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/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
104 #include <linux/if_bridge.h>
105 #include <linux/if_macvlan.h>
107 #include <net/pkt_sched.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/kmod.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/wext.h>
118 #include <net/iw_handler.h>
119 #include <asm/current.h>
120 #include <linux/audit.h>
121 #include <linux/dmaengine.h>
122 #include <linux/err.h>
123 #include <linux/ctype.h>
124 #include <linux/if_arp.h>
125 #include <linux/if_vlan.h>
126 #include <linux/ip.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
133 #include <linux/pci.h>
135 #include "net-sysfs.h"
137 /* Instead of increasing this, you should create a hash table. */
138 #define MAX_GRO_SKBS 8
140 /* This should be increased if a protocol with a bigger head is added. */
141 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 * The list of packet types we will receive (as opposed to discard)
145 * and the routines to invoke.
147 * Why 16. Because with 16 the only overlap we get on a hash of the
148 * low nibble of the protocol value is RARP/SNAP/X.25.
150 * NOTE: That is no longer true with the addition of VLAN tags. Not
151 * sure which should go first, but I bet it won't make much
152 * difference if we are running VLANs. The good news is that
153 * this protocol won't be in the list unless compiled in, so
154 * the average user (w/out VLANs) will not be adversely affected.
171 #define PTYPE_HASH_SIZE (16)
172 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
174 static DEFINE_SPINLOCK(ptype_lock);
175 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
176 static struct list_head ptype_all __read_mostly; /* Taps */
179 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
182 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
184 * Writers must hold the rtnl semaphore while they loop through the
185 * dev_base_head list, and hold dev_base_lock for writing when they do the
186 * actual updates. This allows pure readers to access the list even
187 * while a writer is preparing to update it.
189 * To put it another way, dev_base_lock is held for writing only to
190 * protect against pure readers; the rtnl semaphore provides the
191 * protection against other writers.
193 * See, for example usages, register_netdevice() and
194 * unregister_netdevice(), which must be called with the rtnl
197 DEFINE_RWLOCK(dev_base_lock);
198 EXPORT_SYMBOL(dev_base_lock);
200 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
202 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
203 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
206 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
208 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
211 static inline void rps_lock(struct softnet_data *sd)
214 spin_lock(&sd->input_pkt_queue.lock);
218 static inline void rps_unlock(struct softnet_data *sd)
221 spin_unlock(&sd->input_pkt_queue.lock);
225 /* Device list insertion */
226 static int list_netdevice(struct net_device *dev)
228 struct net *net = dev_net(dev);
232 write_lock_bh(&dev_base_lock);
233 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
234 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
235 hlist_add_head_rcu(&dev->index_hlist,
236 dev_index_hash(net, dev->ifindex));
237 write_unlock_bh(&dev_base_lock);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device *dev)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock);
250 list_del_rcu(&dev->dev_list);
251 hlist_del_rcu(&dev->name_hlist);
252 hlist_del_rcu(&dev->index_hlist);
253 write_unlock_bh(&dev_base_lock);
260 static RAW_NOTIFIER_HEAD(netdev_chain);
263 * Device drivers call our routines to queue packets here. We empty the
264 * queue in the local softnet handler.
267 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
268 EXPORT_PER_CPU_SYMBOL(softnet_data);
270 #ifdef CONFIG_LOCKDEP
272 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
273 * according to dev->type
275 static const unsigned short netdev_lock_type[] =
276 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
277 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
278 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
279 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
280 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
281 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
282 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
283 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
284 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
285 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
286 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
287 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
288 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
289 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
290 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
291 ARPHRD_VOID, ARPHRD_NONE};
293 static const char *const netdev_lock_name[] =
294 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
295 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
296 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
297 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
298 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
299 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
300 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
301 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
302 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
303 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
304 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
305 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
306 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
307 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
308 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
309 "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
318 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
319 if (netdev_lock_type[i] == dev_type)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
326 unsigned short dev_type)
330 i = netdev_lock_pos(dev_type);
331 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
332 netdev_lock_name[i]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
339 i = netdev_lock_pos(dev->type);
340 lockdep_set_class_and_name(&dev->addr_list_lock,
341 &netdev_addr_lock_key[i],
342 netdev_lock_name[i]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
346 unsigned short dev_type)
349 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
377 * dev_add_pack - add packet handler
378 * @pt: packet type declaration
380 * Add a protocol handler to the networking stack. The passed &packet_type
381 * is linked into kernel lists and may not be freed until it has been
382 * removed from the kernel lists.
384 * This call does not sleep therefore it can not
385 * guarantee all CPU's that are in middle of receiving packets
386 * will see the new packet type (until the next received packet).
389 void dev_add_pack(struct packet_type *pt)
393 spin_lock_bh(&ptype_lock);
394 if (pt->type == htons(ETH_P_ALL))
395 list_add_rcu(&pt->list, &ptype_all);
397 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
398 list_add_rcu(&pt->list, &ptype_base[hash]);
400 spin_unlock_bh(&ptype_lock);
402 EXPORT_SYMBOL(dev_add_pack);
405 * __dev_remove_pack - remove packet handler
406 * @pt: packet type declaration
408 * Remove a protocol handler that was previously added to the kernel
409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
410 * from the kernel lists and can be freed or reused once this function
413 * The packet type might still be in use by receivers
414 * and must not be freed until after all the CPU's have gone
415 * through a quiescent state.
417 void __dev_remove_pack(struct packet_type *pt)
419 struct list_head *head;
420 struct packet_type *pt1;
422 spin_lock_bh(&ptype_lock);
424 if (pt->type == htons(ETH_P_ALL))
427 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
429 list_for_each_entry(pt1, head, list) {
431 list_del_rcu(&pt->list);
436 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
438 spin_unlock_bh(&ptype_lock);
440 EXPORT_SYMBOL(__dev_remove_pack);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type *pt)
456 __dev_remove_pack(pt);
460 EXPORT_SYMBOL(dev_remove_pack);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name, struct ifmap *map)
482 struct netdev_boot_setup *s;
486 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
487 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
488 memset(s[i].name, 0, sizeof(s[i].name));
489 strlcpy(s[i].name, name, IFNAMSIZ);
490 memcpy(&s[i].map, map, sizeof(s[i].map));
495 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device *dev)
509 struct netdev_boot_setup *s = dev_boot_setup;
512 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
513 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
514 !strcmp(dev->name, s[i].name)) {
515 dev->irq = s[i].map.irq;
516 dev->base_addr = s[i].map.base_addr;
517 dev->mem_start = s[i].map.mem_start;
518 dev->mem_end = s[i].map.mem_end;
524 EXPORT_SYMBOL(netdev_boot_setup_check);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix, int unit)
539 const struct netdev_boot_setup *s = dev_boot_setup;
543 sprintf(name, "%s%d", prefix, unit);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net, name))
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
553 if (!strcmp(name, s[i].name))
554 return s[i].map.base_addr;
559 * Saves at boot time configured settings for any netdevice.
561 int __init netdev_boot_setup(char *str)
566 str = get_options(str, ARRAY_SIZE(ints), ints);
571 memset(&map, 0, sizeof(map));
575 map.base_addr = ints[2];
577 map.mem_start = ints[3];
579 map.mem_end = ints[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str, &map);
585 __setup("netdev=", netdev_boot_setup);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device *__dev_get_by_name(struct net *net, const char *name)
607 struct hlist_node *p;
608 struct net_device *dev;
609 struct hlist_head *head = dev_name_hash(net, name);
611 hlist_for_each_entry(dev, p, head, name_hlist)
612 if (!strncmp(dev->name, name, IFNAMSIZ))
617 EXPORT_SYMBOL(__dev_get_by_name);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
633 struct hlist_node *p;
634 struct net_device *dev;
635 struct hlist_head *head = dev_name_hash(net, name);
637 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
638 if (!strncmp(dev->name, name, IFNAMSIZ))
643 EXPORT_SYMBOL(dev_get_by_name_rcu);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device *dev_get_by_name(struct net *net, const char *name)
659 struct net_device *dev;
662 dev = dev_get_by_name_rcu(net, name);
668 EXPORT_SYMBOL(dev_get_by_name);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
684 struct hlist_node *p;
685 struct net_device *dev;
686 struct hlist_head *head = dev_index_hash(net, ifindex);
688 hlist_for_each_entry(dev, p, head, index_hlist)
689 if (dev->ifindex == ifindex)
694 EXPORT_SYMBOL(__dev_get_by_index);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
709 struct hlist_node *p;
710 struct net_device *dev;
711 struct hlist_head *head = dev_index_hash(net, ifindex);
713 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
714 if (dev->ifindex == ifindex)
719 EXPORT_SYMBOL(dev_get_by_index_rcu);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device *dev_get_by_index(struct net *net, int ifindex)
735 struct net_device *dev;
738 dev = dev_get_by_index_rcu(net, ifindex);
744 EXPORT_SYMBOL(dev_get_by_index);
747 * dev_getbyhwaddr - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device. The caller must hold the
754 * rtnl semaphore. The returned device has not had its ref count increased
755 * and the caller must therefore be careful about locking
758 * If the API was consistent this would be __dev_get_by_hwaddr
761 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
763 struct net_device *dev;
767 for_each_netdev(net, dev)
768 if (dev->type == type &&
769 !memcmp(dev->dev_addr, ha, dev->addr_len))
774 EXPORT_SYMBOL(dev_getbyhwaddr);
776 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
778 struct net_device *dev;
781 for_each_netdev(net, dev)
782 if (dev->type == type)
787 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
789 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
791 struct net_device *dev, *ret = NULL;
794 for_each_netdev_rcu(net, dev)
795 if (dev->type == type) {
803 EXPORT_SYMBOL(dev_getfirstbyhwtype);
806 * dev_get_by_flags - find any device with given flags
807 * @net: the applicable net namespace
808 * @if_flags: IFF_* values
809 * @mask: bitmask of bits in if_flags to check
811 * Search for any interface with the given flags. Returns NULL if a device
812 * is not found or a pointer to the device. The device returned has
813 * had a reference added and the pointer is safe until the user calls
814 * dev_put to indicate they have finished with it.
817 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
820 struct net_device *dev, *ret;
824 for_each_netdev_rcu(net, dev) {
825 if (((dev->flags ^ if_flags) & mask) == 0) {
834 EXPORT_SYMBOL(dev_get_by_flags);
837 * dev_valid_name - check if name is okay for network device
840 * Network device names need to be valid file names to
841 * to allow sysfs to work. We also disallow any kind of
844 int dev_valid_name(const char *name)
848 if (strlen(name) >= IFNAMSIZ)
850 if (!strcmp(name, ".") || !strcmp(name, ".."))
854 if (*name == '/' || isspace(*name))
860 EXPORT_SYMBOL(dev_valid_name);
863 * __dev_alloc_name - allocate a name for a device
864 * @net: network namespace to allocate the device name in
865 * @name: name format string
866 * @buf: scratch buffer and result name string
868 * Passed a format string - eg "lt%d" it will try and find a suitable
869 * id. It scans list of devices to build up a free map, then chooses
870 * the first empty slot. The caller must hold the dev_base or rtnl lock
871 * while allocating the name and adding the device in order to avoid
873 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
874 * Returns the number of the unit assigned or a negative errno code.
877 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
881 const int max_netdevices = 8*PAGE_SIZE;
882 unsigned long *inuse;
883 struct net_device *d;
885 p = strnchr(name, IFNAMSIZ-1, '%');
888 * Verify the string as this thing may have come from
889 * the user. There must be either one "%d" and no other "%"
892 if (p[1] != 'd' || strchr(p + 2, '%'))
895 /* Use one page as a bit array of possible slots */
896 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
900 for_each_netdev(net, d) {
901 if (!sscanf(d->name, name, &i))
903 if (i < 0 || i >= max_netdevices)
906 /* avoid cases where sscanf is not exact inverse of printf */
907 snprintf(buf, IFNAMSIZ, name, i);
908 if (!strncmp(buf, d->name, IFNAMSIZ))
912 i = find_first_zero_bit(inuse, max_netdevices);
913 free_page((unsigned long) inuse);
917 snprintf(buf, IFNAMSIZ, name, i);
918 if (!__dev_get_by_name(net, buf))
921 /* It is possible to run out of possible slots
922 * when the name is long and there isn't enough space left
923 * for the digits, or if all bits are used.
929 * dev_alloc_name - allocate a name for a device
931 * @name: name format string
933 * Passed a format string - eg "lt%d" it will try and find a suitable
934 * id. It scans list of devices to build up a free map, then chooses
935 * the first empty slot. The caller must hold the dev_base or rtnl lock
936 * while allocating the name and adding the device in order to avoid
938 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
939 * Returns the number of the unit assigned or a negative errno code.
942 int dev_alloc_name(struct net_device *dev, const char *name)
948 BUG_ON(!dev_net(dev));
950 ret = __dev_alloc_name(net, name, buf);
952 strlcpy(dev->name, buf, IFNAMSIZ);
955 EXPORT_SYMBOL(dev_alloc_name);
957 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
961 BUG_ON(!dev_net(dev));
964 if (!dev_valid_name(name))
967 if (fmt && strchr(name, '%'))
968 return dev_alloc_name(dev, name);
969 else if (__dev_get_by_name(net, name))
971 else if (dev->name != name)
972 strlcpy(dev->name, name, IFNAMSIZ);
978 * dev_change_name - change name of a device
980 * @newname: name (or format string) must be at least IFNAMSIZ
982 * Change name of a device, can pass format strings "eth%d".
985 int dev_change_name(struct net_device *dev, const char *newname)
987 char oldname[IFNAMSIZ];
993 BUG_ON(!dev_net(dev));
996 if (dev->flags & IFF_UP)
999 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1002 memcpy(oldname, dev->name, IFNAMSIZ);
1004 err = dev_get_valid_name(dev, newname, 1);
1009 ret = device_rename(&dev->dev, dev->name);
1011 memcpy(dev->name, oldname, IFNAMSIZ);
1015 write_lock_bh(&dev_base_lock);
1016 hlist_del(&dev->name_hlist);
1017 write_unlock_bh(&dev_base_lock);
1021 write_lock_bh(&dev_base_lock);
1022 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1023 write_unlock_bh(&dev_base_lock);
1025 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1026 ret = notifier_to_errno(ret);
1029 /* err >= 0 after dev_alloc_name() or stores the first errno */
1032 memcpy(dev->name, oldname, IFNAMSIZ);
1036 "%s: name change rollback failed: %d.\n",
1045 * dev_set_alias - change ifalias of a device
1047 * @alias: name up to IFALIASZ
1048 * @len: limit of bytes to copy from info
1050 * Set ifalias for a device,
1052 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1056 if (len >= IFALIASZ)
1061 kfree(dev->ifalias);
1062 dev->ifalias = NULL;
1067 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1071 strlcpy(dev->ifalias, alias, len+1);
1077 * netdev_features_change - device changes features
1078 * @dev: device to cause notification
1080 * Called to indicate a device has changed features.
1082 void netdev_features_change(struct net_device *dev)
1084 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1086 EXPORT_SYMBOL(netdev_features_change);
1089 * netdev_state_change - device changes state
1090 * @dev: device to cause notification
1092 * Called to indicate a device has changed state. This function calls
1093 * the notifier chains for netdev_chain and sends a NEWLINK message
1094 * to the routing socket.
1096 void netdev_state_change(struct net_device *dev)
1098 if (dev->flags & IFF_UP) {
1099 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1100 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1103 EXPORT_SYMBOL(netdev_state_change);
1105 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1107 return call_netdevice_notifiers(event, dev);
1109 EXPORT_SYMBOL(netdev_bonding_change);
1112 * dev_load - load a network module
1113 * @net: the applicable net namespace
1114 * @name: name of interface
1116 * If a network interface is not present and the process has suitable
1117 * privileges this function loads the module. If module loading is not
1118 * available in this kernel then it becomes a nop.
1121 void dev_load(struct net *net, const char *name)
1123 struct net_device *dev;
1126 dev = dev_get_by_name_rcu(net, name);
1129 if (!dev && capable(CAP_NET_ADMIN))
1130 request_module("%s", name);
1132 EXPORT_SYMBOL(dev_load);
1134 static int __dev_open(struct net_device *dev)
1136 const struct net_device_ops *ops = dev->netdev_ops;
1142 * Is it even present?
1144 if (!netif_device_present(dev))
1147 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1148 ret = notifier_to_errno(ret);
1153 * Call device private open method
1155 set_bit(__LINK_STATE_START, &dev->state);
1157 if (ops->ndo_validate_addr)
1158 ret = ops->ndo_validate_addr(dev);
1160 if (!ret && ops->ndo_open)
1161 ret = ops->ndo_open(dev);
1164 * If it went open OK then:
1168 clear_bit(__LINK_STATE_START, &dev->state);
1173 dev->flags |= IFF_UP;
1178 net_dmaengine_get();
1181 * Initialize multicasting status
1183 dev_set_rx_mode(dev);
1186 * Wakeup transmit queue engine
1195 * dev_open - prepare an interface for use.
1196 * @dev: device to open
1198 * Takes a device from down to up state. The device's private open
1199 * function is invoked and then the multicast lists are loaded. Finally
1200 * the device is moved into the up state and a %NETDEV_UP message is
1201 * sent to the netdev notifier chain.
1203 * Calling this function on an active interface is a nop. On a failure
1204 * a negative errno code is returned.
1206 int dev_open(struct net_device *dev)
1213 if (dev->flags & IFF_UP)
1219 ret = __dev_open(dev);
1224 * ... and announce new interface.
1226 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1227 call_netdevice_notifiers(NETDEV_UP, dev);
1231 EXPORT_SYMBOL(dev_open);
1233 static int __dev_close(struct net_device *dev)
1235 const struct net_device_ops *ops = dev->netdev_ops;
1241 * Tell people we are going down, so that they can
1242 * prepare to death, when device is still operating.
1244 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1246 clear_bit(__LINK_STATE_START, &dev->state);
1248 /* Synchronize to scheduled poll. We cannot touch poll list,
1249 * it can be even on different cpu. So just clear netif_running().
1251 * dev->stop() will invoke napi_disable() on all of it's
1252 * napi_struct instances on this device.
1254 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1256 dev_deactivate(dev);
1259 * Call the device specific close. This cannot fail.
1260 * Only if device is UP
1262 * We allow it to be called even after a DETACH hot-plug
1269 * Device is now down.
1272 dev->flags &= ~IFF_UP;
1277 net_dmaengine_put();
1283 * dev_close - shutdown an interface.
1284 * @dev: device to shutdown
1286 * This function moves an active device into down state. A
1287 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1288 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1291 int dev_close(struct net_device *dev)
1293 if (!(dev->flags & IFF_UP))
1299 * Tell people we are down
1301 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1302 call_netdevice_notifiers(NETDEV_DOWN, dev);
1306 EXPORT_SYMBOL(dev_close);
1310 * dev_disable_lro - disable Large Receive Offload on a device
1313 * Disable Large Receive Offload (LRO) on a net device. Must be
1314 * called under RTNL. This is needed if received packets may be
1315 * forwarded to another interface.
1317 void dev_disable_lro(struct net_device *dev)
1319 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1320 dev->ethtool_ops->set_flags) {
1321 u32 flags = dev->ethtool_ops->get_flags(dev);
1322 if (flags & ETH_FLAG_LRO) {
1323 flags &= ~ETH_FLAG_LRO;
1324 dev->ethtool_ops->set_flags(dev, flags);
1327 WARN_ON(dev->features & NETIF_F_LRO);
1329 EXPORT_SYMBOL(dev_disable_lro);
1332 static int dev_boot_phase = 1;
1335 * Device change register/unregister. These are not inline or static
1336 * as we export them to the world.
1340 * register_netdevice_notifier - register a network notifier block
1343 * Register a notifier to be called when network device events occur.
1344 * The notifier passed is linked into the kernel structures and must
1345 * not be reused until it has been unregistered. A negative errno code
1346 * is returned on a failure.
1348 * When registered all registration and up events are replayed
1349 * to the new notifier to allow device to have a race free
1350 * view of the network device list.
1353 int register_netdevice_notifier(struct notifier_block *nb)
1355 struct net_device *dev;
1356 struct net_device *last;
1361 err = raw_notifier_chain_register(&netdev_chain, nb);
1367 for_each_netdev(net, dev) {
1368 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1369 err = notifier_to_errno(err);
1373 if (!(dev->flags & IFF_UP))
1376 nb->notifier_call(nb, NETDEV_UP, dev);
1387 for_each_netdev(net, dev) {
1391 if (dev->flags & IFF_UP) {
1392 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1393 nb->notifier_call(nb, NETDEV_DOWN, dev);
1395 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1396 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1400 raw_notifier_chain_unregister(&netdev_chain, nb);
1403 EXPORT_SYMBOL(register_netdevice_notifier);
1406 * unregister_netdevice_notifier - unregister a network notifier block
1409 * Unregister a notifier previously registered by
1410 * register_netdevice_notifier(). The notifier is unlinked into the
1411 * kernel structures and may then be reused. A negative errno code
1412 * is returned on a failure.
1415 int unregister_netdevice_notifier(struct notifier_block *nb)
1420 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1424 EXPORT_SYMBOL(unregister_netdevice_notifier);
1427 * call_netdevice_notifiers - call all network notifier blocks
1428 * @val: value passed unmodified to notifier function
1429 * @dev: net_device pointer passed unmodified to notifier function
1431 * Call all network notifier blocks. Parameters and return value
1432 * are as for raw_notifier_call_chain().
1435 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1438 return raw_notifier_call_chain(&netdev_chain, val, dev);
1441 /* When > 0 there are consumers of rx skb time stamps */
1442 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1444 void net_enable_timestamp(void)
1446 atomic_inc(&netstamp_needed);
1448 EXPORT_SYMBOL(net_enable_timestamp);
1450 void net_disable_timestamp(void)
1452 atomic_dec(&netstamp_needed);
1454 EXPORT_SYMBOL(net_disable_timestamp);
1456 static inline void net_timestamp_set(struct sk_buff *skb)
1458 if (atomic_read(&netstamp_needed))
1459 __net_timestamp(skb);
1461 skb->tstamp.tv64 = 0;
1464 static inline void net_timestamp_check(struct sk_buff *skb)
1466 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1467 __net_timestamp(skb);
1471 * dev_forward_skb - loopback an skb to another netif
1473 * @dev: destination network device
1474 * @skb: buffer to forward
1477 * NET_RX_SUCCESS (no congestion)
1478 * NET_RX_DROP (packet was dropped, but freed)
1480 * dev_forward_skb can be used for injecting an skb from the
1481 * start_xmit function of one device into the receive queue
1482 * of another device.
1484 * The receiving device may be in another namespace, so
1485 * we have to clear all information in the skb that could
1486 * impact namespace isolation.
1488 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1492 if (!(dev->flags & IFF_UP) ||
1493 (skb->len > (dev->mtu + dev->hard_header_len))) {
1497 skb_set_dev(skb, dev);
1498 skb->tstamp.tv64 = 0;
1499 skb->pkt_type = PACKET_HOST;
1500 skb->protocol = eth_type_trans(skb, dev);
1501 return netif_rx(skb);
1503 EXPORT_SYMBOL_GPL(dev_forward_skb);
1506 * Support routine. Sends outgoing frames to any network
1507 * taps currently in use.
1510 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1512 struct packet_type *ptype;
1514 #ifdef CONFIG_NET_CLS_ACT
1515 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1516 net_timestamp_set(skb);
1518 net_timestamp_set(skb);
1522 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1523 /* Never send packets back to the socket
1524 * they originated from - MvS (miquels@drinkel.ow.org)
1526 if ((ptype->dev == dev || !ptype->dev) &&
1527 (ptype->af_packet_priv == NULL ||
1528 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1529 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1533 /* skb->nh should be correctly
1534 set by sender, so that the second statement is
1535 just protection against buggy protocols.
1537 skb_reset_mac_header(skb2);
1539 if (skb_network_header(skb2) < skb2->data ||
1540 skb2->network_header > skb2->tail) {
1541 if (net_ratelimit())
1542 printk(KERN_CRIT "protocol %04x is "
1544 skb2->protocol, dev->name);
1545 skb_reset_network_header(skb2);
1548 skb2->transport_header = skb2->network_header;
1549 skb2->pkt_type = PACKET_OUTGOING;
1550 ptype->func(skb2, skb->dev, ptype, skb->dev);
1557 static inline void __netif_reschedule(struct Qdisc *q)
1559 struct softnet_data *sd;
1560 unsigned long flags;
1562 local_irq_save(flags);
1563 sd = &__get_cpu_var(softnet_data);
1564 q->next_sched = NULL;
1565 *sd->output_queue_tailp = q;
1566 sd->output_queue_tailp = &q->next_sched;
1567 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1568 local_irq_restore(flags);
1571 void __netif_schedule(struct Qdisc *q)
1573 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1574 __netif_reschedule(q);
1576 EXPORT_SYMBOL(__netif_schedule);
1578 void dev_kfree_skb_irq(struct sk_buff *skb)
1580 if (!skb->destructor)
1582 else if (atomic_dec_and_test(&skb->users)) {
1583 struct softnet_data *sd;
1584 unsigned long flags;
1586 local_irq_save(flags);
1587 sd = &__get_cpu_var(softnet_data);
1588 skb->next = sd->completion_queue;
1589 sd->completion_queue = skb;
1590 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1591 local_irq_restore(flags);
1594 EXPORT_SYMBOL(dev_kfree_skb_irq);
1596 void dev_kfree_skb_any(struct sk_buff *skb)
1598 if (in_irq() || irqs_disabled())
1599 dev_kfree_skb_irq(skb);
1603 EXPORT_SYMBOL(dev_kfree_skb_any);
1607 * netif_device_detach - mark device as removed
1608 * @dev: network device
1610 * Mark device as removed from system and therefore no longer available.
1612 void netif_device_detach(struct net_device *dev)
1614 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1615 netif_running(dev)) {
1616 netif_tx_stop_all_queues(dev);
1619 EXPORT_SYMBOL(netif_device_detach);
1622 * netif_device_attach - mark device as attached
1623 * @dev: network device
1625 * Mark device as attached from system and restart if needed.
1627 void netif_device_attach(struct net_device *dev)
1629 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1630 netif_running(dev)) {
1631 netif_tx_wake_all_queues(dev);
1632 __netdev_watchdog_up(dev);
1635 EXPORT_SYMBOL(netif_device_attach);
1637 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1639 return ((features & NETIF_F_GEN_CSUM) ||
1640 ((features & NETIF_F_IP_CSUM) &&
1641 protocol == htons(ETH_P_IP)) ||
1642 ((features & NETIF_F_IPV6_CSUM) &&
1643 protocol == htons(ETH_P_IPV6)) ||
1644 ((features & NETIF_F_FCOE_CRC) &&
1645 protocol == htons(ETH_P_FCOE)));
1648 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1650 if (can_checksum_protocol(dev->features, skb->protocol))
1653 if (skb->protocol == htons(ETH_P_8021Q)) {
1654 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1655 if (can_checksum_protocol(dev->features & dev->vlan_features,
1656 veh->h_vlan_encapsulated_proto))
1664 * skb_dev_set -- assign a new device to a buffer
1665 * @skb: buffer for the new device
1666 * @dev: network device
1668 * If an skb is owned by a device already, we have to reset
1669 * all data private to the namespace a device belongs to
1670 * before assigning it a new device.
1672 #ifdef CONFIG_NET_NS
1673 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1676 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1679 skb_init_secmark(skb);
1683 skb->ipvs_property = 0;
1684 #ifdef CONFIG_NET_SCHED
1690 EXPORT_SYMBOL(skb_set_dev);
1691 #endif /* CONFIG_NET_NS */
1694 * Invalidate hardware checksum when packet is to be mangled, and
1695 * complete checksum manually on outgoing path.
1697 int skb_checksum_help(struct sk_buff *skb)
1700 int ret = 0, offset;
1702 if (skb->ip_summed == CHECKSUM_COMPLETE)
1703 goto out_set_summed;
1705 if (unlikely(skb_shinfo(skb)->gso_size)) {
1706 /* Let GSO fix up the checksum. */
1707 goto out_set_summed;
1710 offset = skb->csum_start - skb_headroom(skb);
1711 BUG_ON(offset >= skb_headlen(skb));
1712 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1714 offset += skb->csum_offset;
1715 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1717 if (skb_cloned(skb) &&
1718 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1719 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1724 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1726 skb->ip_summed = CHECKSUM_NONE;
1730 EXPORT_SYMBOL(skb_checksum_help);
1733 * skb_gso_segment - Perform segmentation on skb.
1734 * @skb: buffer to segment
1735 * @features: features for the output path (see dev->features)
1737 * This function segments the given skb and returns a list of segments.
1739 * It may return NULL if the skb requires no segmentation. This is
1740 * only possible when GSO is used for verifying header integrity.
1742 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1744 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1745 struct packet_type *ptype;
1746 __be16 type = skb->protocol;
1749 skb_reset_mac_header(skb);
1750 skb->mac_len = skb->network_header - skb->mac_header;
1751 __skb_pull(skb, skb->mac_len);
1753 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1754 struct net_device *dev = skb->dev;
1755 struct ethtool_drvinfo info = {};
1757 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1758 dev->ethtool_ops->get_drvinfo(dev, &info);
1760 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1762 info.driver, dev ? dev->features : 0L,
1763 skb->sk ? skb->sk->sk_route_caps : 0L,
1764 skb->len, skb->data_len, skb->ip_summed);
1766 if (skb_header_cloned(skb) &&
1767 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1768 return ERR_PTR(err);
1772 list_for_each_entry_rcu(ptype,
1773 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1774 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1775 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1776 err = ptype->gso_send_check(skb);
1777 segs = ERR_PTR(err);
1778 if (err || skb_gso_ok(skb, features))
1780 __skb_push(skb, (skb->data -
1781 skb_network_header(skb)));
1783 segs = ptype->gso_segment(skb, features);
1789 __skb_push(skb, skb->data - skb_mac_header(skb));
1793 EXPORT_SYMBOL(skb_gso_segment);
1795 /* Take action when hardware reception checksum errors are detected. */
1797 void netdev_rx_csum_fault(struct net_device *dev)
1799 if (net_ratelimit()) {
1800 printk(KERN_ERR "%s: hw csum failure.\n",
1801 dev ? dev->name : "<unknown>");
1805 EXPORT_SYMBOL(netdev_rx_csum_fault);
1808 /* Actually, we should eliminate this check as soon as we know, that:
1809 * 1. IOMMU is present and allows to map all the memory.
1810 * 2. No high memory really exists on this machine.
1813 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1815 #ifdef CONFIG_HIGHMEM
1817 if (!(dev->features & NETIF_F_HIGHDMA)) {
1818 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1819 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1823 if (PCI_DMA_BUS_IS_PHYS) {
1824 struct device *pdev = dev->dev.parent;
1828 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1829 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1830 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1839 void (*destructor)(struct sk_buff *skb);
1842 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1844 static void dev_gso_skb_destructor(struct sk_buff *skb)
1846 struct dev_gso_cb *cb;
1849 struct sk_buff *nskb = skb->next;
1851 skb->next = nskb->next;
1854 } while (skb->next);
1856 cb = DEV_GSO_CB(skb);
1858 cb->destructor(skb);
1862 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1863 * @skb: buffer to segment
1865 * This function segments the given skb and stores the list of segments
1868 static int dev_gso_segment(struct sk_buff *skb)
1870 struct net_device *dev = skb->dev;
1871 struct sk_buff *segs;
1872 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1875 segs = skb_gso_segment(skb, features);
1877 /* Verifying header integrity only. */
1882 return PTR_ERR(segs);
1885 DEV_GSO_CB(skb)->destructor = skb->destructor;
1886 skb->destructor = dev_gso_skb_destructor;
1892 * Try to orphan skb early, right before transmission by the device.
1893 * We cannot orphan skb if tx timestamp is requested, since
1894 * drivers need to call skb_tstamp_tx() to send the timestamp.
1896 static inline void skb_orphan_try(struct sk_buff *skb)
1898 if (!skb_tx(skb)->flags)
1902 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1903 struct netdev_queue *txq)
1905 const struct net_device_ops *ops = dev->netdev_ops;
1906 int rc = NETDEV_TX_OK;
1908 if (likely(!skb->next)) {
1909 if (!list_empty(&ptype_all))
1910 dev_queue_xmit_nit(skb, dev);
1913 * If device doesnt need skb->dst, release it right now while
1914 * its hot in this cpu cache
1916 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1919 skb_orphan_try(skb);
1921 if (netif_needs_gso(dev, skb)) {
1922 if (unlikely(dev_gso_segment(skb)))
1928 rc = ops->ndo_start_xmit(skb, dev);
1929 if (rc == NETDEV_TX_OK)
1930 txq_trans_update(txq);
1936 struct sk_buff *nskb = skb->next;
1938 skb->next = nskb->next;
1942 * If device doesnt need nskb->dst, release it right now while
1943 * its hot in this cpu cache
1945 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1948 rc = ops->ndo_start_xmit(nskb, dev);
1949 if (unlikely(rc != NETDEV_TX_OK)) {
1950 if (rc & ~NETDEV_TX_MASK)
1951 goto out_kfree_gso_skb;
1952 nskb->next = skb->next;
1956 txq_trans_update(txq);
1957 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1958 return NETDEV_TX_BUSY;
1959 } while (skb->next);
1962 if (likely(skb->next == NULL))
1963 skb->destructor = DEV_GSO_CB(skb)->destructor;
1969 static u32 hashrnd __read_mostly;
1971 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1975 if (skb_rx_queue_recorded(skb)) {
1976 hash = skb_get_rx_queue(skb);
1977 while (unlikely(hash >= dev->real_num_tx_queues))
1978 hash -= dev->real_num_tx_queues;
1982 if (skb->sk && skb->sk->sk_hash)
1983 hash = skb->sk->sk_hash;
1985 hash = (__force u16) skb->protocol;
1987 hash = jhash_1word(hash, hashrnd);
1989 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1991 EXPORT_SYMBOL(skb_tx_hash);
1993 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1995 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1996 if (net_ratelimit()) {
1997 pr_warning("%s selects TX queue %d, but "
1998 "real number of TX queues is %d\n",
1999 dev->name, queue_index, dev->real_num_tx_queues);
2006 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2007 struct sk_buff *skb)
2010 struct sock *sk = skb->sk;
2012 if (sk_tx_queue_recorded(sk)) {
2013 queue_index = sk_tx_queue_get(sk);
2015 const struct net_device_ops *ops = dev->netdev_ops;
2017 if (ops->ndo_select_queue) {
2018 queue_index = ops->ndo_select_queue(dev, skb);
2019 queue_index = dev_cap_txqueue(dev, queue_index);
2022 if (dev->real_num_tx_queues > 1)
2023 queue_index = skb_tx_hash(dev, skb);
2026 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2028 if (dst && skb_dst(skb) == dst)
2029 sk_tx_queue_set(sk, queue_index);
2034 skb_set_queue_mapping(skb, queue_index);
2035 return netdev_get_tx_queue(dev, queue_index);
2038 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2039 struct net_device *dev,
2040 struct netdev_queue *txq)
2042 spinlock_t *root_lock = qdisc_lock(q);
2043 bool contended = qdisc_is_running(q);
2047 * Heuristic to force contended enqueues to serialize on a
2048 * separate lock before trying to get qdisc main lock.
2049 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2050 * and dequeue packets faster.
2052 if (unlikely(contended))
2053 spin_lock(&q->busylock);
2055 spin_lock(root_lock);
2056 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2059 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2060 qdisc_run_begin(q)) {
2062 * This is a work-conserving queue; there are no old skbs
2063 * waiting to be sent out; and the qdisc is not running -
2064 * xmit the skb directly.
2066 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2068 __qdisc_update_bstats(q, skb->len);
2069 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2070 if (unlikely(contended)) {
2071 spin_unlock(&q->busylock);
2078 rc = NET_XMIT_SUCCESS;
2081 rc = qdisc_enqueue_root(skb, q);
2082 if (qdisc_run_begin(q)) {
2083 if (unlikely(contended)) {
2084 spin_unlock(&q->busylock);
2090 spin_unlock(root_lock);
2091 if (unlikely(contended))
2092 spin_unlock(&q->busylock);
2097 * Returns true if either:
2098 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2099 * 2. skb is fragmented and the device does not support SG, or if
2100 * at least one of fragments is in highmem and device does not
2101 * support DMA from it.
2103 static inline int skb_needs_linearize(struct sk_buff *skb,
2104 struct net_device *dev)
2106 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
2107 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
2108 illegal_highdma(dev, skb)));
2112 * dev_queue_xmit - transmit a buffer
2113 * @skb: buffer to transmit
2115 * Queue a buffer for transmission to a network device. The caller must
2116 * have set the device and priority and built the buffer before calling
2117 * this function. The function can be called from an interrupt.
2119 * A negative errno code is returned on a failure. A success does not
2120 * guarantee the frame will be transmitted as it may be dropped due
2121 * to congestion or traffic shaping.
2123 * -----------------------------------------------------------------------------------
2124 * I notice this method can also return errors from the queue disciplines,
2125 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2128 * Regardless of the return value, the skb is consumed, so it is currently
2129 * difficult to retry a send to this method. (You can bump the ref count
2130 * before sending to hold a reference for retry if you are careful.)
2132 * When calling this method, interrupts MUST be enabled. This is because
2133 * the BH enable code must have IRQs enabled so that it will not deadlock.
2136 int dev_queue_xmit(struct sk_buff *skb)
2138 struct net_device *dev = skb->dev;
2139 struct netdev_queue *txq;
2143 /* GSO will handle the following emulations directly. */
2144 if (netif_needs_gso(dev, skb))
2147 /* Convert a paged skb to linear, if required */
2148 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2151 /* If packet is not checksummed and device does not support
2152 * checksumming for this protocol, complete checksumming here.
2154 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2155 skb_set_transport_header(skb, skb->csum_start -
2157 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2162 /* Disable soft irqs for various locks below. Also
2163 * stops preemption for RCU.
2167 txq = dev_pick_tx(dev, skb);
2168 q = rcu_dereference_bh(txq->qdisc);
2170 #ifdef CONFIG_NET_CLS_ACT
2171 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2174 rc = __dev_xmit_skb(skb, q, dev, txq);
2178 /* The device has no queue. Common case for software devices:
2179 loopback, all the sorts of tunnels...
2181 Really, it is unlikely that netif_tx_lock protection is necessary
2182 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2184 However, it is possible, that they rely on protection
2187 Check this and shot the lock. It is not prone from deadlocks.
2188 Either shot noqueue qdisc, it is even simpler 8)
2190 if (dev->flags & IFF_UP) {
2191 int cpu = smp_processor_id(); /* ok because BHs are off */
2193 if (txq->xmit_lock_owner != cpu) {
2195 HARD_TX_LOCK(dev, txq, cpu);
2197 if (!netif_tx_queue_stopped(txq)) {
2198 rc = dev_hard_start_xmit(skb, dev, txq);
2199 if (dev_xmit_complete(rc)) {
2200 HARD_TX_UNLOCK(dev, txq);
2204 HARD_TX_UNLOCK(dev, txq);
2205 if (net_ratelimit())
2206 printk(KERN_CRIT "Virtual device %s asks to "
2207 "queue packet!\n", dev->name);
2209 /* Recursion is detected! It is possible,
2211 if (net_ratelimit())
2212 printk(KERN_CRIT "Dead loop on virtual device "
2213 "%s, fix it urgently!\n", dev->name);
2218 rcu_read_unlock_bh();
2224 rcu_read_unlock_bh();
2227 EXPORT_SYMBOL(dev_queue_xmit);
2230 /*=======================================================================
2232 =======================================================================*/
2234 int netdev_max_backlog __read_mostly = 1000;
2235 int netdev_tstamp_prequeue __read_mostly = 1;
2236 int netdev_budget __read_mostly = 300;
2237 int weight_p __read_mostly = 64; /* old backlog weight */
2239 /* Called with irq disabled */
2240 static inline void ____napi_schedule(struct softnet_data *sd,
2241 struct napi_struct *napi)
2243 list_add_tail(&napi->poll_list, &sd->poll_list);
2244 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2249 /* One global table that all flow-based protocols share. */
2250 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2251 EXPORT_SYMBOL(rps_sock_flow_table);
2254 * get_rps_cpu is called from netif_receive_skb and returns the target
2255 * CPU from the RPS map of the receiving queue for a given skb.
2256 * rcu_read_lock must be held on entry.
2258 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2259 struct rps_dev_flow **rflowp)
2261 struct ipv6hdr *ip6;
2263 struct netdev_rx_queue *rxqueue;
2264 struct rps_map *map;
2265 struct rps_dev_flow_table *flow_table;
2266 struct rps_sock_flow_table *sock_flow_table;
2270 u32 addr1, addr2, ihl;
2276 if (skb_rx_queue_recorded(skb)) {
2277 u16 index = skb_get_rx_queue(skb);
2278 if (unlikely(index >= dev->num_rx_queues)) {
2279 if (net_ratelimit()) {
2280 pr_warning("%s received packet on queue "
2281 "%u, but number of RX queues is %u\n",
2282 dev->name, index, dev->num_rx_queues);
2286 rxqueue = dev->_rx + index;
2290 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2294 goto got_hash; /* Skip hash computation on packet header */
2296 switch (skb->protocol) {
2297 case __constant_htons(ETH_P_IP):
2298 if (!pskb_may_pull(skb, sizeof(*ip)))
2301 ip = (struct iphdr *) skb->data;
2302 ip_proto = ip->protocol;
2303 addr1 = (__force u32) ip->saddr;
2304 addr2 = (__force u32) ip->daddr;
2307 case __constant_htons(ETH_P_IPV6):
2308 if (!pskb_may_pull(skb, sizeof(*ip6)))
2311 ip6 = (struct ipv6hdr *) skb->data;
2312 ip_proto = ip6->nexthdr;
2313 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2314 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2327 case IPPROTO_UDPLITE:
2328 if (pskb_may_pull(skb, (ihl * 4) + 4)) {
2329 ports.v32 = * (__force u32 *) (skb->data + (ihl * 4));
2330 if (ports.v16[1] < ports.v16[0])
2331 swap(ports.v16[0], ports.v16[1]);
2339 /* get a consistent hash (same value on both flow directions) */
2342 skb->rxhash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2347 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2348 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2349 if (flow_table && sock_flow_table) {
2351 struct rps_dev_flow *rflow;
2353 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2356 next_cpu = sock_flow_table->ents[skb->rxhash &
2357 sock_flow_table->mask];
2360 * If the desired CPU (where last recvmsg was done) is
2361 * different from current CPU (one in the rx-queue flow
2362 * table entry), switch if one of the following holds:
2363 * - Current CPU is unset (equal to RPS_NO_CPU).
2364 * - Current CPU is offline.
2365 * - The current CPU's queue tail has advanced beyond the
2366 * last packet that was enqueued using this table entry.
2367 * This guarantees that all previous packets for the flow
2368 * have been dequeued, thus preserving in order delivery.
2370 if (unlikely(tcpu != next_cpu) &&
2371 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2372 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2373 rflow->last_qtail)) >= 0)) {
2374 tcpu = rflow->cpu = next_cpu;
2375 if (tcpu != RPS_NO_CPU)
2376 rflow->last_qtail = per_cpu(softnet_data,
2377 tcpu).input_queue_head;
2379 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2386 map = rcu_dereference(rxqueue->rps_map);
2388 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2390 if (cpu_online(tcpu)) {
2400 /* Called from hardirq (IPI) context */
2401 static void rps_trigger_softirq(void *data)
2403 struct softnet_data *sd = data;
2405 ____napi_schedule(sd, &sd->backlog);
2409 #endif /* CONFIG_RPS */
2412 * Check if this softnet_data structure is another cpu one
2413 * If yes, queue it to our IPI list and return 1
2416 static int rps_ipi_queued(struct softnet_data *sd)
2419 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2422 sd->rps_ipi_next = mysd->rps_ipi_list;
2423 mysd->rps_ipi_list = sd;
2425 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2428 #endif /* CONFIG_RPS */
2433 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2434 * queue (may be a remote CPU queue).
2436 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2437 unsigned int *qtail)
2439 struct softnet_data *sd;
2440 unsigned long flags;
2442 sd = &per_cpu(softnet_data, cpu);
2444 local_irq_save(flags);
2447 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2448 if (skb_queue_len(&sd->input_pkt_queue)) {
2450 __skb_queue_tail(&sd->input_pkt_queue, skb);
2451 input_queue_tail_incr_save(sd, qtail);
2453 local_irq_restore(flags);
2454 return NET_RX_SUCCESS;
2457 /* Schedule NAPI for backlog device
2458 * We can use non atomic operation since we own the queue lock
2460 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2461 if (!rps_ipi_queued(sd))
2462 ____napi_schedule(sd, &sd->backlog);
2470 local_irq_restore(flags);
2477 * netif_rx - post buffer to the network code
2478 * @skb: buffer to post
2480 * This function receives a packet from a device driver and queues it for
2481 * the upper (protocol) levels to process. It always succeeds. The buffer
2482 * may be dropped during processing for congestion control or by the
2486 * NET_RX_SUCCESS (no congestion)
2487 * NET_RX_DROP (packet was dropped)
2491 int netif_rx(struct sk_buff *skb)
2495 /* if netpoll wants it, pretend we never saw it */
2496 if (netpoll_rx(skb))
2499 if (netdev_tstamp_prequeue)
2500 net_timestamp_check(skb);
2504 struct rps_dev_flow voidflow, *rflow = &voidflow;
2509 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2511 cpu = smp_processor_id();
2513 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2520 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2526 EXPORT_SYMBOL(netif_rx);
2528 int netif_rx_ni(struct sk_buff *skb)
2533 err = netif_rx(skb);
2534 if (local_softirq_pending())
2540 EXPORT_SYMBOL(netif_rx_ni);
2542 static void net_tx_action(struct softirq_action *h)
2544 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2546 if (sd->completion_queue) {
2547 struct sk_buff *clist;
2549 local_irq_disable();
2550 clist = sd->completion_queue;
2551 sd->completion_queue = NULL;
2555 struct sk_buff *skb = clist;
2556 clist = clist->next;
2558 WARN_ON(atomic_read(&skb->users));
2563 if (sd->output_queue) {
2566 local_irq_disable();
2567 head = sd->output_queue;
2568 sd->output_queue = NULL;
2569 sd->output_queue_tailp = &sd->output_queue;
2573 struct Qdisc *q = head;
2574 spinlock_t *root_lock;
2576 head = head->next_sched;
2578 root_lock = qdisc_lock(q);
2579 if (spin_trylock(root_lock)) {
2580 smp_mb__before_clear_bit();
2581 clear_bit(__QDISC_STATE_SCHED,
2584 spin_unlock(root_lock);
2586 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2588 __netif_reschedule(q);
2590 smp_mb__before_clear_bit();
2591 clear_bit(__QDISC_STATE_SCHED,
2599 static inline int deliver_skb(struct sk_buff *skb,
2600 struct packet_type *pt_prev,
2601 struct net_device *orig_dev)
2603 atomic_inc(&skb->users);
2604 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2607 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2609 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2610 /* This hook is defined here for ATM LANE */
2611 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2612 unsigned char *addr) __read_mostly;
2613 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2617 * If bridge module is loaded call bridging hook.
2618 * returns NULL if packet was consumed.
2620 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2621 struct sk_buff *skb) __read_mostly;
2622 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2624 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2625 struct packet_type **pt_prev, int *ret,
2626 struct net_device *orig_dev)
2628 struct net_bridge_port *port;
2630 if (skb->pkt_type == PACKET_LOOPBACK ||
2631 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2635 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2639 return br_handle_frame_hook(port, skb);
2642 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2645 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2646 struct sk_buff *(*macvlan_handle_frame_hook)(struct macvlan_port *p,
2647 struct sk_buff *skb) __read_mostly;
2648 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2650 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2651 struct packet_type **pt_prev,
2653 struct net_device *orig_dev)
2655 struct macvlan_port *port;
2657 port = rcu_dereference(skb->dev->macvlan_port);
2662 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2665 return macvlan_handle_frame_hook(port, skb);
2668 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2671 #ifdef CONFIG_NET_CLS_ACT
2672 /* TODO: Maybe we should just force sch_ingress to be compiled in
2673 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2674 * a compare and 2 stores extra right now if we dont have it on
2675 * but have CONFIG_NET_CLS_ACT
2676 * NOTE: This doesnt stop any functionality; if you dont have
2677 * the ingress scheduler, you just cant add policies on ingress.
2680 static int ing_filter(struct sk_buff *skb)
2682 struct net_device *dev = skb->dev;
2683 u32 ttl = G_TC_RTTL(skb->tc_verd);
2684 struct netdev_queue *rxq;
2685 int result = TC_ACT_OK;
2688 if (MAX_RED_LOOP < ttl++) {
2690 "Redir loop detected Dropping packet (%d->%d)\n",
2691 skb->skb_iif, dev->ifindex);
2695 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2696 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2698 rxq = &dev->rx_queue;
2701 if (q != &noop_qdisc) {
2702 spin_lock(qdisc_lock(q));
2703 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2704 result = qdisc_enqueue_root(skb, q);
2705 spin_unlock(qdisc_lock(q));
2711 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2712 struct packet_type **pt_prev,
2713 int *ret, struct net_device *orig_dev)
2715 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2719 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2722 /* Huh? Why does turning on AF_PACKET affect this? */
2723 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2726 switch (ing_filter(skb)) {
2740 * netif_nit_deliver - deliver received packets to network taps
2743 * This function is used to deliver incoming packets to network
2744 * taps. It should be used when the normal netif_receive_skb path
2745 * is bypassed, for example because of VLAN acceleration.
2747 void netif_nit_deliver(struct sk_buff *skb)
2749 struct packet_type *ptype;
2751 if (list_empty(&ptype_all))
2754 skb_reset_network_header(skb);
2755 skb_reset_transport_header(skb);
2756 skb->mac_len = skb->network_header - skb->mac_header;
2759 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2760 if (!ptype->dev || ptype->dev == skb->dev)
2761 deliver_skb(skb, ptype, skb->dev);
2766 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2767 struct net_device *master)
2769 if (skb->pkt_type == PACKET_HOST) {
2770 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2772 memcpy(dest, master->dev_addr, ETH_ALEN);
2776 /* On bonding slaves other than the currently active slave, suppress
2777 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2778 * ARP on active-backup slaves with arp_validate enabled.
2780 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2782 struct net_device *dev = skb->dev;
2784 if (master->priv_flags & IFF_MASTER_ARPMON)
2785 dev->last_rx = jiffies;
2787 if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) {
2788 /* Do address unmangle. The local destination address
2789 * will be always the one master has. Provides the right
2790 * functionality in a bridge.
2792 skb_bond_set_mac_by_master(skb, master);
2795 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2796 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2797 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2800 if (master->priv_flags & IFF_MASTER_ALB) {
2801 if (skb->pkt_type != PACKET_BROADCAST &&
2802 skb->pkt_type != PACKET_MULTICAST)
2805 if (master->priv_flags & IFF_MASTER_8023AD &&
2806 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2813 EXPORT_SYMBOL(__skb_bond_should_drop);
2815 static int __netif_receive_skb(struct sk_buff *skb)
2817 struct packet_type *ptype, *pt_prev;
2818 struct net_device *orig_dev;
2819 struct net_device *master;
2820 struct net_device *null_or_orig;
2821 struct net_device *null_or_bond;
2822 int ret = NET_RX_DROP;
2825 if (!netdev_tstamp_prequeue)
2826 net_timestamp_check(skb);
2828 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2829 return NET_RX_SUCCESS;
2831 /* if we've gotten here through NAPI, check netpoll */
2832 if (netpoll_receive_skb(skb))
2836 skb->skb_iif = skb->dev->ifindex;
2838 null_or_orig = NULL;
2839 orig_dev = skb->dev;
2840 master = ACCESS_ONCE(orig_dev->master);
2842 if (skb_bond_should_drop(skb, master))
2843 null_or_orig = orig_dev; /* deliver only exact match */
2848 __this_cpu_inc(softnet_data.processed);
2849 skb_reset_network_header(skb);
2850 skb_reset_transport_header(skb);
2851 skb->mac_len = skb->network_header - skb->mac_header;
2857 #ifdef CONFIG_NET_CLS_ACT
2858 if (skb->tc_verd & TC_NCLS) {
2859 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2864 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2865 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2866 ptype->dev == orig_dev) {
2868 ret = deliver_skb(skb, pt_prev, orig_dev);
2873 #ifdef CONFIG_NET_CLS_ACT
2874 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2880 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2883 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2888 * Make sure frames received on VLAN interfaces stacked on
2889 * bonding interfaces still make their way to any base bonding
2890 * device that may have registered for a specific ptype. The
2891 * handler may have to adjust skb->dev and orig_dev.
2893 null_or_bond = NULL;
2894 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2895 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2896 null_or_bond = vlan_dev_real_dev(skb->dev);
2899 type = skb->protocol;
2900 list_for_each_entry_rcu(ptype,
2901 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2902 if (ptype->type == type && (ptype->dev == null_or_orig ||
2903 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2904 ptype->dev == null_or_bond)) {
2906 ret = deliver_skb(skb, pt_prev, orig_dev);
2912 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2915 /* Jamal, now you will not able to escape explaining
2916 * me how you were going to use this. :-)
2927 * netif_receive_skb - process receive buffer from network
2928 * @skb: buffer to process
2930 * netif_receive_skb() is the main receive data processing function.
2931 * It always succeeds. The buffer may be dropped during processing
2932 * for congestion control or by the protocol layers.
2934 * This function may only be called from softirq context and interrupts
2935 * should be enabled.
2937 * Return values (usually ignored):
2938 * NET_RX_SUCCESS: no congestion
2939 * NET_RX_DROP: packet was dropped
2941 int netif_receive_skb(struct sk_buff *skb)
2943 if (netdev_tstamp_prequeue)
2944 net_timestamp_check(skb);
2948 struct rps_dev_flow voidflow, *rflow = &voidflow;
2953 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2956 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2960 ret = __netif_receive_skb(skb);
2966 return __netif_receive_skb(skb);
2969 EXPORT_SYMBOL(netif_receive_skb);
2971 /* Network device is going away, flush any packets still pending
2972 * Called with irqs disabled.
2974 static void flush_backlog(void *arg)
2976 struct net_device *dev = arg;
2977 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2978 struct sk_buff *skb, *tmp;
2981 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
2982 if (skb->dev == dev) {
2983 __skb_unlink(skb, &sd->input_pkt_queue);
2985 input_queue_head_incr(sd);
2990 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
2991 if (skb->dev == dev) {
2992 __skb_unlink(skb, &sd->process_queue);
2994 input_queue_head_incr(sd);
2999 static int napi_gro_complete(struct sk_buff *skb)
3001 struct packet_type *ptype;
3002 __be16 type = skb->protocol;
3003 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3006 if (NAPI_GRO_CB(skb)->count == 1) {
3007 skb_shinfo(skb)->gso_size = 0;
3012 list_for_each_entry_rcu(ptype, head, list) {
3013 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3016 err = ptype->gro_complete(skb);
3022 WARN_ON(&ptype->list == head);
3024 return NET_RX_SUCCESS;
3028 return netif_receive_skb(skb);
3031 static void napi_gro_flush(struct napi_struct *napi)
3033 struct sk_buff *skb, *next;
3035 for (skb = napi->gro_list; skb; skb = next) {
3038 napi_gro_complete(skb);
3041 napi->gro_count = 0;
3042 napi->gro_list = NULL;
3045 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3047 struct sk_buff **pp = NULL;
3048 struct packet_type *ptype;
3049 __be16 type = skb->protocol;
3050 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3053 enum gro_result ret;
3055 if (!(skb->dev->features & NETIF_F_GRO))
3058 if (skb_is_gso(skb) || skb_has_frags(skb))
3062 list_for_each_entry_rcu(ptype, head, list) {
3063 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3066 skb_set_network_header(skb, skb_gro_offset(skb));
3067 mac_len = skb->network_header - skb->mac_header;
3068 skb->mac_len = mac_len;
3069 NAPI_GRO_CB(skb)->same_flow = 0;
3070 NAPI_GRO_CB(skb)->flush = 0;
3071 NAPI_GRO_CB(skb)->free = 0;
3073 pp = ptype->gro_receive(&napi->gro_list, skb);
3078 if (&ptype->list == head)
3081 same_flow = NAPI_GRO_CB(skb)->same_flow;
3082 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3085 struct sk_buff *nskb = *pp;
3089 napi_gro_complete(nskb);
3096 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3100 NAPI_GRO_CB(skb)->count = 1;
3101 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3102 skb->next = napi->gro_list;
3103 napi->gro_list = skb;
3107 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3108 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3110 BUG_ON(skb->end - skb->tail < grow);
3112 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3115 skb->data_len -= grow;
3117 skb_shinfo(skb)->frags[0].page_offset += grow;
3118 skb_shinfo(skb)->frags[0].size -= grow;
3120 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3121 put_page(skb_shinfo(skb)->frags[0].page);
3122 memmove(skb_shinfo(skb)->frags,
3123 skb_shinfo(skb)->frags + 1,
3124 --skb_shinfo(skb)->nr_frags);
3135 EXPORT_SYMBOL(dev_gro_receive);
3138 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3142 if (netpoll_rx_on(skb))
3145 for (p = napi->gro_list; p; p = p->next) {
3146 NAPI_GRO_CB(p)->same_flow =
3147 (p->dev == skb->dev) &&
3148 !compare_ether_header(skb_mac_header(p),
3149 skb_gro_mac_header(skb));
3150 NAPI_GRO_CB(p)->flush = 0;
3153 return dev_gro_receive(napi, skb);
3156 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3160 if (netif_receive_skb(skb))
3165 case GRO_MERGED_FREE:
3176 EXPORT_SYMBOL(napi_skb_finish);
3178 void skb_gro_reset_offset(struct sk_buff *skb)
3180 NAPI_GRO_CB(skb)->data_offset = 0;
3181 NAPI_GRO_CB(skb)->frag0 = NULL;
3182 NAPI_GRO_CB(skb)->frag0_len = 0;
3184 if (skb->mac_header == skb->tail &&
3185 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3186 NAPI_GRO_CB(skb)->frag0 =
3187 page_address(skb_shinfo(skb)->frags[0].page) +
3188 skb_shinfo(skb)->frags[0].page_offset;
3189 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3192 EXPORT_SYMBOL(skb_gro_reset_offset);
3194 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3196 skb_gro_reset_offset(skb);
3198 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3200 EXPORT_SYMBOL(napi_gro_receive);
3202 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3204 __skb_pull(skb, skb_headlen(skb));
3205 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3209 EXPORT_SYMBOL(napi_reuse_skb);
3211 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3213 struct sk_buff *skb = napi->skb;
3216 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3222 EXPORT_SYMBOL(napi_get_frags);
3224 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3230 skb->protocol = eth_type_trans(skb, skb->dev);
3232 if (ret == GRO_HELD)
3233 skb_gro_pull(skb, -ETH_HLEN);
3234 else if (netif_receive_skb(skb))
3239 case GRO_MERGED_FREE:
3240 napi_reuse_skb(napi, skb);
3249 EXPORT_SYMBOL(napi_frags_finish);
3251 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3253 struct sk_buff *skb = napi->skb;
3260 skb_reset_mac_header(skb);
3261 skb_gro_reset_offset(skb);
3263 off = skb_gro_offset(skb);
3264 hlen = off + sizeof(*eth);
3265 eth = skb_gro_header_fast(skb, off);
3266 if (skb_gro_header_hard(skb, hlen)) {
3267 eth = skb_gro_header_slow(skb, hlen, off);
3268 if (unlikely(!eth)) {
3269 napi_reuse_skb(napi, skb);
3275 skb_gro_pull(skb, sizeof(*eth));
3278 * This works because the only protocols we care about don't require
3279 * special handling. We'll fix it up properly at the end.
3281 skb->protocol = eth->h_proto;
3286 EXPORT_SYMBOL(napi_frags_skb);
3288 gro_result_t napi_gro_frags(struct napi_struct *napi)
3290 struct sk_buff *skb = napi_frags_skb(napi);
3295 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3297 EXPORT_SYMBOL(napi_gro_frags);
3300 * net_rps_action sends any pending IPI's for rps.
3301 * Note: called with local irq disabled, but exits with local irq enabled.
3303 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3306 struct softnet_data *remsd = sd->rps_ipi_list;
3309 sd->rps_ipi_list = NULL;
3313 /* Send pending IPI's to kick RPS processing on remote cpus. */
3315 struct softnet_data *next = remsd->rps_ipi_next;
3317 if (cpu_online(remsd->cpu))
3318 __smp_call_function_single(remsd->cpu,
3327 static int process_backlog(struct napi_struct *napi, int quota)
3330 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3333 /* Check if we have pending ipi, its better to send them now,
3334 * not waiting net_rx_action() end.
3336 if (sd->rps_ipi_list) {
3337 local_irq_disable();
3338 net_rps_action_and_irq_enable(sd);
3341 napi->weight = weight_p;
3342 local_irq_disable();
3343 while (work < quota) {
3344 struct sk_buff *skb;
3347 while ((skb = __skb_dequeue(&sd->process_queue))) {
3349 __netif_receive_skb(skb);
3350 local_irq_disable();
3351 input_queue_head_incr(sd);
3352 if (++work >= quota) {
3359 qlen = skb_queue_len(&sd->input_pkt_queue);
3361 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3362 &sd->process_queue);
3364 if (qlen < quota - work) {
3366 * Inline a custom version of __napi_complete().
3367 * only current cpu owns and manipulates this napi,
3368 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3369 * we can use a plain write instead of clear_bit(),
3370 * and we dont need an smp_mb() memory barrier.
3372 list_del(&napi->poll_list);
3375 quota = work + qlen;
3385 * __napi_schedule - schedule for receive
3386 * @n: entry to schedule
3388 * The entry's receive function will be scheduled to run
3390 void __napi_schedule(struct napi_struct *n)
3392 unsigned long flags;
3394 local_irq_save(flags);
3395 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3396 local_irq_restore(flags);
3398 EXPORT_SYMBOL(__napi_schedule);
3400 void __napi_complete(struct napi_struct *n)
3402 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3403 BUG_ON(n->gro_list);
3405 list_del(&n->poll_list);
3406 smp_mb__before_clear_bit();
3407 clear_bit(NAPI_STATE_SCHED, &n->state);
3409 EXPORT_SYMBOL(__napi_complete);
3411 void napi_complete(struct napi_struct *n)
3413 unsigned long flags;
3416 * don't let napi dequeue from the cpu poll list
3417 * just in case its running on a different cpu
3419 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3423 local_irq_save(flags);
3425 local_irq_restore(flags);
3427 EXPORT_SYMBOL(napi_complete);
3429 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3430 int (*poll)(struct napi_struct *, int), int weight)
3432 INIT_LIST_HEAD(&napi->poll_list);
3433 napi->gro_count = 0;
3434 napi->gro_list = NULL;
3437 napi->weight = weight;
3438 list_add(&napi->dev_list, &dev->napi_list);
3440 #ifdef CONFIG_NETPOLL
3441 spin_lock_init(&napi->poll_lock);
3442 napi->poll_owner = -1;
3444 set_bit(NAPI_STATE_SCHED, &napi->state);
3446 EXPORT_SYMBOL(netif_napi_add);
3448 void netif_napi_del(struct napi_struct *napi)
3450 struct sk_buff *skb, *next;
3452 list_del_init(&napi->dev_list);
3453 napi_free_frags(napi);
3455 for (skb = napi->gro_list; skb; skb = next) {
3461 napi->gro_list = NULL;
3462 napi->gro_count = 0;
3464 EXPORT_SYMBOL(netif_napi_del);
3466 static void net_rx_action(struct softirq_action *h)
3468 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3469 unsigned long time_limit = jiffies + 2;
3470 int budget = netdev_budget;
3473 local_irq_disable();
3475 while (!list_empty(&sd->poll_list)) {
3476 struct napi_struct *n;
3479 /* If softirq window is exhuasted then punt.
3480 * Allow this to run for 2 jiffies since which will allow
3481 * an average latency of 1.5/HZ.
3483 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3488 /* Even though interrupts have been re-enabled, this
3489 * access is safe because interrupts can only add new
3490 * entries to the tail of this list, and only ->poll()
3491 * calls can remove this head entry from the list.
3493 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3495 have = netpoll_poll_lock(n);
3499 /* This NAPI_STATE_SCHED test is for avoiding a race
3500 * with netpoll's poll_napi(). Only the entity which
3501 * obtains the lock and sees NAPI_STATE_SCHED set will
3502 * actually make the ->poll() call. Therefore we avoid
3503 * accidently calling ->poll() when NAPI is not scheduled.
3506 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3507 work = n->poll(n, weight);
3511 WARN_ON_ONCE(work > weight);
3515 local_irq_disable();
3517 /* Drivers must not modify the NAPI state if they
3518 * consume the entire weight. In such cases this code
3519 * still "owns" the NAPI instance and therefore can
3520 * move the instance around on the list at-will.
3522 if (unlikely(work == weight)) {
3523 if (unlikely(napi_disable_pending(n))) {
3526 local_irq_disable();
3528 list_move_tail(&n->poll_list, &sd->poll_list);
3531 netpoll_poll_unlock(have);
3534 net_rps_action_and_irq_enable(sd);
3536 #ifdef CONFIG_NET_DMA
3538 * There may not be any more sk_buffs coming right now, so push
3539 * any pending DMA copies to hardware
3541 dma_issue_pending_all();
3548 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3552 static gifconf_func_t *gifconf_list[NPROTO];
3555 * register_gifconf - register a SIOCGIF handler
3556 * @family: Address family
3557 * @gifconf: Function handler
3559 * Register protocol dependent address dumping routines. The handler
3560 * that is passed must not be freed or reused until it has been replaced
3561 * by another handler.
3563 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3565 if (family >= NPROTO)
3567 gifconf_list[family] = gifconf;
3570 EXPORT_SYMBOL(register_gifconf);
3574 * Map an interface index to its name (SIOCGIFNAME)
3578 * We need this ioctl for efficient implementation of the
3579 * if_indextoname() function required by the IPv6 API. Without
3580 * it, we would have to search all the interfaces to find a
3584 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3586 struct net_device *dev;
3590 * Fetch the caller's info block.
3593 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3597 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3603 strcpy(ifr.ifr_name, dev->name);
3606 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3612 * Perform a SIOCGIFCONF call. This structure will change
3613 * size eventually, and there is nothing I can do about it.
3614 * Thus we will need a 'compatibility mode'.
3617 static int dev_ifconf(struct net *net, char __user *arg)
3620 struct net_device *dev;
3627 * Fetch the caller's info block.
3630 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3637 * Loop over the interfaces, and write an info block for each.
3641 for_each_netdev(net, dev) {
3642 for (i = 0; i < NPROTO; i++) {
3643 if (gifconf_list[i]) {
3646 done = gifconf_list[i](dev, NULL, 0);
3648 done = gifconf_list[i](dev, pos + total,
3658 * All done. Write the updated control block back to the caller.
3660 ifc.ifc_len = total;
3663 * Both BSD and Solaris return 0 here, so we do too.
3665 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3668 #ifdef CONFIG_PROC_FS
3670 * This is invoked by the /proc filesystem handler to display a device
3673 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3676 struct net *net = seq_file_net(seq);
3678 struct net_device *dev;
3682 return SEQ_START_TOKEN;
3685 for_each_netdev_rcu(net, dev)
3692 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3694 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3695 first_net_device(seq_file_net(seq)) :
3696 next_net_device((struct net_device *)v);
3699 return rcu_dereference(dev);
3702 void dev_seq_stop(struct seq_file *seq, void *v)
3708 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3710 const struct net_device_stats *stats = dev_get_stats(dev);
3712 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3713 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3714 dev->name, stats->rx_bytes, stats->rx_packets,
3716 stats->rx_dropped + stats->rx_missed_errors,
3717 stats->rx_fifo_errors,
3718 stats->rx_length_errors + stats->rx_over_errors +
3719 stats->rx_crc_errors + stats->rx_frame_errors,
3720 stats->rx_compressed, stats->multicast,
3721 stats->tx_bytes, stats->tx_packets,
3722 stats->tx_errors, stats->tx_dropped,
3723 stats->tx_fifo_errors, stats->collisions,
3724 stats->tx_carrier_errors +
3725 stats->tx_aborted_errors +
3726 stats->tx_window_errors +
3727 stats->tx_heartbeat_errors,
3728 stats->tx_compressed);
3732 * Called from the PROCfs module. This now uses the new arbitrary sized
3733 * /proc/net interface to create /proc/net/dev
3735 static int dev_seq_show(struct seq_file *seq, void *v)
3737 if (v == SEQ_START_TOKEN)
3738 seq_puts(seq, "Inter-| Receive "
3740 " face |bytes packets errs drop fifo frame "
3741 "compressed multicast|bytes packets errs "
3742 "drop fifo colls carrier compressed\n");
3744 dev_seq_printf_stats(seq, v);
3748 static struct softnet_data *softnet_get_online(loff_t *pos)
3750 struct softnet_data *sd = NULL;
3752 while (*pos < nr_cpu_ids)
3753 if (cpu_online(*pos)) {
3754 sd = &per_cpu(softnet_data, *pos);
3761 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3763 return softnet_get_online(pos);
3766 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3769 return softnet_get_online(pos);
3772 static void softnet_seq_stop(struct seq_file *seq, void *v)
3776 static int softnet_seq_show(struct seq_file *seq, void *v)
3778 struct softnet_data *sd = v;
3780 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3781 sd->processed, sd->dropped, sd->time_squeeze, 0,
3782 0, 0, 0, 0, /* was fastroute */
3783 sd->cpu_collision, sd->received_rps);
3787 static const struct seq_operations dev_seq_ops = {
3788 .start = dev_seq_start,
3789 .next = dev_seq_next,
3790 .stop = dev_seq_stop,
3791 .show = dev_seq_show,
3794 static int dev_seq_open(struct inode *inode, struct file *file)
3796 return seq_open_net(inode, file, &dev_seq_ops,
3797 sizeof(struct seq_net_private));
3800 static const struct file_operations dev_seq_fops = {
3801 .owner = THIS_MODULE,
3802 .open = dev_seq_open,
3804 .llseek = seq_lseek,
3805 .release = seq_release_net,
3808 static const struct seq_operations softnet_seq_ops = {
3809 .start = softnet_seq_start,
3810 .next = softnet_seq_next,
3811 .stop = softnet_seq_stop,
3812 .show = softnet_seq_show,
3815 static int softnet_seq_open(struct inode *inode, struct file *file)
3817 return seq_open(file, &softnet_seq_ops);
3820 static const struct file_operations softnet_seq_fops = {
3821 .owner = THIS_MODULE,
3822 .open = softnet_seq_open,
3824 .llseek = seq_lseek,
3825 .release = seq_release,
3828 static void *ptype_get_idx(loff_t pos)
3830 struct packet_type *pt = NULL;
3834 list_for_each_entry_rcu(pt, &ptype_all, list) {
3840 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3841 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3850 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3854 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3857 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3859 struct packet_type *pt;
3860 struct list_head *nxt;
3864 if (v == SEQ_START_TOKEN)
3865 return ptype_get_idx(0);
3868 nxt = pt->list.next;
3869 if (pt->type == htons(ETH_P_ALL)) {
3870 if (nxt != &ptype_all)
3873 nxt = ptype_base[0].next;
3875 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3877 while (nxt == &ptype_base[hash]) {
3878 if (++hash >= PTYPE_HASH_SIZE)
3880 nxt = ptype_base[hash].next;
3883 return list_entry(nxt, struct packet_type, list);
3886 static void ptype_seq_stop(struct seq_file *seq, void *v)
3892 static int ptype_seq_show(struct seq_file *seq, void *v)
3894 struct packet_type *pt = v;
3896 if (v == SEQ_START_TOKEN)
3897 seq_puts(seq, "Type Device Function\n");
3898 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3899 if (pt->type == htons(ETH_P_ALL))
3900 seq_puts(seq, "ALL ");
3902 seq_printf(seq, "%04x", ntohs(pt->type));
3904 seq_printf(seq, " %-8s %pF\n",
3905 pt->dev ? pt->dev->name : "", pt->func);
3911 static const struct seq_operations ptype_seq_ops = {
3912 .start = ptype_seq_start,
3913 .next = ptype_seq_next,
3914 .stop = ptype_seq_stop,
3915 .show = ptype_seq_show,
3918 static int ptype_seq_open(struct inode *inode, struct file *file)
3920 return seq_open_net(inode, file, &ptype_seq_ops,
3921 sizeof(struct seq_net_private));
3924 static const struct file_operations ptype_seq_fops = {
3925 .owner = THIS_MODULE,
3926 .open = ptype_seq_open,
3928 .llseek = seq_lseek,
3929 .release = seq_release_net,
3933 static int __net_init dev_proc_net_init(struct net *net)
3937 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3939 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3941 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3944 if (wext_proc_init(net))
3950 proc_net_remove(net, "ptype");
3952 proc_net_remove(net, "softnet_stat");
3954 proc_net_remove(net, "dev");
3958 static void __net_exit dev_proc_net_exit(struct net *net)
3960 wext_proc_exit(net);
3962 proc_net_remove(net, "ptype");
3963 proc_net_remove(net, "softnet_stat");
3964 proc_net_remove(net, "dev");
3967 static struct pernet_operations __net_initdata dev_proc_ops = {
3968 .init = dev_proc_net_init,
3969 .exit = dev_proc_net_exit,
3972 static int __init dev_proc_init(void)
3974 return register_pernet_subsys(&dev_proc_ops);
3977 #define dev_proc_init() 0
3978 #endif /* CONFIG_PROC_FS */
3982 * netdev_set_master - set up master/slave pair
3983 * @slave: slave device
3984 * @master: new master device
3986 * Changes the master device of the slave. Pass %NULL to break the
3987 * bonding. The caller must hold the RTNL semaphore. On a failure
3988 * a negative errno code is returned. On success the reference counts
3989 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3990 * function returns zero.
3992 int netdev_set_master(struct net_device *slave, struct net_device *master)
3994 struct net_device *old = slave->master;
4004 slave->master = master;
4011 slave->flags |= IFF_SLAVE;
4013 slave->flags &= ~IFF_SLAVE;
4015 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4018 EXPORT_SYMBOL(netdev_set_master);
4020 static void dev_change_rx_flags(struct net_device *dev, int flags)
4022 const struct net_device_ops *ops = dev->netdev_ops;
4024 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4025 ops->ndo_change_rx_flags(dev, flags);
4028 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4030 unsigned short old_flags = dev->flags;
4036 dev->flags |= IFF_PROMISC;
4037 dev->promiscuity += inc;
4038 if (dev->promiscuity == 0) {
4041 * If inc causes overflow, untouch promisc and return error.
4044 dev->flags &= ~IFF_PROMISC;
4046 dev->promiscuity -= inc;
4047 printk(KERN_WARNING "%s: promiscuity touches roof, "
4048 "set promiscuity failed, promiscuity feature "
4049 "of device might be broken.\n", dev->name);
4053 if (dev->flags != old_flags) {
4054 printk(KERN_INFO "device %s %s promiscuous mode\n",
4055 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4057 if (audit_enabled) {
4058 current_uid_gid(&uid, &gid);
4059 audit_log(current->audit_context, GFP_ATOMIC,
4060 AUDIT_ANOM_PROMISCUOUS,
4061 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4062 dev->name, (dev->flags & IFF_PROMISC),
4063 (old_flags & IFF_PROMISC),
4064 audit_get_loginuid(current),
4066 audit_get_sessionid(current));
4069 dev_change_rx_flags(dev, IFF_PROMISC);
4075 * dev_set_promiscuity - update promiscuity count on a device
4079 * Add or remove promiscuity from a device. While the count in the device
4080 * remains above zero the interface remains promiscuous. Once it hits zero
4081 * the device reverts back to normal filtering operation. A negative inc
4082 * value is used to drop promiscuity on the device.
4083 * Return 0 if successful or a negative errno code on error.
4085 int dev_set_promiscuity(struct net_device *dev, int inc)
4087 unsigned short old_flags = dev->flags;
4090 err = __dev_set_promiscuity(dev, inc);
4093 if (dev->flags != old_flags)
4094 dev_set_rx_mode(dev);
4097 EXPORT_SYMBOL(dev_set_promiscuity);
4100 * dev_set_allmulti - update allmulti count on a device
4104 * Add or remove reception of all multicast frames to a device. While the
4105 * count in the device remains above zero the interface remains listening
4106 * to all interfaces. Once it hits zero the device reverts back to normal
4107 * filtering operation. A negative @inc value is used to drop the counter
4108 * when releasing a resource needing all multicasts.
4109 * Return 0 if successful or a negative errno code on error.
4112 int dev_set_allmulti(struct net_device *dev, int inc)
4114 unsigned short old_flags = dev->flags;
4118 dev->flags |= IFF_ALLMULTI;
4119 dev->allmulti += inc;
4120 if (dev->allmulti == 0) {
4123 * If inc causes overflow, untouch allmulti and return error.
4126 dev->flags &= ~IFF_ALLMULTI;
4128 dev->allmulti -= inc;
4129 printk(KERN_WARNING "%s: allmulti touches roof, "
4130 "set allmulti failed, allmulti feature of "
4131 "device might be broken.\n", dev->name);
4135 if (dev->flags ^ old_flags) {
4136 dev_change_rx_flags(dev, IFF_ALLMULTI);
4137 dev_set_rx_mode(dev);
4141 EXPORT_SYMBOL(dev_set_allmulti);
4144 * Upload unicast and multicast address lists to device and
4145 * configure RX filtering. When the device doesn't support unicast
4146 * filtering it is put in promiscuous mode while unicast addresses
4149 void __dev_set_rx_mode(struct net_device *dev)
4151 const struct net_device_ops *ops = dev->netdev_ops;
4153 /* dev_open will call this function so the list will stay sane. */
4154 if (!(dev->flags&IFF_UP))
4157 if (!netif_device_present(dev))
4160 if (ops->ndo_set_rx_mode)
4161 ops->ndo_set_rx_mode(dev);
4163 /* Unicast addresses changes may only happen under the rtnl,
4164 * therefore calling __dev_set_promiscuity here is safe.
4166 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4167 __dev_set_promiscuity(dev, 1);
4168 dev->uc_promisc = 1;
4169 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4170 __dev_set_promiscuity(dev, -1);
4171 dev->uc_promisc = 0;
4174 if (ops->ndo_set_multicast_list)
4175 ops->ndo_set_multicast_list(dev);
4179 void dev_set_rx_mode(struct net_device *dev)
4181 netif_addr_lock_bh(dev);
4182 __dev_set_rx_mode(dev);
4183 netif_addr_unlock_bh(dev);
4187 * dev_get_flags - get flags reported to userspace
4190 * Get the combination of flag bits exported through APIs to userspace.
4192 unsigned dev_get_flags(const struct net_device *dev)
4196 flags = (dev->flags & ~(IFF_PROMISC |
4201 (dev->gflags & (IFF_PROMISC |
4204 if (netif_running(dev)) {
4205 if (netif_oper_up(dev))
4206 flags |= IFF_RUNNING;
4207 if (netif_carrier_ok(dev))
4208 flags |= IFF_LOWER_UP;
4209 if (netif_dormant(dev))
4210 flags |= IFF_DORMANT;
4215 EXPORT_SYMBOL(dev_get_flags);
4217 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4219 int old_flags = dev->flags;
4225 * Set the flags on our device.
4228 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4229 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4231 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4235 * Load in the correct multicast list now the flags have changed.
4238 if ((old_flags ^ flags) & IFF_MULTICAST)
4239 dev_change_rx_flags(dev, IFF_MULTICAST);
4241 dev_set_rx_mode(dev);
4244 * Have we downed the interface. We handle IFF_UP ourselves
4245 * according to user attempts to set it, rather than blindly
4250 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4251 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4254 dev_set_rx_mode(dev);
4257 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4258 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4260 dev->gflags ^= IFF_PROMISC;
4261 dev_set_promiscuity(dev, inc);
4264 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4265 is important. Some (broken) drivers set IFF_PROMISC, when
4266 IFF_ALLMULTI is requested not asking us and not reporting.
4268 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4269 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4271 dev->gflags ^= IFF_ALLMULTI;
4272 dev_set_allmulti(dev, inc);
4278 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4280 unsigned int changes = dev->flags ^ old_flags;
4282 if (changes & IFF_UP) {
4283 if (dev->flags & IFF_UP)
4284 call_netdevice_notifiers(NETDEV_UP, dev);
4286 call_netdevice_notifiers(NETDEV_DOWN, dev);
4289 if (dev->flags & IFF_UP &&
4290 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4291 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4295 * dev_change_flags - change device settings
4297 * @flags: device state flags
4299 * Change settings on device based state flags. The flags are
4300 * in the userspace exported format.
4302 int dev_change_flags(struct net_device *dev, unsigned flags)
4305 int old_flags = dev->flags;
4307 ret = __dev_change_flags(dev, flags);
4311 changes = old_flags ^ dev->flags;
4313 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4315 __dev_notify_flags(dev, old_flags);
4318 EXPORT_SYMBOL(dev_change_flags);
4321 * dev_set_mtu - Change maximum transfer unit
4323 * @new_mtu: new transfer unit
4325 * Change the maximum transfer size of the network device.
4327 int dev_set_mtu(struct net_device *dev, int new_mtu)
4329 const struct net_device_ops *ops = dev->netdev_ops;
4332 if (new_mtu == dev->mtu)
4335 /* MTU must be positive. */
4339 if (!netif_device_present(dev))
4343 if (ops->ndo_change_mtu)
4344 err = ops->ndo_change_mtu(dev, new_mtu);
4348 if (!err && dev->flags & IFF_UP)
4349 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4352 EXPORT_SYMBOL(dev_set_mtu);
4355 * dev_set_mac_address - Change Media Access Control Address
4359 * Change the hardware (MAC) address of the device
4361 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4363 const struct net_device_ops *ops = dev->netdev_ops;
4366 if (!ops->ndo_set_mac_address)
4368 if (sa->sa_family != dev->type)
4370 if (!netif_device_present(dev))
4372 err = ops->ndo_set_mac_address(dev, sa);
4374 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4377 EXPORT_SYMBOL(dev_set_mac_address);
4380 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4382 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4385 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4391 case SIOCGIFFLAGS: /* Get interface flags */
4392 ifr->ifr_flags = (short) dev_get_flags(dev);
4395 case SIOCGIFMETRIC: /* Get the metric on the interface
4396 (currently unused) */
4397 ifr->ifr_metric = 0;
4400 case SIOCGIFMTU: /* Get the MTU of a device */
4401 ifr->ifr_mtu = dev->mtu;
4406 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4408 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4409 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4410 ifr->ifr_hwaddr.sa_family = dev->type;
4418 ifr->ifr_map.mem_start = dev->mem_start;
4419 ifr->ifr_map.mem_end = dev->mem_end;
4420 ifr->ifr_map.base_addr = dev->base_addr;
4421 ifr->ifr_map.irq = dev->irq;
4422 ifr->ifr_map.dma = dev->dma;
4423 ifr->ifr_map.port = dev->if_port;
4427 ifr->ifr_ifindex = dev->ifindex;
4431 ifr->ifr_qlen = dev->tx_queue_len;
4435 /* dev_ioctl() should ensure this case
4447 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4449 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4452 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4453 const struct net_device_ops *ops;
4458 ops = dev->netdev_ops;
4461 case SIOCSIFFLAGS: /* Set interface flags */
4462 return dev_change_flags(dev, ifr->ifr_flags);
4464 case SIOCSIFMETRIC: /* Set the metric on the interface
4465 (currently unused) */
4468 case SIOCSIFMTU: /* Set the MTU of a device */
4469 return dev_set_mtu(dev, ifr->ifr_mtu);
4472 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4474 case SIOCSIFHWBROADCAST:
4475 if (ifr->ifr_hwaddr.sa_family != dev->type)
4477 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4478 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4479 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4483 if (ops->ndo_set_config) {
4484 if (!netif_device_present(dev))
4486 return ops->ndo_set_config(dev, &ifr->ifr_map);
4491 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4492 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4494 if (!netif_device_present(dev))
4496 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4499 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4500 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4502 if (!netif_device_present(dev))
4504 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4507 if (ifr->ifr_qlen < 0)
4509 dev->tx_queue_len = ifr->ifr_qlen;
4513 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4514 return dev_change_name(dev, ifr->ifr_newname);
4517 * Unknown or private ioctl
4520 if ((cmd >= SIOCDEVPRIVATE &&
4521 cmd <= SIOCDEVPRIVATE + 15) ||
4522 cmd == SIOCBONDENSLAVE ||
4523 cmd == SIOCBONDRELEASE ||
4524 cmd == SIOCBONDSETHWADDR ||
4525 cmd == SIOCBONDSLAVEINFOQUERY ||
4526 cmd == SIOCBONDINFOQUERY ||
4527 cmd == SIOCBONDCHANGEACTIVE ||
4528 cmd == SIOCGMIIPHY ||
4529 cmd == SIOCGMIIREG ||
4530 cmd == SIOCSMIIREG ||
4531 cmd == SIOCBRADDIF ||
4532 cmd == SIOCBRDELIF ||
4533 cmd == SIOCSHWTSTAMP ||
4534 cmd == SIOCWANDEV) {
4536 if (ops->ndo_do_ioctl) {
4537 if (netif_device_present(dev))
4538 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4550 * This function handles all "interface"-type I/O control requests. The actual
4551 * 'doing' part of this is dev_ifsioc above.
4555 * dev_ioctl - network device ioctl
4556 * @net: the applicable net namespace
4557 * @cmd: command to issue
4558 * @arg: pointer to a struct ifreq in user space
4560 * Issue ioctl functions to devices. This is normally called by the
4561 * user space syscall interfaces but can sometimes be useful for
4562 * other purposes. The return value is the return from the syscall if
4563 * positive or a negative errno code on error.
4566 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4572 /* One special case: SIOCGIFCONF takes ifconf argument
4573 and requires shared lock, because it sleeps writing
4577 if (cmd == SIOCGIFCONF) {
4579 ret = dev_ifconf(net, (char __user *) arg);
4583 if (cmd == SIOCGIFNAME)
4584 return dev_ifname(net, (struct ifreq __user *)arg);
4586 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4589 ifr.ifr_name[IFNAMSIZ-1] = 0;
4591 colon = strchr(ifr.ifr_name, ':');
4596 * See which interface the caller is talking about.
4601 * These ioctl calls:
4602 * - can be done by all.
4603 * - atomic and do not require locking.
4614 dev_load(net, ifr.ifr_name);
4616 ret = dev_ifsioc_locked(net, &ifr, cmd);
4621 if (copy_to_user(arg, &ifr,
4622 sizeof(struct ifreq)))
4628 dev_load(net, ifr.ifr_name);
4630 ret = dev_ethtool(net, &ifr);
4635 if (copy_to_user(arg, &ifr,
4636 sizeof(struct ifreq)))
4642 * These ioctl calls:
4643 * - require superuser power.
4644 * - require strict serialization.
4650 if (!capable(CAP_NET_ADMIN))
4652 dev_load(net, ifr.ifr_name);
4654 ret = dev_ifsioc(net, &ifr, cmd);
4659 if (copy_to_user(arg, &ifr,
4660 sizeof(struct ifreq)))
4666 * These ioctl calls:
4667 * - require superuser power.
4668 * - require strict serialization.
4669 * - do not return a value
4679 case SIOCSIFHWBROADCAST:
4682 case SIOCBONDENSLAVE:
4683 case SIOCBONDRELEASE:
4684 case SIOCBONDSETHWADDR:
4685 case SIOCBONDCHANGEACTIVE:
4689 if (!capable(CAP_NET_ADMIN))
4692 case SIOCBONDSLAVEINFOQUERY:
4693 case SIOCBONDINFOQUERY:
4694 dev_load(net, ifr.ifr_name);
4696 ret = dev_ifsioc(net, &ifr, cmd);
4701 /* Get the per device memory space. We can add this but
4702 * currently do not support it */
4704 /* Set the per device memory buffer space.
4705 * Not applicable in our case */
4710 * Unknown or private ioctl.
4713 if (cmd == SIOCWANDEV ||
4714 (cmd >= SIOCDEVPRIVATE &&
4715 cmd <= SIOCDEVPRIVATE + 15)) {
4716 dev_load(net, ifr.ifr_name);
4718 ret = dev_ifsioc(net, &ifr, cmd);
4720 if (!ret && copy_to_user(arg, &ifr,
4721 sizeof(struct ifreq)))
4725 /* Take care of Wireless Extensions */
4726 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4727 return wext_handle_ioctl(net, &ifr, cmd, arg);
4734 * dev_new_index - allocate an ifindex
4735 * @net: the applicable net namespace
4737 * Returns a suitable unique value for a new device interface
4738 * number. The caller must hold the rtnl semaphore or the
4739 * dev_base_lock to be sure it remains unique.
4741 static int dev_new_index(struct net *net)
4747 if (!__dev_get_by_index(net, ifindex))
4752 /* Delayed registration/unregisteration */
4753 static LIST_HEAD(net_todo_list);
4755 static void net_set_todo(struct net_device *dev)
4757 list_add_tail(&dev->todo_list, &net_todo_list);
4760 static void rollback_registered_many(struct list_head *head)
4762 struct net_device *dev, *tmp;
4764 BUG_ON(dev_boot_phase);
4767 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4768 /* Some devices call without registering
4769 * for initialization unwind. Remove those
4770 * devices and proceed with the remaining.
4772 if (dev->reg_state == NETREG_UNINITIALIZED) {
4773 pr_debug("unregister_netdevice: device %s/%p never "
4774 "was registered\n", dev->name, dev);
4777 list_del(&dev->unreg_list);
4781 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4783 /* If device is running, close it first. */
4786 /* And unlink it from device chain. */
4787 unlist_netdevice(dev);
4789 dev->reg_state = NETREG_UNREGISTERING;
4794 list_for_each_entry(dev, head, unreg_list) {
4795 /* Shutdown queueing discipline. */
4799 /* Notify protocols, that we are about to destroy
4800 this device. They should clean all the things.
4802 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4804 if (!dev->rtnl_link_ops ||
4805 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4806 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4809 * Flush the unicast and multicast chains
4814 if (dev->netdev_ops->ndo_uninit)
4815 dev->netdev_ops->ndo_uninit(dev);
4817 /* Notifier chain MUST detach us from master device. */
4818 WARN_ON(dev->master);
4820 /* Remove entries from kobject tree */
4821 netdev_unregister_kobject(dev);
4824 /* Process any work delayed until the end of the batch */
4825 dev = list_first_entry(head, struct net_device, unreg_list);
4826 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4830 list_for_each_entry(dev, head, unreg_list)
4834 static void rollback_registered(struct net_device *dev)
4838 list_add(&dev->unreg_list, &single);
4839 rollback_registered_many(&single);
4842 static void __netdev_init_queue_locks_one(struct net_device *dev,
4843 struct netdev_queue *dev_queue,
4846 spin_lock_init(&dev_queue->_xmit_lock);
4847 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4848 dev_queue->xmit_lock_owner = -1;
4851 static void netdev_init_queue_locks(struct net_device *dev)
4853 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4854 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4857 unsigned long netdev_fix_features(unsigned long features, const char *name)
4859 /* Fix illegal SG+CSUM combinations. */
4860 if ((features & NETIF_F_SG) &&
4861 !(features & NETIF_F_ALL_CSUM)) {
4863 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4864 "checksum feature.\n", name);
4865 features &= ~NETIF_F_SG;
4868 /* TSO requires that SG is present as well. */
4869 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4871 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4872 "SG feature.\n", name);
4873 features &= ~NETIF_F_TSO;
4876 if (features & NETIF_F_UFO) {
4877 if (!(features & NETIF_F_GEN_CSUM)) {
4879 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4880 "since no NETIF_F_HW_CSUM feature.\n",
4882 features &= ~NETIF_F_UFO;
4885 if (!(features & NETIF_F_SG)) {
4887 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4888 "since no NETIF_F_SG feature.\n", name);
4889 features &= ~NETIF_F_UFO;
4895 EXPORT_SYMBOL(netdev_fix_features);
4898 * netif_stacked_transfer_operstate - transfer operstate
4899 * @rootdev: the root or lower level device to transfer state from
4900 * @dev: the device to transfer operstate to
4902 * Transfer operational state from root to device. This is normally
4903 * called when a stacking relationship exists between the root
4904 * device and the device(a leaf device).
4906 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4907 struct net_device *dev)
4909 if (rootdev->operstate == IF_OPER_DORMANT)
4910 netif_dormant_on(dev);
4912 netif_dormant_off(dev);
4914 if (netif_carrier_ok(rootdev)) {
4915 if (!netif_carrier_ok(dev))
4916 netif_carrier_on(dev);
4918 if (netif_carrier_ok(dev))
4919 netif_carrier_off(dev);
4922 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4925 * register_netdevice - register a network device
4926 * @dev: device to register
4928 * Take a completed network device structure and add it to the kernel
4929 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4930 * chain. 0 is returned on success. A negative errno code is returned
4931 * on a failure to set up the device, or if the name is a duplicate.
4933 * Callers must hold the rtnl semaphore. You may want
4934 * register_netdev() instead of this.
4937 * The locking appears insufficient to guarantee two parallel registers
4938 * will not get the same name.
4941 int register_netdevice(struct net_device *dev)
4944 struct net *net = dev_net(dev);
4946 BUG_ON(dev_boot_phase);
4951 /* When net_device's are persistent, this will be fatal. */
4952 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4955 spin_lock_init(&dev->addr_list_lock);
4956 netdev_set_addr_lockdep_class(dev);
4957 netdev_init_queue_locks(dev);
4962 if (!dev->num_rx_queues) {
4964 * Allocate a single RX queue if driver never called
4968 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
4974 dev->_rx->first = dev->_rx;
4975 atomic_set(&dev->_rx->count, 1);
4976 dev->num_rx_queues = 1;
4979 /* Init, if this function is available */
4980 if (dev->netdev_ops->ndo_init) {
4981 ret = dev->netdev_ops->ndo_init(dev);
4989 ret = dev_get_valid_name(dev, dev->name, 0);
4993 dev->ifindex = dev_new_index(net);
4994 if (dev->iflink == -1)
4995 dev->iflink = dev->ifindex;
4997 /* Fix illegal checksum combinations */
4998 if ((dev->features & NETIF_F_HW_CSUM) &&
4999 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5000 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5002 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5005 if ((dev->features & NETIF_F_NO_CSUM) &&
5006 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5007 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5009 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5012 dev->features = netdev_fix_features(dev->features, dev->name);
5014 /* Enable software GSO if SG is supported. */
5015 if (dev->features & NETIF_F_SG)
5016 dev->features |= NETIF_F_GSO;
5018 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5019 ret = notifier_to_errno(ret);
5023 ret = netdev_register_kobject(dev);
5026 dev->reg_state = NETREG_REGISTERED;
5029 * Default initial state at registry is that the
5030 * device is present.
5033 set_bit(__LINK_STATE_PRESENT, &dev->state);
5035 dev_init_scheduler(dev);
5037 list_netdevice(dev);
5039 /* Notify protocols, that a new device appeared. */
5040 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5041 ret = notifier_to_errno(ret);
5043 rollback_registered(dev);
5044 dev->reg_state = NETREG_UNREGISTERED;
5047 * Prevent userspace races by waiting until the network
5048 * device is fully setup before sending notifications.
5050 if (!dev->rtnl_link_ops ||
5051 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5052 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5058 if (dev->netdev_ops->ndo_uninit)
5059 dev->netdev_ops->ndo_uninit(dev);
5062 EXPORT_SYMBOL(register_netdevice);
5065 * init_dummy_netdev - init a dummy network device for NAPI
5066 * @dev: device to init
5068 * This takes a network device structure and initialize the minimum
5069 * amount of fields so it can be used to schedule NAPI polls without
5070 * registering a full blown interface. This is to be used by drivers
5071 * that need to tie several hardware interfaces to a single NAPI
5072 * poll scheduler due to HW limitations.
5074 int init_dummy_netdev(struct net_device *dev)
5076 /* Clear everything. Note we don't initialize spinlocks
5077 * are they aren't supposed to be taken by any of the
5078 * NAPI code and this dummy netdev is supposed to be
5079 * only ever used for NAPI polls
5081 memset(dev, 0, sizeof(struct net_device));
5083 /* make sure we BUG if trying to hit standard
5084 * register/unregister code path
5086 dev->reg_state = NETREG_DUMMY;
5088 /* initialize the ref count */
5089 atomic_set(&dev->refcnt, 1);
5091 /* NAPI wants this */
5092 INIT_LIST_HEAD(&dev->napi_list);
5094 /* a dummy interface is started by default */
5095 set_bit(__LINK_STATE_PRESENT, &dev->state);
5096 set_bit(__LINK_STATE_START, &dev->state);
5100 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5104 * register_netdev - register a network device
5105 * @dev: device to register
5107 * Take a completed network device structure and add it to the kernel
5108 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5109 * chain. 0 is returned on success. A negative errno code is returned
5110 * on a failure to set up the device, or if the name is a duplicate.
5112 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5113 * and expands the device name if you passed a format string to
5116 int register_netdev(struct net_device *dev)
5123 * If the name is a format string the caller wants us to do a
5126 if (strchr(dev->name, '%')) {
5127 err = dev_alloc_name(dev, dev->name);
5132 err = register_netdevice(dev);
5137 EXPORT_SYMBOL(register_netdev);
5140 * netdev_wait_allrefs - wait until all references are gone.
5142 * This is called when unregistering network devices.
5144 * Any protocol or device that holds a reference should register
5145 * for netdevice notification, and cleanup and put back the
5146 * reference if they receive an UNREGISTER event.
5147 * We can get stuck here if buggy protocols don't correctly
5150 static void netdev_wait_allrefs(struct net_device *dev)
5152 unsigned long rebroadcast_time, warning_time;
5154 linkwatch_forget_dev(dev);
5156 rebroadcast_time = warning_time = jiffies;
5157 while (atomic_read(&dev->refcnt) != 0) {
5158 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5161 /* Rebroadcast unregister notification */
5162 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5163 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5164 * should have already handle it the first time */
5166 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5168 /* We must not have linkwatch events
5169 * pending on unregister. If this
5170 * happens, we simply run the queue
5171 * unscheduled, resulting in a noop
5174 linkwatch_run_queue();
5179 rebroadcast_time = jiffies;
5184 if (time_after(jiffies, warning_time + 10 * HZ)) {
5185 printk(KERN_EMERG "unregister_netdevice: "
5186 "waiting for %s to become free. Usage "
5188 dev->name, atomic_read(&dev->refcnt));
5189 warning_time = jiffies;
5198 * register_netdevice(x1);
5199 * register_netdevice(x2);
5201 * unregister_netdevice(y1);
5202 * unregister_netdevice(y2);
5208 * We are invoked by rtnl_unlock().
5209 * This allows us to deal with problems:
5210 * 1) We can delete sysfs objects which invoke hotplug
5211 * without deadlocking with linkwatch via keventd.
5212 * 2) Since we run with the RTNL semaphore not held, we can sleep
5213 * safely in order to wait for the netdev refcnt to drop to zero.
5215 * We must not return until all unregister events added during
5216 * the interval the lock was held have been completed.
5218 void netdev_run_todo(void)
5220 struct list_head list;
5222 /* Snapshot list, allow later requests */
5223 list_replace_init(&net_todo_list, &list);
5227 while (!list_empty(&list)) {
5228 struct net_device *dev
5229 = list_first_entry(&list, struct net_device, todo_list);
5230 list_del(&dev->todo_list);
5232 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5233 printk(KERN_ERR "network todo '%s' but state %d\n",
5234 dev->name, dev->reg_state);
5239 dev->reg_state = NETREG_UNREGISTERED;
5241 on_each_cpu(flush_backlog, dev, 1);
5243 netdev_wait_allrefs(dev);
5246 BUG_ON(atomic_read(&dev->refcnt));
5247 WARN_ON(dev->ip_ptr);
5248 WARN_ON(dev->ip6_ptr);
5249 WARN_ON(dev->dn_ptr);
5251 if (dev->destructor)
5252 dev->destructor(dev);
5254 /* Free network device */
5255 kobject_put(&dev->dev.kobj);
5260 * dev_txq_stats_fold - fold tx_queues stats
5261 * @dev: device to get statistics from
5262 * @stats: struct net_device_stats to hold results
5264 void dev_txq_stats_fold(const struct net_device *dev,
5265 struct net_device_stats *stats)
5267 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5269 struct netdev_queue *txq;
5271 for (i = 0; i < dev->num_tx_queues; i++) {
5272 txq = netdev_get_tx_queue(dev, i);
5273 tx_bytes += txq->tx_bytes;
5274 tx_packets += txq->tx_packets;
5275 tx_dropped += txq->tx_dropped;
5277 if (tx_bytes || tx_packets || tx_dropped) {
5278 stats->tx_bytes = tx_bytes;
5279 stats->tx_packets = tx_packets;
5280 stats->tx_dropped = tx_dropped;
5283 EXPORT_SYMBOL(dev_txq_stats_fold);
5286 * dev_get_stats - get network device statistics
5287 * @dev: device to get statistics from
5289 * Get network statistics from device. The device driver may provide
5290 * its own method by setting dev->netdev_ops->get_stats; otherwise
5291 * the internal statistics structure is used.
5293 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5295 const struct net_device_ops *ops = dev->netdev_ops;
5297 if (ops->ndo_get_stats)
5298 return ops->ndo_get_stats(dev);
5300 dev_txq_stats_fold(dev, &dev->stats);
5303 EXPORT_SYMBOL(dev_get_stats);
5305 static void netdev_init_one_queue(struct net_device *dev,
5306 struct netdev_queue *queue,
5312 static void netdev_init_queues(struct net_device *dev)
5314 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5315 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5316 spin_lock_init(&dev->tx_global_lock);
5320 * alloc_netdev_mq - allocate network device
5321 * @sizeof_priv: size of private data to allocate space for
5322 * @name: device name format string
5323 * @setup: callback to initialize device
5324 * @queue_count: the number of subqueues to allocate
5326 * Allocates a struct net_device with private data area for driver use
5327 * and performs basic initialization. Also allocates subquue structs
5328 * for each queue on the device at the end of the netdevice.
5330 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5331 void (*setup)(struct net_device *), unsigned int queue_count)
5333 struct netdev_queue *tx;
5334 struct net_device *dev;
5336 struct net_device *p;
5338 struct netdev_rx_queue *rx;
5342 BUG_ON(strlen(name) >= sizeof(dev->name));
5344 alloc_size = sizeof(struct net_device);
5346 /* ensure 32-byte alignment of private area */
5347 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5348 alloc_size += sizeof_priv;
5350 /* ensure 32-byte alignment of whole construct */
5351 alloc_size += NETDEV_ALIGN - 1;
5353 p = kzalloc(alloc_size, GFP_KERNEL);
5355 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5359 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5361 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5367 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5369 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5374 atomic_set(&rx->count, queue_count);
5377 * Set a pointer to first element in the array which holds the
5380 for (i = 0; i < queue_count; i++)
5384 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5385 dev->padded = (char *)dev - (char *)p;
5387 if (dev_addr_init(dev))
5393 dev_net_set(dev, &init_net);
5396 dev->num_tx_queues = queue_count;
5397 dev->real_num_tx_queues = queue_count;
5401 dev->num_rx_queues = queue_count;
5404 dev->gso_max_size = GSO_MAX_SIZE;
5406 netdev_init_queues(dev);
5408 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5409 dev->ethtool_ntuple_list.count = 0;
5410 INIT_LIST_HEAD(&dev->napi_list);
5411 INIT_LIST_HEAD(&dev->unreg_list);
5412 INIT_LIST_HEAD(&dev->link_watch_list);
5413 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5415 strcpy(dev->name, name);
5428 EXPORT_SYMBOL(alloc_netdev_mq);
5431 * free_netdev - free network device
5434 * This function does the last stage of destroying an allocated device
5435 * interface. The reference to the device object is released.
5436 * If this is the last reference then it will be freed.
5438 void free_netdev(struct net_device *dev)
5440 struct napi_struct *p, *n;
5442 release_net(dev_net(dev));
5446 /* Flush device addresses */
5447 dev_addr_flush(dev);
5449 /* Clear ethtool n-tuple list */
5450 ethtool_ntuple_flush(dev);
5452 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5455 /* Compatibility with error handling in drivers */
5456 if (dev->reg_state == NETREG_UNINITIALIZED) {
5457 kfree((char *)dev - dev->padded);
5461 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5462 dev->reg_state = NETREG_RELEASED;
5464 /* will free via device release */
5465 put_device(&dev->dev);
5467 EXPORT_SYMBOL(free_netdev);
5470 * synchronize_net - Synchronize with packet receive processing
5472 * Wait for packets currently being received to be done.
5473 * Does not block later packets from starting.
5475 void synchronize_net(void)
5480 EXPORT_SYMBOL(synchronize_net);
5483 * unregister_netdevice_queue - remove device from the kernel
5487 * This function shuts down a device interface and removes it
5488 * from the kernel tables.
5489 * If head not NULL, device is queued to be unregistered later.
5491 * Callers must hold the rtnl semaphore. You may want
5492 * unregister_netdev() instead of this.
5495 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5500 list_move_tail(&dev->unreg_list, head);
5502 rollback_registered(dev);
5503 /* Finish processing unregister after unlock */
5507 EXPORT_SYMBOL(unregister_netdevice_queue);
5510 * unregister_netdevice_many - unregister many devices
5511 * @head: list of devices
5513 void unregister_netdevice_many(struct list_head *head)
5515 struct net_device *dev;
5517 if (!list_empty(head)) {
5518 rollback_registered_many(head);
5519 list_for_each_entry(dev, head, unreg_list)
5523 EXPORT_SYMBOL(unregister_netdevice_many);
5526 * unregister_netdev - remove device from the kernel
5529 * This function shuts down a device interface and removes it
5530 * from the kernel tables.
5532 * This is just a wrapper for unregister_netdevice that takes
5533 * the rtnl semaphore. In general you want to use this and not
5534 * unregister_netdevice.
5536 void unregister_netdev(struct net_device *dev)
5539 unregister_netdevice(dev);
5542 EXPORT_SYMBOL(unregister_netdev);
5545 * dev_change_net_namespace - move device to different nethost namespace
5547 * @net: network namespace
5548 * @pat: If not NULL name pattern to try if the current device name
5549 * is already taken in the destination network namespace.
5551 * This function shuts down a device interface and moves it
5552 * to a new network namespace. On success 0 is returned, on
5553 * a failure a netagive errno code is returned.
5555 * Callers must hold the rtnl semaphore.
5558 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5564 /* Don't allow namespace local devices to be moved. */
5566 if (dev->features & NETIF_F_NETNS_LOCAL)
5569 /* Ensure the device has been registrered */
5571 if (dev->reg_state != NETREG_REGISTERED)
5574 /* Get out if there is nothing todo */
5576 if (net_eq(dev_net(dev), net))
5579 /* Pick the destination device name, and ensure
5580 * we can use it in the destination network namespace.
5583 if (__dev_get_by_name(net, dev->name)) {
5584 /* We get here if we can't use the current device name */
5587 if (dev_get_valid_name(dev, pat, 1))
5592 * And now a mini version of register_netdevice unregister_netdevice.
5595 /* If device is running close it first. */
5598 /* And unlink it from device chain */
5600 unlist_netdevice(dev);
5604 /* Shutdown queueing discipline. */
5607 /* Notify protocols, that we are about to destroy
5608 this device. They should clean all the things.
5610 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5611 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5614 * Flush the unicast and multicast chains
5619 /* Actually switch the network namespace */
5620 dev_net_set(dev, net);
5622 /* If there is an ifindex conflict assign a new one */
5623 if (__dev_get_by_index(net, dev->ifindex)) {
5624 int iflink = (dev->iflink == dev->ifindex);
5625 dev->ifindex = dev_new_index(net);
5627 dev->iflink = dev->ifindex;
5630 /* Fixup kobjects */
5631 err = device_rename(&dev->dev, dev->name);
5634 /* Add the device back in the hashes */
5635 list_netdevice(dev);
5637 /* Notify protocols, that a new device appeared. */
5638 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5641 * Prevent userspace races by waiting until the network
5642 * device is fully setup before sending notifications.
5644 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5651 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5653 static int dev_cpu_callback(struct notifier_block *nfb,
5654 unsigned long action,
5657 struct sk_buff **list_skb;
5658 struct sk_buff *skb;
5659 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5660 struct softnet_data *sd, *oldsd;
5662 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5665 local_irq_disable();
5666 cpu = smp_processor_id();
5667 sd = &per_cpu(softnet_data, cpu);
5668 oldsd = &per_cpu(softnet_data, oldcpu);
5670 /* Find end of our completion_queue. */
5671 list_skb = &sd->completion_queue;
5673 list_skb = &(*list_skb)->next;
5674 /* Append completion queue from offline CPU. */
5675 *list_skb = oldsd->completion_queue;
5676 oldsd->completion_queue = NULL;
5678 /* Append output queue from offline CPU. */
5679 if (oldsd->output_queue) {
5680 *sd->output_queue_tailp = oldsd->output_queue;
5681 sd->output_queue_tailp = oldsd->output_queue_tailp;
5682 oldsd->output_queue = NULL;
5683 oldsd->output_queue_tailp = &oldsd->output_queue;
5686 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5689 /* Process offline CPU's input_pkt_queue */
5690 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5692 input_queue_head_incr(oldsd);
5694 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5696 input_queue_head_incr(oldsd);
5704 * netdev_increment_features - increment feature set by one
5705 * @all: current feature set
5706 * @one: new feature set
5707 * @mask: mask feature set
5709 * Computes a new feature set after adding a device with feature set
5710 * @one to the master device with current feature set @all. Will not
5711 * enable anything that is off in @mask. Returns the new feature set.
5713 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5716 /* If device needs checksumming, downgrade to it. */
5717 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5718 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5719 else if (mask & NETIF_F_ALL_CSUM) {
5720 /* If one device supports v4/v6 checksumming, set for all. */
5721 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5722 !(all & NETIF_F_GEN_CSUM)) {
5723 all &= ~NETIF_F_ALL_CSUM;
5724 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5727 /* If one device supports hw checksumming, set for all. */
5728 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5729 all &= ~NETIF_F_ALL_CSUM;
5730 all |= NETIF_F_HW_CSUM;
5734 one |= NETIF_F_ALL_CSUM;
5736 one |= all & NETIF_F_ONE_FOR_ALL;
5737 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5738 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5742 EXPORT_SYMBOL(netdev_increment_features);
5744 static struct hlist_head *netdev_create_hash(void)
5747 struct hlist_head *hash;
5749 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5751 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5752 INIT_HLIST_HEAD(&hash[i]);
5757 /* Initialize per network namespace state */
5758 static int __net_init netdev_init(struct net *net)
5760 INIT_LIST_HEAD(&net->dev_base_head);
5762 net->dev_name_head = netdev_create_hash();
5763 if (net->dev_name_head == NULL)
5766 net->dev_index_head = netdev_create_hash();
5767 if (net->dev_index_head == NULL)
5773 kfree(net->dev_name_head);
5779 * netdev_drivername - network driver for the device
5780 * @dev: network device
5781 * @buffer: buffer for resulting name
5782 * @len: size of buffer
5784 * Determine network driver for device.
5786 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5788 const struct device_driver *driver;
5789 const struct device *parent;
5791 if (len <= 0 || !buffer)
5795 parent = dev->dev.parent;
5800 driver = parent->driver;
5801 if (driver && driver->name)
5802 strlcpy(buffer, driver->name, len);
5806 static void __net_exit netdev_exit(struct net *net)
5808 kfree(net->dev_name_head);
5809 kfree(net->dev_index_head);
5812 static struct pernet_operations __net_initdata netdev_net_ops = {
5813 .init = netdev_init,
5814 .exit = netdev_exit,
5817 static void __net_exit default_device_exit(struct net *net)
5819 struct net_device *dev, *aux;
5821 * Push all migratable network devices back to the
5822 * initial network namespace
5825 for_each_netdev_safe(net, dev, aux) {
5827 char fb_name[IFNAMSIZ];
5829 /* Ignore unmoveable devices (i.e. loopback) */
5830 if (dev->features & NETIF_F_NETNS_LOCAL)
5833 /* Leave virtual devices for the generic cleanup */
5834 if (dev->rtnl_link_ops)
5837 /* Push remaing network devices to init_net */
5838 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5839 err = dev_change_net_namespace(dev, &init_net, fb_name);
5841 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5842 __func__, dev->name, err);
5849 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5851 /* At exit all network devices most be removed from a network
5852 * namespace. Do this in the reverse order of registeration.
5853 * Do this across as many network namespaces as possible to
5854 * improve batching efficiency.
5856 struct net_device *dev;
5858 LIST_HEAD(dev_kill_list);
5861 list_for_each_entry(net, net_list, exit_list) {
5862 for_each_netdev_reverse(net, dev) {
5863 if (dev->rtnl_link_ops)
5864 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5866 unregister_netdevice_queue(dev, &dev_kill_list);
5869 unregister_netdevice_many(&dev_kill_list);
5873 static struct pernet_operations __net_initdata default_device_ops = {
5874 .exit = default_device_exit,
5875 .exit_batch = default_device_exit_batch,
5879 * Initialize the DEV module. At boot time this walks the device list and
5880 * unhooks any devices that fail to initialise (normally hardware not
5881 * present) and leaves us with a valid list of present and active devices.
5886 * This is called single threaded during boot, so no need
5887 * to take the rtnl semaphore.
5889 static int __init net_dev_init(void)
5891 int i, rc = -ENOMEM;
5893 BUG_ON(!dev_boot_phase);
5895 if (dev_proc_init())
5898 if (netdev_kobject_init())
5901 INIT_LIST_HEAD(&ptype_all);
5902 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5903 INIT_LIST_HEAD(&ptype_base[i]);
5905 if (register_pernet_subsys(&netdev_net_ops))
5909 * Initialise the packet receive queues.
5912 for_each_possible_cpu(i) {
5913 struct softnet_data *sd = &per_cpu(softnet_data, i);
5915 memset(sd, 0, sizeof(*sd));
5916 skb_queue_head_init(&sd->input_pkt_queue);
5917 skb_queue_head_init(&sd->process_queue);
5918 sd->completion_queue = NULL;
5919 INIT_LIST_HEAD(&sd->poll_list);
5920 sd->output_queue = NULL;
5921 sd->output_queue_tailp = &sd->output_queue;
5923 sd->csd.func = rps_trigger_softirq;
5929 sd->backlog.poll = process_backlog;
5930 sd->backlog.weight = weight_p;
5931 sd->backlog.gro_list = NULL;
5932 sd->backlog.gro_count = 0;
5937 /* The loopback device is special if any other network devices
5938 * is present in a network namespace the loopback device must
5939 * be present. Since we now dynamically allocate and free the
5940 * loopback device ensure this invariant is maintained by
5941 * keeping the loopback device as the first device on the
5942 * list of network devices. Ensuring the loopback devices
5943 * is the first device that appears and the last network device
5946 if (register_pernet_device(&loopback_net_ops))
5949 if (register_pernet_device(&default_device_ops))
5952 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5953 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5955 hotcpu_notifier(dev_cpu_callback, 0);
5963 subsys_initcall(net_dev_init);
5965 static int __init initialize_hashrnd(void)
5967 get_random_bytes(&hashrnd, sizeof(hashrnd));
5971 late_initcall_sync(initialize_hashrnd);