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>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <linux/pci.h>
133 #include "net-sysfs.h"
135 /* Instead of increasing this, you should create a hash table. */
136 #define MAX_GRO_SKBS 8
138 /* This should be increased if a protocol with a bigger head is added. */
139 #define GRO_MAX_HEAD (MAX_HEADER + 128)
142 * The list of packet types we will receive (as opposed to discard)
143 * and the routines to invoke.
145 * Why 16. Because with 16 the only overlap we get on a hash of the
146 * low nibble of the protocol value is RARP/SNAP/X.25.
148 * NOTE: That is no longer true with the addition of VLAN tags. Not
149 * sure which should go first, but I bet it won't make much
150 * difference if we are running VLANs. The good news is that
151 * this protocol won't be in the list unless compiled in, so
152 * the average user (w/out VLANs) will not be adversely affected.
169 #define PTYPE_HASH_SIZE (16)
170 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
172 static DEFINE_SPINLOCK(ptype_lock);
173 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
174 static struct list_head ptype_all __read_mostly; /* Taps */
177 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
180 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
182 * Writers must hold the rtnl semaphore while they loop through the
183 * dev_base_head list, and hold dev_base_lock for writing when they do the
184 * actual updates. This allows pure readers to access the list even
185 * while a writer is preparing to update it.
187 * To put it another way, dev_base_lock is held for writing only to
188 * protect against pure readers; the rtnl semaphore provides the
189 * protection against other writers.
191 * See, for example usages, register_netdevice() and
192 * unregister_netdevice(), which must be called with the rtnl
195 DEFINE_RWLOCK(dev_base_lock);
196 EXPORT_SYMBOL(dev_base_lock);
198 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
200 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
201 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
204 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
206 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
209 static inline void rps_lock(struct softnet_data *sd)
212 spin_lock(&sd->input_pkt_queue.lock);
216 static inline void rps_unlock(struct softnet_data *sd)
219 spin_unlock(&sd->input_pkt_queue.lock);
223 /* Device list insertion */
224 static int list_netdevice(struct net_device *dev)
226 struct net *net = dev_net(dev);
230 write_lock_bh(&dev_base_lock);
231 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
232 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
233 hlist_add_head_rcu(&dev->index_hlist,
234 dev_index_hash(net, dev->ifindex));
235 write_unlock_bh(&dev_base_lock);
239 /* Device list removal
240 * caller must respect a RCU grace period before freeing/reusing dev
242 static void unlist_netdevice(struct net_device *dev)
246 /* Unlink dev from the device chain */
247 write_lock_bh(&dev_base_lock);
248 list_del_rcu(&dev->dev_list);
249 hlist_del_rcu(&dev->name_hlist);
250 hlist_del_rcu(&dev->index_hlist);
251 write_unlock_bh(&dev_base_lock);
258 static RAW_NOTIFIER_HEAD(netdev_chain);
261 * Device drivers call our routines to queue packets here. We empty the
262 * queue in the local softnet handler.
265 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
266 EXPORT_PER_CPU_SYMBOL(softnet_data);
268 #ifdef CONFIG_LOCKDEP
270 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
271 * according to dev->type
273 static const unsigned short netdev_lock_type[] =
274 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
275 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
276 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
277 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
278 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
279 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
280 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
281 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
282 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
283 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
284 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
285 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
286 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
287 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
288 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
289 ARPHRD_VOID, ARPHRD_NONE};
291 static const char *const netdev_lock_name[] =
292 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
293 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
294 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
295 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
296 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
297 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
298 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
299 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
300 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
301 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
302 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
303 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
304 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
305 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
306 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
307 "_xmit_VOID", "_xmit_NONE"};
309 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
310 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
316 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
317 if (netdev_lock_type[i] == dev_type)
319 /* the last key is used by default */
320 return ARRAY_SIZE(netdev_lock_type) - 1;
323 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
324 unsigned short dev_type)
328 i = netdev_lock_pos(dev_type);
329 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
330 netdev_lock_name[i]);
333 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
337 i = netdev_lock_pos(dev->type);
338 lockdep_set_class_and_name(&dev->addr_list_lock,
339 &netdev_addr_lock_key[i],
340 netdev_lock_name[i]);
343 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
344 unsigned short dev_type)
347 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
352 /*******************************************************************************
354 Protocol management and registration routines
356 *******************************************************************************/
359 * Add a protocol ID to the list. Now that the input handler is
360 * smarter we can dispense with all the messy stuff that used to be
363 * BEWARE!!! Protocol handlers, mangling input packets,
364 * MUST BE last in hash buckets and checking protocol handlers
365 * MUST start from promiscuous ptype_all chain in net_bh.
366 * It is true now, do not change it.
367 * Explanation follows: if protocol handler, mangling packet, will
368 * be the first on list, it is not able to sense, that packet
369 * is cloned and should be copied-on-write, so that it will
370 * change it and subsequent readers will get broken packet.
375 * dev_add_pack - add packet handler
376 * @pt: packet type declaration
378 * Add a protocol handler to the networking stack. The passed &packet_type
379 * is linked into kernel lists and may not be freed until it has been
380 * removed from the kernel lists.
382 * This call does not sleep therefore it can not
383 * guarantee all CPU's that are in middle of receiving packets
384 * will see the new packet type (until the next received packet).
387 void dev_add_pack(struct packet_type *pt)
391 spin_lock_bh(&ptype_lock);
392 if (pt->type == htons(ETH_P_ALL))
393 list_add_rcu(&pt->list, &ptype_all);
395 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
396 list_add_rcu(&pt->list, &ptype_base[hash]);
398 spin_unlock_bh(&ptype_lock);
400 EXPORT_SYMBOL(dev_add_pack);
403 * __dev_remove_pack - remove packet handler
404 * @pt: packet type declaration
406 * Remove a protocol handler that was previously added to the kernel
407 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
408 * from the kernel lists and can be freed or reused once this function
411 * The packet type might still be in use by receivers
412 * and must not be freed until after all the CPU's have gone
413 * through a quiescent state.
415 void __dev_remove_pack(struct packet_type *pt)
417 struct list_head *head;
418 struct packet_type *pt1;
420 spin_lock_bh(&ptype_lock);
422 if (pt->type == htons(ETH_P_ALL))
425 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
427 list_for_each_entry(pt1, head, list) {
429 list_del_rcu(&pt->list);
434 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
436 spin_unlock_bh(&ptype_lock);
438 EXPORT_SYMBOL(__dev_remove_pack);
441 * dev_remove_pack - remove packet handler
442 * @pt: packet type declaration
444 * Remove a protocol handler that was previously added to the kernel
445 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
446 * from the kernel lists and can be freed or reused once this function
449 * This call sleeps to guarantee that no CPU is looking at the packet
452 void dev_remove_pack(struct packet_type *pt)
454 __dev_remove_pack(pt);
458 EXPORT_SYMBOL(dev_remove_pack);
460 /******************************************************************************
462 Device Boot-time Settings Routines
464 *******************************************************************************/
466 /* Boot time configuration table */
467 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
470 * netdev_boot_setup_add - add new setup entry
471 * @name: name of the device
472 * @map: configured settings for the device
474 * Adds new setup entry to the dev_boot_setup list. The function
475 * returns 0 on error and 1 on success. This is a generic routine to
478 static int netdev_boot_setup_add(char *name, struct ifmap *map)
480 struct netdev_boot_setup *s;
484 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
485 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
486 memset(s[i].name, 0, sizeof(s[i].name));
487 strlcpy(s[i].name, name, IFNAMSIZ);
488 memcpy(&s[i].map, map, sizeof(s[i].map));
493 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
497 * netdev_boot_setup_check - check boot time settings
498 * @dev: the netdevice
500 * Check boot time settings for the device.
501 * The found settings are set for the device to be used
502 * later in the device probing.
503 * Returns 0 if no settings found, 1 if they are.
505 int netdev_boot_setup_check(struct net_device *dev)
507 struct netdev_boot_setup *s = dev_boot_setup;
510 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
511 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
512 !strcmp(dev->name, s[i].name)) {
513 dev->irq = s[i].map.irq;
514 dev->base_addr = s[i].map.base_addr;
515 dev->mem_start = s[i].map.mem_start;
516 dev->mem_end = s[i].map.mem_end;
522 EXPORT_SYMBOL(netdev_boot_setup_check);
526 * netdev_boot_base - get address from boot time settings
527 * @prefix: prefix for network device
528 * @unit: id for network device
530 * Check boot time settings for the base address of device.
531 * The found settings are set for the device to be used
532 * later in the device probing.
533 * Returns 0 if no settings found.
535 unsigned long netdev_boot_base(const char *prefix, int unit)
537 const struct netdev_boot_setup *s = dev_boot_setup;
541 sprintf(name, "%s%d", prefix, unit);
544 * If device already registered then return base of 1
545 * to indicate not to probe for this interface
547 if (__dev_get_by_name(&init_net, name))
550 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
551 if (!strcmp(name, s[i].name))
552 return s[i].map.base_addr;
557 * Saves at boot time configured settings for any netdevice.
559 int __init netdev_boot_setup(char *str)
564 str = get_options(str, ARRAY_SIZE(ints), ints);
569 memset(&map, 0, sizeof(map));
573 map.base_addr = ints[2];
575 map.mem_start = ints[3];
577 map.mem_end = ints[4];
579 /* Add new entry to the list */
580 return netdev_boot_setup_add(str, &map);
583 __setup("netdev=", netdev_boot_setup);
585 /*******************************************************************************
587 Device Interface Subroutines
589 *******************************************************************************/
592 * __dev_get_by_name - find a device by its name
593 * @net: the applicable net namespace
594 * @name: name to find
596 * Find an interface by name. Must be called under RTNL semaphore
597 * or @dev_base_lock. If the name is found a pointer to the device
598 * is returned. If the name is not found then %NULL is returned. The
599 * reference counters are not incremented so the caller must be
600 * careful with locks.
603 struct net_device *__dev_get_by_name(struct net *net, const char *name)
605 struct hlist_node *p;
606 struct net_device *dev;
607 struct hlist_head *head = dev_name_hash(net, name);
609 hlist_for_each_entry(dev, p, head, name_hlist)
610 if (!strncmp(dev->name, name, IFNAMSIZ))
615 EXPORT_SYMBOL(__dev_get_by_name);
618 * dev_get_by_name_rcu - find a device by its name
619 * @net: the applicable net namespace
620 * @name: name to find
622 * Find an interface by name.
623 * If the name is found a pointer to the device is returned.
624 * If the name is not found then %NULL is returned.
625 * The reference counters are not incremented so the caller must be
626 * careful with locks. The caller must hold RCU lock.
629 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
631 struct hlist_node *p;
632 struct net_device *dev;
633 struct hlist_head *head = dev_name_hash(net, name);
635 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
636 if (!strncmp(dev->name, name, IFNAMSIZ))
641 EXPORT_SYMBOL(dev_get_by_name_rcu);
644 * dev_get_by_name - find a device by its name
645 * @net: the applicable net namespace
646 * @name: name to find
648 * Find an interface by name. This can be called from any
649 * context and does its own locking. The returned handle has
650 * the usage count incremented and the caller must use dev_put() to
651 * release it when it is no longer needed. %NULL is returned if no
652 * matching device is found.
655 struct net_device *dev_get_by_name(struct net *net, const char *name)
657 struct net_device *dev;
660 dev = dev_get_by_name_rcu(net, name);
666 EXPORT_SYMBOL(dev_get_by_name);
669 * __dev_get_by_index - find a device by its ifindex
670 * @net: the applicable net namespace
671 * @ifindex: index of device
673 * Search for an interface by index. Returns %NULL if the device
674 * is not found or a pointer to the device. The device has not
675 * had its reference counter increased so the caller must be careful
676 * about locking. The caller must hold either the RTNL semaphore
680 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
682 struct hlist_node *p;
683 struct net_device *dev;
684 struct hlist_head *head = dev_index_hash(net, ifindex);
686 hlist_for_each_entry(dev, p, head, index_hlist)
687 if (dev->ifindex == ifindex)
692 EXPORT_SYMBOL(__dev_get_by_index);
695 * dev_get_by_index_rcu - find a device by its ifindex
696 * @net: the applicable net namespace
697 * @ifindex: index of device
699 * Search for an interface by index. Returns %NULL if the device
700 * is not found or a pointer to the device. The device has not
701 * had its reference counter increased so the caller must be careful
702 * about locking. The caller must hold RCU lock.
705 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
707 struct hlist_node *p;
708 struct net_device *dev;
709 struct hlist_head *head = dev_index_hash(net, ifindex);
711 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
712 if (dev->ifindex == ifindex)
717 EXPORT_SYMBOL(dev_get_by_index_rcu);
721 * dev_get_by_index - find a device by its ifindex
722 * @net: the applicable net namespace
723 * @ifindex: index of device
725 * Search for an interface by index. Returns NULL if the device
726 * is not found or a pointer to the device. The device returned has
727 * had a reference added and the pointer is safe until the user calls
728 * dev_put to indicate they have finished with it.
731 struct net_device *dev_get_by_index(struct net *net, int ifindex)
733 struct net_device *dev;
736 dev = dev_get_by_index_rcu(net, ifindex);
742 EXPORT_SYMBOL(dev_get_by_index);
745 * dev_getbyhwaddr - find a device by its hardware address
746 * @net: the applicable net namespace
747 * @type: media type of device
748 * @ha: hardware address
750 * Search for an interface by MAC address. Returns NULL if the device
751 * is not found or a pointer to the device. The caller must hold the
752 * rtnl semaphore. The returned device has not had its ref count increased
753 * and the caller must therefore be careful about locking
756 * If the API was consistent this would be __dev_get_by_hwaddr
759 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
761 struct net_device *dev;
765 for_each_netdev(net, dev)
766 if (dev->type == type &&
767 !memcmp(dev->dev_addr, ha, dev->addr_len))
772 EXPORT_SYMBOL(dev_getbyhwaddr);
774 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
776 struct net_device *dev;
779 for_each_netdev(net, dev)
780 if (dev->type == type)
785 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
787 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
789 struct net_device *dev, *ret = NULL;
792 for_each_netdev_rcu(net, dev)
793 if (dev->type == type) {
801 EXPORT_SYMBOL(dev_getfirstbyhwtype);
804 * dev_get_by_flags_rcu - find any device with given flags
805 * @net: the applicable net namespace
806 * @if_flags: IFF_* values
807 * @mask: bitmask of bits in if_flags to check
809 * Search for any interface with the given flags. Returns NULL if a device
810 * is not found or a pointer to the device. Must be called inside
811 * rcu_read_lock(), and result refcount is unchanged.
814 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
817 struct net_device *dev, *ret;
820 for_each_netdev_rcu(net, dev) {
821 if (((dev->flags ^ if_flags) & mask) == 0) {
828 EXPORT_SYMBOL(dev_get_by_flags_rcu);
831 * dev_valid_name - check if name is okay for network device
834 * Network device names need to be valid file names to
835 * to allow sysfs to work. We also disallow any kind of
838 int dev_valid_name(const char *name)
842 if (strlen(name) >= IFNAMSIZ)
844 if (!strcmp(name, ".") || !strcmp(name, ".."))
848 if (*name == '/' || isspace(*name))
854 EXPORT_SYMBOL(dev_valid_name);
857 * __dev_alloc_name - allocate a name for a device
858 * @net: network namespace to allocate the device name in
859 * @name: name format string
860 * @buf: scratch buffer and result name string
862 * Passed a format string - eg "lt%d" it will try and find a suitable
863 * id. It scans list of devices to build up a free map, then chooses
864 * the first empty slot. The caller must hold the dev_base or rtnl lock
865 * while allocating the name and adding the device in order to avoid
867 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
868 * Returns the number of the unit assigned or a negative errno code.
871 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
875 const int max_netdevices = 8*PAGE_SIZE;
876 unsigned long *inuse;
877 struct net_device *d;
879 p = strnchr(name, IFNAMSIZ-1, '%');
882 * Verify the string as this thing may have come from
883 * the user. There must be either one "%d" and no other "%"
886 if (p[1] != 'd' || strchr(p + 2, '%'))
889 /* Use one page as a bit array of possible slots */
890 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
894 for_each_netdev(net, d) {
895 if (!sscanf(d->name, name, &i))
897 if (i < 0 || i >= max_netdevices)
900 /* avoid cases where sscanf is not exact inverse of printf */
901 snprintf(buf, IFNAMSIZ, name, i);
902 if (!strncmp(buf, d->name, IFNAMSIZ))
906 i = find_first_zero_bit(inuse, max_netdevices);
907 free_page((unsigned long) inuse);
911 snprintf(buf, IFNAMSIZ, name, i);
912 if (!__dev_get_by_name(net, buf))
915 /* It is possible to run out of possible slots
916 * when the name is long and there isn't enough space left
917 * for the digits, or if all bits are used.
923 * dev_alloc_name - allocate a name for a device
925 * @name: name format string
927 * Passed a format string - eg "lt%d" it will try and find a suitable
928 * id. It scans list of devices to build up a free map, then chooses
929 * the first empty slot. The caller must hold the dev_base or rtnl lock
930 * while allocating the name and adding the device in order to avoid
932 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
933 * Returns the number of the unit assigned or a negative errno code.
936 int dev_alloc_name(struct net_device *dev, const char *name)
942 BUG_ON(!dev_net(dev));
944 ret = __dev_alloc_name(net, name, buf);
946 strlcpy(dev->name, buf, IFNAMSIZ);
949 EXPORT_SYMBOL(dev_alloc_name);
951 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
955 BUG_ON(!dev_net(dev));
958 if (!dev_valid_name(name))
961 if (fmt && strchr(name, '%'))
962 return dev_alloc_name(dev, name);
963 else if (__dev_get_by_name(net, name))
965 else if (dev->name != name)
966 strlcpy(dev->name, name, IFNAMSIZ);
972 * dev_change_name - change name of a device
974 * @newname: name (or format string) must be at least IFNAMSIZ
976 * Change name of a device, can pass format strings "eth%d".
979 int dev_change_name(struct net_device *dev, const char *newname)
981 char oldname[IFNAMSIZ];
987 BUG_ON(!dev_net(dev));
990 if (dev->flags & IFF_UP)
993 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
996 memcpy(oldname, dev->name, IFNAMSIZ);
998 err = dev_get_valid_name(dev, newname, 1);
1003 ret = device_rename(&dev->dev, dev->name);
1005 memcpy(dev->name, oldname, IFNAMSIZ);
1009 write_lock_bh(&dev_base_lock);
1010 hlist_del(&dev->name_hlist);
1011 write_unlock_bh(&dev_base_lock);
1015 write_lock_bh(&dev_base_lock);
1016 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1017 write_unlock_bh(&dev_base_lock);
1019 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1020 ret = notifier_to_errno(ret);
1023 /* err >= 0 after dev_alloc_name() or stores the first errno */
1026 memcpy(dev->name, oldname, IFNAMSIZ);
1030 "%s: name change rollback failed: %d.\n",
1039 * dev_set_alias - change ifalias of a device
1041 * @alias: name up to IFALIASZ
1042 * @len: limit of bytes to copy from info
1044 * Set ifalias for a device,
1046 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1050 if (len >= IFALIASZ)
1055 kfree(dev->ifalias);
1056 dev->ifalias = NULL;
1061 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1065 strlcpy(dev->ifalias, alias, len+1);
1071 * netdev_features_change - device changes features
1072 * @dev: device to cause notification
1074 * Called to indicate a device has changed features.
1076 void netdev_features_change(struct net_device *dev)
1078 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1080 EXPORT_SYMBOL(netdev_features_change);
1083 * netdev_state_change - device changes state
1084 * @dev: device to cause notification
1086 * Called to indicate a device has changed state. This function calls
1087 * the notifier chains for netdev_chain and sends a NEWLINK message
1088 * to the routing socket.
1090 void netdev_state_change(struct net_device *dev)
1092 if (dev->flags & IFF_UP) {
1093 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1094 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1097 EXPORT_SYMBOL(netdev_state_change);
1099 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1101 return call_netdevice_notifiers(event, dev);
1103 EXPORT_SYMBOL(netdev_bonding_change);
1106 * dev_load - load a network module
1107 * @net: the applicable net namespace
1108 * @name: name of interface
1110 * If a network interface is not present and the process has suitable
1111 * privileges this function loads the module. If module loading is not
1112 * available in this kernel then it becomes a nop.
1115 void dev_load(struct net *net, const char *name)
1117 struct net_device *dev;
1120 dev = dev_get_by_name_rcu(net, name);
1123 if (!dev && capable(CAP_NET_ADMIN))
1124 request_module("%s", name);
1126 EXPORT_SYMBOL(dev_load);
1128 static int __dev_open(struct net_device *dev)
1130 const struct net_device_ops *ops = dev->netdev_ops;
1136 * Is it even present?
1138 if (!netif_device_present(dev))
1141 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1142 ret = notifier_to_errno(ret);
1147 * Call device private open method
1149 set_bit(__LINK_STATE_START, &dev->state);
1151 if (ops->ndo_validate_addr)
1152 ret = ops->ndo_validate_addr(dev);
1154 if (!ret && ops->ndo_open)
1155 ret = ops->ndo_open(dev);
1158 * If it went open OK then:
1162 clear_bit(__LINK_STATE_START, &dev->state);
1167 dev->flags |= IFF_UP;
1172 net_dmaengine_get();
1175 * Initialize multicasting status
1177 dev_set_rx_mode(dev);
1180 * Wakeup transmit queue engine
1189 * dev_open - prepare an interface for use.
1190 * @dev: device to open
1192 * Takes a device from down to up state. The device's private open
1193 * function is invoked and then the multicast lists are loaded. Finally
1194 * the device is moved into the up state and a %NETDEV_UP message is
1195 * sent to the netdev notifier chain.
1197 * Calling this function on an active interface is a nop. On a failure
1198 * a negative errno code is returned.
1200 int dev_open(struct net_device *dev)
1207 if (dev->flags & IFF_UP)
1213 ret = __dev_open(dev);
1218 * ... and announce new interface.
1220 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1221 call_netdevice_notifiers(NETDEV_UP, dev);
1225 EXPORT_SYMBOL(dev_open);
1227 static int __dev_close(struct net_device *dev)
1229 const struct net_device_ops *ops = dev->netdev_ops;
1235 * Tell people we are going down, so that they can
1236 * prepare to death, when device is still operating.
1238 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1240 clear_bit(__LINK_STATE_START, &dev->state);
1242 /* Synchronize to scheduled poll. We cannot touch poll list,
1243 * it can be even on different cpu. So just clear netif_running().
1245 * dev->stop() will invoke napi_disable() on all of it's
1246 * napi_struct instances on this device.
1248 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1250 dev_deactivate(dev);
1253 * Call the device specific close. This cannot fail.
1254 * Only if device is UP
1256 * We allow it to be called even after a DETACH hot-plug
1263 * Device is now down.
1266 dev->flags &= ~IFF_UP;
1271 net_dmaengine_put();
1277 * dev_close - shutdown an interface.
1278 * @dev: device to shutdown
1280 * This function moves an active device into down state. A
1281 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1282 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1285 int dev_close(struct net_device *dev)
1287 if (!(dev->flags & IFF_UP))
1293 * Tell people we are down
1295 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1296 call_netdevice_notifiers(NETDEV_DOWN, dev);
1300 EXPORT_SYMBOL(dev_close);
1304 * dev_disable_lro - disable Large Receive Offload on a device
1307 * Disable Large Receive Offload (LRO) on a net device. Must be
1308 * called under RTNL. This is needed if received packets may be
1309 * forwarded to another interface.
1311 void dev_disable_lro(struct net_device *dev)
1313 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1314 dev->ethtool_ops->set_flags) {
1315 u32 flags = dev->ethtool_ops->get_flags(dev);
1316 if (flags & ETH_FLAG_LRO) {
1317 flags &= ~ETH_FLAG_LRO;
1318 dev->ethtool_ops->set_flags(dev, flags);
1321 WARN_ON(dev->features & NETIF_F_LRO);
1323 EXPORT_SYMBOL(dev_disable_lro);
1326 static int dev_boot_phase = 1;
1329 * Device change register/unregister. These are not inline or static
1330 * as we export them to the world.
1334 * register_netdevice_notifier - register a network notifier block
1337 * Register a notifier to be called when network device events occur.
1338 * The notifier passed is linked into the kernel structures and must
1339 * not be reused until it has been unregistered. A negative errno code
1340 * is returned on a failure.
1342 * When registered all registration and up events are replayed
1343 * to the new notifier to allow device to have a race free
1344 * view of the network device list.
1347 int register_netdevice_notifier(struct notifier_block *nb)
1349 struct net_device *dev;
1350 struct net_device *last;
1355 err = raw_notifier_chain_register(&netdev_chain, nb);
1361 for_each_netdev(net, dev) {
1362 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1363 err = notifier_to_errno(err);
1367 if (!(dev->flags & IFF_UP))
1370 nb->notifier_call(nb, NETDEV_UP, dev);
1381 for_each_netdev(net, dev) {
1385 if (dev->flags & IFF_UP) {
1386 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1387 nb->notifier_call(nb, NETDEV_DOWN, dev);
1389 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1390 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1394 raw_notifier_chain_unregister(&netdev_chain, nb);
1397 EXPORT_SYMBOL(register_netdevice_notifier);
1400 * unregister_netdevice_notifier - unregister a network notifier block
1403 * Unregister a notifier previously registered by
1404 * register_netdevice_notifier(). The notifier is unlinked into the
1405 * kernel structures and may then be reused. A negative errno code
1406 * is returned on a failure.
1409 int unregister_netdevice_notifier(struct notifier_block *nb)
1414 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1418 EXPORT_SYMBOL(unregister_netdevice_notifier);
1421 * call_netdevice_notifiers - call all network notifier blocks
1422 * @val: value passed unmodified to notifier function
1423 * @dev: net_device pointer passed unmodified to notifier function
1425 * Call all network notifier blocks. Parameters and return value
1426 * are as for raw_notifier_call_chain().
1429 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1432 return raw_notifier_call_chain(&netdev_chain, val, dev);
1435 /* When > 0 there are consumers of rx skb time stamps */
1436 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1438 void net_enable_timestamp(void)
1440 atomic_inc(&netstamp_needed);
1442 EXPORT_SYMBOL(net_enable_timestamp);
1444 void net_disable_timestamp(void)
1446 atomic_dec(&netstamp_needed);
1448 EXPORT_SYMBOL(net_disable_timestamp);
1450 static inline void net_timestamp_set(struct sk_buff *skb)
1452 if (atomic_read(&netstamp_needed))
1453 __net_timestamp(skb);
1455 skb->tstamp.tv64 = 0;
1458 static inline void net_timestamp_check(struct sk_buff *skb)
1460 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1461 __net_timestamp(skb);
1465 * dev_forward_skb - loopback an skb to another netif
1467 * @dev: destination network device
1468 * @skb: buffer to forward
1471 * NET_RX_SUCCESS (no congestion)
1472 * NET_RX_DROP (packet was dropped, but freed)
1474 * dev_forward_skb can be used for injecting an skb from the
1475 * start_xmit function of one device into the receive queue
1476 * of another device.
1478 * The receiving device may be in another namespace, so
1479 * we have to clear all information in the skb that could
1480 * impact namespace isolation.
1482 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1487 if (!(dev->flags & IFF_UP) ||
1488 (skb->len > (dev->mtu + dev->hard_header_len))) {
1492 skb_set_dev(skb, dev);
1493 skb->tstamp.tv64 = 0;
1494 skb->pkt_type = PACKET_HOST;
1495 skb->protocol = eth_type_trans(skb, dev);
1496 return netif_rx(skb);
1498 EXPORT_SYMBOL_GPL(dev_forward_skb);
1501 * Support routine. Sends outgoing frames to any network
1502 * taps currently in use.
1505 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1507 struct packet_type *ptype;
1509 #ifdef CONFIG_NET_CLS_ACT
1510 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1511 net_timestamp_set(skb);
1513 net_timestamp_set(skb);
1517 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1518 /* Never send packets back to the socket
1519 * they originated from - MvS (miquels@drinkel.ow.org)
1521 if ((ptype->dev == dev || !ptype->dev) &&
1522 (ptype->af_packet_priv == NULL ||
1523 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1524 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1528 /* skb->nh should be correctly
1529 set by sender, so that the second statement is
1530 just protection against buggy protocols.
1532 skb_reset_mac_header(skb2);
1534 if (skb_network_header(skb2) < skb2->data ||
1535 skb2->network_header > skb2->tail) {
1536 if (net_ratelimit())
1537 printk(KERN_CRIT "protocol %04x is "
1539 ntohs(skb2->protocol),
1541 skb_reset_network_header(skb2);
1544 skb2->transport_header = skb2->network_header;
1545 skb2->pkt_type = PACKET_OUTGOING;
1546 ptype->func(skb2, skb->dev, ptype, skb->dev);
1553 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1554 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1556 void netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1558 unsigned int real_num = dev->real_num_tx_queues;
1560 if (unlikely(txq > dev->num_tx_queues))
1562 else if (txq > real_num)
1563 dev->real_num_tx_queues = txq;
1564 else if (txq < real_num) {
1565 dev->real_num_tx_queues = txq;
1566 qdisc_reset_all_tx_gt(dev, txq);
1569 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1571 static inline void __netif_reschedule(struct Qdisc *q)
1573 struct softnet_data *sd;
1574 unsigned long flags;
1576 local_irq_save(flags);
1577 sd = &__get_cpu_var(softnet_data);
1578 q->next_sched = NULL;
1579 *sd->output_queue_tailp = q;
1580 sd->output_queue_tailp = &q->next_sched;
1581 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1582 local_irq_restore(flags);
1585 void __netif_schedule(struct Qdisc *q)
1587 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1588 __netif_reschedule(q);
1590 EXPORT_SYMBOL(__netif_schedule);
1592 void dev_kfree_skb_irq(struct sk_buff *skb)
1594 if (!skb->destructor)
1596 else if (atomic_dec_and_test(&skb->users)) {
1597 struct softnet_data *sd;
1598 unsigned long flags;
1600 local_irq_save(flags);
1601 sd = &__get_cpu_var(softnet_data);
1602 skb->next = sd->completion_queue;
1603 sd->completion_queue = skb;
1604 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1605 local_irq_restore(flags);
1608 EXPORT_SYMBOL(dev_kfree_skb_irq);
1610 void dev_kfree_skb_any(struct sk_buff *skb)
1612 if (in_irq() || irqs_disabled())
1613 dev_kfree_skb_irq(skb);
1617 EXPORT_SYMBOL(dev_kfree_skb_any);
1621 * netif_device_detach - mark device as removed
1622 * @dev: network device
1624 * Mark device as removed from system and therefore no longer available.
1626 void netif_device_detach(struct net_device *dev)
1628 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1629 netif_running(dev)) {
1630 netif_tx_stop_all_queues(dev);
1633 EXPORT_SYMBOL(netif_device_detach);
1636 * netif_device_attach - mark device as attached
1637 * @dev: network device
1639 * Mark device as attached from system and restart if needed.
1641 void netif_device_attach(struct net_device *dev)
1643 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1644 netif_running(dev)) {
1645 netif_tx_wake_all_queues(dev);
1646 __netdev_watchdog_up(dev);
1649 EXPORT_SYMBOL(netif_device_attach);
1651 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1653 return ((features & NETIF_F_GEN_CSUM) ||
1654 ((features & NETIF_F_IP_CSUM) &&
1655 protocol == htons(ETH_P_IP)) ||
1656 ((features & NETIF_F_IPV6_CSUM) &&
1657 protocol == htons(ETH_P_IPV6)) ||
1658 ((features & NETIF_F_FCOE_CRC) &&
1659 protocol == htons(ETH_P_FCOE)));
1662 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1664 if (can_checksum_protocol(dev->features, skb->protocol))
1667 if (skb->protocol == htons(ETH_P_8021Q)) {
1668 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1669 if (can_checksum_protocol(dev->features & dev->vlan_features,
1670 veh->h_vlan_encapsulated_proto))
1678 * skb_dev_set -- assign a new device to a buffer
1679 * @skb: buffer for the new device
1680 * @dev: network device
1682 * If an skb is owned by a device already, we have to reset
1683 * all data private to the namespace a device belongs to
1684 * before assigning it a new device.
1686 #ifdef CONFIG_NET_NS
1687 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1690 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1693 skb_init_secmark(skb);
1697 skb->ipvs_property = 0;
1698 #ifdef CONFIG_NET_SCHED
1704 EXPORT_SYMBOL(skb_set_dev);
1705 #endif /* CONFIG_NET_NS */
1708 * Invalidate hardware checksum when packet is to be mangled, and
1709 * complete checksum manually on outgoing path.
1711 int skb_checksum_help(struct sk_buff *skb)
1714 int ret = 0, offset;
1716 if (skb->ip_summed == CHECKSUM_COMPLETE)
1717 goto out_set_summed;
1719 if (unlikely(skb_shinfo(skb)->gso_size)) {
1720 /* Let GSO fix up the checksum. */
1721 goto out_set_summed;
1724 offset = skb->csum_start - skb_headroom(skb);
1725 BUG_ON(offset >= skb_headlen(skb));
1726 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1728 offset += skb->csum_offset;
1729 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1731 if (skb_cloned(skb) &&
1732 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1733 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1738 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1740 skb->ip_summed = CHECKSUM_NONE;
1744 EXPORT_SYMBOL(skb_checksum_help);
1747 * skb_gso_segment - Perform segmentation on skb.
1748 * @skb: buffer to segment
1749 * @features: features for the output path (see dev->features)
1751 * This function segments the given skb and returns a list of segments.
1753 * It may return NULL if the skb requires no segmentation. This is
1754 * only possible when GSO is used for verifying header integrity.
1756 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1758 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1759 struct packet_type *ptype;
1760 __be16 type = skb->protocol;
1763 skb_reset_mac_header(skb);
1764 skb->mac_len = skb->network_header - skb->mac_header;
1765 __skb_pull(skb, skb->mac_len);
1767 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1768 struct net_device *dev = skb->dev;
1769 struct ethtool_drvinfo info = {};
1771 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1772 dev->ethtool_ops->get_drvinfo(dev, &info);
1774 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1776 info.driver, dev ? dev->features : 0L,
1777 skb->sk ? skb->sk->sk_route_caps : 0L,
1778 skb->len, skb->data_len, skb->ip_summed);
1780 if (skb_header_cloned(skb) &&
1781 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1782 return ERR_PTR(err);
1786 list_for_each_entry_rcu(ptype,
1787 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1788 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1789 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1790 err = ptype->gso_send_check(skb);
1791 segs = ERR_PTR(err);
1792 if (err || skb_gso_ok(skb, features))
1794 __skb_push(skb, (skb->data -
1795 skb_network_header(skb)));
1797 segs = ptype->gso_segment(skb, features);
1803 __skb_push(skb, skb->data - skb_mac_header(skb));
1807 EXPORT_SYMBOL(skb_gso_segment);
1809 /* Take action when hardware reception checksum errors are detected. */
1811 void netdev_rx_csum_fault(struct net_device *dev)
1813 if (net_ratelimit()) {
1814 printk(KERN_ERR "%s: hw csum failure.\n",
1815 dev ? dev->name : "<unknown>");
1819 EXPORT_SYMBOL(netdev_rx_csum_fault);
1822 /* Actually, we should eliminate this check as soon as we know, that:
1823 * 1. IOMMU is present and allows to map all the memory.
1824 * 2. No high memory really exists on this machine.
1827 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1829 #ifdef CONFIG_HIGHMEM
1831 if (!(dev->features & NETIF_F_HIGHDMA)) {
1832 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1833 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1837 if (PCI_DMA_BUS_IS_PHYS) {
1838 struct device *pdev = dev->dev.parent;
1842 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1843 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1844 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1853 void (*destructor)(struct sk_buff *skb);
1856 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1858 static void dev_gso_skb_destructor(struct sk_buff *skb)
1860 struct dev_gso_cb *cb;
1863 struct sk_buff *nskb = skb->next;
1865 skb->next = nskb->next;
1868 } while (skb->next);
1870 cb = DEV_GSO_CB(skb);
1872 cb->destructor(skb);
1876 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1877 * @skb: buffer to segment
1879 * This function segments the given skb and stores the list of segments
1882 static int dev_gso_segment(struct sk_buff *skb)
1884 struct net_device *dev = skb->dev;
1885 struct sk_buff *segs;
1886 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1889 segs = skb_gso_segment(skb, features);
1891 /* Verifying header integrity only. */
1896 return PTR_ERR(segs);
1899 DEV_GSO_CB(skb)->destructor = skb->destructor;
1900 skb->destructor = dev_gso_skb_destructor;
1906 * Try to orphan skb early, right before transmission by the device.
1907 * We cannot orphan skb if tx timestamp is requested, since
1908 * drivers need to call skb_tstamp_tx() to send the timestamp.
1910 static inline void skb_orphan_try(struct sk_buff *skb)
1912 struct sock *sk = skb->sk;
1914 if (sk && !skb_tx(skb)->flags) {
1915 /* skb_tx_hash() wont be able to get sk.
1916 * We copy sk_hash into skb->rxhash
1919 skb->rxhash = sk->sk_hash;
1925 * Returns true if either:
1926 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1927 * 2. skb is fragmented and the device does not support SG, or if
1928 * at least one of fragments is in highmem and device does not
1929 * support DMA from it.
1931 static inline int skb_needs_linearize(struct sk_buff *skb,
1932 struct net_device *dev)
1934 return skb_is_nonlinear(skb) &&
1935 ((skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
1936 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
1937 illegal_highdma(dev, skb))));
1940 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1941 struct netdev_queue *txq)
1943 const struct net_device_ops *ops = dev->netdev_ops;
1944 int rc = NETDEV_TX_OK;
1946 if (likely(!skb->next)) {
1947 if (!list_empty(&ptype_all))
1948 dev_queue_xmit_nit(skb, dev);
1951 * If device doesnt need skb->dst, release it right now while
1952 * its hot in this cpu cache
1954 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1957 skb_orphan_try(skb);
1959 if (netif_needs_gso(dev, skb)) {
1960 if (unlikely(dev_gso_segment(skb)))
1965 if (skb_needs_linearize(skb, dev) &&
1966 __skb_linearize(skb))
1969 /* If packet is not checksummed and device does not
1970 * support checksumming for this protocol, complete
1971 * checksumming here.
1973 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1974 skb_set_transport_header(skb, skb->csum_start -
1976 if (!dev_can_checksum(dev, skb) &&
1977 skb_checksum_help(skb))
1982 rc = ops->ndo_start_xmit(skb, dev);
1983 if (rc == NETDEV_TX_OK)
1984 txq_trans_update(txq);
1990 struct sk_buff *nskb = skb->next;
1992 skb->next = nskb->next;
1996 * If device doesnt need nskb->dst, release it right now while
1997 * its hot in this cpu cache
1999 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2002 rc = ops->ndo_start_xmit(nskb, dev);
2003 if (unlikely(rc != NETDEV_TX_OK)) {
2004 if (rc & ~NETDEV_TX_MASK)
2005 goto out_kfree_gso_skb;
2006 nskb->next = skb->next;
2010 txq_trans_update(txq);
2011 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2012 return NETDEV_TX_BUSY;
2013 } while (skb->next);
2016 if (likely(skb->next == NULL))
2017 skb->destructor = DEV_GSO_CB(skb)->destructor;
2023 static u32 hashrnd __read_mostly;
2025 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
2029 if (skb_rx_queue_recorded(skb)) {
2030 hash = skb_get_rx_queue(skb);
2031 while (unlikely(hash >= dev->real_num_tx_queues))
2032 hash -= dev->real_num_tx_queues;
2036 if (skb->sk && skb->sk->sk_hash)
2037 hash = skb->sk->sk_hash;
2039 hash = (__force u16) skb->protocol ^ skb->rxhash;
2040 hash = jhash_1word(hash, hashrnd);
2042 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2044 EXPORT_SYMBOL(skb_tx_hash);
2046 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2048 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2049 if (net_ratelimit()) {
2050 pr_warning("%s selects TX queue %d, but "
2051 "real number of TX queues is %d\n",
2052 dev->name, queue_index, dev->real_num_tx_queues);
2059 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2060 struct sk_buff *skb)
2063 struct sock *sk = skb->sk;
2065 queue_index = sk_tx_queue_get(sk);
2066 if (queue_index < 0) {
2067 const struct net_device_ops *ops = dev->netdev_ops;
2069 if (ops->ndo_select_queue) {
2070 queue_index = ops->ndo_select_queue(dev, skb);
2071 queue_index = dev_cap_txqueue(dev, queue_index);
2074 if (dev->real_num_tx_queues > 1)
2075 queue_index = skb_tx_hash(dev, skb);
2078 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2080 if (dst && skb_dst(skb) == dst)
2081 sk_tx_queue_set(sk, queue_index);
2086 skb_set_queue_mapping(skb, queue_index);
2087 return netdev_get_tx_queue(dev, queue_index);
2090 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2091 struct net_device *dev,
2092 struct netdev_queue *txq)
2094 spinlock_t *root_lock = qdisc_lock(q);
2095 bool contended = qdisc_is_running(q);
2099 * Heuristic to force contended enqueues to serialize on a
2100 * separate lock before trying to get qdisc main lock.
2101 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2102 * and dequeue packets faster.
2104 if (unlikely(contended))
2105 spin_lock(&q->busylock);
2107 spin_lock(root_lock);
2108 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2111 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2112 qdisc_run_begin(q)) {
2114 * This is a work-conserving queue; there are no old skbs
2115 * waiting to be sent out; and the qdisc is not running -
2116 * xmit the skb directly.
2118 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2120 __qdisc_update_bstats(q, skb->len);
2121 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2122 if (unlikely(contended)) {
2123 spin_unlock(&q->busylock);
2130 rc = NET_XMIT_SUCCESS;
2133 rc = qdisc_enqueue_root(skb, q);
2134 if (qdisc_run_begin(q)) {
2135 if (unlikely(contended)) {
2136 spin_unlock(&q->busylock);
2142 spin_unlock(root_lock);
2143 if (unlikely(contended))
2144 spin_unlock(&q->busylock);
2149 * dev_queue_xmit - transmit a buffer
2150 * @skb: buffer to transmit
2152 * Queue a buffer for transmission to a network device. The caller must
2153 * have set the device and priority and built the buffer before calling
2154 * this function. The function can be called from an interrupt.
2156 * A negative errno code is returned on a failure. A success does not
2157 * guarantee the frame will be transmitted as it may be dropped due
2158 * to congestion or traffic shaping.
2160 * -----------------------------------------------------------------------------------
2161 * I notice this method can also return errors from the queue disciplines,
2162 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2165 * Regardless of the return value, the skb is consumed, so it is currently
2166 * difficult to retry a send to this method. (You can bump the ref count
2167 * before sending to hold a reference for retry if you are careful.)
2169 * When calling this method, interrupts MUST be enabled. This is because
2170 * the BH enable code must have IRQs enabled so that it will not deadlock.
2173 int dev_queue_xmit(struct sk_buff *skb)
2175 struct net_device *dev = skb->dev;
2176 struct netdev_queue *txq;
2180 /* Disable soft irqs for various locks below. Also
2181 * stops preemption for RCU.
2185 txq = dev_pick_tx(dev, skb);
2186 q = rcu_dereference_bh(txq->qdisc);
2188 #ifdef CONFIG_NET_CLS_ACT
2189 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2192 rc = __dev_xmit_skb(skb, q, dev, txq);
2196 /* The device has no queue. Common case for software devices:
2197 loopback, all the sorts of tunnels...
2199 Really, it is unlikely that netif_tx_lock protection is necessary
2200 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2202 However, it is possible, that they rely on protection
2205 Check this and shot the lock. It is not prone from deadlocks.
2206 Either shot noqueue qdisc, it is even simpler 8)
2208 if (dev->flags & IFF_UP) {
2209 int cpu = smp_processor_id(); /* ok because BHs are off */
2211 if (txq->xmit_lock_owner != cpu) {
2213 HARD_TX_LOCK(dev, txq, cpu);
2215 if (!netif_tx_queue_stopped(txq)) {
2216 rc = dev_hard_start_xmit(skb, dev, txq);
2217 if (dev_xmit_complete(rc)) {
2218 HARD_TX_UNLOCK(dev, txq);
2222 HARD_TX_UNLOCK(dev, txq);
2223 if (net_ratelimit())
2224 printk(KERN_CRIT "Virtual device %s asks to "
2225 "queue packet!\n", dev->name);
2227 /* Recursion is detected! It is possible,
2229 if (net_ratelimit())
2230 printk(KERN_CRIT "Dead loop on virtual device "
2231 "%s, fix it urgently!\n", dev->name);
2236 rcu_read_unlock_bh();
2241 rcu_read_unlock_bh();
2244 EXPORT_SYMBOL(dev_queue_xmit);
2247 /*=======================================================================
2249 =======================================================================*/
2251 int netdev_max_backlog __read_mostly = 1000;
2252 int netdev_tstamp_prequeue __read_mostly = 1;
2253 int netdev_budget __read_mostly = 300;
2254 int weight_p __read_mostly = 64; /* old backlog weight */
2256 /* Called with irq disabled */
2257 static inline void ____napi_schedule(struct softnet_data *sd,
2258 struct napi_struct *napi)
2260 list_add_tail(&napi->poll_list, &sd->poll_list);
2261 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2266 /* One global table that all flow-based protocols share. */
2267 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2268 EXPORT_SYMBOL(rps_sock_flow_table);
2271 * get_rps_cpu is called from netif_receive_skb and returns the target
2272 * CPU from the RPS map of the receiving queue for a given skb.
2273 * rcu_read_lock must be held on entry.
2275 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2276 struct rps_dev_flow **rflowp)
2278 struct ipv6hdr *ip6;
2280 struct netdev_rx_queue *rxqueue;
2281 struct rps_map *map;
2282 struct rps_dev_flow_table *flow_table;
2283 struct rps_sock_flow_table *sock_flow_table;
2287 u32 addr1, addr2, ihl;
2293 if (skb_rx_queue_recorded(skb)) {
2294 u16 index = skb_get_rx_queue(skb);
2295 if (unlikely(index >= dev->num_rx_queues)) {
2296 WARN_ONCE(dev->num_rx_queues > 1, "%s received packet "
2297 "on queue %u, but number of RX queues is %u\n",
2298 dev->name, index, dev->num_rx_queues);
2301 rxqueue = dev->_rx + index;
2305 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2309 goto got_hash; /* Skip hash computation on packet header */
2311 switch (skb->protocol) {
2312 case __constant_htons(ETH_P_IP):
2313 if (!pskb_may_pull(skb, sizeof(*ip)))
2316 ip = (struct iphdr *) skb->data;
2317 ip_proto = ip->protocol;
2318 addr1 = (__force u32) ip->saddr;
2319 addr2 = (__force u32) ip->daddr;
2322 case __constant_htons(ETH_P_IPV6):
2323 if (!pskb_may_pull(skb, sizeof(*ip6)))
2326 ip6 = (struct ipv6hdr *) skb->data;
2327 ip_proto = ip6->nexthdr;
2328 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2329 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2342 case IPPROTO_UDPLITE:
2343 if (pskb_may_pull(skb, (ihl * 4) + 4)) {
2344 ports.v32 = * (__force u32 *) (skb->data + (ihl * 4));
2345 if (ports.v16[1] < ports.v16[0])
2346 swap(ports.v16[0], ports.v16[1]);
2354 /* get a consistent hash (same value on both flow directions) */
2357 skb->rxhash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2362 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2363 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2364 if (flow_table && sock_flow_table) {
2366 struct rps_dev_flow *rflow;
2368 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2371 next_cpu = sock_flow_table->ents[skb->rxhash &
2372 sock_flow_table->mask];
2375 * If the desired CPU (where last recvmsg was done) is
2376 * different from current CPU (one in the rx-queue flow
2377 * table entry), switch if one of the following holds:
2378 * - Current CPU is unset (equal to RPS_NO_CPU).
2379 * - Current CPU is offline.
2380 * - The current CPU's queue tail has advanced beyond the
2381 * last packet that was enqueued using this table entry.
2382 * This guarantees that all previous packets for the flow
2383 * have been dequeued, thus preserving in order delivery.
2385 if (unlikely(tcpu != next_cpu) &&
2386 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2387 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2388 rflow->last_qtail)) >= 0)) {
2389 tcpu = rflow->cpu = next_cpu;
2390 if (tcpu != RPS_NO_CPU)
2391 rflow->last_qtail = per_cpu(softnet_data,
2392 tcpu).input_queue_head;
2394 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2401 map = rcu_dereference(rxqueue->rps_map);
2403 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2405 if (cpu_online(tcpu)) {
2415 /* Called from hardirq (IPI) context */
2416 static void rps_trigger_softirq(void *data)
2418 struct softnet_data *sd = data;
2420 ____napi_schedule(sd, &sd->backlog);
2424 #endif /* CONFIG_RPS */
2427 * Check if this softnet_data structure is another cpu one
2428 * If yes, queue it to our IPI list and return 1
2431 static int rps_ipi_queued(struct softnet_data *sd)
2434 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2437 sd->rps_ipi_next = mysd->rps_ipi_list;
2438 mysd->rps_ipi_list = sd;
2440 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2443 #endif /* CONFIG_RPS */
2448 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2449 * queue (may be a remote CPU queue).
2451 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2452 unsigned int *qtail)
2454 struct softnet_data *sd;
2455 unsigned long flags;
2457 sd = &per_cpu(softnet_data, cpu);
2459 local_irq_save(flags);
2462 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2463 if (skb_queue_len(&sd->input_pkt_queue)) {
2465 __skb_queue_tail(&sd->input_pkt_queue, skb);
2466 input_queue_tail_incr_save(sd, qtail);
2468 local_irq_restore(flags);
2469 return NET_RX_SUCCESS;
2472 /* Schedule NAPI for backlog device
2473 * We can use non atomic operation since we own the queue lock
2475 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2476 if (!rps_ipi_queued(sd))
2477 ____napi_schedule(sd, &sd->backlog);
2485 local_irq_restore(flags);
2492 * netif_rx - post buffer to the network code
2493 * @skb: buffer to post
2495 * This function receives a packet from a device driver and queues it for
2496 * the upper (protocol) levels to process. It always succeeds. The buffer
2497 * may be dropped during processing for congestion control or by the
2501 * NET_RX_SUCCESS (no congestion)
2502 * NET_RX_DROP (packet was dropped)
2506 int netif_rx(struct sk_buff *skb)
2510 /* if netpoll wants it, pretend we never saw it */
2511 if (netpoll_rx(skb))
2514 if (netdev_tstamp_prequeue)
2515 net_timestamp_check(skb);
2519 struct rps_dev_flow voidflow, *rflow = &voidflow;
2524 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2526 cpu = smp_processor_id();
2528 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2535 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2541 EXPORT_SYMBOL(netif_rx);
2543 int netif_rx_ni(struct sk_buff *skb)
2548 err = netif_rx(skb);
2549 if (local_softirq_pending())
2555 EXPORT_SYMBOL(netif_rx_ni);
2557 static void net_tx_action(struct softirq_action *h)
2559 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2561 if (sd->completion_queue) {
2562 struct sk_buff *clist;
2564 local_irq_disable();
2565 clist = sd->completion_queue;
2566 sd->completion_queue = NULL;
2570 struct sk_buff *skb = clist;
2571 clist = clist->next;
2573 WARN_ON(atomic_read(&skb->users));
2578 if (sd->output_queue) {
2581 local_irq_disable();
2582 head = sd->output_queue;
2583 sd->output_queue = NULL;
2584 sd->output_queue_tailp = &sd->output_queue;
2588 struct Qdisc *q = head;
2589 spinlock_t *root_lock;
2591 head = head->next_sched;
2593 root_lock = qdisc_lock(q);
2594 if (spin_trylock(root_lock)) {
2595 smp_mb__before_clear_bit();
2596 clear_bit(__QDISC_STATE_SCHED,
2599 spin_unlock(root_lock);
2601 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2603 __netif_reschedule(q);
2605 smp_mb__before_clear_bit();
2606 clear_bit(__QDISC_STATE_SCHED,
2614 static inline int deliver_skb(struct sk_buff *skb,
2615 struct packet_type *pt_prev,
2616 struct net_device *orig_dev)
2618 atomic_inc(&skb->users);
2619 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2622 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2623 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2624 /* This hook is defined here for ATM LANE */
2625 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2626 unsigned char *addr) __read_mostly;
2627 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2630 #ifdef CONFIG_NET_CLS_ACT
2631 /* TODO: Maybe we should just force sch_ingress to be compiled in
2632 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2633 * a compare and 2 stores extra right now if we dont have it on
2634 * but have CONFIG_NET_CLS_ACT
2635 * NOTE: This doesnt stop any functionality; if you dont have
2636 * the ingress scheduler, you just cant add policies on ingress.
2639 static int ing_filter(struct sk_buff *skb)
2641 struct net_device *dev = skb->dev;
2642 u32 ttl = G_TC_RTTL(skb->tc_verd);
2643 struct netdev_queue *rxq;
2644 int result = TC_ACT_OK;
2647 if (unlikely(MAX_RED_LOOP < ttl++)) {
2648 if (net_ratelimit())
2649 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2650 skb->skb_iif, dev->ifindex);
2654 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2655 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2657 rxq = &dev->rx_queue;
2660 if (q != &noop_qdisc) {
2661 spin_lock(qdisc_lock(q));
2662 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2663 result = qdisc_enqueue_root(skb, q);
2664 spin_unlock(qdisc_lock(q));
2670 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2671 struct packet_type **pt_prev,
2672 int *ret, struct net_device *orig_dev)
2674 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2678 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2682 switch (ing_filter(skb)) {
2696 * netif_nit_deliver - deliver received packets to network taps
2699 * This function is used to deliver incoming packets to network
2700 * taps. It should be used when the normal netif_receive_skb path
2701 * is bypassed, for example because of VLAN acceleration.
2703 void netif_nit_deliver(struct sk_buff *skb)
2705 struct packet_type *ptype;
2707 if (list_empty(&ptype_all))
2710 skb_reset_network_header(skb);
2711 skb_reset_transport_header(skb);
2712 skb->mac_len = skb->network_header - skb->mac_header;
2715 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2716 if (!ptype->dev || ptype->dev == skb->dev)
2717 deliver_skb(skb, ptype, skb->dev);
2723 * netdev_rx_handler_register - register receive handler
2724 * @dev: device to register a handler for
2725 * @rx_handler: receive handler to register
2726 * @rx_handler_data: data pointer that is used by rx handler
2728 * Register a receive hander for a device. This handler will then be
2729 * called from __netif_receive_skb. A negative errno code is returned
2732 * The caller must hold the rtnl_mutex.
2734 int netdev_rx_handler_register(struct net_device *dev,
2735 rx_handler_func_t *rx_handler,
2736 void *rx_handler_data)
2740 if (dev->rx_handler)
2743 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2744 rcu_assign_pointer(dev->rx_handler, rx_handler);
2748 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2751 * netdev_rx_handler_unregister - unregister receive handler
2752 * @dev: device to unregister a handler from
2754 * Unregister a receive hander from a device.
2756 * The caller must hold the rtnl_mutex.
2758 void netdev_rx_handler_unregister(struct net_device *dev)
2762 rcu_assign_pointer(dev->rx_handler, NULL);
2763 rcu_assign_pointer(dev->rx_handler_data, NULL);
2765 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2767 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2768 struct net_device *master)
2770 if (skb->pkt_type == PACKET_HOST) {
2771 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2773 memcpy(dest, master->dev_addr, ETH_ALEN);
2777 /* On bonding slaves other than the currently active slave, suppress
2778 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2779 * ARP on active-backup slaves with arp_validate enabled.
2781 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2783 struct net_device *dev = skb->dev;
2785 if (master->priv_flags & IFF_MASTER_ARPMON)
2786 dev->last_rx = jiffies;
2788 if ((master->priv_flags & IFF_MASTER_ALB) &&
2789 (master->priv_flags & IFF_BRIDGE_PORT)) {
2790 /* Do address unmangle. The local destination address
2791 * will be always the one master has. Provides the right
2792 * functionality in a bridge.
2794 skb_bond_set_mac_by_master(skb, master);
2797 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2798 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2799 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2802 if (master->priv_flags & IFF_MASTER_ALB) {
2803 if (skb->pkt_type != PACKET_BROADCAST &&
2804 skb->pkt_type != PACKET_MULTICAST)
2807 if (master->priv_flags & IFF_MASTER_8023AD &&
2808 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2815 EXPORT_SYMBOL(__skb_bond_should_drop);
2817 static int __netif_receive_skb(struct sk_buff *skb)
2819 struct packet_type *ptype, *pt_prev;
2820 rx_handler_func_t *rx_handler;
2821 struct net_device *orig_dev;
2822 struct net_device *master;
2823 struct net_device *null_or_orig;
2824 struct net_device *orig_or_bond;
2825 int ret = NET_RX_DROP;
2828 if (!netdev_tstamp_prequeue)
2829 net_timestamp_check(skb);
2831 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2832 return NET_RX_SUCCESS;
2834 /* if we've gotten here through NAPI, check netpoll */
2835 if (netpoll_receive_skb(skb))
2839 skb->skb_iif = skb->dev->ifindex;
2842 * bonding note: skbs received on inactive slaves should only
2843 * be delivered to pkt handlers that are exact matches. Also
2844 * the deliver_no_wcard flag will be set. If packet handlers
2845 * are sensitive to duplicate packets these skbs will need to
2846 * be dropped at the handler. The vlan accel path may have
2847 * already set the deliver_no_wcard flag.
2849 null_or_orig = NULL;
2850 orig_dev = skb->dev;
2851 master = ACCESS_ONCE(orig_dev->master);
2852 if (skb->deliver_no_wcard)
2853 null_or_orig = orig_dev;
2855 if (skb_bond_should_drop(skb, master)) {
2856 skb->deliver_no_wcard = 1;
2857 null_or_orig = orig_dev; /* deliver only exact match */
2862 __this_cpu_inc(softnet_data.processed);
2863 skb_reset_network_header(skb);
2864 skb_reset_transport_header(skb);
2865 skb->mac_len = skb->network_header - skb->mac_header;
2871 #ifdef CONFIG_NET_CLS_ACT
2872 if (skb->tc_verd & TC_NCLS) {
2873 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2878 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2879 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2880 ptype->dev == orig_dev) {
2882 ret = deliver_skb(skb, pt_prev, orig_dev);
2887 #ifdef CONFIG_NET_CLS_ACT
2888 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2894 /* Handle special case of bridge or macvlan */
2895 rx_handler = rcu_dereference(skb->dev->rx_handler);
2898 ret = deliver_skb(skb, pt_prev, orig_dev);
2901 skb = rx_handler(skb);
2907 * Make sure frames received on VLAN interfaces stacked on
2908 * bonding interfaces still make their way to any base bonding
2909 * device that may have registered for a specific ptype. The
2910 * handler may have to adjust skb->dev and orig_dev.
2912 orig_or_bond = orig_dev;
2913 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2914 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2915 orig_or_bond = vlan_dev_real_dev(skb->dev);
2918 type = skb->protocol;
2919 list_for_each_entry_rcu(ptype,
2920 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2921 if (ptype->type == type && (ptype->dev == null_or_orig ||
2922 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2923 ptype->dev == orig_or_bond)) {
2925 ret = deliver_skb(skb, pt_prev, orig_dev);
2931 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2934 /* Jamal, now you will not able to escape explaining
2935 * me how you were going to use this. :-)
2946 * netif_receive_skb - process receive buffer from network
2947 * @skb: buffer to process
2949 * netif_receive_skb() is the main receive data processing function.
2950 * It always succeeds. The buffer may be dropped during processing
2951 * for congestion control or by the protocol layers.
2953 * This function may only be called from softirq context and interrupts
2954 * should be enabled.
2956 * Return values (usually ignored):
2957 * NET_RX_SUCCESS: no congestion
2958 * NET_RX_DROP: packet was dropped
2960 int netif_receive_skb(struct sk_buff *skb)
2962 if (netdev_tstamp_prequeue)
2963 net_timestamp_check(skb);
2965 if (skb_defer_rx_timestamp(skb))
2966 return NET_RX_SUCCESS;
2970 struct rps_dev_flow voidflow, *rflow = &voidflow;
2975 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2978 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2982 ret = __netif_receive_skb(skb);
2988 return __netif_receive_skb(skb);
2991 EXPORT_SYMBOL(netif_receive_skb);
2993 /* Network device is going away, flush any packets still pending
2994 * Called with irqs disabled.
2996 static void flush_backlog(void *arg)
2998 struct net_device *dev = arg;
2999 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3000 struct sk_buff *skb, *tmp;
3003 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3004 if (skb->dev == dev) {
3005 __skb_unlink(skb, &sd->input_pkt_queue);
3007 input_queue_head_incr(sd);
3012 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3013 if (skb->dev == dev) {
3014 __skb_unlink(skb, &sd->process_queue);
3016 input_queue_head_incr(sd);
3021 static int napi_gro_complete(struct sk_buff *skb)
3023 struct packet_type *ptype;
3024 __be16 type = skb->protocol;
3025 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3028 if (NAPI_GRO_CB(skb)->count == 1) {
3029 skb_shinfo(skb)->gso_size = 0;
3034 list_for_each_entry_rcu(ptype, head, list) {
3035 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3038 err = ptype->gro_complete(skb);
3044 WARN_ON(&ptype->list == head);
3046 return NET_RX_SUCCESS;
3050 return netif_receive_skb(skb);
3053 static void napi_gro_flush(struct napi_struct *napi)
3055 struct sk_buff *skb, *next;
3057 for (skb = napi->gro_list; skb; skb = next) {
3060 napi_gro_complete(skb);
3063 napi->gro_count = 0;
3064 napi->gro_list = NULL;
3067 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3069 struct sk_buff **pp = NULL;
3070 struct packet_type *ptype;
3071 __be16 type = skb->protocol;
3072 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3075 enum gro_result ret;
3077 if (!(skb->dev->features & NETIF_F_GRO))
3080 if (skb_is_gso(skb) || skb_has_frags(skb))
3084 list_for_each_entry_rcu(ptype, head, list) {
3085 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3088 skb_set_network_header(skb, skb_gro_offset(skb));
3089 mac_len = skb->network_header - skb->mac_header;
3090 skb->mac_len = mac_len;
3091 NAPI_GRO_CB(skb)->same_flow = 0;
3092 NAPI_GRO_CB(skb)->flush = 0;
3093 NAPI_GRO_CB(skb)->free = 0;
3095 pp = ptype->gro_receive(&napi->gro_list, skb);
3100 if (&ptype->list == head)
3103 same_flow = NAPI_GRO_CB(skb)->same_flow;
3104 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3107 struct sk_buff *nskb = *pp;
3111 napi_gro_complete(nskb);
3118 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3122 NAPI_GRO_CB(skb)->count = 1;
3123 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3124 skb->next = napi->gro_list;
3125 napi->gro_list = skb;
3129 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3130 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3132 BUG_ON(skb->end - skb->tail < grow);
3134 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3137 skb->data_len -= grow;
3139 skb_shinfo(skb)->frags[0].page_offset += grow;
3140 skb_shinfo(skb)->frags[0].size -= grow;
3142 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3143 put_page(skb_shinfo(skb)->frags[0].page);
3144 memmove(skb_shinfo(skb)->frags,
3145 skb_shinfo(skb)->frags + 1,
3146 --skb_shinfo(skb)->nr_frags);
3157 EXPORT_SYMBOL(dev_gro_receive);
3160 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3164 if (netpoll_rx_on(skb))
3167 for (p = napi->gro_list; p; p = p->next) {
3168 NAPI_GRO_CB(p)->same_flow =
3169 (p->dev == skb->dev) &&
3170 !compare_ether_header(skb_mac_header(p),
3171 skb_gro_mac_header(skb));
3172 NAPI_GRO_CB(p)->flush = 0;
3175 return dev_gro_receive(napi, skb);
3178 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3182 if (netif_receive_skb(skb))
3187 case GRO_MERGED_FREE:
3198 EXPORT_SYMBOL(napi_skb_finish);
3200 void skb_gro_reset_offset(struct sk_buff *skb)
3202 NAPI_GRO_CB(skb)->data_offset = 0;
3203 NAPI_GRO_CB(skb)->frag0 = NULL;
3204 NAPI_GRO_CB(skb)->frag0_len = 0;
3206 if (skb->mac_header == skb->tail &&
3207 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3208 NAPI_GRO_CB(skb)->frag0 =
3209 page_address(skb_shinfo(skb)->frags[0].page) +
3210 skb_shinfo(skb)->frags[0].page_offset;
3211 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3214 EXPORT_SYMBOL(skb_gro_reset_offset);
3216 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3218 skb_gro_reset_offset(skb);
3220 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3222 EXPORT_SYMBOL(napi_gro_receive);
3224 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3226 __skb_pull(skb, skb_headlen(skb));
3227 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3231 EXPORT_SYMBOL(napi_reuse_skb);
3233 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3235 struct sk_buff *skb = napi->skb;
3238 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3244 EXPORT_SYMBOL(napi_get_frags);
3246 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3252 skb->protocol = eth_type_trans(skb, skb->dev);
3254 if (ret == GRO_HELD)
3255 skb_gro_pull(skb, -ETH_HLEN);
3256 else if (netif_receive_skb(skb))
3261 case GRO_MERGED_FREE:
3262 napi_reuse_skb(napi, skb);
3271 EXPORT_SYMBOL(napi_frags_finish);
3273 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3275 struct sk_buff *skb = napi->skb;
3282 skb_reset_mac_header(skb);
3283 skb_gro_reset_offset(skb);
3285 off = skb_gro_offset(skb);
3286 hlen = off + sizeof(*eth);
3287 eth = skb_gro_header_fast(skb, off);
3288 if (skb_gro_header_hard(skb, hlen)) {
3289 eth = skb_gro_header_slow(skb, hlen, off);
3290 if (unlikely(!eth)) {
3291 napi_reuse_skb(napi, skb);
3297 skb_gro_pull(skb, sizeof(*eth));
3300 * This works because the only protocols we care about don't require
3301 * special handling. We'll fix it up properly at the end.
3303 skb->protocol = eth->h_proto;
3308 EXPORT_SYMBOL(napi_frags_skb);
3310 gro_result_t napi_gro_frags(struct napi_struct *napi)
3312 struct sk_buff *skb = napi_frags_skb(napi);
3317 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3319 EXPORT_SYMBOL(napi_gro_frags);
3322 * net_rps_action sends any pending IPI's for rps.
3323 * Note: called with local irq disabled, but exits with local irq enabled.
3325 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3328 struct softnet_data *remsd = sd->rps_ipi_list;
3331 sd->rps_ipi_list = NULL;
3335 /* Send pending IPI's to kick RPS processing on remote cpus. */
3337 struct softnet_data *next = remsd->rps_ipi_next;
3339 if (cpu_online(remsd->cpu))
3340 __smp_call_function_single(remsd->cpu,
3349 static int process_backlog(struct napi_struct *napi, int quota)
3352 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3355 /* Check if we have pending ipi, its better to send them now,
3356 * not waiting net_rx_action() end.
3358 if (sd->rps_ipi_list) {
3359 local_irq_disable();
3360 net_rps_action_and_irq_enable(sd);
3363 napi->weight = weight_p;
3364 local_irq_disable();
3365 while (work < quota) {
3366 struct sk_buff *skb;
3369 while ((skb = __skb_dequeue(&sd->process_queue))) {
3371 __netif_receive_skb(skb);
3372 local_irq_disable();
3373 input_queue_head_incr(sd);
3374 if (++work >= quota) {
3381 qlen = skb_queue_len(&sd->input_pkt_queue);
3383 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3384 &sd->process_queue);
3386 if (qlen < quota - work) {
3388 * Inline a custom version of __napi_complete().
3389 * only current cpu owns and manipulates this napi,
3390 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3391 * we can use a plain write instead of clear_bit(),
3392 * and we dont need an smp_mb() memory barrier.
3394 list_del(&napi->poll_list);
3397 quota = work + qlen;
3407 * __napi_schedule - schedule for receive
3408 * @n: entry to schedule
3410 * The entry's receive function will be scheduled to run
3412 void __napi_schedule(struct napi_struct *n)
3414 unsigned long flags;
3416 local_irq_save(flags);
3417 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3418 local_irq_restore(flags);
3420 EXPORT_SYMBOL(__napi_schedule);
3422 void __napi_complete(struct napi_struct *n)
3424 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3425 BUG_ON(n->gro_list);
3427 list_del(&n->poll_list);
3428 smp_mb__before_clear_bit();
3429 clear_bit(NAPI_STATE_SCHED, &n->state);
3431 EXPORT_SYMBOL(__napi_complete);
3433 void napi_complete(struct napi_struct *n)
3435 unsigned long flags;
3438 * don't let napi dequeue from the cpu poll list
3439 * just in case its running on a different cpu
3441 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3445 local_irq_save(flags);
3447 local_irq_restore(flags);
3449 EXPORT_SYMBOL(napi_complete);
3451 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3452 int (*poll)(struct napi_struct *, int), int weight)
3454 INIT_LIST_HEAD(&napi->poll_list);
3455 napi->gro_count = 0;
3456 napi->gro_list = NULL;
3459 napi->weight = weight;
3460 list_add(&napi->dev_list, &dev->napi_list);
3462 #ifdef CONFIG_NETPOLL
3463 spin_lock_init(&napi->poll_lock);
3464 napi->poll_owner = -1;
3466 set_bit(NAPI_STATE_SCHED, &napi->state);
3468 EXPORT_SYMBOL(netif_napi_add);
3470 void netif_napi_del(struct napi_struct *napi)
3472 struct sk_buff *skb, *next;
3474 list_del_init(&napi->dev_list);
3475 napi_free_frags(napi);
3477 for (skb = napi->gro_list; skb; skb = next) {
3483 napi->gro_list = NULL;
3484 napi->gro_count = 0;
3486 EXPORT_SYMBOL(netif_napi_del);
3488 static void net_rx_action(struct softirq_action *h)
3490 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3491 unsigned long time_limit = jiffies + 2;
3492 int budget = netdev_budget;
3495 local_irq_disable();
3497 while (!list_empty(&sd->poll_list)) {
3498 struct napi_struct *n;
3501 /* If softirq window is exhuasted then punt.
3502 * Allow this to run for 2 jiffies since which will allow
3503 * an average latency of 1.5/HZ.
3505 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3510 /* Even though interrupts have been re-enabled, this
3511 * access is safe because interrupts can only add new
3512 * entries to the tail of this list, and only ->poll()
3513 * calls can remove this head entry from the list.
3515 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3517 have = netpoll_poll_lock(n);
3521 /* This NAPI_STATE_SCHED test is for avoiding a race
3522 * with netpoll's poll_napi(). Only the entity which
3523 * obtains the lock and sees NAPI_STATE_SCHED set will
3524 * actually make the ->poll() call. Therefore we avoid
3525 * accidently calling ->poll() when NAPI is not scheduled.
3528 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3529 work = n->poll(n, weight);
3533 WARN_ON_ONCE(work > weight);
3537 local_irq_disable();
3539 /* Drivers must not modify the NAPI state if they
3540 * consume the entire weight. In such cases this code
3541 * still "owns" the NAPI instance and therefore can
3542 * move the instance around on the list at-will.
3544 if (unlikely(work == weight)) {
3545 if (unlikely(napi_disable_pending(n))) {
3548 local_irq_disable();
3550 list_move_tail(&n->poll_list, &sd->poll_list);
3553 netpoll_poll_unlock(have);
3556 net_rps_action_and_irq_enable(sd);
3558 #ifdef CONFIG_NET_DMA
3560 * There may not be any more sk_buffs coming right now, so push
3561 * any pending DMA copies to hardware
3563 dma_issue_pending_all();
3570 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3574 static gifconf_func_t *gifconf_list[NPROTO];
3577 * register_gifconf - register a SIOCGIF handler
3578 * @family: Address family
3579 * @gifconf: Function handler
3581 * Register protocol dependent address dumping routines. The handler
3582 * that is passed must not be freed or reused until it has been replaced
3583 * by another handler.
3585 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3587 if (family >= NPROTO)
3589 gifconf_list[family] = gifconf;
3592 EXPORT_SYMBOL(register_gifconf);
3596 * Map an interface index to its name (SIOCGIFNAME)
3600 * We need this ioctl for efficient implementation of the
3601 * if_indextoname() function required by the IPv6 API. Without
3602 * it, we would have to search all the interfaces to find a
3606 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3608 struct net_device *dev;
3612 * Fetch the caller's info block.
3615 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3619 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3625 strcpy(ifr.ifr_name, dev->name);
3628 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3634 * Perform a SIOCGIFCONF call. This structure will change
3635 * size eventually, and there is nothing I can do about it.
3636 * Thus we will need a 'compatibility mode'.
3639 static int dev_ifconf(struct net *net, char __user *arg)
3642 struct net_device *dev;
3649 * Fetch the caller's info block.
3652 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3659 * Loop over the interfaces, and write an info block for each.
3663 for_each_netdev(net, dev) {
3664 for (i = 0; i < NPROTO; i++) {
3665 if (gifconf_list[i]) {
3668 done = gifconf_list[i](dev, NULL, 0);
3670 done = gifconf_list[i](dev, pos + total,
3680 * All done. Write the updated control block back to the caller.
3682 ifc.ifc_len = total;
3685 * Both BSD and Solaris return 0 here, so we do too.
3687 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3690 #ifdef CONFIG_PROC_FS
3692 * This is invoked by the /proc filesystem handler to display a device
3695 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3698 struct net *net = seq_file_net(seq);
3700 struct net_device *dev;
3704 return SEQ_START_TOKEN;
3707 for_each_netdev_rcu(net, dev)
3714 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3716 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3717 first_net_device(seq_file_net(seq)) :
3718 next_net_device((struct net_device *)v);
3721 return rcu_dereference(dev);
3724 void dev_seq_stop(struct seq_file *seq, void *v)
3730 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3732 struct rtnl_link_stats64 temp;
3733 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3735 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3736 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3737 dev->name, stats->rx_bytes, stats->rx_packets,
3739 stats->rx_dropped + stats->rx_missed_errors,
3740 stats->rx_fifo_errors,
3741 stats->rx_length_errors + stats->rx_over_errors +
3742 stats->rx_crc_errors + stats->rx_frame_errors,
3743 stats->rx_compressed, stats->multicast,
3744 stats->tx_bytes, stats->tx_packets,
3745 stats->tx_errors, stats->tx_dropped,
3746 stats->tx_fifo_errors, stats->collisions,
3747 stats->tx_carrier_errors +
3748 stats->tx_aborted_errors +
3749 stats->tx_window_errors +
3750 stats->tx_heartbeat_errors,
3751 stats->tx_compressed);
3755 * Called from the PROCfs module. This now uses the new arbitrary sized
3756 * /proc/net interface to create /proc/net/dev
3758 static int dev_seq_show(struct seq_file *seq, void *v)
3760 if (v == SEQ_START_TOKEN)
3761 seq_puts(seq, "Inter-| Receive "
3763 " face |bytes packets errs drop fifo frame "
3764 "compressed multicast|bytes packets errs "
3765 "drop fifo colls carrier compressed\n");
3767 dev_seq_printf_stats(seq, v);
3771 static struct softnet_data *softnet_get_online(loff_t *pos)
3773 struct softnet_data *sd = NULL;
3775 while (*pos < nr_cpu_ids)
3776 if (cpu_online(*pos)) {
3777 sd = &per_cpu(softnet_data, *pos);
3784 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3786 return softnet_get_online(pos);
3789 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3792 return softnet_get_online(pos);
3795 static void softnet_seq_stop(struct seq_file *seq, void *v)
3799 static int softnet_seq_show(struct seq_file *seq, void *v)
3801 struct softnet_data *sd = v;
3803 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3804 sd->processed, sd->dropped, sd->time_squeeze, 0,
3805 0, 0, 0, 0, /* was fastroute */
3806 sd->cpu_collision, sd->received_rps);
3810 static const struct seq_operations dev_seq_ops = {
3811 .start = dev_seq_start,
3812 .next = dev_seq_next,
3813 .stop = dev_seq_stop,
3814 .show = dev_seq_show,
3817 static int dev_seq_open(struct inode *inode, struct file *file)
3819 return seq_open_net(inode, file, &dev_seq_ops,
3820 sizeof(struct seq_net_private));
3823 static const struct file_operations dev_seq_fops = {
3824 .owner = THIS_MODULE,
3825 .open = dev_seq_open,
3827 .llseek = seq_lseek,
3828 .release = seq_release_net,
3831 static const struct seq_operations softnet_seq_ops = {
3832 .start = softnet_seq_start,
3833 .next = softnet_seq_next,
3834 .stop = softnet_seq_stop,
3835 .show = softnet_seq_show,
3838 static int softnet_seq_open(struct inode *inode, struct file *file)
3840 return seq_open(file, &softnet_seq_ops);
3843 static const struct file_operations softnet_seq_fops = {
3844 .owner = THIS_MODULE,
3845 .open = softnet_seq_open,
3847 .llseek = seq_lseek,
3848 .release = seq_release,
3851 static void *ptype_get_idx(loff_t pos)
3853 struct packet_type *pt = NULL;
3857 list_for_each_entry_rcu(pt, &ptype_all, list) {
3863 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3864 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3873 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3877 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3880 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3882 struct packet_type *pt;
3883 struct list_head *nxt;
3887 if (v == SEQ_START_TOKEN)
3888 return ptype_get_idx(0);
3891 nxt = pt->list.next;
3892 if (pt->type == htons(ETH_P_ALL)) {
3893 if (nxt != &ptype_all)
3896 nxt = ptype_base[0].next;
3898 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3900 while (nxt == &ptype_base[hash]) {
3901 if (++hash >= PTYPE_HASH_SIZE)
3903 nxt = ptype_base[hash].next;
3906 return list_entry(nxt, struct packet_type, list);
3909 static void ptype_seq_stop(struct seq_file *seq, void *v)
3915 static int ptype_seq_show(struct seq_file *seq, void *v)
3917 struct packet_type *pt = v;
3919 if (v == SEQ_START_TOKEN)
3920 seq_puts(seq, "Type Device Function\n");
3921 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3922 if (pt->type == htons(ETH_P_ALL))
3923 seq_puts(seq, "ALL ");
3925 seq_printf(seq, "%04x", ntohs(pt->type));
3927 seq_printf(seq, " %-8s %pF\n",
3928 pt->dev ? pt->dev->name : "", pt->func);
3934 static const struct seq_operations ptype_seq_ops = {
3935 .start = ptype_seq_start,
3936 .next = ptype_seq_next,
3937 .stop = ptype_seq_stop,
3938 .show = ptype_seq_show,
3941 static int ptype_seq_open(struct inode *inode, struct file *file)
3943 return seq_open_net(inode, file, &ptype_seq_ops,
3944 sizeof(struct seq_net_private));
3947 static const struct file_operations ptype_seq_fops = {
3948 .owner = THIS_MODULE,
3949 .open = ptype_seq_open,
3951 .llseek = seq_lseek,
3952 .release = seq_release_net,
3956 static int __net_init dev_proc_net_init(struct net *net)
3960 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3962 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3964 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3967 if (wext_proc_init(net))
3973 proc_net_remove(net, "ptype");
3975 proc_net_remove(net, "softnet_stat");
3977 proc_net_remove(net, "dev");
3981 static void __net_exit dev_proc_net_exit(struct net *net)
3983 wext_proc_exit(net);
3985 proc_net_remove(net, "ptype");
3986 proc_net_remove(net, "softnet_stat");
3987 proc_net_remove(net, "dev");
3990 static struct pernet_operations __net_initdata dev_proc_ops = {
3991 .init = dev_proc_net_init,
3992 .exit = dev_proc_net_exit,
3995 static int __init dev_proc_init(void)
3997 return register_pernet_subsys(&dev_proc_ops);
4000 #define dev_proc_init() 0
4001 #endif /* CONFIG_PROC_FS */
4005 * netdev_set_master - set up master/slave pair
4006 * @slave: slave device
4007 * @master: new master device
4009 * Changes the master device of the slave. Pass %NULL to break the
4010 * bonding. The caller must hold the RTNL semaphore. On a failure
4011 * a negative errno code is returned. On success the reference counts
4012 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4013 * function returns zero.
4015 int netdev_set_master(struct net_device *slave, struct net_device *master)
4017 struct net_device *old = slave->master;
4027 slave->master = master;
4034 slave->flags |= IFF_SLAVE;
4036 slave->flags &= ~IFF_SLAVE;
4038 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4041 EXPORT_SYMBOL(netdev_set_master);
4043 static void dev_change_rx_flags(struct net_device *dev, int flags)
4045 const struct net_device_ops *ops = dev->netdev_ops;
4047 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4048 ops->ndo_change_rx_flags(dev, flags);
4051 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4053 unsigned short old_flags = dev->flags;
4059 dev->flags |= IFF_PROMISC;
4060 dev->promiscuity += inc;
4061 if (dev->promiscuity == 0) {
4064 * If inc causes overflow, untouch promisc and return error.
4067 dev->flags &= ~IFF_PROMISC;
4069 dev->promiscuity -= inc;
4070 printk(KERN_WARNING "%s: promiscuity touches roof, "
4071 "set promiscuity failed, promiscuity feature "
4072 "of device might be broken.\n", dev->name);
4076 if (dev->flags != old_flags) {
4077 printk(KERN_INFO "device %s %s promiscuous mode\n",
4078 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4080 if (audit_enabled) {
4081 current_uid_gid(&uid, &gid);
4082 audit_log(current->audit_context, GFP_ATOMIC,
4083 AUDIT_ANOM_PROMISCUOUS,
4084 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4085 dev->name, (dev->flags & IFF_PROMISC),
4086 (old_flags & IFF_PROMISC),
4087 audit_get_loginuid(current),
4089 audit_get_sessionid(current));
4092 dev_change_rx_flags(dev, IFF_PROMISC);
4098 * dev_set_promiscuity - update promiscuity count on a device
4102 * Add or remove promiscuity from a device. While the count in the device
4103 * remains above zero the interface remains promiscuous. Once it hits zero
4104 * the device reverts back to normal filtering operation. A negative inc
4105 * value is used to drop promiscuity on the device.
4106 * Return 0 if successful or a negative errno code on error.
4108 int dev_set_promiscuity(struct net_device *dev, int inc)
4110 unsigned short old_flags = dev->flags;
4113 err = __dev_set_promiscuity(dev, inc);
4116 if (dev->flags != old_flags)
4117 dev_set_rx_mode(dev);
4120 EXPORT_SYMBOL(dev_set_promiscuity);
4123 * dev_set_allmulti - update allmulti count on a device
4127 * Add or remove reception of all multicast frames to a device. While the
4128 * count in the device remains above zero the interface remains listening
4129 * to all interfaces. Once it hits zero the device reverts back to normal
4130 * filtering operation. A negative @inc value is used to drop the counter
4131 * when releasing a resource needing all multicasts.
4132 * Return 0 if successful or a negative errno code on error.
4135 int dev_set_allmulti(struct net_device *dev, int inc)
4137 unsigned short old_flags = dev->flags;
4141 dev->flags |= IFF_ALLMULTI;
4142 dev->allmulti += inc;
4143 if (dev->allmulti == 0) {
4146 * If inc causes overflow, untouch allmulti and return error.
4149 dev->flags &= ~IFF_ALLMULTI;
4151 dev->allmulti -= inc;
4152 printk(KERN_WARNING "%s: allmulti touches roof, "
4153 "set allmulti failed, allmulti feature of "
4154 "device might be broken.\n", dev->name);
4158 if (dev->flags ^ old_flags) {
4159 dev_change_rx_flags(dev, IFF_ALLMULTI);
4160 dev_set_rx_mode(dev);
4164 EXPORT_SYMBOL(dev_set_allmulti);
4167 * Upload unicast and multicast address lists to device and
4168 * configure RX filtering. When the device doesn't support unicast
4169 * filtering it is put in promiscuous mode while unicast addresses
4172 void __dev_set_rx_mode(struct net_device *dev)
4174 const struct net_device_ops *ops = dev->netdev_ops;
4176 /* dev_open will call this function so the list will stay sane. */
4177 if (!(dev->flags&IFF_UP))
4180 if (!netif_device_present(dev))
4183 if (ops->ndo_set_rx_mode)
4184 ops->ndo_set_rx_mode(dev);
4186 /* Unicast addresses changes may only happen under the rtnl,
4187 * therefore calling __dev_set_promiscuity here is safe.
4189 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4190 __dev_set_promiscuity(dev, 1);
4191 dev->uc_promisc = 1;
4192 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4193 __dev_set_promiscuity(dev, -1);
4194 dev->uc_promisc = 0;
4197 if (ops->ndo_set_multicast_list)
4198 ops->ndo_set_multicast_list(dev);
4202 void dev_set_rx_mode(struct net_device *dev)
4204 netif_addr_lock_bh(dev);
4205 __dev_set_rx_mode(dev);
4206 netif_addr_unlock_bh(dev);
4210 * dev_get_flags - get flags reported to userspace
4213 * Get the combination of flag bits exported through APIs to userspace.
4215 unsigned dev_get_flags(const struct net_device *dev)
4219 flags = (dev->flags & ~(IFF_PROMISC |
4224 (dev->gflags & (IFF_PROMISC |
4227 if (netif_running(dev)) {
4228 if (netif_oper_up(dev))
4229 flags |= IFF_RUNNING;
4230 if (netif_carrier_ok(dev))
4231 flags |= IFF_LOWER_UP;
4232 if (netif_dormant(dev))
4233 flags |= IFF_DORMANT;
4238 EXPORT_SYMBOL(dev_get_flags);
4240 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4242 int old_flags = dev->flags;
4248 * Set the flags on our device.
4251 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4252 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4254 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4258 * Load in the correct multicast list now the flags have changed.
4261 if ((old_flags ^ flags) & IFF_MULTICAST)
4262 dev_change_rx_flags(dev, IFF_MULTICAST);
4264 dev_set_rx_mode(dev);
4267 * Have we downed the interface. We handle IFF_UP ourselves
4268 * according to user attempts to set it, rather than blindly
4273 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4274 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4277 dev_set_rx_mode(dev);
4280 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4281 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4283 dev->gflags ^= IFF_PROMISC;
4284 dev_set_promiscuity(dev, inc);
4287 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4288 is important. Some (broken) drivers set IFF_PROMISC, when
4289 IFF_ALLMULTI is requested not asking us and not reporting.
4291 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4292 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4294 dev->gflags ^= IFF_ALLMULTI;
4295 dev_set_allmulti(dev, inc);
4301 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4303 unsigned int changes = dev->flags ^ old_flags;
4305 if (changes & IFF_UP) {
4306 if (dev->flags & IFF_UP)
4307 call_netdevice_notifiers(NETDEV_UP, dev);
4309 call_netdevice_notifiers(NETDEV_DOWN, dev);
4312 if (dev->flags & IFF_UP &&
4313 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4314 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4318 * dev_change_flags - change device settings
4320 * @flags: device state flags
4322 * Change settings on device based state flags. The flags are
4323 * in the userspace exported format.
4325 int dev_change_flags(struct net_device *dev, unsigned flags)
4328 int old_flags = dev->flags;
4330 ret = __dev_change_flags(dev, flags);
4334 changes = old_flags ^ dev->flags;
4336 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4338 __dev_notify_flags(dev, old_flags);
4341 EXPORT_SYMBOL(dev_change_flags);
4344 * dev_set_mtu - Change maximum transfer unit
4346 * @new_mtu: new transfer unit
4348 * Change the maximum transfer size of the network device.
4350 int dev_set_mtu(struct net_device *dev, int new_mtu)
4352 const struct net_device_ops *ops = dev->netdev_ops;
4355 if (new_mtu == dev->mtu)
4358 /* MTU must be positive. */
4362 if (!netif_device_present(dev))
4366 if (ops->ndo_change_mtu)
4367 err = ops->ndo_change_mtu(dev, new_mtu);
4371 if (!err && dev->flags & IFF_UP)
4372 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4375 EXPORT_SYMBOL(dev_set_mtu);
4378 * dev_set_mac_address - Change Media Access Control Address
4382 * Change the hardware (MAC) address of the device
4384 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4386 const struct net_device_ops *ops = dev->netdev_ops;
4389 if (!ops->ndo_set_mac_address)
4391 if (sa->sa_family != dev->type)
4393 if (!netif_device_present(dev))
4395 err = ops->ndo_set_mac_address(dev, sa);
4397 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4400 EXPORT_SYMBOL(dev_set_mac_address);
4403 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4405 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4408 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4414 case SIOCGIFFLAGS: /* Get interface flags */
4415 ifr->ifr_flags = (short) dev_get_flags(dev);
4418 case SIOCGIFMETRIC: /* Get the metric on the interface
4419 (currently unused) */
4420 ifr->ifr_metric = 0;
4423 case SIOCGIFMTU: /* Get the MTU of a device */
4424 ifr->ifr_mtu = dev->mtu;
4429 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4431 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4432 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4433 ifr->ifr_hwaddr.sa_family = dev->type;
4441 ifr->ifr_map.mem_start = dev->mem_start;
4442 ifr->ifr_map.mem_end = dev->mem_end;
4443 ifr->ifr_map.base_addr = dev->base_addr;
4444 ifr->ifr_map.irq = dev->irq;
4445 ifr->ifr_map.dma = dev->dma;
4446 ifr->ifr_map.port = dev->if_port;
4450 ifr->ifr_ifindex = dev->ifindex;
4454 ifr->ifr_qlen = dev->tx_queue_len;
4458 /* dev_ioctl() should ensure this case
4470 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4472 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4475 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4476 const struct net_device_ops *ops;
4481 ops = dev->netdev_ops;
4484 case SIOCSIFFLAGS: /* Set interface flags */
4485 return dev_change_flags(dev, ifr->ifr_flags);
4487 case SIOCSIFMETRIC: /* Set the metric on the interface
4488 (currently unused) */
4491 case SIOCSIFMTU: /* Set the MTU of a device */
4492 return dev_set_mtu(dev, ifr->ifr_mtu);
4495 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4497 case SIOCSIFHWBROADCAST:
4498 if (ifr->ifr_hwaddr.sa_family != dev->type)
4500 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4501 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4502 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4506 if (ops->ndo_set_config) {
4507 if (!netif_device_present(dev))
4509 return ops->ndo_set_config(dev, &ifr->ifr_map);
4514 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4515 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4517 if (!netif_device_present(dev))
4519 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4522 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4523 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4525 if (!netif_device_present(dev))
4527 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4530 if (ifr->ifr_qlen < 0)
4532 dev->tx_queue_len = ifr->ifr_qlen;
4536 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4537 return dev_change_name(dev, ifr->ifr_newname);
4540 * Unknown or private ioctl
4543 if ((cmd >= SIOCDEVPRIVATE &&
4544 cmd <= SIOCDEVPRIVATE + 15) ||
4545 cmd == SIOCBONDENSLAVE ||
4546 cmd == SIOCBONDRELEASE ||
4547 cmd == SIOCBONDSETHWADDR ||
4548 cmd == SIOCBONDSLAVEINFOQUERY ||
4549 cmd == SIOCBONDINFOQUERY ||
4550 cmd == SIOCBONDCHANGEACTIVE ||
4551 cmd == SIOCGMIIPHY ||
4552 cmd == SIOCGMIIREG ||
4553 cmd == SIOCSMIIREG ||
4554 cmd == SIOCBRADDIF ||
4555 cmd == SIOCBRDELIF ||
4556 cmd == SIOCSHWTSTAMP ||
4557 cmd == SIOCWANDEV) {
4559 if (ops->ndo_do_ioctl) {
4560 if (netif_device_present(dev))
4561 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4573 * This function handles all "interface"-type I/O control requests. The actual
4574 * 'doing' part of this is dev_ifsioc above.
4578 * dev_ioctl - network device ioctl
4579 * @net: the applicable net namespace
4580 * @cmd: command to issue
4581 * @arg: pointer to a struct ifreq in user space
4583 * Issue ioctl functions to devices. This is normally called by the
4584 * user space syscall interfaces but can sometimes be useful for
4585 * other purposes. The return value is the return from the syscall if
4586 * positive or a negative errno code on error.
4589 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4595 /* One special case: SIOCGIFCONF takes ifconf argument
4596 and requires shared lock, because it sleeps writing
4600 if (cmd == SIOCGIFCONF) {
4602 ret = dev_ifconf(net, (char __user *) arg);
4606 if (cmd == SIOCGIFNAME)
4607 return dev_ifname(net, (struct ifreq __user *)arg);
4609 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4612 ifr.ifr_name[IFNAMSIZ-1] = 0;
4614 colon = strchr(ifr.ifr_name, ':');
4619 * See which interface the caller is talking about.
4624 * These ioctl calls:
4625 * - can be done by all.
4626 * - atomic and do not require locking.
4637 dev_load(net, ifr.ifr_name);
4639 ret = dev_ifsioc_locked(net, &ifr, cmd);
4644 if (copy_to_user(arg, &ifr,
4645 sizeof(struct ifreq)))
4651 dev_load(net, ifr.ifr_name);
4653 ret = dev_ethtool(net, &ifr);
4658 if (copy_to_user(arg, &ifr,
4659 sizeof(struct ifreq)))
4665 * These ioctl calls:
4666 * - require superuser power.
4667 * - require strict serialization.
4673 if (!capable(CAP_NET_ADMIN))
4675 dev_load(net, ifr.ifr_name);
4677 ret = dev_ifsioc(net, &ifr, cmd);
4682 if (copy_to_user(arg, &ifr,
4683 sizeof(struct ifreq)))
4689 * These ioctl calls:
4690 * - require superuser power.
4691 * - require strict serialization.
4692 * - do not return a value
4702 case SIOCSIFHWBROADCAST:
4705 case SIOCBONDENSLAVE:
4706 case SIOCBONDRELEASE:
4707 case SIOCBONDSETHWADDR:
4708 case SIOCBONDCHANGEACTIVE:
4712 if (!capable(CAP_NET_ADMIN))
4715 case SIOCBONDSLAVEINFOQUERY:
4716 case SIOCBONDINFOQUERY:
4717 dev_load(net, ifr.ifr_name);
4719 ret = dev_ifsioc(net, &ifr, cmd);
4724 /* Get the per device memory space. We can add this but
4725 * currently do not support it */
4727 /* Set the per device memory buffer space.
4728 * Not applicable in our case */
4733 * Unknown or private ioctl.
4736 if (cmd == SIOCWANDEV ||
4737 (cmd >= SIOCDEVPRIVATE &&
4738 cmd <= SIOCDEVPRIVATE + 15)) {
4739 dev_load(net, ifr.ifr_name);
4741 ret = dev_ifsioc(net, &ifr, cmd);
4743 if (!ret && copy_to_user(arg, &ifr,
4744 sizeof(struct ifreq)))
4748 /* Take care of Wireless Extensions */
4749 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4750 return wext_handle_ioctl(net, &ifr, cmd, arg);
4757 * dev_new_index - allocate an ifindex
4758 * @net: the applicable net namespace
4760 * Returns a suitable unique value for a new device interface
4761 * number. The caller must hold the rtnl semaphore or the
4762 * dev_base_lock to be sure it remains unique.
4764 static int dev_new_index(struct net *net)
4770 if (!__dev_get_by_index(net, ifindex))
4775 /* Delayed registration/unregisteration */
4776 static LIST_HEAD(net_todo_list);
4778 static void net_set_todo(struct net_device *dev)
4780 list_add_tail(&dev->todo_list, &net_todo_list);
4783 static void rollback_registered_many(struct list_head *head)
4785 struct net_device *dev, *tmp;
4787 BUG_ON(dev_boot_phase);
4790 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4791 /* Some devices call without registering
4792 * for initialization unwind. Remove those
4793 * devices and proceed with the remaining.
4795 if (dev->reg_state == NETREG_UNINITIALIZED) {
4796 pr_debug("unregister_netdevice: device %s/%p never "
4797 "was registered\n", dev->name, dev);
4800 list_del(&dev->unreg_list);
4804 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4806 /* If device is running, close it first. */
4809 /* And unlink it from device chain. */
4810 unlist_netdevice(dev);
4812 dev->reg_state = NETREG_UNREGISTERING;
4817 list_for_each_entry(dev, head, unreg_list) {
4818 /* Shutdown queueing discipline. */
4822 /* Notify protocols, that we are about to destroy
4823 this device. They should clean all the things.
4825 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4827 if (!dev->rtnl_link_ops ||
4828 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4829 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4832 * Flush the unicast and multicast chains
4837 if (dev->netdev_ops->ndo_uninit)
4838 dev->netdev_ops->ndo_uninit(dev);
4840 /* Notifier chain MUST detach us from master device. */
4841 WARN_ON(dev->master);
4843 /* Remove entries from kobject tree */
4844 netdev_unregister_kobject(dev);
4847 /* Process any work delayed until the end of the batch */
4848 dev = list_first_entry(head, struct net_device, unreg_list);
4849 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4853 list_for_each_entry(dev, head, unreg_list)
4857 static void rollback_registered(struct net_device *dev)
4861 list_add(&dev->unreg_list, &single);
4862 rollback_registered_many(&single);
4865 static void __netdev_init_queue_locks_one(struct net_device *dev,
4866 struct netdev_queue *dev_queue,
4869 spin_lock_init(&dev_queue->_xmit_lock);
4870 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4871 dev_queue->xmit_lock_owner = -1;
4874 static void netdev_init_queue_locks(struct net_device *dev)
4876 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4877 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4880 unsigned long netdev_fix_features(unsigned long features, const char *name)
4882 /* Fix illegal SG+CSUM combinations. */
4883 if ((features & NETIF_F_SG) &&
4884 !(features & NETIF_F_ALL_CSUM)) {
4886 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4887 "checksum feature.\n", name);
4888 features &= ~NETIF_F_SG;
4891 /* TSO requires that SG is present as well. */
4892 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4894 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4895 "SG feature.\n", name);
4896 features &= ~NETIF_F_TSO;
4899 if (features & NETIF_F_UFO) {
4900 if (!(features & NETIF_F_GEN_CSUM)) {
4902 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4903 "since no NETIF_F_HW_CSUM feature.\n",
4905 features &= ~NETIF_F_UFO;
4908 if (!(features & NETIF_F_SG)) {
4910 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4911 "since no NETIF_F_SG feature.\n", name);
4912 features &= ~NETIF_F_UFO;
4918 EXPORT_SYMBOL(netdev_fix_features);
4921 * netif_stacked_transfer_operstate - transfer operstate
4922 * @rootdev: the root or lower level device to transfer state from
4923 * @dev: the device to transfer operstate to
4925 * Transfer operational state from root to device. This is normally
4926 * called when a stacking relationship exists between the root
4927 * device and the device(a leaf device).
4929 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4930 struct net_device *dev)
4932 if (rootdev->operstate == IF_OPER_DORMANT)
4933 netif_dormant_on(dev);
4935 netif_dormant_off(dev);
4937 if (netif_carrier_ok(rootdev)) {
4938 if (!netif_carrier_ok(dev))
4939 netif_carrier_on(dev);
4941 if (netif_carrier_ok(dev))
4942 netif_carrier_off(dev);
4945 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4948 * register_netdevice - register a network device
4949 * @dev: device to register
4951 * Take a completed network device structure and add it to the kernel
4952 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4953 * chain. 0 is returned on success. A negative errno code is returned
4954 * on a failure to set up the device, or if the name is a duplicate.
4956 * Callers must hold the rtnl semaphore. You may want
4957 * register_netdev() instead of this.
4960 * The locking appears insufficient to guarantee two parallel registers
4961 * will not get the same name.
4964 int register_netdevice(struct net_device *dev)
4967 struct net *net = dev_net(dev);
4969 BUG_ON(dev_boot_phase);
4974 /* When net_device's are persistent, this will be fatal. */
4975 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4978 spin_lock_init(&dev->addr_list_lock);
4979 netdev_set_addr_lockdep_class(dev);
4980 netdev_init_queue_locks(dev);
4985 if (!dev->num_rx_queues) {
4987 * Allocate a single RX queue if driver never called
4991 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
4997 dev->_rx->first = dev->_rx;
4998 atomic_set(&dev->_rx->count, 1);
4999 dev->num_rx_queues = 1;
5002 /* Init, if this function is available */
5003 if (dev->netdev_ops->ndo_init) {
5004 ret = dev->netdev_ops->ndo_init(dev);
5012 ret = dev_get_valid_name(dev, dev->name, 0);
5016 dev->ifindex = dev_new_index(net);
5017 if (dev->iflink == -1)
5018 dev->iflink = dev->ifindex;
5020 /* Fix illegal checksum combinations */
5021 if ((dev->features & NETIF_F_HW_CSUM) &&
5022 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5023 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5025 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5028 if ((dev->features & NETIF_F_NO_CSUM) &&
5029 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5030 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5032 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5035 dev->features = netdev_fix_features(dev->features, dev->name);
5037 /* Enable software GSO if SG is supported. */
5038 if (dev->features & NETIF_F_SG)
5039 dev->features |= NETIF_F_GSO;
5041 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5042 ret = notifier_to_errno(ret);
5046 ret = netdev_register_kobject(dev);
5049 dev->reg_state = NETREG_REGISTERED;
5052 * Default initial state at registry is that the
5053 * device is present.
5056 set_bit(__LINK_STATE_PRESENT, &dev->state);
5058 dev_init_scheduler(dev);
5060 list_netdevice(dev);
5062 /* Notify protocols, that a new device appeared. */
5063 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5064 ret = notifier_to_errno(ret);
5066 rollback_registered(dev);
5067 dev->reg_state = NETREG_UNREGISTERED;
5070 * Prevent userspace races by waiting until the network
5071 * device is fully setup before sending notifications.
5073 if (!dev->rtnl_link_ops ||
5074 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5075 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5081 if (dev->netdev_ops->ndo_uninit)
5082 dev->netdev_ops->ndo_uninit(dev);
5085 EXPORT_SYMBOL(register_netdevice);
5088 * init_dummy_netdev - init a dummy network device for NAPI
5089 * @dev: device to init
5091 * This takes a network device structure and initialize the minimum
5092 * amount of fields so it can be used to schedule NAPI polls without
5093 * registering a full blown interface. This is to be used by drivers
5094 * that need to tie several hardware interfaces to a single NAPI
5095 * poll scheduler due to HW limitations.
5097 int init_dummy_netdev(struct net_device *dev)
5099 /* Clear everything. Note we don't initialize spinlocks
5100 * are they aren't supposed to be taken by any of the
5101 * NAPI code and this dummy netdev is supposed to be
5102 * only ever used for NAPI polls
5104 memset(dev, 0, sizeof(struct net_device));
5106 /* make sure we BUG if trying to hit standard
5107 * register/unregister code path
5109 dev->reg_state = NETREG_DUMMY;
5111 /* initialize the ref count */
5112 atomic_set(&dev->refcnt, 1);
5114 /* NAPI wants this */
5115 INIT_LIST_HEAD(&dev->napi_list);
5117 /* a dummy interface is started by default */
5118 set_bit(__LINK_STATE_PRESENT, &dev->state);
5119 set_bit(__LINK_STATE_START, &dev->state);
5123 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5127 * register_netdev - register a network device
5128 * @dev: device to register
5130 * Take a completed network device structure and add it to the kernel
5131 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5132 * chain. 0 is returned on success. A negative errno code is returned
5133 * on a failure to set up the device, or if the name is a duplicate.
5135 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5136 * and expands the device name if you passed a format string to
5139 int register_netdev(struct net_device *dev)
5146 * If the name is a format string the caller wants us to do a
5149 if (strchr(dev->name, '%')) {
5150 err = dev_alloc_name(dev, dev->name);
5155 err = register_netdevice(dev);
5160 EXPORT_SYMBOL(register_netdev);
5163 * netdev_wait_allrefs - wait until all references are gone.
5165 * This is called when unregistering network devices.
5167 * Any protocol or device that holds a reference should register
5168 * for netdevice notification, and cleanup and put back the
5169 * reference if they receive an UNREGISTER event.
5170 * We can get stuck here if buggy protocols don't correctly
5173 static void netdev_wait_allrefs(struct net_device *dev)
5175 unsigned long rebroadcast_time, warning_time;
5177 linkwatch_forget_dev(dev);
5179 rebroadcast_time = warning_time = jiffies;
5180 while (atomic_read(&dev->refcnt) != 0) {
5181 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5184 /* Rebroadcast unregister notification */
5185 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5186 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5187 * should have already handle it the first time */
5189 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5191 /* We must not have linkwatch events
5192 * pending on unregister. If this
5193 * happens, we simply run the queue
5194 * unscheduled, resulting in a noop
5197 linkwatch_run_queue();
5202 rebroadcast_time = jiffies;
5207 if (time_after(jiffies, warning_time + 10 * HZ)) {
5208 printk(KERN_EMERG "unregister_netdevice: "
5209 "waiting for %s to become free. Usage "
5211 dev->name, atomic_read(&dev->refcnt));
5212 warning_time = jiffies;
5221 * register_netdevice(x1);
5222 * register_netdevice(x2);
5224 * unregister_netdevice(y1);
5225 * unregister_netdevice(y2);
5231 * We are invoked by rtnl_unlock().
5232 * This allows us to deal with problems:
5233 * 1) We can delete sysfs objects which invoke hotplug
5234 * without deadlocking with linkwatch via keventd.
5235 * 2) Since we run with the RTNL semaphore not held, we can sleep
5236 * safely in order to wait for the netdev refcnt to drop to zero.
5238 * We must not return until all unregister events added during
5239 * the interval the lock was held have been completed.
5241 void netdev_run_todo(void)
5243 struct list_head list;
5245 /* Snapshot list, allow later requests */
5246 list_replace_init(&net_todo_list, &list);
5250 while (!list_empty(&list)) {
5251 struct net_device *dev
5252 = list_first_entry(&list, struct net_device, todo_list);
5253 list_del(&dev->todo_list);
5255 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5256 printk(KERN_ERR "network todo '%s' but state %d\n",
5257 dev->name, dev->reg_state);
5262 dev->reg_state = NETREG_UNREGISTERED;
5264 on_each_cpu(flush_backlog, dev, 1);
5266 netdev_wait_allrefs(dev);
5269 BUG_ON(atomic_read(&dev->refcnt));
5270 WARN_ON(dev->ip_ptr);
5271 WARN_ON(dev->ip6_ptr);
5272 WARN_ON(dev->dn_ptr);
5274 if (dev->destructor)
5275 dev->destructor(dev);
5277 /* Free network device */
5278 kobject_put(&dev->dev.kobj);
5283 * dev_txq_stats_fold - fold tx_queues stats
5284 * @dev: device to get statistics from
5285 * @stats: struct rtnl_link_stats64 to hold results
5287 void dev_txq_stats_fold(const struct net_device *dev,
5288 struct rtnl_link_stats64 *stats)
5290 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5292 struct netdev_queue *txq;
5294 for (i = 0; i < dev->num_tx_queues; i++) {
5295 txq = netdev_get_tx_queue(dev, i);
5296 spin_lock_bh(&txq->_xmit_lock);
5297 tx_bytes += txq->tx_bytes;
5298 tx_packets += txq->tx_packets;
5299 tx_dropped += txq->tx_dropped;
5300 spin_unlock_bh(&txq->_xmit_lock);
5302 if (tx_bytes || tx_packets || tx_dropped) {
5303 stats->tx_bytes = tx_bytes;
5304 stats->tx_packets = tx_packets;
5305 stats->tx_dropped = tx_dropped;
5308 EXPORT_SYMBOL(dev_txq_stats_fold);
5310 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5311 * fields in the same order, with only the type differing.
5313 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5314 const struct net_device_stats *netdev_stats)
5316 #if BITS_PER_LONG == 64
5317 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5318 memcpy(stats64, netdev_stats, sizeof(*stats64));
5320 size_t i, n = sizeof(*stats64) / sizeof(u64);
5321 const unsigned long *src = (const unsigned long *)netdev_stats;
5322 u64 *dst = (u64 *)stats64;
5324 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5325 sizeof(*stats64) / sizeof(u64));
5326 for (i = 0; i < n; i++)
5332 * dev_get_stats - get network device statistics
5333 * @dev: device to get statistics from
5334 * @storage: place to store stats
5336 * Get network statistics from device. Return @storage.
5337 * The device driver may provide its own method by setting
5338 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5339 * otherwise the internal statistics structure is used.
5341 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5342 struct rtnl_link_stats64 *storage)
5344 const struct net_device_ops *ops = dev->netdev_ops;
5346 if (ops->ndo_get_stats64) {
5347 memset(storage, 0, sizeof(*storage));
5348 return ops->ndo_get_stats64(dev, storage);
5350 if (ops->ndo_get_stats) {
5351 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5354 netdev_stats_to_stats64(storage, &dev->stats);
5355 dev_txq_stats_fold(dev, storage);
5358 EXPORT_SYMBOL(dev_get_stats);
5360 static void netdev_init_one_queue(struct net_device *dev,
5361 struct netdev_queue *queue,
5367 static void netdev_init_queues(struct net_device *dev)
5369 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5370 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5371 spin_lock_init(&dev->tx_global_lock);
5375 * alloc_netdev_mq - allocate network device
5376 * @sizeof_priv: size of private data to allocate space for
5377 * @name: device name format string
5378 * @setup: callback to initialize device
5379 * @queue_count: the number of subqueues to allocate
5381 * Allocates a struct net_device with private data area for driver use
5382 * and performs basic initialization. Also allocates subquue structs
5383 * for each queue on the device at the end of the netdevice.
5385 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5386 void (*setup)(struct net_device *), unsigned int queue_count)
5388 struct netdev_queue *tx;
5389 struct net_device *dev;
5391 struct net_device *p;
5393 struct netdev_rx_queue *rx;
5397 BUG_ON(strlen(name) >= sizeof(dev->name));
5399 alloc_size = sizeof(struct net_device);
5401 /* ensure 32-byte alignment of private area */
5402 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5403 alloc_size += sizeof_priv;
5405 /* ensure 32-byte alignment of whole construct */
5406 alloc_size += NETDEV_ALIGN - 1;
5408 p = kzalloc(alloc_size, GFP_KERNEL);
5410 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5414 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5416 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5422 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5424 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5429 atomic_set(&rx->count, queue_count);
5432 * Set a pointer to first element in the array which holds the
5435 for (i = 0; i < queue_count; i++)
5439 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5440 dev->padded = (char *)dev - (char *)p;
5442 if (dev_addr_init(dev))
5448 dev_net_set(dev, &init_net);
5451 dev->num_tx_queues = queue_count;
5452 dev->real_num_tx_queues = queue_count;
5456 dev->num_rx_queues = queue_count;
5459 dev->gso_max_size = GSO_MAX_SIZE;
5461 netdev_init_queues(dev);
5463 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5464 dev->ethtool_ntuple_list.count = 0;
5465 INIT_LIST_HEAD(&dev->napi_list);
5466 INIT_LIST_HEAD(&dev->unreg_list);
5467 INIT_LIST_HEAD(&dev->link_watch_list);
5468 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5470 strcpy(dev->name, name);
5483 EXPORT_SYMBOL(alloc_netdev_mq);
5486 * free_netdev - free network device
5489 * This function does the last stage of destroying an allocated device
5490 * interface. The reference to the device object is released.
5491 * If this is the last reference then it will be freed.
5493 void free_netdev(struct net_device *dev)
5495 struct napi_struct *p, *n;
5497 release_net(dev_net(dev));
5501 /* Flush device addresses */
5502 dev_addr_flush(dev);
5504 /* Clear ethtool n-tuple list */
5505 ethtool_ntuple_flush(dev);
5507 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5510 /* Compatibility with error handling in drivers */
5511 if (dev->reg_state == NETREG_UNINITIALIZED) {
5512 kfree((char *)dev - dev->padded);
5516 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5517 dev->reg_state = NETREG_RELEASED;
5519 /* will free via device release */
5520 put_device(&dev->dev);
5522 EXPORT_SYMBOL(free_netdev);
5525 * synchronize_net - Synchronize with packet receive processing
5527 * Wait for packets currently being received to be done.
5528 * Does not block later packets from starting.
5530 void synchronize_net(void)
5535 EXPORT_SYMBOL(synchronize_net);
5538 * unregister_netdevice_queue - remove device from the kernel
5542 * This function shuts down a device interface and removes it
5543 * from the kernel tables.
5544 * If head not NULL, device is queued to be unregistered later.
5546 * Callers must hold the rtnl semaphore. You may want
5547 * unregister_netdev() instead of this.
5550 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5555 list_move_tail(&dev->unreg_list, head);
5557 rollback_registered(dev);
5558 /* Finish processing unregister after unlock */
5562 EXPORT_SYMBOL(unregister_netdevice_queue);
5565 * unregister_netdevice_many - unregister many devices
5566 * @head: list of devices
5568 void unregister_netdevice_many(struct list_head *head)
5570 struct net_device *dev;
5572 if (!list_empty(head)) {
5573 rollback_registered_many(head);
5574 list_for_each_entry(dev, head, unreg_list)
5578 EXPORT_SYMBOL(unregister_netdevice_many);
5581 * unregister_netdev - remove device from the kernel
5584 * This function shuts down a device interface and removes it
5585 * from the kernel tables.
5587 * This is just a wrapper for unregister_netdevice that takes
5588 * the rtnl semaphore. In general you want to use this and not
5589 * unregister_netdevice.
5591 void unregister_netdev(struct net_device *dev)
5594 unregister_netdevice(dev);
5597 EXPORT_SYMBOL(unregister_netdev);
5600 * dev_change_net_namespace - move device to different nethost namespace
5602 * @net: network namespace
5603 * @pat: If not NULL name pattern to try if the current device name
5604 * is already taken in the destination network namespace.
5606 * This function shuts down a device interface and moves it
5607 * to a new network namespace. On success 0 is returned, on
5608 * a failure a netagive errno code is returned.
5610 * Callers must hold the rtnl semaphore.
5613 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5619 /* Don't allow namespace local devices to be moved. */
5621 if (dev->features & NETIF_F_NETNS_LOCAL)
5624 /* Ensure the device has been registrered */
5626 if (dev->reg_state != NETREG_REGISTERED)
5629 /* Get out if there is nothing todo */
5631 if (net_eq(dev_net(dev), net))
5634 /* Pick the destination device name, and ensure
5635 * we can use it in the destination network namespace.
5638 if (__dev_get_by_name(net, dev->name)) {
5639 /* We get here if we can't use the current device name */
5642 if (dev_get_valid_name(dev, pat, 1))
5647 * And now a mini version of register_netdevice unregister_netdevice.
5650 /* If device is running close it first. */
5653 /* And unlink it from device chain */
5655 unlist_netdevice(dev);
5659 /* Shutdown queueing discipline. */
5662 /* Notify protocols, that we are about to destroy
5663 this device. They should clean all the things.
5665 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5666 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5669 * Flush the unicast and multicast chains
5674 /* Actually switch the network namespace */
5675 dev_net_set(dev, net);
5677 /* If there is an ifindex conflict assign a new one */
5678 if (__dev_get_by_index(net, dev->ifindex)) {
5679 int iflink = (dev->iflink == dev->ifindex);
5680 dev->ifindex = dev_new_index(net);
5682 dev->iflink = dev->ifindex;
5685 /* Fixup kobjects */
5686 err = device_rename(&dev->dev, dev->name);
5689 /* Add the device back in the hashes */
5690 list_netdevice(dev);
5692 /* Notify protocols, that a new device appeared. */
5693 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5696 * Prevent userspace races by waiting until the network
5697 * device is fully setup before sending notifications.
5699 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5706 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5708 static int dev_cpu_callback(struct notifier_block *nfb,
5709 unsigned long action,
5712 struct sk_buff **list_skb;
5713 struct sk_buff *skb;
5714 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5715 struct softnet_data *sd, *oldsd;
5717 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5720 local_irq_disable();
5721 cpu = smp_processor_id();
5722 sd = &per_cpu(softnet_data, cpu);
5723 oldsd = &per_cpu(softnet_data, oldcpu);
5725 /* Find end of our completion_queue. */
5726 list_skb = &sd->completion_queue;
5728 list_skb = &(*list_skb)->next;
5729 /* Append completion queue from offline CPU. */
5730 *list_skb = oldsd->completion_queue;
5731 oldsd->completion_queue = NULL;
5733 /* Append output queue from offline CPU. */
5734 if (oldsd->output_queue) {
5735 *sd->output_queue_tailp = oldsd->output_queue;
5736 sd->output_queue_tailp = oldsd->output_queue_tailp;
5737 oldsd->output_queue = NULL;
5738 oldsd->output_queue_tailp = &oldsd->output_queue;
5741 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5744 /* Process offline CPU's input_pkt_queue */
5745 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5747 input_queue_head_incr(oldsd);
5749 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5751 input_queue_head_incr(oldsd);
5759 * netdev_increment_features - increment feature set by one
5760 * @all: current feature set
5761 * @one: new feature set
5762 * @mask: mask feature set
5764 * Computes a new feature set after adding a device with feature set
5765 * @one to the master device with current feature set @all. Will not
5766 * enable anything that is off in @mask. Returns the new feature set.
5768 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5771 /* If device needs checksumming, downgrade to it. */
5772 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5773 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5774 else if (mask & NETIF_F_ALL_CSUM) {
5775 /* If one device supports v4/v6 checksumming, set for all. */
5776 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5777 !(all & NETIF_F_GEN_CSUM)) {
5778 all &= ~NETIF_F_ALL_CSUM;
5779 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5782 /* If one device supports hw checksumming, set for all. */
5783 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5784 all &= ~NETIF_F_ALL_CSUM;
5785 all |= NETIF_F_HW_CSUM;
5789 one |= NETIF_F_ALL_CSUM;
5791 one |= all & NETIF_F_ONE_FOR_ALL;
5792 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5793 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5797 EXPORT_SYMBOL(netdev_increment_features);
5799 static struct hlist_head *netdev_create_hash(void)
5802 struct hlist_head *hash;
5804 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5806 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5807 INIT_HLIST_HEAD(&hash[i]);
5812 /* Initialize per network namespace state */
5813 static int __net_init netdev_init(struct net *net)
5815 INIT_LIST_HEAD(&net->dev_base_head);
5817 net->dev_name_head = netdev_create_hash();
5818 if (net->dev_name_head == NULL)
5821 net->dev_index_head = netdev_create_hash();
5822 if (net->dev_index_head == NULL)
5828 kfree(net->dev_name_head);
5834 * netdev_drivername - network driver for the device
5835 * @dev: network device
5836 * @buffer: buffer for resulting name
5837 * @len: size of buffer
5839 * Determine network driver for device.
5841 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5843 const struct device_driver *driver;
5844 const struct device *parent;
5846 if (len <= 0 || !buffer)
5850 parent = dev->dev.parent;
5855 driver = parent->driver;
5856 if (driver && driver->name)
5857 strlcpy(buffer, driver->name, len);
5861 static int __netdev_printk(const char *level, const struct net_device *dev,
5862 struct va_format *vaf)
5866 if (dev && dev->dev.parent)
5867 r = dev_printk(level, dev->dev.parent, "%s: %pV",
5868 netdev_name(dev), vaf);
5870 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
5872 r = printk("%s(NULL net_device): %pV", level, vaf);
5877 int netdev_printk(const char *level, const struct net_device *dev,
5878 const char *format, ...)
5880 struct va_format vaf;
5884 va_start(args, format);
5889 r = __netdev_printk(level, dev, &vaf);
5894 EXPORT_SYMBOL(netdev_printk);
5896 #define define_netdev_printk_level(func, level) \
5897 int func(const struct net_device *dev, const char *fmt, ...) \
5900 struct va_format vaf; \
5903 va_start(args, fmt); \
5908 r = __netdev_printk(level, dev, &vaf); \
5913 EXPORT_SYMBOL(func);
5915 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
5916 define_netdev_printk_level(netdev_alert, KERN_ALERT);
5917 define_netdev_printk_level(netdev_crit, KERN_CRIT);
5918 define_netdev_printk_level(netdev_err, KERN_ERR);
5919 define_netdev_printk_level(netdev_warn, KERN_WARNING);
5920 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
5921 define_netdev_printk_level(netdev_info, KERN_INFO);
5923 static void __net_exit netdev_exit(struct net *net)
5925 kfree(net->dev_name_head);
5926 kfree(net->dev_index_head);
5929 static struct pernet_operations __net_initdata netdev_net_ops = {
5930 .init = netdev_init,
5931 .exit = netdev_exit,
5934 static void __net_exit default_device_exit(struct net *net)
5936 struct net_device *dev, *aux;
5938 * Push all migratable network devices back to the
5939 * initial network namespace
5942 for_each_netdev_safe(net, dev, aux) {
5944 char fb_name[IFNAMSIZ];
5946 /* Ignore unmoveable devices (i.e. loopback) */
5947 if (dev->features & NETIF_F_NETNS_LOCAL)
5950 /* Leave virtual devices for the generic cleanup */
5951 if (dev->rtnl_link_ops)
5954 /* Push remaing network devices to init_net */
5955 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5956 err = dev_change_net_namespace(dev, &init_net, fb_name);
5958 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5959 __func__, dev->name, err);
5966 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5968 /* At exit all network devices most be removed from a network
5969 * namespace. Do this in the reverse order of registeration.
5970 * Do this across as many network namespaces as possible to
5971 * improve batching efficiency.
5973 struct net_device *dev;
5975 LIST_HEAD(dev_kill_list);
5978 list_for_each_entry(net, net_list, exit_list) {
5979 for_each_netdev_reverse(net, dev) {
5980 if (dev->rtnl_link_ops)
5981 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5983 unregister_netdevice_queue(dev, &dev_kill_list);
5986 unregister_netdevice_many(&dev_kill_list);
5990 static struct pernet_operations __net_initdata default_device_ops = {
5991 .exit = default_device_exit,
5992 .exit_batch = default_device_exit_batch,
5996 * Initialize the DEV module. At boot time this walks the device list and
5997 * unhooks any devices that fail to initialise (normally hardware not
5998 * present) and leaves us with a valid list of present and active devices.
6003 * This is called single threaded during boot, so no need
6004 * to take the rtnl semaphore.
6006 static int __init net_dev_init(void)
6008 int i, rc = -ENOMEM;
6010 BUG_ON(!dev_boot_phase);
6012 if (dev_proc_init())
6015 if (netdev_kobject_init())
6018 INIT_LIST_HEAD(&ptype_all);
6019 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6020 INIT_LIST_HEAD(&ptype_base[i]);
6022 if (register_pernet_subsys(&netdev_net_ops))
6026 * Initialise the packet receive queues.
6029 for_each_possible_cpu(i) {
6030 struct softnet_data *sd = &per_cpu(softnet_data, i);
6032 memset(sd, 0, sizeof(*sd));
6033 skb_queue_head_init(&sd->input_pkt_queue);
6034 skb_queue_head_init(&sd->process_queue);
6035 sd->completion_queue = NULL;
6036 INIT_LIST_HEAD(&sd->poll_list);
6037 sd->output_queue = NULL;
6038 sd->output_queue_tailp = &sd->output_queue;
6040 sd->csd.func = rps_trigger_softirq;
6046 sd->backlog.poll = process_backlog;
6047 sd->backlog.weight = weight_p;
6048 sd->backlog.gro_list = NULL;
6049 sd->backlog.gro_count = 0;
6054 /* The loopback device is special if any other network devices
6055 * is present in a network namespace the loopback device must
6056 * be present. Since we now dynamically allocate and free the
6057 * loopback device ensure this invariant is maintained by
6058 * keeping the loopback device as the first device on the
6059 * list of network devices. Ensuring the loopback devices
6060 * is the first device that appears and the last network device
6063 if (register_pernet_device(&loopback_net_ops))
6066 if (register_pernet_device(&default_device_ops))
6069 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6070 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6072 hotcpu_notifier(dev_cpu_callback, 0);
6080 subsys_initcall(net_dev_init);
6082 static int __init initialize_hashrnd(void)
6084 get_random_bytes(&hashrnd, sizeof(hashrnd));
6088 late_initcall_sync(initialize_hashrnd);