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 /* For now only devices in the initial network namespace
1012 if (net_eq(net, &init_net)) {
1013 ret = device_rename(&dev->dev, dev->name);
1015 memcpy(dev->name, oldname, IFNAMSIZ);
1020 write_lock_bh(&dev_base_lock);
1021 hlist_del(&dev->name_hlist);
1022 write_unlock_bh(&dev_base_lock);
1026 write_lock_bh(&dev_base_lock);
1027 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1028 write_unlock_bh(&dev_base_lock);
1030 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1031 ret = notifier_to_errno(ret);
1034 /* err >= 0 after dev_alloc_name() or stores the first errno */
1037 memcpy(dev->name, oldname, IFNAMSIZ);
1041 "%s: name change rollback failed: %d.\n",
1050 * dev_set_alias - change ifalias of a device
1052 * @alias: name up to IFALIASZ
1053 * @len: limit of bytes to copy from info
1055 * Set ifalias for a device,
1057 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1061 if (len >= IFALIASZ)
1066 kfree(dev->ifalias);
1067 dev->ifalias = NULL;
1072 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1076 strlcpy(dev->ifalias, alias, len+1);
1082 * netdev_features_change - device changes features
1083 * @dev: device to cause notification
1085 * Called to indicate a device has changed features.
1087 void netdev_features_change(struct net_device *dev)
1089 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1091 EXPORT_SYMBOL(netdev_features_change);
1094 * netdev_state_change - device changes state
1095 * @dev: device to cause notification
1097 * Called to indicate a device has changed state. This function calls
1098 * the notifier chains for netdev_chain and sends a NEWLINK message
1099 * to the routing socket.
1101 void netdev_state_change(struct net_device *dev)
1103 if (dev->flags & IFF_UP) {
1104 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1105 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1108 EXPORT_SYMBOL(netdev_state_change);
1110 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1112 return call_netdevice_notifiers(event, dev);
1114 EXPORT_SYMBOL(netdev_bonding_change);
1117 * dev_load - load a network module
1118 * @net: the applicable net namespace
1119 * @name: name of interface
1121 * If a network interface is not present and the process has suitable
1122 * privileges this function loads the module. If module loading is not
1123 * available in this kernel then it becomes a nop.
1126 void dev_load(struct net *net, const char *name)
1128 struct net_device *dev;
1131 dev = dev_get_by_name_rcu(net, name);
1134 if (!dev && capable(CAP_NET_ADMIN))
1135 request_module("%s", name);
1137 EXPORT_SYMBOL(dev_load);
1139 static int __dev_open(struct net_device *dev)
1141 const struct net_device_ops *ops = dev->netdev_ops;
1147 * Is it even present?
1149 if (!netif_device_present(dev))
1152 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1153 ret = notifier_to_errno(ret);
1158 * Call device private open method
1160 set_bit(__LINK_STATE_START, &dev->state);
1162 if (ops->ndo_validate_addr)
1163 ret = ops->ndo_validate_addr(dev);
1165 if (!ret && ops->ndo_open)
1166 ret = ops->ndo_open(dev);
1169 * If it went open OK then:
1173 clear_bit(__LINK_STATE_START, &dev->state);
1178 dev->flags |= IFF_UP;
1183 net_dmaengine_get();
1186 * Initialize multicasting status
1188 dev_set_rx_mode(dev);
1191 * Wakeup transmit queue engine
1200 * dev_open - prepare an interface for use.
1201 * @dev: device to open
1203 * Takes a device from down to up state. The device's private open
1204 * function is invoked and then the multicast lists are loaded. Finally
1205 * the device is moved into the up state and a %NETDEV_UP message is
1206 * sent to the netdev notifier chain.
1208 * Calling this function on an active interface is a nop. On a failure
1209 * a negative errno code is returned.
1211 int dev_open(struct net_device *dev)
1218 if (dev->flags & IFF_UP)
1224 ret = __dev_open(dev);
1229 * ... and announce new interface.
1231 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1232 call_netdevice_notifiers(NETDEV_UP, dev);
1236 EXPORT_SYMBOL(dev_open);
1238 static int __dev_close(struct net_device *dev)
1240 const struct net_device_ops *ops = dev->netdev_ops;
1246 * Tell people we are going down, so that they can
1247 * prepare to death, when device is still operating.
1249 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1251 clear_bit(__LINK_STATE_START, &dev->state);
1253 /* Synchronize to scheduled poll. We cannot touch poll list,
1254 * it can be even on different cpu. So just clear netif_running().
1256 * dev->stop() will invoke napi_disable() on all of it's
1257 * napi_struct instances on this device.
1259 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1261 dev_deactivate(dev);
1264 * Call the device specific close. This cannot fail.
1265 * Only if device is UP
1267 * We allow it to be called even after a DETACH hot-plug
1274 * Device is now down.
1277 dev->flags &= ~IFF_UP;
1282 net_dmaengine_put();
1288 * dev_close - shutdown an interface.
1289 * @dev: device to shutdown
1291 * This function moves an active device into down state. A
1292 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1293 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1296 int dev_close(struct net_device *dev)
1298 if (!(dev->flags & IFF_UP))
1304 * Tell people we are down
1306 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1307 call_netdevice_notifiers(NETDEV_DOWN, dev);
1311 EXPORT_SYMBOL(dev_close);
1315 * dev_disable_lro - disable Large Receive Offload on a device
1318 * Disable Large Receive Offload (LRO) on a net device. Must be
1319 * called under RTNL. This is needed if received packets may be
1320 * forwarded to another interface.
1322 void dev_disable_lro(struct net_device *dev)
1324 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1325 dev->ethtool_ops->set_flags) {
1326 u32 flags = dev->ethtool_ops->get_flags(dev);
1327 if (flags & ETH_FLAG_LRO) {
1328 flags &= ~ETH_FLAG_LRO;
1329 dev->ethtool_ops->set_flags(dev, flags);
1332 WARN_ON(dev->features & NETIF_F_LRO);
1334 EXPORT_SYMBOL(dev_disable_lro);
1337 static int dev_boot_phase = 1;
1340 * Device change register/unregister. These are not inline or static
1341 * as we export them to the world.
1345 * register_netdevice_notifier - register a network notifier block
1348 * Register a notifier to be called when network device events occur.
1349 * The notifier passed is linked into the kernel structures and must
1350 * not be reused until it has been unregistered. A negative errno code
1351 * is returned on a failure.
1353 * When registered all registration and up events are replayed
1354 * to the new notifier to allow device to have a race free
1355 * view of the network device list.
1358 int register_netdevice_notifier(struct notifier_block *nb)
1360 struct net_device *dev;
1361 struct net_device *last;
1366 err = raw_notifier_chain_register(&netdev_chain, nb);
1372 for_each_netdev(net, dev) {
1373 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1374 err = notifier_to_errno(err);
1378 if (!(dev->flags & IFF_UP))
1381 nb->notifier_call(nb, NETDEV_UP, dev);
1392 for_each_netdev(net, dev) {
1396 if (dev->flags & IFF_UP) {
1397 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1398 nb->notifier_call(nb, NETDEV_DOWN, dev);
1400 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1401 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1405 raw_notifier_chain_unregister(&netdev_chain, nb);
1408 EXPORT_SYMBOL(register_netdevice_notifier);
1411 * unregister_netdevice_notifier - unregister a network notifier block
1414 * Unregister a notifier previously registered by
1415 * register_netdevice_notifier(). The notifier is unlinked into the
1416 * kernel structures and may then be reused. A negative errno code
1417 * is returned on a failure.
1420 int unregister_netdevice_notifier(struct notifier_block *nb)
1425 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1429 EXPORT_SYMBOL(unregister_netdevice_notifier);
1432 * call_netdevice_notifiers - call all network notifier blocks
1433 * @val: value passed unmodified to notifier function
1434 * @dev: net_device pointer passed unmodified to notifier function
1436 * Call all network notifier blocks. Parameters and return value
1437 * are as for raw_notifier_call_chain().
1440 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1443 return raw_notifier_call_chain(&netdev_chain, val, dev);
1446 /* When > 0 there are consumers of rx skb time stamps */
1447 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1449 void net_enable_timestamp(void)
1451 atomic_inc(&netstamp_needed);
1453 EXPORT_SYMBOL(net_enable_timestamp);
1455 void net_disable_timestamp(void)
1457 atomic_dec(&netstamp_needed);
1459 EXPORT_SYMBOL(net_disable_timestamp);
1461 static inline void net_timestamp_set(struct sk_buff *skb)
1463 if (atomic_read(&netstamp_needed))
1464 __net_timestamp(skb);
1466 skb->tstamp.tv64 = 0;
1469 static inline void net_timestamp_check(struct sk_buff *skb)
1471 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1472 __net_timestamp(skb);
1476 * dev_forward_skb - loopback an skb to another netif
1478 * @dev: destination network device
1479 * @skb: buffer to forward
1482 * NET_RX_SUCCESS (no congestion)
1483 * NET_RX_DROP (packet was dropped, but freed)
1485 * dev_forward_skb can be used for injecting an skb from the
1486 * start_xmit function of one device into the receive queue
1487 * of another device.
1489 * The receiving device may be in another namespace, so
1490 * we have to clear all information in the skb that could
1491 * impact namespace isolation.
1493 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1497 if (!(dev->flags & IFF_UP) ||
1498 (skb->len > (dev->mtu + dev->hard_header_len))) {
1502 skb_set_dev(skb, dev);
1503 skb->tstamp.tv64 = 0;
1504 skb->pkt_type = PACKET_HOST;
1505 skb->protocol = eth_type_trans(skb, dev);
1506 return netif_rx(skb);
1508 EXPORT_SYMBOL_GPL(dev_forward_skb);
1511 * Support routine. Sends outgoing frames to any network
1512 * taps currently in use.
1515 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1517 struct packet_type *ptype;
1519 #ifdef CONFIG_NET_CLS_ACT
1520 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1521 net_timestamp_set(skb);
1523 net_timestamp_set(skb);
1527 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1528 /* Never send packets back to the socket
1529 * they originated from - MvS (miquels@drinkel.ow.org)
1531 if ((ptype->dev == dev || !ptype->dev) &&
1532 (ptype->af_packet_priv == NULL ||
1533 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1534 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1538 /* skb->nh should be correctly
1539 set by sender, so that the second statement is
1540 just protection against buggy protocols.
1542 skb_reset_mac_header(skb2);
1544 if (skb_network_header(skb2) < skb2->data ||
1545 skb2->network_header > skb2->tail) {
1546 if (net_ratelimit())
1547 printk(KERN_CRIT "protocol %04x is "
1549 skb2->protocol, dev->name);
1550 skb_reset_network_header(skb2);
1553 skb2->transport_header = skb2->network_header;
1554 skb2->pkt_type = PACKET_OUTGOING;
1555 ptype->func(skb2, skb->dev, ptype, skb->dev);
1562 static inline void __netif_reschedule(struct Qdisc *q)
1564 struct softnet_data *sd;
1565 unsigned long flags;
1567 local_irq_save(flags);
1568 sd = &__get_cpu_var(softnet_data);
1569 q->next_sched = NULL;
1570 *sd->output_queue_tailp = q;
1571 sd->output_queue_tailp = &q->next_sched;
1572 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1573 local_irq_restore(flags);
1576 void __netif_schedule(struct Qdisc *q)
1578 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1579 __netif_reschedule(q);
1581 EXPORT_SYMBOL(__netif_schedule);
1583 void dev_kfree_skb_irq(struct sk_buff *skb)
1585 if (atomic_dec_and_test(&skb->users)) {
1586 struct softnet_data *sd;
1587 unsigned long flags;
1589 local_irq_save(flags);
1590 sd = &__get_cpu_var(softnet_data);
1591 skb->next = sd->completion_queue;
1592 sd->completion_queue = skb;
1593 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1594 local_irq_restore(flags);
1597 EXPORT_SYMBOL(dev_kfree_skb_irq);
1599 void dev_kfree_skb_any(struct sk_buff *skb)
1601 if (in_irq() || irqs_disabled())
1602 dev_kfree_skb_irq(skb);
1606 EXPORT_SYMBOL(dev_kfree_skb_any);
1610 * netif_device_detach - mark device as removed
1611 * @dev: network device
1613 * Mark device as removed from system and therefore no longer available.
1615 void netif_device_detach(struct net_device *dev)
1617 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1618 netif_running(dev)) {
1619 netif_tx_stop_all_queues(dev);
1622 EXPORT_SYMBOL(netif_device_detach);
1625 * netif_device_attach - mark device as attached
1626 * @dev: network device
1628 * Mark device as attached from system and restart if needed.
1630 void netif_device_attach(struct net_device *dev)
1632 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1633 netif_running(dev)) {
1634 netif_tx_wake_all_queues(dev);
1635 __netdev_watchdog_up(dev);
1638 EXPORT_SYMBOL(netif_device_attach);
1640 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1642 return ((features & NETIF_F_GEN_CSUM) ||
1643 ((features & NETIF_F_IP_CSUM) &&
1644 protocol == htons(ETH_P_IP)) ||
1645 ((features & NETIF_F_IPV6_CSUM) &&
1646 protocol == htons(ETH_P_IPV6)) ||
1647 ((features & NETIF_F_FCOE_CRC) &&
1648 protocol == htons(ETH_P_FCOE)));
1651 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1653 if (can_checksum_protocol(dev->features, skb->protocol))
1656 if (skb->protocol == htons(ETH_P_8021Q)) {
1657 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1658 if (can_checksum_protocol(dev->features & dev->vlan_features,
1659 veh->h_vlan_encapsulated_proto))
1667 * skb_dev_set -- assign a new device to a buffer
1668 * @skb: buffer for the new device
1669 * @dev: network device
1671 * If an skb is owned by a device already, we have to reset
1672 * all data private to the namespace a device belongs to
1673 * before assigning it a new device.
1675 #ifdef CONFIG_NET_NS
1676 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1679 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1682 skb_init_secmark(skb);
1686 skb->ipvs_property = 0;
1687 #ifdef CONFIG_NET_SCHED
1693 EXPORT_SYMBOL(skb_set_dev);
1694 #endif /* CONFIG_NET_NS */
1697 * Invalidate hardware checksum when packet is to be mangled, and
1698 * complete checksum manually on outgoing path.
1700 int skb_checksum_help(struct sk_buff *skb)
1703 int ret = 0, offset;
1705 if (skb->ip_summed == CHECKSUM_COMPLETE)
1706 goto out_set_summed;
1708 if (unlikely(skb_shinfo(skb)->gso_size)) {
1709 /* Let GSO fix up the checksum. */
1710 goto out_set_summed;
1713 offset = skb->csum_start - skb_headroom(skb);
1714 BUG_ON(offset >= skb_headlen(skb));
1715 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1717 offset += skb->csum_offset;
1718 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1720 if (skb_cloned(skb) &&
1721 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1722 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1727 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1729 skb->ip_summed = CHECKSUM_NONE;
1733 EXPORT_SYMBOL(skb_checksum_help);
1736 * skb_gso_segment - Perform segmentation on skb.
1737 * @skb: buffer to segment
1738 * @features: features for the output path (see dev->features)
1740 * This function segments the given skb and returns a list of segments.
1742 * It may return NULL if the skb requires no segmentation. This is
1743 * only possible when GSO is used for verifying header integrity.
1745 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1747 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1748 struct packet_type *ptype;
1749 __be16 type = skb->protocol;
1752 skb_reset_mac_header(skb);
1753 skb->mac_len = skb->network_header - skb->mac_header;
1754 __skb_pull(skb, skb->mac_len);
1756 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1757 struct net_device *dev = skb->dev;
1758 struct ethtool_drvinfo info = {};
1760 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1761 dev->ethtool_ops->get_drvinfo(dev, &info);
1763 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1765 info.driver, dev ? dev->features : 0L,
1766 skb->sk ? skb->sk->sk_route_caps : 0L,
1767 skb->len, skb->data_len, skb->ip_summed);
1769 if (skb_header_cloned(skb) &&
1770 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1771 return ERR_PTR(err);
1775 list_for_each_entry_rcu(ptype,
1776 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1777 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1778 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1779 err = ptype->gso_send_check(skb);
1780 segs = ERR_PTR(err);
1781 if (err || skb_gso_ok(skb, features))
1783 __skb_push(skb, (skb->data -
1784 skb_network_header(skb)));
1786 segs = ptype->gso_segment(skb, features);
1792 __skb_push(skb, skb->data - skb_mac_header(skb));
1796 EXPORT_SYMBOL(skb_gso_segment);
1798 /* Take action when hardware reception checksum errors are detected. */
1800 void netdev_rx_csum_fault(struct net_device *dev)
1802 if (net_ratelimit()) {
1803 printk(KERN_ERR "%s: hw csum failure.\n",
1804 dev ? dev->name : "<unknown>");
1808 EXPORT_SYMBOL(netdev_rx_csum_fault);
1811 /* Actually, we should eliminate this check as soon as we know, that:
1812 * 1. IOMMU is present and allows to map all the memory.
1813 * 2. No high memory really exists on this machine.
1816 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1818 #ifdef CONFIG_HIGHMEM
1820 if (!(dev->features & NETIF_F_HIGHDMA)) {
1821 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1822 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1826 if (PCI_DMA_BUS_IS_PHYS) {
1827 struct device *pdev = dev->dev.parent;
1831 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1832 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1833 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1842 void (*destructor)(struct sk_buff *skb);
1845 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1847 static void dev_gso_skb_destructor(struct sk_buff *skb)
1849 struct dev_gso_cb *cb;
1852 struct sk_buff *nskb = skb->next;
1854 skb->next = nskb->next;
1857 } while (skb->next);
1859 cb = DEV_GSO_CB(skb);
1861 cb->destructor(skb);
1865 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1866 * @skb: buffer to segment
1868 * This function segments the given skb and stores the list of segments
1871 static int dev_gso_segment(struct sk_buff *skb)
1873 struct net_device *dev = skb->dev;
1874 struct sk_buff *segs;
1875 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1878 segs = skb_gso_segment(skb, features);
1880 /* Verifying header integrity only. */
1885 return PTR_ERR(segs);
1888 DEV_GSO_CB(skb)->destructor = skb->destructor;
1889 skb->destructor = dev_gso_skb_destructor;
1895 * Try to orphan skb early, right before transmission by the device.
1896 * We cannot orphan skb if tx timestamp is requested, since
1897 * drivers need to call skb_tstamp_tx() to send the timestamp.
1899 static inline void skb_orphan_try(struct sk_buff *skb)
1901 if (!skb_tx(skb)->flags)
1905 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1906 struct netdev_queue *txq)
1908 const struct net_device_ops *ops = dev->netdev_ops;
1909 int rc = NETDEV_TX_OK;
1911 if (likely(!skb->next)) {
1912 if (!list_empty(&ptype_all))
1913 dev_queue_xmit_nit(skb, dev);
1916 * If device doesnt need skb->dst, release it right now while
1917 * its hot in this cpu cache
1919 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1922 skb_orphan_try(skb);
1924 if (netif_needs_gso(dev, skb)) {
1925 if (unlikely(dev_gso_segment(skb)))
1931 rc = ops->ndo_start_xmit(skb, dev);
1932 if (rc == NETDEV_TX_OK)
1933 txq_trans_update(txq);
1939 struct sk_buff *nskb = skb->next;
1941 skb->next = nskb->next;
1945 * If device doesnt need nskb->dst, release it right now while
1946 * its hot in this cpu cache
1948 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1951 rc = ops->ndo_start_xmit(nskb, dev);
1952 if (unlikely(rc != NETDEV_TX_OK)) {
1953 if (rc & ~NETDEV_TX_MASK)
1954 goto out_kfree_gso_skb;
1955 nskb->next = skb->next;
1959 txq_trans_update(txq);
1960 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1961 return NETDEV_TX_BUSY;
1962 } while (skb->next);
1965 if (likely(skb->next == NULL))
1966 skb->destructor = DEV_GSO_CB(skb)->destructor;
1972 static u32 hashrnd __read_mostly;
1974 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1978 if (skb_rx_queue_recorded(skb)) {
1979 hash = skb_get_rx_queue(skb);
1980 while (unlikely(hash >= dev->real_num_tx_queues))
1981 hash -= dev->real_num_tx_queues;
1985 if (skb->sk && skb->sk->sk_hash)
1986 hash = skb->sk->sk_hash;
1988 hash = (__force u16) skb->protocol;
1990 hash = jhash_1word(hash, hashrnd);
1992 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1994 EXPORT_SYMBOL(skb_tx_hash);
1996 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1998 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1999 if (net_ratelimit()) {
2000 pr_warning("%s selects TX queue %d, but "
2001 "real number of TX queues is %d\n",
2002 dev->name, queue_index, dev->real_num_tx_queues);
2009 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2010 struct sk_buff *skb)
2013 struct sock *sk = skb->sk;
2015 if (sk_tx_queue_recorded(sk)) {
2016 queue_index = sk_tx_queue_get(sk);
2018 const struct net_device_ops *ops = dev->netdev_ops;
2020 if (ops->ndo_select_queue) {
2021 queue_index = ops->ndo_select_queue(dev, skb);
2022 queue_index = dev_cap_txqueue(dev, queue_index);
2025 if (dev->real_num_tx_queues > 1)
2026 queue_index = skb_tx_hash(dev, skb);
2029 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2031 if (dst && skb_dst(skb) == dst)
2032 sk_tx_queue_set(sk, queue_index);
2037 skb_set_queue_mapping(skb, queue_index);
2038 return netdev_get_tx_queue(dev, queue_index);
2041 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2042 struct net_device *dev,
2043 struct netdev_queue *txq)
2045 spinlock_t *root_lock = qdisc_lock(q);
2048 spin_lock(root_lock);
2049 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2052 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2053 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
2055 * This is a work-conserving queue; there are no old skbs
2056 * waiting to be sent out; and the qdisc is not running -
2057 * xmit the skb directly.
2059 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2061 __qdisc_update_bstats(q, skb->len);
2062 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
2065 clear_bit(__QDISC_STATE_RUNNING, &q->state);
2067 rc = NET_XMIT_SUCCESS;
2070 rc = qdisc_enqueue_root(skb, q);
2073 spin_unlock(root_lock);
2079 * Returns true if either:
2080 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2081 * 2. skb is fragmented and the device does not support SG, or if
2082 * at least one of fragments is in highmem and device does not
2083 * support DMA from it.
2085 static inline int skb_needs_linearize(struct sk_buff *skb,
2086 struct net_device *dev)
2088 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
2089 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
2090 illegal_highdma(dev, skb)));
2094 * dev_queue_xmit - transmit a buffer
2095 * @skb: buffer to transmit
2097 * Queue a buffer for transmission to a network device. The caller must
2098 * have set the device and priority and built the buffer before calling
2099 * this function. The function can be called from an interrupt.
2101 * A negative errno code is returned on a failure. A success does not
2102 * guarantee the frame will be transmitted as it may be dropped due
2103 * to congestion or traffic shaping.
2105 * -----------------------------------------------------------------------------------
2106 * I notice this method can also return errors from the queue disciplines,
2107 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2110 * Regardless of the return value, the skb is consumed, so it is currently
2111 * difficult to retry a send to this method. (You can bump the ref count
2112 * before sending to hold a reference for retry if you are careful.)
2114 * When calling this method, interrupts MUST be enabled. This is because
2115 * the BH enable code must have IRQs enabled so that it will not deadlock.
2118 int dev_queue_xmit(struct sk_buff *skb)
2120 struct net_device *dev = skb->dev;
2121 struct netdev_queue *txq;
2125 /* GSO will handle the following emulations directly. */
2126 if (netif_needs_gso(dev, skb))
2129 /* Convert a paged skb to linear, if required */
2130 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2133 /* If packet is not checksummed and device does not support
2134 * checksumming for this protocol, complete checksumming here.
2136 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2137 skb_set_transport_header(skb, skb->csum_start -
2139 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2144 /* Disable soft irqs for various locks below. Also
2145 * stops preemption for RCU.
2149 txq = dev_pick_tx(dev, skb);
2150 q = rcu_dereference_bh(txq->qdisc);
2152 #ifdef CONFIG_NET_CLS_ACT
2153 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2156 rc = __dev_xmit_skb(skb, q, dev, txq);
2160 /* The device has no queue. Common case for software devices:
2161 loopback, all the sorts of tunnels...
2163 Really, it is unlikely that netif_tx_lock protection is necessary
2164 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2166 However, it is possible, that they rely on protection
2169 Check this and shot the lock. It is not prone from deadlocks.
2170 Either shot noqueue qdisc, it is even simpler 8)
2172 if (dev->flags & IFF_UP) {
2173 int cpu = smp_processor_id(); /* ok because BHs are off */
2175 if (txq->xmit_lock_owner != cpu) {
2177 HARD_TX_LOCK(dev, txq, cpu);
2179 if (!netif_tx_queue_stopped(txq)) {
2180 rc = dev_hard_start_xmit(skb, dev, txq);
2181 if (dev_xmit_complete(rc)) {
2182 HARD_TX_UNLOCK(dev, txq);
2186 HARD_TX_UNLOCK(dev, txq);
2187 if (net_ratelimit())
2188 printk(KERN_CRIT "Virtual device %s asks to "
2189 "queue packet!\n", dev->name);
2191 /* Recursion is detected! It is possible,
2193 if (net_ratelimit())
2194 printk(KERN_CRIT "Dead loop on virtual device "
2195 "%s, fix it urgently!\n", dev->name);
2200 rcu_read_unlock_bh();
2206 rcu_read_unlock_bh();
2209 EXPORT_SYMBOL(dev_queue_xmit);
2212 /*=======================================================================
2214 =======================================================================*/
2216 int netdev_max_backlog __read_mostly = 1000;
2217 int netdev_tstamp_prequeue __read_mostly = 1;
2218 int netdev_budget __read_mostly = 300;
2219 int weight_p __read_mostly = 64; /* old backlog weight */
2221 /* Called with irq disabled */
2222 static inline void ____napi_schedule(struct softnet_data *sd,
2223 struct napi_struct *napi)
2225 list_add_tail(&napi->poll_list, &sd->poll_list);
2226 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2231 /* One global table that all flow-based protocols share. */
2232 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2233 EXPORT_SYMBOL(rps_sock_flow_table);
2236 * get_rps_cpu is called from netif_receive_skb and returns the target
2237 * CPU from the RPS map of the receiving queue for a given skb.
2238 * rcu_read_lock must be held on entry.
2240 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2241 struct rps_dev_flow **rflowp)
2243 struct ipv6hdr *ip6;
2245 struct netdev_rx_queue *rxqueue;
2246 struct rps_map *map;
2247 struct rps_dev_flow_table *flow_table;
2248 struct rps_sock_flow_table *sock_flow_table;
2252 u32 addr1, addr2, ihl;
2258 if (skb_rx_queue_recorded(skb)) {
2259 u16 index = skb_get_rx_queue(skb);
2260 if (unlikely(index >= dev->num_rx_queues)) {
2261 if (net_ratelimit()) {
2262 pr_warning("%s received packet on queue "
2263 "%u, but number of RX queues is %u\n",
2264 dev->name, index, dev->num_rx_queues);
2268 rxqueue = dev->_rx + index;
2272 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2276 goto got_hash; /* Skip hash computation on packet header */
2278 switch (skb->protocol) {
2279 case __constant_htons(ETH_P_IP):
2280 if (!pskb_may_pull(skb, sizeof(*ip)))
2283 ip = (struct iphdr *) skb->data;
2284 ip_proto = ip->protocol;
2285 addr1 = (__force u32) ip->saddr;
2286 addr2 = (__force u32) ip->daddr;
2289 case __constant_htons(ETH_P_IPV6):
2290 if (!pskb_may_pull(skb, sizeof(*ip6)))
2293 ip6 = (struct ipv6hdr *) skb->data;
2294 ip_proto = ip6->nexthdr;
2295 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2296 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2309 case IPPROTO_UDPLITE:
2310 if (pskb_may_pull(skb, (ihl * 4) + 4)) {
2311 ports.v32 = * (__force u32 *) (skb->data + (ihl * 4));
2312 if (ports.v16[1] < ports.v16[0])
2313 swap(ports.v16[0], ports.v16[1]);
2321 /* get a consistent hash (same value on both flow directions) */
2324 skb->rxhash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2329 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2330 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2331 if (flow_table && sock_flow_table) {
2333 struct rps_dev_flow *rflow;
2335 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2338 next_cpu = sock_flow_table->ents[skb->rxhash &
2339 sock_flow_table->mask];
2342 * If the desired CPU (where last recvmsg was done) is
2343 * different from current CPU (one in the rx-queue flow
2344 * table entry), switch if one of the following holds:
2345 * - Current CPU is unset (equal to RPS_NO_CPU).
2346 * - Current CPU is offline.
2347 * - The current CPU's queue tail has advanced beyond the
2348 * last packet that was enqueued using this table entry.
2349 * This guarantees that all previous packets for the flow
2350 * have been dequeued, thus preserving in order delivery.
2352 if (unlikely(tcpu != next_cpu) &&
2353 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2354 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2355 rflow->last_qtail)) >= 0)) {
2356 tcpu = rflow->cpu = next_cpu;
2357 if (tcpu != RPS_NO_CPU)
2358 rflow->last_qtail = per_cpu(softnet_data,
2359 tcpu).input_queue_head;
2361 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2368 map = rcu_dereference(rxqueue->rps_map);
2370 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2372 if (cpu_online(tcpu)) {
2382 /* Called from hardirq (IPI) context */
2383 static void rps_trigger_softirq(void *data)
2385 struct softnet_data *sd = data;
2387 ____napi_schedule(sd, &sd->backlog);
2391 #endif /* CONFIG_RPS */
2394 * Check if this softnet_data structure is another cpu one
2395 * If yes, queue it to our IPI list and return 1
2398 static int rps_ipi_queued(struct softnet_data *sd)
2401 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2404 sd->rps_ipi_next = mysd->rps_ipi_list;
2405 mysd->rps_ipi_list = sd;
2407 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2410 #endif /* CONFIG_RPS */
2415 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2416 * queue (may be a remote CPU queue).
2418 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2419 unsigned int *qtail)
2421 struct softnet_data *sd;
2422 unsigned long flags;
2424 sd = &per_cpu(softnet_data, cpu);
2426 local_irq_save(flags);
2429 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2430 if (skb_queue_len(&sd->input_pkt_queue)) {
2432 __skb_queue_tail(&sd->input_pkt_queue, skb);
2433 input_queue_tail_incr_save(sd, qtail);
2435 local_irq_restore(flags);
2436 return NET_RX_SUCCESS;
2439 /* Schedule NAPI for backlog device
2440 * We can use non atomic operation since we own the queue lock
2442 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2443 if (!rps_ipi_queued(sd))
2444 ____napi_schedule(sd, &sd->backlog);
2452 local_irq_restore(flags);
2459 * netif_rx - post buffer to the network code
2460 * @skb: buffer to post
2462 * This function receives a packet from a device driver and queues it for
2463 * the upper (protocol) levels to process. It always succeeds. The buffer
2464 * may be dropped during processing for congestion control or by the
2468 * NET_RX_SUCCESS (no congestion)
2469 * NET_RX_DROP (packet was dropped)
2473 int netif_rx(struct sk_buff *skb)
2477 /* if netpoll wants it, pretend we never saw it */
2478 if (netpoll_rx(skb))
2481 if (netdev_tstamp_prequeue)
2482 net_timestamp_check(skb);
2486 struct rps_dev_flow voidflow, *rflow = &voidflow;
2491 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2493 cpu = smp_processor_id();
2495 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2502 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2508 EXPORT_SYMBOL(netif_rx);
2510 int netif_rx_ni(struct sk_buff *skb)
2515 err = netif_rx(skb);
2516 if (local_softirq_pending())
2522 EXPORT_SYMBOL(netif_rx_ni);
2524 static void net_tx_action(struct softirq_action *h)
2526 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2528 if (sd->completion_queue) {
2529 struct sk_buff *clist;
2531 local_irq_disable();
2532 clist = sd->completion_queue;
2533 sd->completion_queue = NULL;
2537 struct sk_buff *skb = clist;
2538 clist = clist->next;
2540 WARN_ON(atomic_read(&skb->users));
2545 if (sd->output_queue) {
2548 local_irq_disable();
2549 head = sd->output_queue;
2550 sd->output_queue = NULL;
2551 sd->output_queue_tailp = &sd->output_queue;
2555 struct Qdisc *q = head;
2556 spinlock_t *root_lock;
2558 head = head->next_sched;
2560 root_lock = qdisc_lock(q);
2561 if (spin_trylock(root_lock)) {
2562 smp_mb__before_clear_bit();
2563 clear_bit(__QDISC_STATE_SCHED,
2566 spin_unlock(root_lock);
2568 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2570 __netif_reschedule(q);
2572 smp_mb__before_clear_bit();
2573 clear_bit(__QDISC_STATE_SCHED,
2581 static inline int deliver_skb(struct sk_buff *skb,
2582 struct packet_type *pt_prev,
2583 struct net_device *orig_dev)
2585 atomic_inc(&skb->users);
2586 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2589 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2591 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2592 /* This hook is defined here for ATM LANE */
2593 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2594 unsigned char *addr) __read_mostly;
2595 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2599 * If bridge module is loaded call bridging hook.
2600 * returns NULL if packet was consumed.
2602 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2603 struct sk_buff *skb) __read_mostly;
2604 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2606 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2607 struct packet_type **pt_prev, int *ret,
2608 struct net_device *orig_dev)
2610 struct net_bridge_port *port;
2612 if (skb->pkt_type == PACKET_LOOPBACK ||
2613 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2617 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2621 return br_handle_frame_hook(port, skb);
2624 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2627 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2628 struct sk_buff *(*macvlan_handle_frame_hook)(struct macvlan_port *p,
2629 struct sk_buff *skb) __read_mostly;
2630 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2632 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2633 struct packet_type **pt_prev,
2635 struct net_device *orig_dev)
2637 struct macvlan_port *port;
2639 port = rcu_dereference(skb->dev->macvlan_port);
2644 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2647 return macvlan_handle_frame_hook(port, skb);
2650 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2653 #ifdef CONFIG_NET_CLS_ACT
2654 /* TODO: Maybe we should just force sch_ingress to be compiled in
2655 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2656 * a compare and 2 stores extra right now if we dont have it on
2657 * but have CONFIG_NET_CLS_ACT
2658 * NOTE: This doesnt stop any functionality; if you dont have
2659 * the ingress scheduler, you just cant add policies on ingress.
2662 static int ing_filter(struct sk_buff *skb)
2664 struct net_device *dev = skb->dev;
2665 u32 ttl = G_TC_RTTL(skb->tc_verd);
2666 struct netdev_queue *rxq;
2667 int result = TC_ACT_OK;
2670 if (MAX_RED_LOOP < ttl++) {
2672 "Redir loop detected Dropping packet (%d->%d)\n",
2673 skb->skb_iif, dev->ifindex);
2677 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2678 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2680 rxq = &dev->rx_queue;
2683 if (q != &noop_qdisc) {
2684 spin_lock(qdisc_lock(q));
2685 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2686 result = qdisc_enqueue_root(skb, q);
2687 spin_unlock(qdisc_lock(q));
2693 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2694 struct packet_type **pt_prev,
2695 int *ret, struct net_device *orig_dev)
2697 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2701 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2704 /* Huh? Why does turning on AF_PACKET affect this? */
2705 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2708 switch (ing_filter(skb)) {
2722 * netif_nit_deliver - deliver received packets to network taps
2725 * This function is used to deliver incoming packets to network
2726 * taps. It should be used when the normal netif_receive_skb path
2727 * is bypassed, for example because of VLAN acceleration.
2729 void netif_nit_deliver(struct sk_buff *skb)
2731 struct packet_type *ptype;
2733 if (list_empty(&ptype_all))
2736 skb_reset_network_header(skb);
2737 skb_reset_transport_header(skb);
2738 skb->mac_len = skb->network_header - skb->mac_header;
2741 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2742 if (!ptype->dev || ptype->dev == skb->dev)
2743 deliver_skb(skb, ptype, skb->dev);
2748 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2749 struct net_device *master)
2751 if (skb->pkt_type == PACKET_HOST) {
2752 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2754 memcpy(dest, master->dev_addr, ETH_ALEN);
2758 /* On bonding slaves other than the currently active slave, suppress
2759 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2760 * ARP on active-backup slaves with arp_validate enabled.
2762 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2764 struct net_device *dev = skb->dev;
2766 if (master->priv_flags & IFF_MASTER_ARPMON)
2767 dev->last_rx = jiffies;
2769 if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) {
2770 /* Do address unmangle. The local destination address
2771 * will be always the one master has. Provides the right
2772 * functionality in a bridge.
2774 skb_bond_set_mac_by_master(skb, master);
2777 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2778 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2779 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2782 if (master->priv_flags & IFF_MASTER_ALB) {
2783 if (skb->pkt_type != PACKET_BROADCAST &&
2784 skb->pkt_type != PACKET_MULTICAST)
2787 if (master->priv_flags & IFF_MASTER_8023AD &&
2788 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2795 EXPORT_SYMBOL(__skb_bond_should_drop);
2797 static int __netif_receive_skb(struct sk_buff *skb)
2799 struct packet_type *ptype, *pt_prev;
2800 struct net_device *orig_dev;
2801 struct net_device *master;
2802 struct net_device *null_or_orig;
2803 struct net_device *null_or_bond;
2804 int ret = NET_RX_DROP;
2807 if (!netdev_tstamp_prequeue)
2808 net_timestamp_check(skb);
2810 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2811 return NET_RX_SUCCESS;
2813 /* if we've gotten here through NAPI, check netpoll */
2814 if (netpoll_receive_skb(skb))
2818 skb->skb_iif = skb->dev->ifindex;
2820 null_or_orig = NULL;
2821 orig_dev = skb->dev;
2822 master = ACCESS_ONCE(orig_dev->master);
2824 if (skb_bond_should_drop(skb, master))
2825 null_or_orig = orig_dev; /* deliver only exact match */
2830 __get_cpu_var(softnet_data).processed++;
2832 skb_reset_network_header(skb);
2833 skb_reset_transport_header(skb);
2834 skb->mac_len = skb->network_header - skb->mac_header;
2840 #ifdef CONFIG_NET_CLS_ACT
2841 if (skb->tc_verd & TC_NCLS) {
2842 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2847 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2848 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2849 ptype->dev == orig_dev) {
2851 ret = deliver_skb(skb, pt_prev, orig_dev);
2856 #ifdef CONFIG_NET_CLS_ACT
2857 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2863 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2866 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2871 * Make sure frames received on VLAN interfaces stacked on
2872 * bonding interfaces still make their way to any base bonding
2873 * device that may have registered for a specific ptype. The
2874 * handler may have to adjust skb->dev and orig_dev.
2876 null_or_bond = NULL;
2877 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2878 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2879 null_or_bond = vlan_dev_real_dev(skb->dev);
2882 type = skb->protocol;
2883 list_for_each_entry_rcu(ptype,
2884 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2885 if (ptype->type == type && (ptype->dev == null_or_orig ||
2886 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2887 ptype->dev == null_or_bond)) {
2889 ret = deliver_skb(skb, pt_prev, orig_dev);
2895 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2898 /* Jamal, now you will not able to escape explaining
2899 * me how you were going to use this. :-)
2910 * netif_receive_skb - process receive buffer from network
2911 * @skb: buffer to process
2913 * netif_receive_skb() is the main receive data processing function.
2914 * It always succeeds. The buffer may be dropped during processing
2915 * for congestion control or by the protocol layers.
2917 * This function may only be called from softirq context and interrupts
2918 * should be enabled.
2920 * Return values (usually ignored):
2921 * NET_RX_SUCCESS: no congestion
2922 * NET_RX_DROP: packet was dropped
2924 int netif_receive_skb(struct sk_buff *skb)
2926 if (netdev_tstamp_prequeue)
2927 net_timestamp_check(skb);
2931 struct rps_dev_flow voidflow, *rflow = &voidflow;
2936 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2939 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2943 ret = __netif_receive_skb(skb);
2949 return __netif_receive_skb(skb);
2952 EXPORT_SYMBOL(netif_receive_skb);
2954 /* Network device is going away, flush any packets still pending
2955 * Called with irqs disabled.
2957 static void flush_backlog(void *arg)
2959 struct net_device *dev = arg;
2960 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2961 struct sk_buff *skb, *tmp;
2964 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
2965 if (skb->dev == dev) {
2966 __skb_unlink(skb, &sd->input_pkt_queue);
2968 input_queue_head_incr(sd);
2973 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
2974 if (skb->dev == dev) {
2975 __skb_unlink(skb, &sd->process_queue);
2977 input_queue_head_incr(sd);
2982 static int napi_gro_complete(struct sk_buff *skb)
2984 struct packet_type *ptype;
2985 __be16 type = skb->protocol;
2986 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2989 if (NAPI_GRO_CB(skb)->count == 1) {
2990 skb_shinfo(skb)->gso_size = 0;
2995 list_for_each_entry_rcu(ptype, head, list) {
2996 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2999 err = ptype->gro_complete(skb);
3005 WARN_ON(&ptype->list == head);
3007 return NET_RX_SUCCESS;
3011 return netif_receive_skb(skb);
3014 static void napi_gro_flush(struct napi_struct *napi)
3016 struct sk_buff *skb, *next;
3018 for (skb = napi->gro_list; skb; skb = next) {
3021 napi_gro_complete(skb);
3024 napi->gro_count = 0;
3025 napi->gro_list = NULL;
3028 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3030 struct sk_buff **pp = NULL;
3031 struct packet_type *ptype;
3032 __be16 type = skb->protocol;
3033 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3036 enum gro_result ret;
3038 if (!(skb->dev->features & NETIF_F_GRO))
3041 if (skb_is_gso(skb) || skb_has_frags(skb))
3045 list_for_each_entry_rcu(ptype, head, list) {
3046 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3049 skb_set_network_header(skb, skb_gro_offset(skb));
3050 mac_len = skb->network_header - skb->mac_header;
3051 skb->mac_len = mac_len;
3052 NAPI_GRO_CB(skb)->same_flow = 0;
3053 NAPI_GRO_CB(skb)->flush = 0;
3054 NAPI_GRO_CB(skb)->free = 0;
3056 pp = ptype->gro_receive(&napi->gro_list, skb);
3061 if (&ptype->list == head)
3064 same_flow = NAPI_GRO_CB(skb)->same_flow;
3065 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3068 struct sk_buff *nskb = *pp;
3072 napi_gro_complete(nskb);
3079 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3083 NAPI_GRO_CB(skb)->count = 1;
3084 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3085 skb->next = napi->gro_list;
3086 napi->gro_list = skb;
3090 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3091 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3093 BUG_ON(skb->end - skb->tail < grow);
3095 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3098 skb->data_len -= grow;
3100 skb_shinfo(skb)->frags[0].page_offset += grow;
3101 skb_shinfo(skb)->frags[0].size -= grow;
3103 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3104 put_page(skb_shinfo(skb)->frags[0].page);
3105 memmove(skb_shinfo(skb)->frags,
3106 skb_shinfo(skb)->frags + 1,
3107 --skb_shinfo(skb)->nr_frags);
3118 EXPORT_SYMBOL(dev_gro_receive);
3121 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3125 if (netpoll_rx_on(skb))
3128 for (p = napi->gro_list; p; p = p->next) {
3129 NAPI_GRO_CB(p)->same_flow =
3130 (p->dev == skb->dev) &&
3131 !compare_ether_header(skb_mac_header(p),
3132 skb_gro_mac_header(skb));
3133 NAPI_GRO_CB(p)->flush = 0;
3136 return dev_gro_receive(napi, skb);
3139 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3143 if (netif_receive_skb(skb))
3148 case GRO_MERGED_FREE:
3159 EXPORT_SYMBOL(napi_skb_finish);
3161 void skb_gro_reset_offset(struct sk_buff *skb)
3163 NAPI_GRO_CB(skb)->data_offset = 0;
3164 NAPI_GRO_CB(skb)->frag0 = NULL;
3165 NAPI_GRO_CB(skb)->frag0_len = 0;
3167 if (skb->mac_header == skb->tail &&
3168 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3169 NAPI_GRO_CB(skb)->frag0 =
3170 page_address(skb_shinfo(skb)->frags[0].page) +
3171 skb_shinfo(skb)->frags[0].page_offset;
3172 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3175 EXPORT_SYMBOL(skb_gro_reset_offset);
3177 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3179 skb_gro_reset_offset(skb);
3181 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3183 EXPORT_SYMBOL(napi_gro_receive);
3185 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3187 __skb_pull(skb, skb_headlen(skb));
3188 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3192 EXPORT_SYMBOL(napi_reuse_skb);
3194 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3196 struct sk_buff *skb = napi->skb;
3199 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3205 EXPORT_SYMBOL(napi_get_frags);
3207 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3213 skb->protocol = eth_type_trans(skb, skb->dev);
3215 if (ret == GRO_HELD)
3216 skb_gro_pull(skb, -ETH_HLEN);
3217 else if (netif_receive_skb(skb))
3222 case GRO_MERGED_FREE:
3223 napi_reuse_skb(napi, skb);
3232 EXPORT_SYMBOL(napi_frags_finish);
3234 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3236 struct sk_buff *skb = napi->skb;
3243 skb_reset_mac_header(skb);
3244 skb_gro_reset_offset(skb);
3246 off = skb_gro_offset(skb);
3247 hlen = off + sizeof(*eth);
3248 eth = skb_gro_header_fast(skb, off);
3249 if (skb_gro_header_hard(skb, hlen)) {
3250 eth = skb_gro_header_slow(skb, hlen, off);
3251 if (unlikely(!eth)) {
3252 napi_reuse_skb(napi, skb);
3258 skb_gro_pull(skb, sizeof(*eth));
3261 * This works because the only protocols we care about don't require
3262 * special handling. We'll fix it up properly at the end.
3264 skb->protocol = eth->h_proto;
3269 EXPORT_SYMBOL(napi_frags_skb);
3271 gro_result_t napi_gro_frags(struct napi_struct *napi)
3273 struct sk_buff *skb = napi_frags_skb(napi);
3278 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3280 EXPORT_SYMBOL(napi_gro_frags);
3283 * net_rps_action sends any pending IPI's for rps.
3284 * Note: called with local irq disabled, but exits with local irq enabled.
3286 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3289 struct softnet_data *remsd = sd->rps_ipi_list;
3292 sd->rps_ipi_list = NULL;
3296 /* Send pending IPI's to kick RPS processing on remote cpus. */
3298 struct softnet_data *next = remsd->rps_ipi_next;
3300 if (cpu_online(remsd->cpu))
3301 __smp_call_function_single(remsd->cpu,
3310 static int process_backlog(struct napi_struct *napi, int quota)
3313 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3316 /* Check if we have pending ipi, its better to send them now,
3317 * not waiting net_rx_action() end.
3319 if (sd->rps_ipi_list) {
3320 local_irq_disable();
3321 net_rps_action_and_irq_enable(sd);
3324 napi->weight = weight_p;
3325 local_irq_disable();
3326 while (work < quota) {
3327 struct sk_buff *skb;
3330 while ((skb = __skb_dequeue(&sd->process_queue))) {
3332 __netif_receive_skb(skb);
3333 local_irq_disable();
3334 input_queue_head_incr(sd);
3335 if (++work >= quota) {
3342 qlen = skb_queue_len(&sd->input_pkt_queue);
3344 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3345 &sd->process_queue);
3347 if (qlen < quota - work) {
3349 * Inline a custom version of __napi_complete().
3350 * only current cpu owns and manipulates this napi,
3351 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3352 * we can use a plain write instead of clear_bit(),
3353 * and we dont need an smp_mb() memory barrier.
3355 list_del(&napi->poll_list);
3358 quota = work + qlen;
3368 * __napi_schedule - schedule for receive
3369 * @n: entry to schedule
3371 * The entry's receive function will be scheduled to run
3373 void __napi_schedule(struct napi_struct *n)
3375 unsigned long flags;
3377 local_irq_save(flags);
3378 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3379 local_irq_restore(flags);
3381 EXPORT_SYMBOL(__napi_schedule);
3383 void __napi_complete(struct napi_struct *n)
3385 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3386 BUG_ON(n->gro_list);
3388 list_del(&n->poll_list);
3389 smp_mb__before_clear_bit();
3390 clear_bit(NAPI_STATE_SCHED, &n->state);
3392 EXPORT_SYMBOL(__napi_complete);
3394 void napi_complete(struct napi_struct *n)
3396 unsigned long flags;
3399 * don't let napi dequeue from the cpu poll list
3400 * just in case its running on a different cpu
3402 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3406 local_irq_save(flags);
3408 local_irq_restore(flags);
3410 EXPORT_SYMBOL(napi_complete);
3412 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3413 int (*poll)(struct napi_struct *, int), int weight)
3415 INIT_LIST_HEAD(&napi->poll_list);
3416 napi->gro_count = 0;
3417 napi->gro_list = NULL;
3420 napi->weight = weight;
3421 list_add(&napi->dev_list, &dev->napi_list);
3423 #ifdef CONFIG_NETPOLL
3424 spin_lock_init(&napi->poll_lock);
3425 napi->poll_owner = -1;
3427 set_bit(NAPI_STATE_SCHED, &napi->state);
3429 EXPORT_SYMBOL(netif_napi_add);
3431 void netif_napi_del(struct napi_struct *napi)
3433 struct sk_buff *skb, *next;
3435 list_del_init(&napi->dev_list);
3436 napi_free_frags(napi);
3438 for (skb = napi->gro_list; skb; skb = next) {
3444 napi->gro_list = NULL;
3445 napi->gro_count = 0;
3447 EXPORT_SYMBOL(netif_napi_del);
3449 static void net_rx_action(struct softirq_action *h)
3451 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3452 unsigned long time_limit = jiffies + 2;
3453 int budget = netdev_budget;
3456 local_irq_disable();
3458 while (!list_empty(&sd->poll_list)) {
3459 struct napi_struct *n;
3462 /* If softirq window is exhuasted then punt.
3463 * Allow this to run for 2 jiffies since which will allow
3464 * an average latency of 1.5/HZ.
3466 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3471 /* Even though interrupts have been re-enabled, this
3472 * access is safe because interrupts can only add new
3473 * entries to the tail of this list, and only ->poll()
3474 * calls can remove this head entry from the list.
3476 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3478 have = netpoll_poll_lock(n);
3482 /* This NAPI_STATE_SCHED test is for avoiding a race
3483 * with netpoll's poll_napi(). Only the entity which
3484 * obtains the lock and sees NAPI_STATE_SCHED set will
3485 * actually make the ->poll() call. Therefore we avoid
3486 * accidently calling ->poll() when NAPI is not scheduled.
3489 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3490 work = n->poll(n, weight);
3494 WARN_ON_ONCE(work > weight);
3498 local_irq_disable();
3500 /* Drivers must not modify the NAPI state if they
3501 * consume the entire weight. In such cases this code
3502 * still "owns" the NAPI instance and therefore can
3503 * move the instance around on the list at-will.
3505 if (unlikely(work == weight)) {
3506 if (unlikely(napi_disable_pending(n))) {
3509 local_irq_disable();
3511 list_move_tail(&n->poll_list, &sd->poll_list);
3514 netpoll_poll_unlock(have);
3517 net_rps_action_and_irq_enable(sd);
3519 #ifdef CONFIG_NET_DMA
3521 * There may not be any more sk_buffs coming right now, so push
3522 * any pending DMA copies to hardware
3524 dma_issue_pending_all();
3531 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3535 static gifconf_func_t *gifconf_list[NPROTO];
3538 * register_gifconf - register a SIOCGIF handler
3539 * @family: Address family
3540 * @gifconf: Function handler
3542 * Register protocol dependent address dumping routines. The handler
3543 * that is passed must not be freed or reused until it has been replaced
3544 * by another handler.
3546 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3548 if (family >= NPROTO)
3550 gifconf_list[family] = gifconf;
3553 EXPORT_SYMBOL(register_gifconf);
3557 * Map an interface index to its name (SIOCGIFNAME)
3561 * We need this ioctl for efficient implementation of the
3562 * if_indextoname() function required by the IPv6 API. Without
3563 * it, we would have to search all the interfaces to find a
3567 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3569 struct net_device *dev;
3573 * Fetch the caller's info block.
3576 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3580 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3586 strcpy(ifr.ifr_name, dev->name);
3589 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3595 * Perform a SIOCGIFCONF call. This structure will change
3596 * size eventually, and there is nothing I can do about it.
3597 * Thus we will need a 'compatibility mode'.
3600 static int dev_ifconf(struct net *net, char __user *arg)
3603 struct net_device *dev;
3610 * Fetch the caller's info block.
3613 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3620 * Loop over the interfaces, and write an info block for each.
3624 for_each_netdev(net, dev) {
3625 for (i = 0; i < NPROTO; i++) {
3626 if (gifconf_list[i]) {
3629 done = gifconf_list[i](dev, NULL, 0);
3631 done = gifconf_list[i](dev, pos + total,
3641 * All done. Write the updated control block back to the caller.
3643 ifc.ifc_len = total;
3646 * Both BSD and Solaris return 0 here, so we do too.
3648 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3651 #ifdef CONFIG_PROC_FS
3653 * This is invoked by the /proc filesystem handler to display a device
3656 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3659 struct net *net = seq_file_net(seq);
3661 struct net_device *dev;
3665 return SEQ_START_TOKEN;
3668 for_each_netdev_rcu(net, dev)
3675 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3677 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3678 first_net_device(seq_file_net(seq)) :
3679 next_net_device((struct net_device *)v);
3682 return rcu_dereference(dev);
3685 void dev_seq_stop(struct seq_file *seq, void *v)
3691 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3693 const struct net_device_stats *stats = dev_get_stats(dev);
3695 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3696 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3697 dev->name, stats->rx_bytes, stats->rx_packets,
3699 stats->rx_dropped + stats->rx_missed_errors,
3700 stats->rx_fifo_errors,
3701 stats->rx_length_errors + stats->rx_over_errors +
3702 stats->rx_crc_errors + stats->rx_frame_errors,
3703 stats->rx_compressed, stats->multicast,
3704 stats->tx_bytes, stats->tx_packets,
3705 stats->tx_errors, stats->tx_dropped,
3706 stats->tx_fifo_errors, stats->collisions,
3707 stats->tx_carrier_errors +
3708 stats->tx_aborted_errors +
3709 stats->tx_window_errors +
3710 stats->tx_heartbeat_errors,
3711 stats->tx_compressed);
3715 * Called from the PROCfs module. This now uses the new arbitrary sized
3716 * /proc/net interface to create /proc/net/dev
3718 static int dev_seq_show(struct seq_file *seq, void *v)
3720 if (v == SEQ_START_TOKEN)
3721 seq_puts(seq, "Inter-| Receive "
3723 " face |bytes packets errs drop fifo frame "
3724 "compressed multicast|bytes packets errs "
3725 "drop fifo colls carrier compressed\n");
3727 dev_seq_printf_stats(seq, v);
3731 static struct softnet_data *softnet_get_online(loff_t *pos)
3733 struct softnet_data *sd = NULL;
3735 while (*pos < nr_cpu_ids)
3736 if (cpu_online(*pos)) {
3737 sd = &per_cpu(softnet_data, *pos);
3744 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3746 return softnet_get_online(pos);
3749 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3752 return softnet_get_online(pos);
3755 static void softnet_seq_stop(struct seq_file *seq, void *v)
3759 static int softnet_seq_show(struct seq_file *seq, void *v)
3761 struct softnet_data *sd = v;
3763 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3764 sd->processed, sd->dropped, sd->time_squeeze, 0,
3765 0, 0, 0, 0, /* was fastroute */
3766 sd->cpu_collision, sd->received_rps);
3770 static const struct seq_operations dev_seq_ops = {
3771 .start = dev_seq_start,
3772 .next = dev_seq_next,
3773 .stop = dev_seq_stop,
3774 .show = dev_seq_show,
3777 static int dev_seq_open(struct inode *inode, struct file *file)
3779 return seq_open_net(inode, file, &dev_seq_ops,
3780 sizeof(struct seq_net_private));
3783 static const struct file_operations dev_seq_fops = {
3784 .owner = THIS_MODULE,
3785 .open = dev_seq_open,
3787 .llseek = seq_lseek,
3788 .release = seq_release_net,
3791 static const struct seq_operations softnet_seq_ops = {
3792 .start = softnet_seq_start,
3793 .next = softnet_seq_next,
3794 .stop = softnet_seq_stop,
3795 .show = softnet_seq_show,
3798 static int softnet_seq_open(struct inode *inode, struct file *file)
3800 return seq_open(file, &softnet_seq_ops);
3803 static const struct file_operations softnet_seq_fops = {
3804 .owner = THIS_MODULE,
3805 .open = softnet_seq_open,
3807 .llseek = seq_lseek,
3808 .release = seq_release,
3811 static void *ptype_get_idx(loff_t pos)
3813 struct packet_type *pt = NULL;
3817 list_for_each_entry_rcu(pt, &ptype_all, list) {
3823 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3824 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3833 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3837 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3840 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3842 struct packet_type *pt;
3843 struct list_head *nxt;
3847 if (v == SEQ_START_TOKEN)
3848 return ptype_get_idx(0);
3851 nxt = pt->list.next;
3852 if (pt->type == htons(ETH_P_ALL)) {
3853 if (nxt != &ptype_all)
3856 nxt = ptype_base[0].next;
3858 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3860 while (nxt == &ptype_base[hash]) {
3861 if (++hash >= PTYPE_HASH_SIZE)
3863 nxt = ptype_base[hash].next;
3866 return list_entry(nxt, struct packet_type, list);
3869 static void ptype_seq_stop(struct seq_file *seq, void *v)
3875 static int ptype_seq_show(struct seq_file *seq, void *v)
3877 struct packet_type *pt = v;
3879 if (v == SEQ_START_TOKEN)
3880 seq_puts(seq, "Type Device Function\n");
3881 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3882 if (pt->type == htons(ETH_P_ALL))
3883 seq_puts(seq, "ALL ");
3885 seq_printf(seq, "%04x", ntohs(pt->type));
3887 seq_printf(seq, " %-8s %pF\n",
3888 pt->dev ? pt->dev->name : "", pt->func);
3894 static const struct seq_operations ptype_seq_ops = {
3895 .start = ptype_seq_start,
3896 .next = ptype_seq_next,
3897 .stop = ptype_seq_stop,
3898 .show = ptype_seq_show,
3901 static int ptype_seq_open(struct inode *inode, struct file *file)
3903 return seq_open_net(inode, file, &ptype_seq_ops,
3904 sizeof(struct seq_net_private));
3907 static const struct file_operations ptype_seq_fops = {
3908 .owner = THIS_MODULE,
3909 .open = ptype_seq_open,
3911 .llseek = seq_lseek,
3912 .release = seq_release_net,
3916 static int __net_init dev_proc_net_init(struct net *net)
3920 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3922 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3924 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3927 if (wext_proc_init(net))
3933 proc_net_remove(net, "ptype");
3935 proc_net_remove(net, "softnet_stat");
3937 proc_net_remove(net, "dev");
3941 static void __net_exit dev_proc_net_exit(struct net *net)
3943 wext_proc_exit(net);
3945 proc_net_remove(net, "ptype");
3946 proc_net_remove(net, "softnet_stat");
3947 proc_net_remove(net, "dev");
3950 static struct pernet_operations __net_initdata dev_proc_ops = {
3951 .init = dev_proc_net_init,
3952 .exit = dev_proc_net_exit,
3955 static int __init dev_proc_init(void)
3957 return register_pernet_subsys(&dev_proc_ops);
3960 #define dev_proc_init() 0
3961 #endif /* CONFIG_PROC_FS */
3965 * netdev_set_master - set up master/slave pair
3966 * @slave: slave device
3967 * @master: new master device
3969 * Changes the master device of the slave. Pass %NULL to break the
3970 * bonding. The caller must hold the RTNL semaphore. On a failure
3971 * a negative errno code is returned. On success the reference counts
3972 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3973 * function returns zero.
3975 int netdev_set_master(struct net_device *slave, struct net_device *master)
3977 struct net_device *old = slave->master;
3987 slave->master = master;
3994 slave->flags |= IFF_SLAVE;
3996 slave->flags &= ~IFF_SLAVE;
3998 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4001 EXPORT_SYMBOL(netdev_set_master);
4003 static void dev_change_rx_flags(struct net_device *dev, int flags)
4005 const struct net_device_ops *ops = dev->netdev_ops;
4007 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4008 ops->ndo_change_rx_flags(dev, flags);
4011 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4013 unsigned short old_flags = dev->flags;
4019 dev->flags |= IFF_PROMISC;
4020 dev->promiscuity += inc;
4021 if (dev->promiscuity == 0) {
4024 * If inc causes overflow, untouch promisc and return error.
4027 dev->flags &= ~IFF_PROMISC;
4029 dev->promiscuity -= inc;
4030 printk(KERN_WARNING "%s: promiscuity touches roof, "
4031 "set promiscuity failed, promiscuity feature "
4032 "of device might be broken.\n", dev->name);
4036 if (dev->flags != old_flags) {
4037 printk(KERN_INFO "device %s %s promiscuous mode\n",
4038 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4040 if (audit_enabled) {
4041 current_uid_gid(&uid, &gid);
4042 audit_log(current->audit_context, GFP_ATOMIC,
4043 AUDIT_ANOM_PROMISCUOUS,
4044 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4045 dev->name, (dev->flags & IFF_PROMISC),
4046 (old_flags & IFF_PROMISC),
4047 audit_get_loginuid(current),
4049 audit_get_sessionid(current));
4052 dev_change_rx_flags(dev, IFF_PROMISC);
4058 * dev_set_promiscuity - update promiscuity count on a device
4062 * Add or remove promiscuity from a device. While the count in the device
4063 * remains above zero the interface remains promiscuous. Once it hits zero
4064 * the device reverts back to normal filtering operation. A negative inc
4065 * value is used to drop promiscuity on the device.
4066 * Return 0 if successful or a negative errno code on error.
4068 int dev_set_promiscuity(struct net_device *dev, int inc)
4070 unsigned short old_flags = dev->flags;
4073 err = __dev_set_promiscuity(dev, inc);
4076 if (dev->flags != old_flags)
4077 dev_set_rx_mode(dev);
4080 EXPORT_SYMBOL(dev_set_promiscuity);
4083 * dev_set_allmulti - update allmulti count on a device
4087 * Add or remove reception of all multicast frames to a device. While the
4088 * count in the device remains above zero the interface remains listening
4089 * to all interfaces. Once it hits zero the device reverts back to normal
4090 * filtering operation. A negative @inc value is used to drop the counter
4091 * when releasing a resource needing all multicasts.
4092 * Return 0 if successful or a negative errno code on error.
4095 int dev_set_allmulti(struct net_device *dev, int inc)
4097 unsigned short old_flags = dev->flags;
4101 dev->flags |= IFF_ALLMULTI;
4102 dev->allmulti += inc;
4103 if (dev->allmulti == 0) {
4106 * If inc causes overflow, untouch allmulti and return error.
4109 dev->flags &= ~IFF_ALLMULTI;
4111 dev->allmulti -= inc;
4112 printk(KERN_WARNING "%s: allmulti touches roof, "
4113 "set allmulti failed, allmulti feature of "
4114 "device might be broken.\n", dev->name);
4118 if (dev->flags ^ old_flags) {
4119 dev_change_rx_flags(dev, IFF_ALLMULTI);
4120 dev_set_rx_mode(dev);
4124 EXPORT_SYMBOL(dev_set_allmulti);
4127 * Upload unicast and multicast address lists to device and
4128 * configure RX filtering. When the device doesn't support unicast
4129 * filtering it is put in promiscuous mode while unicast addresses
4132 void __dev_set_rx_mode(struct net_device *dev)
4134 const struct net_device_ops *ops = dev->netdev_ops;
4136 /* dev_open will call this function so the list will stay sane. */
4137 if (!(dev->flags&IFF_UP))
4140 if (!netif_device_present(dev))
4143 if (ops->ndo_set_rx_mode)
4144 ops->ndo_set_rx_mode(dev);
4146 /* Unicast addresses changes may only happen under the rtnl,
4147 * therefore calling __dev_set_promiscuity here is safe.
4149 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4150 __dev_set_promiscuity(dev, 1);
4151 dev->uc_promisc = 1;
4152 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4153 __dev_set_promiscuity(dev, -1);
4154 dev->uc_promisc = 0;
4157 if (ops->ndo_set_multicast_list)
4158 ops->ndo_set_multicast_list(dev);
4162 void dev_set_rx_mode(struct net_device *dev)
4164 netif_addr_lock_bh(dev);
4165 __dev_set_rx_mode(dev);
4166 netif_addr_unlock_bh(dev);
4170 * dev_get_flags - get flags reported to userspace
4173 * Get the combination of flag bits exported through APIs to userspace.
4175 unsigned dev_get_flags(const struct net_device *dev)
4179 flags = (dev->flags & ~(IFF_PROMISC |
4184 (dev->gflags & (IFF_PROMISC |
4187 if (netif_running(dev)) {
4188 if (netif_oper_up(dev))
4189 flags |= IFF_RUNNING;
4190 if (netif_carrier_ok(dev))
4191 flags |= IFF_LOWER_UP;
4192 if (netif_dormant(dev))
4193 flags |= IFF_DORMANT;
4198 EXPORT_SYMBOL(dev_get_flags);
4200 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4202 int old_flags = dev->flags;
4208 * Set the flags on our device.
4211 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4212 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4214 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4218 * Load in the correct multicast list now the flags have changed.
4221 if ((old_flags ^ flags) & IFF_MULTICAST)
4222 dev_change_rx_flags(dev, IFF_MULTICAST);
4224 dev_set_rx_mode(dev);
4227 * Have we downed the interface. We handle IFF_UP ourselves
4228 * according to user attempts to set it, rather than blindly
4233 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4234 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4237 dev_set_rx_mode(dev);
4240 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4241 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4243 dev->gflags ^= IFF_PROMISC;
4244 dev_set_promiscuity(dev, inc);
4247 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4248 is important. Some (broken) drivers set IFF_PROMISC, when
4249 IFF_ALLMULTI is requested not asking us and not reporting.
4251 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4252 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4254 dev->gflags ^= IFF_ALLMULTI;
4255 dev_set_allmulti(dev, inc);
4261 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4263 unsigned int changes = dev->flags ^ old_flags;
4265 if (changes & IFF_UP) {
4266 if (dev->flags & IFF_UP)
4267 call_netdevice_notifiers(NETDEV_UP, dev);
4269 call_netdevice_notifiers(NETDEV_DOWN, dev);
4272 if (dev->flags & IFF_UP &&
4273 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4274 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4278 * dev_change_flags - change device settings
4280 * @flags: device state flags
4282 * Change settings on device based state flags. The flags are
4283 * in the userspace exported format.
4285 int dev_change_flags(struct net_device *dev, unsigned flags)
4288 int old_flags = dev->flags;
4290 ret = __dev_change_flags(dev, flags);
4294 changes = old_flags ^ dev->flags;
4296 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4298 __dev_notify_flags(dev, old_flags);
4301 EXPORT_SYMBOL(dev_change_flags);
4304 * dev_set_mtu - Change maximum transfer unit
4306 * @new_mtu: new transfer unit
4308 * Change the maximum transfer size of the network device.
4310 int dev_set_mtu(struct net_device *dev, int new_mtu)
4312 const struct net_device_ops *ops = dev->netdev_ops;
4315 if (new_mtu == dev->mtu)
4318 /* MTU must be positive. */
4322 if (!netif_device_present(dev))
4326 if (ops->ndo_change_mtu)
4327 err = ops->ndo_change_mtu(dev, new_mtu);
4331 if (!err && dev->flags & IFF_UP)
4332 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4335 EXPORT_SYMBOL(dev_set_mtu);
4338 * dev_set_mac_address - Change Media Access Control Address
4342 * Change the hardware (MAC) address of the device
4344 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4346 const struct net_device_ops *ops = dev->netdev_ops;
4349 if (!ops->ndo_set_mac_address)
4351 if (sa->sa_family != dev->type)
4353 if (!netif_device_present(dev))
4355 err = ops->ndo_set_mac_address(dev, sa);
4357 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4360 EXPORT_SYMBOL(dev_set_mac_address);
4363 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4365 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4368 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4374 case SIOCGIFFLAGS: /* Get interface flags */
4375 ifr->ifr_flags = (short) dev_get_flags(dev);
4378 case SIOCGIFMETRIC: /* Get the metric on the interface
4379 (currently unused) */
4380 ifr->ifr_metric = 0;
4383 case SIOCGIFMTU: /* Get the MTU of a device */
4384 ifr->ifr_mtu = dev->mtu;
4389 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4391 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4392 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4393 ifr->ifr_hwaddr.sa_family = dev->type;
4401 ifr->ifr_map.mem_start = dev->mem_start;
4402 ifr->ifr_map.mem_end = dev->mem_end;
4403 ifr->ifr_map.base_addr = dev->base_addr;
4404 ifr->ifr_map.irq = dev->irq;
4405 ifr->ifr_map.dma = dev->dma;
4406 ifr->ifr_map.port = dev->if_port;
4410 ifr->ifr_ifindex = dev->ifindex;
4414 ifr->ifr_qlen = dev->tx_queue_len;
4418 /* dev_ioctl() should ensure this case
4430 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4432 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4435 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4436 const struct net_device_ops *ops;
4441 ops = dev->netdev_ops;
4444 case SIOCSIFFLAGS: /* Set interface flags */
4445 return dev_change_flags(dev, ifr->ifr_flags);
4447 case SIOCSIFMETRIC: /* Set the metric on the interface
4448 (currently unused) */
4451 case SIOCSIFMTU: /* Set the MTU of a device */
4452 return dev_set_mtu(dev, ifr->ifr_mtu);
4455 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4457 case SIOCSIFHWBROADCAST:
4458 if (ifr->ifr_hwaddr.sa_family != dev->type)
4460 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4461 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4462 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4466 if (ops->ndo_set_config) {
4467 if (!netif_device_present(dev))
4469 return ops->ndo_set_config(dev, &ifr->ifr_map);
4474 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4475 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4477 if (!netif_device_present(dev))
4479 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4482 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4483 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4485 if (!netif_device_present(dev))
4487 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4490 if (ifr->ifr_qlen < 0)
4492 dev->tx_queue_len = ifr->ifr_qlen;
4496 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4497 return dev_change_name(dev, ifr->ifr_newname);
4500 * Unknown or private ioctl
4503 if ((cmd >= SIOCDEVPRIVATE &&
4504 cmd <= SIOCDEVPRIVATE + 15) ||
4505 cmd == SIOCBONDENSLAVE ||
4506 cmd == SIOCBONDRELEASE ||
4507 cmd == SIOCBONDSETHWADDR ||
4508 cmd == SIOCBONDSLAVEINFOQUERY ||
4509 cmd == SIOCBONDINFOQUERY ||
4510 cmd == SIOCBONDCHANGEACTIVE ||
4511 cmd == SIOCGMIIPHY ||
4512 cmd == SIOCGMIIREG ||
4513 cmd == SIOCSMIIREG ||
4514 cmd == SIOCBRADDIF ||
4515 cmd == SIOCBRDELIF ||
4516 cmd == SIOCSHWTSTAMP ||
4517 cmd == SIOCWANDEV) {
4519 if (ops->ndo_do_ioctl) {
4520 if (netif_device_present(dev))
4521 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4533 * This function handles all "interface"-type I/O control requests. The actual
4534 * 'doing' part of this is dev_ifsioc above.
4538 * dev_ioctl - network device ioctl
4539 * @net: the applicable net namespace
4540 * @cmd: command to issue
4541 * @arg: pointer to a struct ifreq in user space
4543 * Issue ioctl functions to devices. This is normally called by the
4544 * user space syscall interfaces but can sometimes be useful for
4545 * other purposes. The return value is the return from the syscall if
4546 * positive or a negative errno code on error.
4549 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4555 /* One special case: SIOCGIFCONF takes ifconf argument
4556 and requires shared lock, because it sleeps writing
4560 if (cmd == SIOCGIFCONF) {
4562 ret = dev_ifconf(net, (char __user *) arg);
4566 if (cmd == SIOCGIFNAME)
4567 return dev_ifname(net, (struct ifreq __user *)arg);
4569 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4572 ifr.ifr_name[IFNAMSIZ-1] = 0;
4574 colon = strchr(ifr.ifr_name, ':');
4579 * See which interface the caller is talking about.
4584 * These ioctl calls:
4585 * - can be done by all.
4586 * - atomic and do not require locking.
4597 dev_load(net, ifr.ifr_name);
4599 ret = dev_ifsioc_locked(net, &ifr, cmd);
4604 if (copy_to_user(arg, &ifr,
4605 sizeof(struct ifreq)))
4611 dev_load(net, ifr.ifr_name);
4613 ret = dev_ethtool(net, &ifr);
4618 if (copy_to_user(arg, &ifr,
4619 sizeof(struct ifreq)))
4625 * These ioctl calls:
4626 * - require superuser power.
4627 * - require strict serialization.
4633 if (!capable(CAP_NET_ADMIN))
4635 dev_load(net, ifr.ifr_name);
4637 ret = dev_ifsioc(net, &ifr, cmd);
4642 if (copy_to_user(arg, &ifr,
4643 sizeof(struct ifreq)))
4649 * These ioctl calls:
4650 * - require superuser power.
4651 * - require strict serialization.
4652 * - do not return a value
4662 case SIOCSIFHWBROADCAST:
4665 case SIOCBONDENSLAVE:
4666 case SIOCBONDRELEASE:
4667 case SIOCBONDSETHWADDR:
4668 case SIOCBONDCHANGEACTIVE:
4672 if (!capable(CAP_NET_ADMIN))
4675 case SIOCBONDSLAVEINFOQUERY:
4676 case SIOCBONDINFOQUERY:
4677 dev_load(net, ifr.ifr_name);
4679 ret = dev_ifsioc(net, &ifr, cmd);
4684 /* Get the per device memory space. We can add this but
4685 * currently do not support it */
4687 /* Set the per device memory buffer space.
4688 * Not applicable in our case */
4693 * Unknown or private ioctl.
4696 if (cmd == SIOCWANDEV ||
4697 (cmd >= SIOCDEVPRIVATE &&
4698 cmd <= SIOCDEVPRIVATE + 15)) {
4699 dev_load(net, ifr.ifr_name);
4701 ret = dev_ifsioc(net, &ifr, cmd);
4703 if (!ret && copy_to_user(arg, &ifr,
4704 sizeof(struct ifreq)))
4708 /* Take care of Wireless Extensions */
4709 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4710 return wext_handle_ioctl(net, &ifr, cmd, arg);
4717 * dev_new_index - allocate an ifindex
4718 * @net: the applicable net namespace
4720 * Returns a suitable unique value for a new device interface
4721 * number. The caller must hold the rtnl semaphore or the
4722 * dev_base_lock to be sure it remains unique.
4724 static int dev_new_index(struct net *net)
4730 if (!__dev_get_by_index(net, ifindex))
4735 /* Delayed registration/unregisteration */
4736 static LIST_HEAD(net_todo_list);
4738 static void net_set_todo(struct net_device *dev)
4740 list_add_tail(&dev->todo_list, &net_todo_list);
4743 static void rollback_registered_many(struct list_head *head)
4745 struct net_device *dev, *tmp;
4747 BUG_ON(dev_boot_phase);
4750 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4751 /* Some devices call without registering
4752 * for initialization unwind. Remove those
4753 * devices and proceed with the remaining.
4755 if (dev->reg_state == NETREG_UNINITIALIZED) {
4756 pr_debug("unregister_netdevice: device %s/%p never "
4757 "was registered\n", dev->name, dev);
4760 list_del(&dev->unreg_list);
4764 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4766 /* If device is running, close it first. */
4769 /* And unlink it from device chain. */
4770 unlist_netdevice(dev);
4772 dev->reg_state = NETREG_UNREGISTERING;
4777 list_for_each_entry(dev, head, unreg_list) {
4778 /* Shutdown queueing discipline. */
4782 /* Notify protocols, that we are about to destroy
4783 this device. They should clean all the things.
4785 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4787 if (!dev->rtnl_link_ops ||
4788 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4789 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4792 * Flush the unicast and multicast chains
4797 if (dev->netdev_ops->ndo_uninit)
4798 dev->netdev_ops->ndo_uninit(dev);
4800 /* Notifier chain MUST detach us from master device. */
4801 WARN_ON(dev->master);
4803 /* Remove entries from kobject tree */
4804 netdev_unregister_kobject(dev);
4807 /* Process any work delayed until the end of the batch */
4808 dev = list_first_entry(head, struct net_device, unreg_list);
4809 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4813 list_for_each_entry(dev, head, unreg_list)
4817 static void rollback_registered(struct net_device *dev)
4821 list_add(&dev->unreg_list, &single);
4822 rollback_registered_many(&single);
4825 static void __netdev_init_queue_locks_one(struct net_device *dev,
4826 struct netdev_queue *dev_queue,
4829 spin_lock_init(&dev_queue->_xmit_lock);
4830 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4831 dev_queue->xmit_lock_owner = -1;
4834 static void netdev_init_queue_locks(struct net_device *dev)
4836 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4837 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4840 unsigned long netdev_fix_features(unsigned long features, const char *name)
4842 /* Fix illegal SG+CSUM combinations. */
4843 if ((features & NETIF_F_SG) &&
4844 !(features & NETIF_F_ALL_CSUM)) {
4846 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4847 "checksum feature.\n", name);
4848 features &= ~NETIF_F_SG;
4851 /* TSO requires that SG is present as well. */
4852 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4854 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4855 "SG feature.\n", name);
4856 features &= ~NETIF_F_TSO;
4859 if (features & NETIF_F_UFO) {
4860 if (!(features & NETIF_F_GEN_CSUM)) {
4862 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4863 "since no NETIF_F_HW_CSUM feature.\n",
4865 features &= ~NETIF_F_UFO;
4868 if (!(features & NETIF_F_SG)) {
4870 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4871 "since no NETIF_F_SG feature.\n", name);
4872 features &= ~NETIF_F_UFO;
4878 EXPORT_SYMBOL(netdev_fix_features);
4881 * netif_stacked_transfer_operstate - transfer operstate
4882 * @rootdev: the root or lower level device to transfer state from
4883 * @dev: the device to transfer operstate to
4885 * Transfer operational state from root to device. This is normally
4886 * called when a stacking relationship exists between the root
4887 * device and the device(a leaf device).
4889 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4890 struct net_device *dev)
4892 if (rootdev->operstate == IF_OPER_DORMANT)
4893 netif_dormant_on(dev);
4895 netif_dormant_off(dev);
4897 if (netif_carrier_ok(rootdev)) {
4898 if (!netif_carrier_ok(dev))
4899 netif_carrier_on(dev);
4901 if (netif_carrier_ok(dev))
4902 netif_carrier_off(dev);
4905 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4908 * register_netdevice - register a network device
4909 * @dev: device to register
4911 * Take a completed network device structure and add it to the kernel
4912 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4913 * chain. 0 is returned on success. A negative errno code is returned
4914 * on a failure to set up the device, or if the name is a duplicate.
4916 * Callers must hold the rtnl semaphore. You may want
4917 * register_netdev() instead of this.
4920 * The locking appears insufficient to guarantee two parallel registers
4921 * will not get the same name.
4924 int register_netdevice(struct net_device *dev)
4927 struct net *net = dev_net(dev);
4929 BUG_ON(dev_boot_phase);
4934 /* When net_device's are persistent, this will be fatal. */
4935 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4938 spin_lock_init(&dev->addr_list_lock);
4939 netdev_set_addr_lockdep_class(dev);
4940 netdev_init_queue_locks(dev);
4945 if (!dev->num_rx_queues) {
4947 * Allocate a single RX queue if driver never called
4951 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
4957 dev->_rx->first = dev->_rx;
4958 atomic_set(&dev->_rx->count, 1);
4959 dev->num_rx_queues = 1;
4962 /* Init, if this function is available */
4963 if (dev->netdev_ops->ndo_init) {
4964 ret = dev->netdev_ops->ndo_init(dev);
4972 ret = dev_get_valid_name(dev, dev->name, 0);
4976 dev->ifindex = dev_new_index(net);
4977 if (dev->iflink == -1)
4978 dev->iflink = dev->ifindex;
4980 /* Fix illegal checksum combinations */
4981 if ((dev->features & NETIF_F_HW_CSUM) &&
4982 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4983 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4985 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4988 if ((dev->features & NETIF_F_NO_CSUM) &&
4989 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4990 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4992 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4995 dev->features = netdev_fix_features(dev->features, dev->name);
4997 /* Enable software GSO if SG is supported. */
4998 if (dev->features & NETIF_F_SG)
4999 dev->features |= NETIF_F_GSO;
5001 netdev_initialize_kobject(dev);
5003 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5004 ret = notifier_to_errno(ret);
5008 ret = netdev_register_kobject(dev);
5011 dev->reg_state = NETREG_REGISTERED;
5014 * Default initial state at registry is that the
5015 * device is present.
5018 set_bit(__LINK_STATE_PRESENT, &dev->state);
5020 dev_init_scheduler(dev);
5022 list_netdevice(dev);
5024 /* Notify protocols, that a new device appeared. */
5025 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5026 ret = notifier_to_errno(ret);
5028 rollback_registered(dev);
5029 dev->reg_state = NETREG_UNREGISTERED;
5032 * Prevent userspace races by waiting until the network
5033 * device is fully setup before sending notifications.
5035 if (!dev->rtnl_link_ops ||
5036 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5037 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5043 if (dev->netdev_ops->ndo_uninit)
5044 dev->netdev_ops->ndo_uninit(dev);
5047 EXPORT_SYMBOL(register_netdevice);
5050 * init_dummy_netdev - init a dummy network device for NAPI
5051 * @dev: device to init
5053 * This takes a network device structure and initialize the minimum
5054 * amount of fields so it can be used to schedule NAPI polls without
5055 * registering a full blown interface. This is to be used by drivers
5056 * that need to tie several hardware interfaces to a single NAPI
5057 * poll scheduler due to HW limitations.
5059 int init_dummy_netdev(struct net_device *dev)
5061 /* Clear everything. Note we don't initialize spinlocks
5062 * are they aren't supposed to be taken by any of the
5063 * NAPI code and this dummy netdev is supposed to be
5064 * only ever used for NAPI polls
5066 memset(dev, 0, sizeof(struct net_device));
5068 /* make sure we BUG if trying to hit standard
5069 * register/unregister code path
5071 dev->reg_state = NETREG_DUMMY;
5073 /* initialize the ref count */
5074 atomic_set(&dev->refcnt, 1);
5076 /* NAPI wants this */
5077 INIT_LIST_HEAD(&dev->napi_list);
5079 /* a dummy interface is started by default */
5080 set_bit(__LINK_STATE_PRESENT, &dev->state);
5081 set_bit(__LINK_STATE_START, &dev->state);
5085 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5089 * register_netdev - register a network device
5090 * @dev: device to register
5092 * Take a completed network device structure and add it to the kernel
5093 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5094 * chain. 0 is returned on success. A negative errno code is returned
5095 * on a failure to set up the device, or if the name is a duplicate.
5097 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5098 * and expands the device name if you passed a format string to
5101 int register_netdev(struct net_device *dev)
5108 * If the name is a format string the caller wants us to do a
5111 if (strchr(dev->name, '%')) {
5112 err = dev_alloc_name(dev, dev->name);
5117 err = register_netdevice(dev);
5122 EXPORT_SYMBOL(register_netdev);
5125 * netdev_wait_allrefs - wait until all references are gone.
5127 * This is called when unregistering network devices.
5129 * Any protocol or device that holds a reference should register
5130 * for netdevice notification, and cleanup and put back the
5131 * reference if they receive an UNREGISTER event.
5132 * We can get stuck here if buggy protocols don't correctly
5135 static void netdev_wait_allrefs(struct net_device *dev)
5137 unsigned long rebroadcast_time, warning_time;
5139 linkwatch_forget_dev(dev);
5141 rebroadcast_time = warning_time = jiffies;
5142 while (atomic_read(&dev->refcnt) != 0) {
5143 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5146 /* Rebroadcast unregister notification */
5147 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5148 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5149 * should have already handle it the first time */
5151 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5153 /* We must not have linkwatch events
5154 * pending on unregister. If this
5155 * happens, we simply run the queue
5156 * unscheduled, resulting in a noop
5159 linkwatch_run_queue();
5164 rebroadcast_time = jiffies;
5169 if (time_after(jiffies, warning_time + 10 * HZ)) {
5170 printk(KERN_EMERG "unregister_netdevice: "
5171 "waiting for %s to become free. Usage "
5173 dev->name, atomic_read(&dev->refcnt));
5174 warning_time = jiffies;
5183 * register_netdevice(x1);
5184 * register_netdevice(x2);
5186 * unregister_netdevice(y1);
5187 * unregister_netdevice(y2);
5193 * We are invoked by rtnl_unlock().
5194 * This allows us to deal with problems:
5195 * 1) We can delete sysfs objects which invoke hotplug
5196 * without deadlocking with linkwatch via keventd.
5197 * 2) Since we run with the RTNL semaphore not held, we can sleep
5198 * safely in order to wait for the netdev refcnt to drop to zero.
5200 * We must not return until all unregister events added during
5201 * the interval the lock was held have been completed.
5203 void netdev_run_todo(void)
5205 struct list_head list;
5207 /* Snapshot list, allow later requests */
5208 list_replace_init(&net_todo_list, &list);
5212 while (!list_empty(&list)) {
5213 struct net_device *dev
5214 = list_first_entry(&list, struct net_device, todo_list);
5215 list_del(&dev->todo_list);
5217 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5218 printk(KERN_ERR "network todo '%s' but state %d\n",
5219 dev->name, dev->reg_state);
5224 dev->reg_state = NETREG_UNREGISTERED;
5226 on_each_cpu(flush_backlog, dev, 1);
5228 netdev_wait_allrefs(dev);
5231 BUG_ON(atomic_read(&dev->refcnt));
5232 WARN_ON(dev->ip_ptr);
5233 WARN_ON(dev->ip6_ptr);
5234 WARN_ON(dev->dn_ptr);
5236 if (dev->destructor)
5237 dev->destructor(dev);
5239 /* Free network device */
5240 kobject_put(&dev->dev.kobj);
5245 * dev_txq_stats_fold - fold tx_queues stats
5246 * @dev: device to get statistics from
5247 * @stats: struct net_device_stats to hold results
5249 void dev_txq_stats_fold(const struct net_device *dev,
5250 struct net_device_stats *stats)
5252 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5254 struct netdev_queue *txq;
5256 for (i = 0; i < dev->num_tx_queues; i++) {
5257 txq = netdev_get_tx_queue(dev, i);
5258 tx_bytes += txq->tx_bytes;
5259 tx_packets += txq->tx_packets;
5260 tx_dropped += txq->tx_dropped;
5262 if (tx_bytes || tx_packets || tx_dropped) {
5263 stats->tx_bytes = tx_bytes;
5264 stats->tx_packets = tx_packets;
5265 stats->tx_dropped = tx_dropped;
5268 EXPORT_SYMBOL(dev_txq_stats_fold);
5271 * dev_get_stats - get network device statistics
5272 * @dev: device to get statistics from
5274 * Get network statistics from device. The device driver may provide
5275 * its own method by setting dev->netdev_ops->get_stats; otherwise
5276 * the internal statistics structure is used.
5278 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5280 const struct net_device_ops *ops = dev->netdev_ops;
5282 if (ops->ndo_get_stats)
5283 return ops->ndo_get_stats(dev);
5285 dev_txq_stats_fold(dev, &dev->stats);
5288 EXPORT_SYMBOL(dev_get_stats);
5290 static void netdev_init_one_queue(struct net_device *dev,
5291 struct netdev_queue *queue,
5297 static void netdev_init_queues(struct net_device *dev)
5299 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5300 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5301 spin_lock_init(&dev->tx_global_lock);
5305 * alloc_netdev_mq - allocate network device
5306 * @sizeof_priv: size of private data to allocate space for
5307 * @name: device name format string
5308 * @setup: callback to initialize device
5309 * @queue_count: the number of subqueues to allocate
5311 * Allocates a struct net_device with private data area for driver use
5312 * and performs basic initialization. Also allocates subquue structs
5313 * for each queue on the device at the end of the netdevice.
5315 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5316 void (*setup)(struct net_device *), unsigned int queue_count)
5318 struct netdev_queue *tx;
5319 struct net_device *dev;
5321 struct net_device *p;
5323 struct netdev_rx_queue *rx;
5327 BUG_ON(strlen(name) >= sizeof(dev->name));
5329 alloc_size = sizeof(struct net_device);
5331 /* ensure 32-byte alignment of private area */
5332 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5333 alloc_size += sizeof_priv;
5335 /* ensure 32-byte alignment of whole construct */
5336 alloc_size += NETDEV_ALIGN - 1;
5338 p = kzalloc(alloc_size, GFP_KERNEL);
5340 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5344 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5346 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5352 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5354 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5359 atomic_set(&rx->count, queue_count);
5362 * Set a pointer to first element in the array which holds the
5365 for (i = 0; i < queue_count; i++)
5369 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5370 dev->padded = (char *)dev - (char *)p;
5372 if (dev_addr_init(dev))
5378 dev_net_set(dev, &init_net);
5381 dev->num_tx_queues = queue_count;
5382 dev->real_num_tx_queues = queue_count;
5386 dev->num_rx_queues = queue_count;
5389 dev->gso_max_size = GSO_MAX_SIZE;
5391 netdev_init_queues(dev);
5393 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5394 dev->ethtool_ntuple_list.count = 0;
5395 INIT_LIST_HEAD(&dev->napi_list);
5396 INIT_LIST_HEAD(&dev->unreg_list);
5397 INIT_LIST_HEAD(&dev->link_watch_list);
5398 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5400 strcpy(dev->name, name);
5413 EXPORT_SYMBOL(alloc_netdev_mq);
5416 * free_netdev - free network device
5419 * This function does the last stage of destroying an allocated device
5420 * interface. The reference to the device object is released.
5421 * If this is the last reference then it will be freed.
5423 void free_netdev(struct net_device *dev)
5425 struct napi_struct *p, *n;
5427 release_net(dev_net(dev));
5431 /* Flush device addresses */
5432 dev_addr_flush(dev);
5434 /* Clear ethtool n-tuple list */
5435 ethtool_ntuple_flush(dev);
5437 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5440 /* Compatibility with error handling in drivers */
5441 if (dev->reg_state == NETREG_UNINITIALIZED) {
5442 kfree((char *)dev - dev->padded);
5446 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5447 dev->reg_state = NETREG_RELEASED;
5449 /* will free via device release */
5450 put_device(&dev->dev);
5452 EXPORT_SYMBOL(free_netdev);
5455 * synchronize_net - Synchronize with packet receive processing
5457 * Wait for packets currently being received to be done.
5458 * Does not block later packets from starting.
5460 void synchronize_net(void)
5465 EXPORT_SYMBOL(synchronize_net);
5468 * unregister_netdevice_queue - remove device from the kernel
5472 * This function shuts down a device interface and removes it
5473 * from the kernel tables.
5474 * If head not NULL, device is queued to be unregistered later.
5476 * Callers must hold the rtnl semaphore. You may want
5477 * unregister_netdev() instead of this.
5480 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5485 list_move_tail(&dev->unreg_list, head);
5487 rollback_registered(dev);
5488 /* Finish processing unregister after unlock */
5492 EXPORT_SYMBOL(unregister_netdevice_queue);
5495 * unregister_netdevice_many - unregister many devices
5496 * @head: list of devices
5498 void unregister_netdevice_many(struct list_head *head)
5500 struct net_device *dev;
5502 if (!list_empty(head)) {
5503 rollback_registered_many(head);
5504 list_for_each_entry(dev, head, unreg_list)
5508 EXPORT_SYMBOL(unregister_netdevice_many);
5511 * unregister_netdev - remove device from the kernel
5514 * This function shuts down a device interface and removes it
5515 * from the kernel tables.
5517 * This is just a wrapper for unregister_netdevice that takes
5518 * the rtnl semaphore. In general you want to use this and not
5519 * unregister_netdevice.
5521 void unregister_netdev(struct net_device *dev)
5524 unregister_netdevice(dev);
5527 EXPORT_SYMBOL(unregister_netdev);
5530 * dev_change_net_namespace - move device to different nethost namespace
5532 * @net: network namespace
5533 * @pat: If not NULL name pattern to try if the current device name
5534 * is already taken in the destination network namespace.
5536 * This function shuts down a device interface and moves it
5537 * to a new network namespace. On success 0 is returned, on
5538 * a failure a netagive errno code is returned.
5540 * Callers must hold the rtnl semaphore.
5543 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5549 /* Don't allow namespace local devices to be moved. */
5551 if (dev->features & NETIF_F_NETNS_LOCAL)
5555 /* Don't allow real devices to be moved when sysfs
5559 if (dev->dev.parent)
5563 /* Ensure the device has been registrered */
5565 if (dev->reg_state != NETREG_REGISTERED)
5568 /* Get out if there is nothing todo */
5570 if (net_eq(dev_net(dev), net))
5573 /* Pick the destination device name, and ensure
5574 * we can use it in the destination network namespace.
5577 if (__dev_get_by_name(net, dev->name)) {
5578 /* We get here if we can't use the current device name */
5581 if (dev_get_valid_name(dev, pat, 1))
5586 * And now a mini version of register_netdevice unregister_netdevice.
5589 /* If device is running close it first. */
5592 /* And unlink it from device chain */
5594 unlist_netdevice(dev);
5598 /* Shutdown queueing discipline. */
5601 /* Notify protocols, that we are about to destroy
5602 this device. They should clean all the things.
5604 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5605 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5608 * Flush the unicast and multicast chains
5613 netdev_unregister_kobject(dev);
5615 /* Actually switch the network namespace */
5616 dev_net_set(dev, net);
5618 /* If there is an ifindex conflict assign a new one */
5619 if (__dev_get_by_index(net, dev->ifindex)) {
5620 int iflink = (dev->iflink == dev->ifindex);
5621 dev->ifindex = dev_new_index(net);
5623 dev->iflink = dev->ifindex;
5626 /* Fixup kobjects */
5627 err = netdev_register_kobject(dev);
5630 /* Add the device back in the hashes */
5631 list_netdevice(dev);
5633 /* Notify protocols, that a new device appeared. */
5634 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5637 * Prevent userspace races by waiting until the network
5638 * device is fully setup before sending notifications.
5640 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5647 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5649 static int dev_cpu_callback(struct notifier_block *nfb,
5650 unsigned long action,
5653 struct sk_buff **list_skb;
5654 struct sk_buff *skb;
5655 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5656 struct softnet_data *sd, *oldsd;
5658 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5661 local_irq_disable();
5662 cpu = smp_processor_id();
5663 sd = &per_cpu(softnet_data, cpu);
5664 oldsd = &per_cpu(softnet_data, oldcpu);
5666 /* Find end of our completion_queue. */
5667 list_skb = &sd->completion_queue;
5669 list_skb = &(*list_skb)->next;
5670 /* Append completion queue from offline CPU. */
5671 *list_skb = oldsd->completion_queue;
5672 oldsd->completion_queue = NULL;
5674 /* Append output queue from offline CPU. */
5675 if (oldsd->output_queue) {
5676 *sd->output_queue_tailp = oldsd->output_queue;
5677 sd->output_queue_tailp = oldsd->output_queue_tailp;
5678 oldsd->output_queue = NULL;
5679 oldsd->output_queue_tailp = &oldsd->output_queue;
5682 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5685 /* Process offline CPU's input_pkt_queue */
5686 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5688 input_queue_head_incr(oldsd);
5690 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5692 input_queue_head_incr(oldsd);
5700 * netdev_increment_features - increment feature set by one
5701 * @all: current feature set
5702 * @one: new feature set
5703 * @mask: mask feature set
5705 * Computes a new feature set after adding a device with feature set
5706 * @one to the master device with current feature set @all. Will not
5707 * enable anything that is off in @mask. Returns the new feature set.
5709 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5712 /* If device needs checksumming, downgrade to it. */
5713 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5714 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5715 else if (mask & NETIF_F_ALL_CSUM) {
5716 /* If one device supports v4/v6 checksumming, set for all. */
5717 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5718 !(all & NETIF_F_GEN_CSUM)) {
5719 all &= ~NETIF_F_ALL_CSUM;
5720 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5723 /* If one device supports hw checksumming, set for all. */
5724 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5725 all &= ~NETIF_F_ALL_CSUM;
5726 all |= NETIF_F_HW_CSUM;
5730 one |= NETIF_F_ALL_CSUM;
5732 one |= all & NETIF_F_ONE_FOR_ALL;
5733 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5734 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5738 EXPORT_SYMBOL(netdev_increment_features);
5740 static struct hlist_head *netdev_create_hash(void)
5743 struct hlist_head *hash;
5745 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5747 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5748 INIT_HLIST_HEAD(&hash[i]);
5753 /* Initialize per network namespace state */
5754 static int __net_init netdev_init(struct net *net)
5756 INIT_LIST_HEAD(&net->dev_base_head);
5758 net->dev_name_head = netdev_create_hash();
5759 if (net->dev_name_head == NULL)
5762 net->dev_index_head = netdev_create_hash();
5763 if (net->dev_index_head == NULL)
5769 kfree(net->dev_name_head);
5775 * netdev_drivername - network driver for the device
5776 * @dev: network device
5777 * @buffer: buffer for resulting name
5778 * @len: size of buffer
5780 * Determine network driver for device.
5782 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5784 const struct device_driver *driver;
5785 const struct device *parent;
5787 if (len <= 0 || !buffer)
5791 parent = dev->dev.parent;
5796 driver = parent->driver;
5797 if (driver && driver->name)
5798 strlcpy(buffer, driver->name, len);
5802 static void __net_exit netdev_exit(struct net *net)
5804 kfree(net->dev_name_head);
5805 kfree(net->dev_index_head);
5808 static struct pernet_operations __net_initdata netdev_net_ops = {
5809 .init = netdev_init,
5810 .exit = netdev_exit,
5813 static void __net_exit default_device_exit(struct net *net)
5815 struct net_device *dev, *aux;
5817 * Push all migratable network devices back to the
5818 * initial network namespace
5821 for_each_netdev_safe(net, dev, aux) {
5823 char fb_name[IFNAMSIZ];
5825 /* Ignore unmoveable devices (i.e. loopback) */
5826 if (dev->features & NETIF_F_NETNS_LOCAL)
5829 /* Leave virtual devices for the generic cleanup */
5830 if (dev->rtnl_link_ops)
5833 /* Push remaing network devices to init_net */
5834 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5835 err = dev_change_net_namespace(dev, &init_net, fb_name);
5837 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5838 __func__, dev->name, err);
5845 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5847 /* At exit all network devices most be removed from a network
5848 * namespace. Do this in the reverse order of registeration.
5849 * Do this across as many network namespaces as possible to
5850 * improve batching efficiency.
5852 struct net_device *dev;
5854 LIST_HEAD(dev_kill_list);
5857 list_for_each_entry(net, net_list, exit_list) {
5858 for_each_netdev_reverse(net, dev) {
5859 if (dev->rtnl_link_ops)
5860 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5862 unregister_netdevice_queue(dev, &dev_kill_list);
5865 unregister_netdevice_many(&dev_kill_list);
5869 static struct pernet_operations __net_initdata default_device_ops = {
5870 .exit = default_device_exit,
5871 .exit_batch = default_device_exit_batch,
5875 * Initialize the DEV module. At boot time this walks the device list and
5876 * unhooks any devices that fail to initialise (normally hardware not
5877 * present) and leaves us with a valid list of present and active devices.
5882 * This is called single threaded during boot, so no need
5883 * to take the rtnl semaphore.
5885 static int __init net_dev_init(void)
5887 int i, rc = -ENOMEM;
5889 BUG_ON(!dev_boot_phase);
5891 if (dev_proc_init())
5894 if (netdev_kobject_init())
5897 INIT_LIST_HEAD(&ptype_all);
5898 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5899 INIT_LIST_HEAD(&ptype_base[i]);
5901 if (register_pernet_subsys(&netdev_net_ops))
5905 * Initialise the packet receive queues.
5908 for_each_possible_cpu(i) {
5909 struct softnet_data *sd = &per_cpu(softnet_data, i);
5911 memset(sd, 0, sizeof(*sd));
5912 skb_queue_head_init(&sd->input_pkt_queue);
5913 skb_queue_head_init(&sd->process_queue);
5914 sd->completion_queue = NULL;
5915 INIT_LIST_HEAD(&sd->poll_list);
5916 sd->output_queue = NULL;
5917 sd->output_queue_tailp = &sd->output_queue;
5919 sd->csd.func = rps_trigger_softirq;
5925 sd->backlog.poll = process_backlog;
5926 sd->backlog.weight = weight_p;
5927 sd->backlog.gro_list = NULL;
5928 sd->backlog.gro_count = 0;
5933 /* The loopback device is special if any other network devices
5934 * is present in a network namespace the loopback device must
5935 * be present. Since we now dynamically allocate and free the
5936 * loopback device ensure this invariant is maintained by
5937 * keeping the loopback device as the first device on the
5938 * list of network devices. Ensuring the loopback devices
5939 * is the first device that appears and the last network device
5942 if (register_pernet_device(&loopback_net_ops))
5945 if (register_pernet_device(&default_device_ops))
5948 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5949 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5951 hotcpu_notifier(dev_cpu_callback, 0);
5959 subsys_initcall(net_dev_init);
5961 static int __init initialize_hashrnd(void)
5963 get_random_bytes(&hashrnd, sizeof(hashrnd));
5967 late_initcall_sync(initialize_hashrnd);