load(net, ifr.ifr_name);
4082 ret = dev_ethtool(net, &ifr);
4087 if (copy_to_user(arg, &ifr,
4088 sizeof(struct ifreq)))
4094 * These ioctl calls:
4095 * - require superuser power.
4096 * - require strict serialization.
4102 if (!capable(CAP_NET_ADMIN))
4104 dev_load(net, ifr.ifr_name);
4106 ret = dev_ifsioc(net, &ifr, cmd);
4111 if (copy_to_user(arg, &ifr,
4112 sizeof(struct ifreq)))
4118 * These ioctl calls:
4119 * - require superuser power.
4120 * - require strict serialization.
4121 * - do not return a value
4131 case SIOCSIFHWBROADCAST:
4134 case SIOCBONDENSLAVE:
4135 case SIOCBONDRELEASE:
4136 case SIOCBONDSETHWADDR:
4137 case SIOCBONDCHANGEACTIVE:
4141 if (!capable(CAP_NET_ADMIN))
4144 case SIOCBONDSLAVEINFOQUERY:
4145 case SIOCBONDINFOQUERY:
4146 dev_load(net, ifr.ifr_name);
4148 ret = dev_ifsioc(net, &ifr, cmd);
4153 /* Get the per device memory space. We can add this but
4154 * currently do not support it */
4156 /* Set the per device memory buffer space.
4157 * Not applicable in our case */
4162 * Unknown or private ioctl.
4165 if (cmd == SIOCWANDEV ||
4166 (cmd >= SIOCDEVPRIVATE &&
4167 cmd <= SIOCDEVPRIVATE + 15)) {
4168 dev_load(net, ifr.ifr_name);
4170 ret = dev_ifsioc(net, &ifr, cmd);
4172 if (!ret && copy_to_user(arg, &ifr,
4173 sizeof(struct ifreq)))
4177 /* Take care of Wireless Extensions */
4178 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4179 return wext_handle_ioctl(net, &ifr, cmd, arg);
4186 * dev_new_index - allocate an ifindex
4187 * @net: the applicable net namespace
4189 * Returns a suitable unique value for a new device interface
4190 * number. The caller must hold the rtnl semaphore or the
4191 * dev_base_lock to be sure it remains unique.
4193 static int dev_new_index(struct net *net)
4199 if (!__dev_get_by_index(net, ifindex))
4204 /* Delayed registration/unregisteration */
4205 static LIST_HEAD(net_todo_list);
4207 static void net_set_todo(struct net_device *dev)
4209 list_add_tail(&dev->todo_list, &net_todo_list);
4212 static void rollback_registered(struct net_device *dev)
4214 BUG_ON(dev_boot_phase);
4217 /* Some devices call without registering for initialization unwind. */
4218 if (dev->reg_state == NETREG_UNINITIALIZED) {
4219 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
4220 "was registered\n", dev->name, dev);
4226 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4228 /* If device is running, close it first. */
4231 /* And unlink it from device chain. */
4232 unlist_netdevice(dev);
4234 dev->reg_state = NETREG_UNREGISTERING;
4238 /* Shutdown queueing discipline. */
4242 /* Notify protocols, that we are about to destroy
4243 this device. They should clean all the things.
4245 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4248 * Flush the unicast and multicast chains
4250 dev_addr_discard(dev);
4252 if (dev->netdev_ops->ndo_uninit)
4253 dev->netdev_ops->ndo_uninit(dev);
4255 /* Notifier chain MUST detach us from master device. */
4256 WARN_ON(dev->master);
4258 /* Remove entries from kobject tree */
4259 netdev_unregister_kobject(dev);
4266 static void __netdev_init_queue_locks_one(struct net_device *dev,
4267 struct netdev_queue *dev_queue,
4270 spin_lock_init(&dev_queue->_xmit_lock);
4271 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4272 dev_queue->xmit_lock_owner = -1;
4275 static void netdev_init_queue_locks(struct net_device *dev)
4277 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4278 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4281 unsigned long netdev_fix_features(unsigned long features, const char *name)
4283 /* Fix illegal SG+CSUM combinations. */
4284 if ((features & NETIF_F_SG) &&
4285 !(features & NETIF_F_ALL_CSUM)) {
4287 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4288 "checksum feature.\n", name);
4289 features &= ~NETIF_F_SG;
4292 /* TSO requires that SG is present as well. */
4293 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4295 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4296 "SG feature.\n", name);
4297 features &= ~NETIF_F_TSO;
4300 if (features & NETIF_F_UFO) {
4301 if (!(features & NETIF_F_GEN_CSUM)) {
4303 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4304 "since no NETIF_F_HW_CSUM feature.\n",
4306 features &= ~NETIF_F_UFO;
4309 if (!(features & NETIF_F_SG)) {
4311 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4312 "since no NETIF_F_SG feature.\n", name);
4313 features &= ~NETIF_F_UFO;
4319 EXPORT_SYMBOL(netdev_fix_features);
4321 /* Some devices need to (re-)set their netdev_ops inside
4322 * ->init() or similar. If that happens, we have to setup
4323 * the compat pointers again.
4325 void netdev_resync_ops(struct net_device *dev)
4327 #ifdef CONFIG_COMPAT_NET_DEV_OPS
4328 const struct net_device_ops *ops = dev->netdev_ops;
4330 dev->init = ops->ndo_init;
4331 dev->uninit = ops->ndo_uninit;
4332 dev->open = ops->ndo_open;
4333 dev->change_rx_flags = ops->ndo_change_rx_flags;
4334 dev->set_rx_mode = ops->ndo_set_rx_mode;
4335 dev->set_multicast_list = ops->ndo_set_multicast_list;
4336 dev->set_mac_address = ops->ndo_set_mac_address;
4337 dev->validate_addr = ops->ndo_validate_addr;
4338 dev->do_ioctl = ops->ndo_do_ioctl;
4339 dev->set_config = ops->ndo_set_config;
4340 dev->change_mtu = ops->ndo_change_mtu;
4341 dev->neigh_setup = ops->ndo_neigh_setup;
4342 dev->tx_timeout = ops->ndo_tx_timeout;
4343 dev->get_stats = ops->ndo_get_stats;
4344 dev->vlan_rx_register = ops->ndo_vlan_rx_register;
4345 dev->vlan_rx_add_vid = ops->ndo_vlan_rx_add_vid;
4346 dev->vlan_rx_kill_vid = ops->ndo_vlan_rx_kill_vid;
4347 #ifdef CONFIG_NET_POLL_CONTROLLER
4348 dev->poll_controller = ops->ndo_poll_controller;
4352 EXPORT_SYMBOL(netdev_resync_ops);
4355 * register_netdevice - register a network device
4356 * @dev: device to register
4358 * Take a completed network device structure and add it to the kernel
4359 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4360 * chain. 0 is returned on success. A negative errno code is returned
4361 * on a failure to set up the device, or if the name is a duplicate.
4363 * Callers must hold the rtnl semaphore. You may want
4364 * register_netdev() instead of this.
4367 * The locking appears insufficient to guarantee two parallel registers
4368 * will not get the same name.
4371 int register_netdevice(struct net_device *dev)
4373 struct hlist_head *head;
4374 struct hlist_node *p;
4376 struct net *net = dev_net(dev);
4378 BUG_ON(dev_boot_phase);
4383 /* When net_device's are persistent, this will be fatal. */
4384 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4387 spin_lock_init(&dev->addr_list_lock);
4388 netdev_set_addr_lockdep_class(dev);
4389 netdev_init_queue_locks(dev);
4393 #ifdef CONFIG_COMPAT_NET_DEV_OPS
4394 /* Netdevice_ops API compatibility support.
4395 * This is temporary until all network devices are converted.
4397 if (dev->netdev_ops) {
4398 netdev_resync_ops(dev);
4400 char drivername[64];
4401 pr_info("%s (%s): not using net_device_ops yet\n",
4402 dev->name, netdev_drivername(dev, drivername, 64));
4404 /* This works only because net_device_ops and the
4405 compatibility structure are the same. */
4406 dev->netdev_ops = (void *) &(dev->init);
4410 /* Init, if this function is available */
4411 if (dev->netdev_ops->ndo_init) {
4412 ret = dev->netdev_ops->ndo_init(dev);
4420 if (!dev_valid_name(dev->name)) {
4425 dev->ifindex = dev_new_index(net);
4426 if (dev->iflink == -1)
4427 dev->iflink = dev->ifindex;
4429 /* Check for existence of name */
4430 head = dev_name_hash(net, dev->name);
4431 hlist_for_each(p, head) {
4432 struct net_device *d
4433 = hlist_entry(p, struct net_device, name_hlist);
4434 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4440 /* Fix illegal checksum combinations */
4441 if ((dev->features & NETIF_F_HW_CSUM) &&
4442 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4443 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4445 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4448 if ((dev->features & NETIF_F_NO_CSUM) &&
4449 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4450 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4452 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4455 dev->features = netdev_fix_features(dev->features, dev->name);
4457 /* Enable software GSO if SG is supported. */
4458 if (dev->features & NETIF_F_SG)
4459 dev->features |= NETIF_F_GSO;
4461 netdev_initialize_kobject(dev);
4462 ret = netdev_register_kobject(dev);
4465 dev->reg_state = NETREG_REGISTERED;
4468 * Default initial state at registry is that the
4469 * device is present.
4472 set_bit(__LINK_STATE_PRESENT, &dev->state);
4474 dev_init_scheduler(dev);
4476 list_netdevice(dev);
4478 /* Notify protocols, that a new device appeared. */
4479 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4480 ret = notifier_to_errno(ret);
4482 rollback_registered(dev);
4483 dev->reg_state = NETREG_UNREGISTERED;
4490 if (dev->netdev_ops->ndo_uninit)
4491 dev->netdev_ops->ndo_uninit(dev);
4496 * init_dummy_netdev - init a dummy network device for NAPI
4497 * @dev: device to init
4499 * This takes a network device structure and initialize the minimum
4500 * amount of fields so it can be used to schedule NAPI polls without
4501 * registering a full blown interface. This is to be used by drivers
4502 * that need to tie several hardware interfaces to a single NAPI
4503 * poll scheduler due to HW limitations.
4505 int init_dummy_netdev(struct net_device *dev)
4507 /* Clear everything. Note we don't initialize spinlocks
4508 * are they aren't supposed to be taken by any of the
4509 * NAPI code and this dummy netdev is supposed to be
4510 * only ever used for NAPI polls
4512 memset(dev, 0, sizeof(struct net_device));
4514 /* make sure we BUG if trying to hit standard
4515 * register/unregister code path
4517 dev->reg_state = NETREG_DUMMY;
4519 /* initialize the ref count */
4520 atomic_set(&dev->refcnt, 1);
4522 /* NAPI wants this */
4523 INIT_LIST_HEAD(&dev->napi_list);
4525 /* a dummy interface is started by default */
4526 set_bit(__LINK_STATE_PRESENT, &dev->state);
4527 set_bit(__LINK_STATE_START, &dev->state);
4531 EXPORT_SYMBOL_GPL(init_dummy_netdev);
4535 * register_netdev - register a network device
4536 * @dev: device to register
4538 * Take a completed network device structure and add it to the kernel
4539 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4540 * chain. 0 is returned on success. A negative errno code is returned
4541 * on a failure to set up the device, or if the name is a duplicate.
4543 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4544 * and expands the device name if you passed a format string to
4547 int register_netdev(struct net_device *dev)
4554 * If the name is a format string the caller wants us to do a
4557 if (strchr(dev->name, '%')) {
4558 err = dev_alloc_name(dev, dev->name);
4563 err = register_netdevice(dev);
4568 EXPORT_SYMBOL(register_netdev);
4571 * netdev_wait_allrefs - wait until all references are gone.
4573 * This is called when unregistering network devices.
4575 * Any protocol or device that holds a reference should register
4576 * for netdevice notification, and cleanup and put back the
4577 * reference if they receive an UNREGISTER event.
4578 * We can get stuck here if buggy protocols don't correctly
4581 static void netdev_wait_allrefs(struct net_device *dev)
4583 unsigned long rebroadcast_time, warning_time;
4585 rebroadcast_time = warning_time = jiffies;
4586 while (atomic_read(&dev->refcnt) != 0) {
4587 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4590 /* Rebroadcast unregister notification */
4591 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4593 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4595 /* We must not have linkwatch events
4596 * pending on unregister. If this
4597 * happens, we simply run the queue
4598 * unscheduled, resulting in a noop
4601 linkwatch_run_queue();
4606 rebroadcast_time = jiffies;
4611 if (time_after(jiffies, warning_time + 10 * HZ)) {
4612 printk(KERN_EMERG "unregister_netdevice: "
4613 "waiting for %s to become free. Usage "
4615 dev->name, atomic_read(&dev->refcnt));
4616 warning_time = jiffies;
4625 * register_netdevice(x1);
4626 * register_netdevice(x2);
4628 * unregister_netdevice(y1);
4629 * unregister_netdevice(y2);
4635 * We are invoked by rtnl_unlock().
4636 * This allows us to deal with problems:
4637 * 1) We can delete sysfs objects which invoke hotplug
4638 * without deadlocking with linkwatch via keventd.
4639 * 2) Since we run with the RTNL semaphore not held, we can sleep
4640 * safely in order to wait for the netdev refcnt to drop to zero.
4642 * We must not return until all unregister events added during
4643 * the interval the lock was held have been completed.
4645 void netdev_run_todo(void)
4647 struct list_head list;
4649 /* Snapshot list, allow later requests */
4650 list_replace_init(&net_todo_list, &list);
4654 while (!list_empty(&list)) {
4655 struct net_device *dev
4656 = list_entry(list.next, struct net_device, todo_list);
4657 list_del(&dev->todo_list);
4659 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4660 printk(KERN_ERR "network todo '%s' but state %d\n",
4661 dev->name, dev->reg_state);
4666 dev->reg_state = NETREG_UNREGISTERED;
4668 on_each_cpu(flush_backlog, dev, 1);
4670 netdev_wait_allrefs(dev);
4673 BUG_ON(atomic_read(&dev->refcnt));
4674 WARN_ON(dev->ip_ptr);
4675 WARN_ON(dev->ip6_ptr);
4676 WARN_ON(dev->dn_ptr);
4678 if (dev->destructor)
4679 dev->destructor(dev);
4681 /* Free network device */
4682 kobject_put(&dev->dev.kobj);
4687 * dev_get_stats - get network device statistics
4688 * @dev: device to get statistics from
4690 * Get network statistics from device. The device driver may provide
4691 * its own method by setting dev->netdev_ops->get_stats; otherwise
4692 * the internal statistics structure is used.
4694 const struct net_device_stats *dev_get_stats(struct net_device *dev)
4696 const struct net_device_ops *ops = dev->netdev_ops;
4698 if (ops->ndo_get_stats)
4699 return ops->ndo_get_stats(dev);
4703 EXPORT_SYMBOL(dev_get_stats);
4705 static void netdev_init_one_queue(struct net_device *dev,
4706 struct netdev_queue *queue,
4712 static void netdev_init_queues(struct net_device *dev)
4714 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4715 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4716 spin_lock_init(&dev->tx_global_lock);
4720 * alloc_netdev_mq - allocate network device
4721 * @sizeof_priv: size of private data to allocate space for
4722 * @name: device name format string
4723 * @setup: callback to initialize device
4724 * @queue_count: the number of subqueues to allocate
4726 * Allocates a struct net_device with private data area for driver use
4727 * and performs basic initialization. Also allocates subquue structs
4728 * for each queue on the device at the end of the netdevice.
4730 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4731 void (*setup)(struct net_device *), unsigned int queue_count)
4733 struct netdev_queue *tx;
4734 struct net_device *dev;
4738 BUG_ON(strlen(name) >= sizeof(dev->name));
4740 alloc_size = sizeof(struct net_device);
4742 /* ensure 32-byte alignment of private area */
4743 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4744 alloc_size += sizeof_priv;
4746 /* ensure 32-byte alignment of whole construct */
4747 alloc_size += NETDEV_ALIGN_CONST;
4749 p = kzalloc(alloc_size, GFP_KERNEL);
4751 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4755 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
4757 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4763 dev = (struct net_device *)
4764 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4765 dev->padded = (char *)dev - (char *)p;
4766 dev_net_set(dev, &init_net);
4769 dev->num_tx_queues = queue_count;
4770 dev->real_num_tx_queues = queue_count;
4772 dev->gso_max_size = GSO_MAX_SIZE;
4774 netdev_init_queues(dev);
4776 INIT_LIST_HEAD(&dev->napi_list);
4778 strcpy(dev->name, name);
4781 EXPORT_SYMBOL(alloc_netdev_mq);
4784 * free_netdev - free network device
4787 * This function does the last stage of destroying an allocated device
4788 * interface. The reference to the device object is released.
4789 * If this is the last reference then it will be freed.
4791 void free_netdev(struct net_device *dev)
4793 struct napi_struct *p, *n;
4795 release_net(dev_net(dev));
4799 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
4802 /* Compatibility with error handling in drivers */
4803 if (dev->reg_state == NETREG_UNINITIALIZED) {
4804 kfree((char *)dev - dev->padded);
4808 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4809 dev->reg_state = NETREG_RELEASED;
4811 /* will free via device release */
4812 put_device(&dev->dev);
4816 * synchronize_net - Synchronize with packet receive processing
4818 * Wait for packets currently being received to be done.
4819 * Does not block later packets from starting.
4821 void synchronize_net(void)
4828 * unregister_netdevice - remove device from the kernel
4831 * This function shuts down a device interface and removes it
4832 * from the kernel tables.
4834 * Callers must hold the rtnl semaphore. You may want
4835 * unregister_netdev() instead of this.
4838 void unregister_netdevice(struct net_device *dev)
4842 rollback_registered(dev);
4843 /* Finish processing unregister after unlock */
4848 * unregister_netdev - remove device from the kernel
4851 * This function shuts down a device interface and removes it
4852 * from the kernel tables.
4854 * This is just a wrapper for unregister_netdevice that takes
4855 * the rtnl semaphore. In general you want to use this and not
4856 * unregister_netdevice.
4858 void unregister_netdev(struct net_device *dev)
4861 unregister_netdevice(dev);
4865 EXPORT_SYMBOL(unregister_netdev);
4868 * dev_change_net_namespace - move device to different nethost namespace
4870 * @net: network namespace
4871 * @pat: If not NULL name pattern to try if the current device name
4872 * is already taken in the destination network namespace.
4874 * This function shuts down a device interface and moves it
4875 * to a new network namespace. On success 0 is returned, on
4876 * a failure a netagive errno code is returned.
4878 * Callers must hold the rtnl semaphore.
4881 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4884 const char *destname;
4889 /* Don't allow namespace local devices to be moved. */
4891 if (dev->features & NETIF_F_NETNS_LOCAL)
4895 /* Don't allow real devices to be moved when sysfs
4899 if (dev->dev.parent)
4903 /* Ensure the device has been registrered */
4905 if (dev->reg_state != NETREG_REGISTERED)
4908 /* Get out if there is nothing todo */
4910 if (net_eq(dev_net(dev), net))
4913 /* Pick the destination device name, and ensure
4914 * we can use it in the destination network namespace.
4917 destname = dev->name;
4918 if (__dev_get_by_name(net, destname)) {
4919 /* We get here if we can't use the current device name */
4922 if (!dev_valid_name(pat))
4924 if (strchr(pat, '%')) {
4925 if (__dev_alloc_name(net, pat, buf) < 0)
4930 if (__dev_get_by_name(net, destname))
4935 * And now a mini version of register_netdevice unregister_netdevice.
4938 /* If device is running close it first. */
4941 /* And unlink it from device chain */
4943 unlist_netdevice(dev);
4947 /* Shutdown queueing discipline. */
4950 /* Notify protocols, that we are about to destroy
4951 this device. They should clean all the things.
4953 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4956 * Flush the unicast and multicast chains
4958 dev_addr_discard(dev);
4960 netdev_unregister_kobject(dev);
4962 /* Actually switch the network namespace */
4963 dev_net_set(dev, net);
4965 /* Assign the new device name */
4966 if (destname != dev->name)
4967 strcpy(dev->name, destname);
4969 /* If there is an ifindex conflict assign a new one */
4970 if (__dev_get_by_index(net, dev->ifindex)) {
4971 int iflink = (dev->iflink == dev->ifindex);
4972 dev->ifindex = dev_new_index(net);
4974 dev->iflink = dev->ifindex;
4977 /* Fixup kobjects */
4978 err = netdev_register_kobject(dev);
4981 /* Add the device back in the hashes */
4982 list_netdevice(dev);
4984 /* Notify protocols, that a new device appeared. */
4985 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4993 static int dev_cpu_callback(struct notifier_block *nfb,
4994 unsigned long action,
4997 struct sk_buff **list_skb;
4998 struct Qdisc **list_net;
4999 struct sk_buff *skb;
5000 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5001 struct softnet_data *sd, *oldsd;
5003 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5006 local_irq_disable();
5007 cpu = smp_processor_id();
5008 sd = &per_cpu(softnet_data, cpu);
5009 oldsd = &per_cpu(softnet_data, oldcpu);
5011 /* Find end of our completion_queue. */
5012 list_skb = &sd->completion_queue;
5014 list_skb = &(*list_skb)->next;
5015 /* Append completion queue from offline CPU. */
5016 *list_skb = oldsd->completion_queue;
5017 oldsd->completion_queue = NULL;
5019 /* Find end of our output_queue. */
5020 list_net = &sd->output_queue;
5022 list_net = &(*list_net)->next_sched;
5023 /* Append output queue from offline CPU. */
5024 *list_net = oldsd->output_queue;
5025 oldsd->output_queue = NULL;
5027 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5030 /* Process offline CPU's input_pkt_queue */
5031 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5039 * netdev_increment_features - increment feature set by one
5040 * @all: current feature set
5041 * @one: new feature set
5042 * @mask: mask feature set
5044 * Computes a new feature set after adding a device with feature set
5045 * @one to the master device with current feature set @all. Will not
5046 * enable anything that is off in @mask. Returns the new feature set.
5048 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5051 /* If device needs checksumming, downgrade to it. */
5052 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5053 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5054 else if (mask & NETIF_F_ALL_CSUM) {
5055 /* If one device supports v4/v6 checksumming, set for all. */
5056 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5057 !(all & NETIF_F_GEN_CSUM)) {
5058 all &= ~NETIF_F_ALL_CSUM;
5059 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5062 /* If one device supports hw checksumming, set for all. */
5063 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5064 all &= ~NETIF_F_ALL_CSUM;
5065 all |= NETIF_F_HW_CSUM;
5069 one |= NETIF_F_ALL_CSUM;
5071 one |= all & NETIF_F_ONE_FOR_ALL;
5072 all &= one | NETIF_F_LLTX | NETIF_F_GSO;
5073 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5077 EXPORT_SYMBOL(netdev_increment_features);
5079 static struct hlist_head *netdev_create_hash(void)
5082 struct hlist_head *hash;
5084 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5086 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5087 INIT_HLIST_HEAD(&hash[i]);
5092 /* Initialize per network namespace state */
5093 static int __net_init netdev_init(struct net *net)
5095 INIT_LIST_HEAD(&net->dev_base_head);
5097 net->dev_name_head = netdev_create_hash();
5098 if (net->dev_name_head == NULL)
5101 net->dev_index_head = netdev_create_hash();
5102 if (net->dev_index_head == NULL)
5108 kfree(net->dev_name_head);
5114 * netdev_drivername - network driver for the device
5115 * @dev: network device
5116 * @buffer: buffer for resulting name
5117 * @len: size of buffer
5119 * Determine network driver for device.
5121 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5123 const struct device_driver *driver;
5124 const struct device *parent;
5126 if (len <= 0 || !buffer)
5130 parent = dev->dev.parent;
5135 driver = parent->driver;
5136 if (driver && driver->name)
5137 strlcpy(buffer, driver->name, len);
5141 static void __net_exit netdev_exit(struct net *net)
5143 kfree(net->dev_name_head);
5144 kfree(net->dev_index_head);
5147 static struct pernet_operations __net_initdata netdev_net_ops = {
5148 .init = netdev_init,
5149 .exit = netdev_exit,
5152 static void __net_exit default_device_exit(struct net *net)
5154 struct net_device *dev;
5156 * Push all migratable of the network devices back to the
5157 * initial network namespace
5161 for_each_netdev(net, dev) {
5163 char fb_name[IFNAMSIZ];
5165 /* Ignore unmoveable devices (i.e. loopback) */
5166 if (dev->features & NETIF_F_NETNS_LOCAL)
5169 /* Delete virtual devices */
5170 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5171 dev->rtnl_link_ops->dellink(dev);
5175 /* Push remaing network devices to init_net */
5176 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5177 err = dev_change_net_namespace(dev, &init_net, fb_name);
5179 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5180 __func__, dev->name, err);
5188 static struct pernet_operations __net_initdata default_device_ops = {
5189 .exit = default_device_exit,
5193 * Initialize the DEV module. At boot time this walks the device list and
5194 * unhooks any devices that fail to initialise (normally hardware not
5195 * present) and leaves us with a valid list of present and active devices.
5200 * This is called single threaded during boot, so no need
5201 * to take the rtnl semaphore.
5203 static int __init net_dev_init(void)
5205 int i, rc = -ENOMEM;
5207 BUG_ON(!dev_boot_phase);
5209 if (dev_proc_init())
5212 if (netdev_kobject_init())
5215 INIT_LIST_HEAD(&ptype_all);
5216 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5217 INIT_LIST_HEAD(&ptype_base[i]);
5219 if (register_pernet_subsys(&netdev_net_ops))
5223 * Initialise the packet receive queues.
5226 for_each_possible_cpu(i) {
5227 struct softnet_data *queue;
5229 queue = &per_cpu(softnet_data, i);
5230 skb_queue_head_init(&queue->input_pkt_queue);
5231 queue->completion_queue = NULL;
5232 INIT_LIST_HEAD(&queue->poll_list);
5234 queue->backlog.poll = process_backlog;
5235 queue->backlog.weight = weight_p;
5236 queue->backlog.gro_list = NULL;
5237 queue->backlog.gro_count = 0;
5242 /* The loopback device is special if any other network devices
5243 * is present in a network namespace the loopback device must
5244 * be present. Since we now dynamically allocate and free the
5245 * loopback device ensure this invariant is maintained by
5246 * keeping the loopback device as the first device on the
5247 * list of network devices. Ensuring the loopback devices
5248 * is the first device that appears and the last network device
5251 if (register_pernet_device(&loopback_net_ops))
5254 if (register_pernet_device(&default_device_ops))
5257 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5258 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5260 hotcpu_notifier(dev_cpu_callback, 0);
5268 subsys_initcall(net_dev_init);
5270 static int __init initialize_hashrnd(void)
5272 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5276 late_initcall_sync(initialize_hashrnd);
5278 EXPORT_SYMBOL(__dev_get_by_index);
5279 EXPORT_SYMBOL(__dev_get_by_name);
5280 EXPORT_SYMBOL(__dev_remove_pack);
5281 EXPORT_SYMBOL(dev_valid_name);
5282 EXPORT_SYMBOL(dev_add_pack);
5283 EXPORT_SYMBOL(dev_alloc_name);
5284 EXPORT_SYMBOL(dev_close);
5285 EXPORT_SYMBOL(dev_get_by_flags);
5286 EXPORT_SYMBOL(dev_get_by_index);
5287 EXPORT_SYMBOL(dev_get_by_name);
5288 EXPORT_SYMBOL(dev_open);
5289 EXPORT_SYMBOL(dev_queue_xmit);
5290 EXPORT_SYMBOL(dev_remove_pack);
5291 EXPORT_SYMBOL(dev_set_allmulti);
5292 EXPORT_SYMBOL(dev_set_promiscuity);
5293 EXPORT_SYMBOL(dev_change_flags);
5294 EXPORT_SYMBOL(dev_set_mtu);
5295 EXPORT_SYMBOL(dev_set_mac_address);
5296 EXPORT_SYMBOL(free_netdev);
5297 EXPORT_SYMBOL(netdev_boot_setup_check);
5298 EXPORT_SYMBOL(netdev_set_master);
5299 EXPORT_SYMBOL(netdev_state_change);
5300 EXPORT_SYMBOL(netif_receive_skb);
5301 EXPORT_SYMBOL(netif_rx);
5302 EXPORT_SYMBOL(register_gifconf);
5303 EXPORT_SYMBOL(register_netdevice);
5304 EXPORT_SYMBOL(register_netdevice_notifier);
5305 EXPORT_SYMBOL(skb_checksum_help);
5306 EXPORT_SYMBOL(synchronize_net);
5307 EXPORT_SYMBOL(unregister_netdevice);
5308 EXPORT_SYMBOL(unregister_netdevice_notifier);
5309 EXPORT_SYMBOL(net_enable_timestamp);
5310 EXPORT_SYMBOL(net_disable_timestamp);
5311 EXPORT_SYMBOL(dev_get_flags);
5313 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
5314 EXPORT_SYMBOL(br_handle_frame_hook);
5315 EXPORT_SYMBOL(br_fdb_get_hook);
5316 EXPORT_SYMBOL(br_fdb_put_hook);
5319 EXPORT_SYMBOL(dev_load);
5321 EXPORT_PER_CPU_SYMBOL(softnet_data);