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/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
103 #include <linux/if_bridge.h>
104 #include <linux/if_macvlan.h>
106 #include <net/pkt_sched.h>
107 #include <net/checksum.h>
108 #include <net/xfrm.h>
109 #include <linux/highmem.h>
110 #include <linux/init.h>
111 #include <linux/kmod.h>
112 #include <linux/module.h>
113 #include <linux/netpoll.h>
114 #include <linux/rcupdate.h>
115 #include <linux/delay.h>
116 #include <net/wext.h>
117 #include <net/iw_handler.h>
118 #include <asm/current.h>
119 #include <linux/audit.h>
120 #include <linux/dmaengine.h>
121 #include <linux/err.h>
122 #include <linux/ctype.h>
123 #include <linux/if_arp.h>
124 #include <linux/if_vlan.h>
125 #include <linux/ip.h>
127 #include <linux/ipv6.h>
128 #include <linux/in.h>
129 #include <linux/jhash.h>
130 #include <linux/random.h>
131 #include <trace/events/napi.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 /* Device list insertion */
210 static int list_netdevice(struct net_device *dev)
212 struct net *net = dev_net(dev);
216 write_lock_bh(&dev_base_lock);
217 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
218 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
219 hlist_add_head_rcu(&dev->index_hlist,
220 dev_index_hash(net, dev->ifindex));
221 write_unlock_bh(&dev_base_lock);
225 /* Device list removal
226 * caller must respect a RCU grace period before freeing/reusing dev
228 static void unlist_netdevice(struct net_device *dev)
232 /* Unlink dev from the device chain */
233 write_lock_bh(&dev_base_lock);
234 list_del_rcu(&dev->dev_list);
235 hlist_del_rcu(&dev->name_hlist);
236 hlist_del_rcu(&dev->index_hlist);
237 write_unlock_bh(&dev_base_lock);
244 static RAW_NOTIFIER_HEAD(netdev_chain);
247 * Device drivers call our routines to queue packets here. We empty the
248 * queue in the local softnet handler.
251 DEFINE_PER_CPU(struct softnet_data, softnet_data);
252 EXPORT_PER_CPU_SYMBOL(softnet_data);
254 #ifdef CONFIG_LOCKDEP
256 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
257 * according to dev->type
259 static const unsigned short netdev_lock_type[] =
260 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
261 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
262 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
263 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
264 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
265 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
266 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
267 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
268 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
269 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
270 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
271 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
272 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
273 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
274 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
275 ARPHRD_VOID, ARPHRD_NONE};
277 static const char *const netdev_lock_name[] =
278 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
279 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
280 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
281 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
282 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
283 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
284 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
285 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
286 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
287 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
288 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
289 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
290 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
291 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
292 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
293 "_xmit_VOID", "_xmit_NONE"};
295 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
296 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
298 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
302 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
303 if (netdev_lock_type[i] == dev_type)
305 /* the last key is used by default */
306 return ARRAY_SIZE(netdev_lock_type) - 1;
309 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
310 unsigned short dev_type)
314 i = netdev_lock_pos(dev_type);
315 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
316 netdev_lock_name[i]);
319 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
323 i = netdev_lock_pos(dev->type);
324 lockdep_set_class_and_name(&dev->addr_list_lock,
325 &netdev_addr_lock_key[i],
326 netdev_lock_name[i]);
329 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
330 unsigned short dev_type)
333 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
338 /*******************************************************************************
340 Protocol management and registration routines
342 *******************************************************************************/
345 * Add a protocol ID to the list. Now that the input handler is
346 * smarter we can dispense with all the messy stuff that used to be
349 * BEWARE!!! Protocol handlers, mangling input packets,
350 * MUST BE last in hash buckets and checking protocol handlers
351 * MUST start from promiscuous ptype_all chain in net_bh.
352 * It is true now, do not change it.
353 * Explanation follows: if protocol handler, mangling packet, will
354 * be the first on list, it is not able to sense, that packet
355 * is cloned and should be copied-on-write, so that it will
356 * change it and subsequent readers will get broken packet.
361 * dev_add_pack - add packet handler
362 * @pt: packet type declaration
364 * Add a protocol handler to the networking stack. The passed &packet_type
365 * is linked into kernel lists and may not be freed until it has been
366 * removed from the kernel lists.
368 * This call does not sleep therefore it can not
369 * guarantee all CPU's that are in middle of receiving packets
370 * will see the new packet type (until the next received packet).
373 void dev_add_pack(struct packet_type *pt)
377 spin_lock_bh(&ptype_lock);
378 if (pt->type == htons(ETH_P_ALL))
379 list_add_rcu(&pt->list, &ptype_all);
381 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
382 list_add_rcu(&pt->list, &ptype_base[hash]);
384 spin_unlock_bh(&ptype_lock);
386 EXPORT_SYMBOL(dev_add_pack);
389 * __dev_remove_pack - remove packet handler
390 * @pt: packet type declaration
392 * Remove a protocol handler that was previously added to the kernel
393 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
394 * from the kernel lists and can be freed or reused once this function
397 * The packet type might still be in use by receivers
398 * and must not be freed until after all the CPU's have gone
399 * through a quiescent state.
401 void __dev_remove_pack(struct packet_type *pt)
403 struct list_head *head;
404 struct packet_type *pt1;
406 spin_lock_bh(&ptype_lock);
408 if (pt->type == htons(ETH_P_ALL))
411 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
413 list_for_each_entry(pt1, head, list) {
415 list_del_rcu(&pt->list);
420 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
422 spin_unlock_bh(&ptype_lock);
424 EXPORT_SYMBOL(__dev_remove_pack);
427 * dev_remove_pack - remove packet handler
428 * @pt: packet type declaration
430 * Remove a protocol handler that was previously added to the kernel
431 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
432 * from the kernel lists and can be freed or reused once this function
435 * This call sleeps to guarantee that no CPU is looking at the packet
438 void dev_remove_pack(struct packet_type *pt)
440 __dev_remove_pack(pt);
444 EXPORT_SYMBOL(dev_remove_pack);
446 /******************************************************************************
448 Device Boot-time Settings Routines
450 *******************************************************************************/
452 /* Boot time configuration table */
453 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
456 * netdev_boot_setup_add - add new setup entry
457 * @name: name of the device
458 * @map: configured settings for the device
460 * Adds new setup entry to the dev_boot_setup list. The function
461 * returns 0 on error and 1 on success. This is a generic routine to
464 static int netdev_boot_setup_add(char *name, struct ifmap *map)
466 struct netdev_boot_setup *s;
470 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
471 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
472 memset(s[i].name, 0, sizeof(s[i].name));
473 strlcpy(s[i].name, name, IFNAMSIZ);
474 memcpy(&s[i].map, map, sizeof(s[i].map));
479 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
483 * netdev_boot_setup_check - check boot time settings
484 * @dev: the netdevice
486 * Check boot time settings for the device.
487 * The found settings are set for the device to be used
488 * later in the device probing.
489 * Returns 0 if no settings found, 1 if they are.
491 int netdev_boot_setup_check(struct net_device *dev)
493 struct netdev_boot_setup *s = dev_boot_setup;
496 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
497 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
498 !strcmp(dev->name, s[i].name)) {
499 dev->irq = s[i].map.irq;
500 dev->base_addr = s[i].map.base_addr;
501 dev->mem_start = s[i].map.mem_start;
502 dev->mem_end = s[i].map.mem_end;
508 EXPORT_SYMBOL(netdev_boot_setup_check);
512 * netdev_boot_base - get address from boot time settings
513 * @prefix: prefix for network device
514 * @unit: id for network device
516 * Check boot time settings for the base address of device.
517 * The found settings are set for the device to be used
518 * later in the device probing.
519 * Returns 0 if no settings found.
521 unsigned long netdev_boot_base(const char *prefix, int unit)
523 const struct netdev_boot_setup *s = dev_boot_setup;
527 sprintf(name, "%s%d", prefix, unit);
530 * If device already registered then return base of 1
531 * to indicate not to probe for this interface
533 if (__dev_get_by_name(&init_net, name))
536 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
537 if (!strcmp(name, s[i].name))
538 return s[i].map.base_addr;
543 * Saves at boot time configured settings for any netdevice.
545 int __init netdev_boot_setup(char *str)
550 str = get_options(str, ARRAY_SIZE(ints), ints);
555 memset(&map, 0, sizeof(map));
559 map.base_addr = ints[2];
561 map.mem_start = ints[3];
563 map.mem_end = ints[4];
565 /* Add new entry to the list */
566 return netdev_boot_setup_add(str, &map);
569 __setup("netdev=", netdev_boot_setup);
571 /*******************************************************************************
573 Device Interface Subroutines
575 *******************************************************************************/
578 * __dev_get_by_name - find a device by its name
579 * @net: the applicable net namespace
580 * @name: name to find
582 * Find an interface by name. Must be called under RTNL semaphore
583 * or @dev_base_lock. If the name is found a pointer to the device
584 * is returned. If the name is not found then %NULL is returned. The
585 * reference counters are not incremented so the caller must be
586 * careful with locks.
589 struct net_device *__dev_get_by_name(struct net *net, const char *name)
591 struct hlist_node *p;
592 struct net_device *dev;
593 struct hlist_head *head = dev_name_hash(net, name);
595 hlist_for_each_entry(dev, p, head, name_hlist)
596 if (!strncmp(dev->name, name, IFNAMSIZ))
601 EXPORT_SYMBOL(__dev_get_by_name);
604 * dev_get_by_name_rcu - find a device by its name
605 * @net: the applicable net namespace
606 * @name: name to find
608 * Find an interface by name.
609 * If the name is found a pointer to the device is returned.
610 * If the name is not found then %NULL is returned.
611 * The reference counters are not incremented so the caller must be
612 * careful with locks. The caller must hold RCU lock.
615 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
617 struct hlist_node *p;
618 struct net_device *dev;
619 struct hlist_head *head = dev_name_hash(net, name);
621 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
622 if (!strncmp(dev->name, name, IFNAMSIZ))
627 EXPORT_SYMBOL(dev_get_by_name_rcu);
630 * dev_get_by_name - find a device by its name
631 * @net: the applicable net namespace
632 * @name: name to find
634 * Find an interface by name. This can be called from any
635 * context and does its own locking. The returned handle has
636 * the usage count incremented and the caller must use dev_put() to
637 * release it when it is no longer needed. %NULL is returned if no
638 * matching device is found.
641 struct net_device *dev_get_by_name(struct net *net, const char *name)
643 struct net_device *dev;
646 dev = dev_get_by_name_rcu(net, name);
652 EXPORT_SYMBOL(dev_get_by_name);
655 * __dev_get_by_index - find a device by its ifindex
656 * @net: the applicable net namespace
657 * @ifindex: index of device
659 * Search for an interface by index. Returns %NULL if the device
660 * is not found or a pointer to the device. The device has not
661 * had its reference counter increased so the caller must be careful
662 * about locking. The caller must hold either the RTNL semaphore
666 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
668 struct hlist_node *p;
669 struct net_device *dev;
670 struct hlist_head *head = dev_index_hash(net, ifindex);
672 hlist_for_each_entry(dev, p, head, index_hlist)
673 if (dev->ifindex == ifindex)
678 EXPORT_SYMBOL(__dev_get_by_index);
681 * dev_get_by_index_rcu - find a device by its ifindex
682 * @net: the applicable net namespace
683 * @ifindex: index of device
685 * Search for an interface by index. Returns %NULL if the device
686 * is not found or a pointer to the device. The device has not
687 * had its reference counter increased so the caller must be careful
688 * about locking. The caller must hold RCU lock.
691 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
693 struct hlist_node *p;
694 struct net_device *dev;
695 struct hlist_head *head = dev_index_hash(net, ifindex);
697 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
698 if (dev->ifindex == ifindex)
703 EXPORT_SYMBOL(dev_get_by_index_rcu);
707 * dev_get_by_index - find a device by its ifindex
708 * @net: the applicable net namespace
709 * @ifindex: index of device
711 * Search for an interface by index. Returns NULL if the device
712 * is not found or a pointer to the device. The device returned has
713 * had a reference added and the pointer is safe until the user calls
714 * dev_put to indicate they have finished with it.
717 struct net_device *dev_get_by_index(struct net *net, int ifindex)
719 struct net_device *dev;
722 dev = dev_get_by_index_rcu(net, ifindex);
728 EXPORT_SYMBOL(dev_get_by_index);
731 * dev_getbyhwaddr - find a device by its hardware address
732 * @net: the applicable net namespace
733 * @type: media type of device
734 * @ha: hardware address
736 * Search for an interface by MAC address. Returns NULL if the device
737 * is not found or a pointer to the device. The caller must hold the
738 * rtnl semaphore. The returned device has not had its ref count increased
739 * and the caller must therefore be careful about locking
742 * If the API was consistent this would be __dev_get_by_hwaddr
745 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
747 struct net_device *dev;
751 for_each_netdev(net, dev)
752 if (dev->type == type &&
753 !memcmp(dev->dev_addr, ha, dev->addr_len))
758 EXPORT_SYMBOL(dev_getbyhwaddr);
760 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
762 struct net_device *dev;
765 for_each_netdev(net, dev)
766 if (dev->type == type)
771 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
773 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
775 struct net_device *dev, *ret = NULL;
778 for_each_netdev_rcu(net, dev)
779 if (dev->type == type) {
787 EXPORT_SYMBOL(dev_getfirstbyhwtype);
790 * dev_get_by_flags - find any device with given flags
791 * @net: the applicable net namespace
792 * @if_flags: IFF_* values
793 * @mask: bitmask of bits in if_flags to check
795 * Search for any interface with the given flags. Returns NULL if a device
796 * is not found or a pointer to the device. The device returned has
797 * had a reference added and the pointer is safe until the user calls
798 * dev_put to indicate they have finished with it.
801 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
804 struct net_device *dev, *ret;
808 for_each_netdev_rcu(net, dev) {
809 if (((dev->flags ^ if_flags) & mask) == 0) {
818 EXPORT_SYMBOL(dev_get_by_flags);
821 * dev_valid_name - check if name is okay for network device
824 * Network device names need to be valid file names to
825 * to allow sysfs to work. We also disallow any kind of
828 int dev_valid_name(const char *name)
832 if (strlen(name) >= IFNAMSIZ)
834 if (!strcmp(name, ".") || !strcmp(name, ".."))
838 if (*name == '/' || isspace(*name))
844 EXPORT_SYMBOL(dev_valid_name);
847 * __dev_alloc_name - allocate a name for a device
848 * @net: network namespace to allocate the device name in
849 * @name: name format string
850 * @buf: scratch buffer and result name string
852 * Passed a format string - eg "lt%d" it will try and find a suitable
853 * id. It scans list of devices to build up a free map, then chooses
854 * the first empty slot. The caller must hold the dev_base or rtnl lock
855 * while allocating the name and adding the device in order to avoid
857 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
858 * Returns the number of the unit assigned or a negative errno code.
861 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
865 const int max_netdevices = 8*PAGE_SIZE;
866 unsigned long *inuse;
867 struct net_device *d;
869 p = strnchr(name, IFNAMSIZ-1, '%');
872 * Verify the string as this thing may have come from
873 * the user. There must be either one "%d" and no other "%"
876 if (p[1] != 'd' || strchr(p + 2, '%'))
879 /* Use one page as a bit array of possible slots */
880 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
884 for_each_netdev(net, d) {
885 if (!sscanf(d->name, name, &i))
887 if (i < 0 || i >= max_netdevices)
890 /* avoid cases where sscanf is not exact inverse of printf */
891 snprintf(buf, IFNAMSIZ, name, i);
892 if (!strncmp(buf, d->name, IFNAMSIZ))
896 i = find_first_zero_bit(inuse, max_netdevices);
897 free_page((unsigned long) inuse);
901 snprintf(buf, IFNAMSIZ, name, i);
902 if (!__dev_get_by_name(net, buf))
905 /* It is possible to run out of possible slots
906 * when the name is long and there isn't enough space left
907 * for the digits, or if all bits are used.
913 * dev_alloc_name - allocate a name for a device
915 * @name: name format string
917 * Passed a format string - eg "lt%d" it will try and find a suitable
918 * id. It scans list of devices to build up a free map, then chooses
919 * the first empty slot. The caller must hold the dev_base or rtnl lock
920 * while allocating the name and adding the device in order to avoid
922 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
923 * Returns the number of the unit assigned or a negative errno code.
926 int dev_alloc_name(struct net_device *dev, const char *name)
932 BUG_ON(!dev_net(dev));
934 ret = __dev_alloc_name(net, name, buf);
936 strlcpy(dev->name, buf, IFNAMSIZ);
939 EXPORT_SYMBOL(dev_alloc_name);
941 static int dev_get_valid_name(struct net *net, const char *name, char *buf,
944 if (!dev_valid_name(name))
947 if (fmt && strchr(name, '%'))
948 return __dev_alloc_name(net, name, buf);
949 else if (__dev_get_by_name(net, name))
951 else if (buf != name)
952 strlcpy(buf, name, IFNAMSIZ);
958 * dev_change_name - change name of a device
960 * @newname: name (or format string) must be at least IFNAMSIZ
962 * Change name of a device, can pass format strings "eth%d".
965 int dev_change_name(struct net_device *dev, const char *newname)
967 char oldname[IFNAMSIZ];
973 BUG_ON(!dev_net(dev));
976 if (dev->flags & IFF_UP)
979 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
982 memcpy(oldname, dev->name, IFNAMSIZ);
984 err = dev_get_valid_name(net, newname, dev->name, 1);
989 /* For now only devices in the initial network namespace
992 if (net_eq(net, &init_net)) {
993 ret = device_rename(&dev->dev, dev->name);
995 memcpy(dev->name, oldname, IFNAMSIZ);
1000 write_lock_bh(&dev_base_lock);
1001 hlist_del(&dev->name_hlist);
1002 write_unlock_bh(&dev_base_lock);
1006 write_lock_bh(&dev_base_lock);
1007 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1008 write_unlock_bh(&dev_base_lock);
1010 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1011 ret = notifier_to_errno(ret);
1014 /* err >= 0 after dev_alloc_name() or stores the first errno */
1017 memcpy(dev->name, oldname, IFNAMSIZ);
1021 "%s: name change rollback failed: %d.\n",
1030 * dev_set_alias - change ifalias of a device
1032 * @alias: name up to IFALIASZ
1033 * @len: limit of bytes to copy from info
1035 * Set ifalias for a device,
1037 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1041 if (len >= IFALIASZ)
1046 kfree(dev->ifalias);
1047 dev->ifalias = NULL;
1052 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1056 strlcpy(dev->ifalias, alias, len+1);
1062 * netdev_features_change - device changes features
1063 * @dev: device to cause notification
1065 * Called to indicate a device has changed features.
1067 void netdev_features_change(struct net_device *dev)
1069 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1071 EXPORT_SYMBOL(netdev_features_change);
1074 * netdev_state_change - device changes state
1075 * @dev: device to cause notification
1077 * Called to indicate a device has changed state. This function calls
1078 * the notifier chains for netdev_chain and sends a NEWLINK message
1079 * to the routing socket.
1081 void netdev_state_change(struct net_device *dev)
1083 if (dev->flags & IFF_UP) {
1084 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1085 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1088 EXPORT_SYMBOL(netdev_state_change);
1090 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1092 return call_netdevice_notifiers(event, dev);
1094 EXPORT_SYMBOL(netdev_bonding_change);
1097 * dev_load - load a network module
1098 * @net: the applicable net namespace
1099 * @name: name of interface
1101 * If a network interface is not present and the process has suitable
1102 * privileges this function loads the module. If module loading is not
1103 * available in this kernel then it becomes a nop.
1106 void dev_load(struct net *net, const char *name)
1108 struct net_device *dev;
1111 dev = dev_get_by_name_rcu(net, name);
1114 if (!dev && capable(CAP_NET_ADMIN))
1115 request_module("%s", name);
1117 EXPORT_SYMBOL(dev_load);
1119 static int __dev_open(struct net_device *dev)
1121 const struct net_device_ops *ops = dev->netdev_ops;
1127 * Is it even present?
1129 if (!netif_device_present(dev))
1132 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1133 ret = notifier_to_errno(ret);
1138 * Call device private open method
1140 set_bit(__LINK_STATE_START, &dev->state);
1142 if (ops->ndo_validate_addr)
1143 ret = ops->ndo_validate_addr(dev);
1145 if (!ret && ops->ndo_open)
1146 ret = ops->ndo_open(dev);
1149 * If it went open OK then:
1153 clear_bit(__LINK_STATE_START, &dev->state);
1158 dev->flags |= IFF_UP;
1163 net_dmaengine_get();
1166 * Initialize multicasting status
1168 dev_set_rx_mode(dev);
1171 * Wakeup transmit queue engine
1180 * dev_open - prepare an interface for use.
1181 * @dev: device to open
1183 * Takes a device from down to up state. The device's private open
1184 * function is invoked and then the multicast lists are loaded. Finally
1185 * the device is moved into the up state and a %NETDEV_UP message is
1186 * sent to the netdev notifier chain.
1188 * Calling this function on an active interface is a nop. On a failure
1189 * a negative errno code is returned.
1191 int dev_open(struct net_device *dev)
1198 if (dev->flags & IFF_UP)
1204 ret = __dev_open(dev);
1209 * ... and announce new interface.
1211 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1212 call_netdevice_notifiers(NETDEV_UP, dev);
1216 EXPORT_SYMBOL(dev_open);
1218 static int __dev_close(struct net_device *dev)
1220 const struct net_device_ops *ops = dev->netdev_ops;
1226 * Tell people we are going down, so that they can
1227 * prepare to death, when device is still operating.
1229 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1231 clear_bit(__LINK_STATE_START, &dev->state);
1233 /* Synchronize to scheduled poll. We cannot touch poll list,
1234 * it can be even on different cpu. So just clear netif_running().
1236 * dev->stop() will invoke napi_disable() on all of it's
1237 * napi_struct instances on this device.
1239 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1241 dev_deactivate(dev);
1244 * Call the device specific close. This cannot fail.
1245 * Only if device is UP
1247 * We allow it to be called even after a DETACH hot-plug
1254 * Device is now down.
1257 dev->flags &= ~IFF_UP;
1262 net_dmaengine_put();
1268 * dev_close - shutdown an interface.
1269 * @dev: device to shutdown
1271 * This function moves an active device into down state. A
1272 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1273 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1276 int dev_close(struct net_device *dev)
1278 if (!(dev->flags & IFF_UP))
1284 * Tell people we are down
1286 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1287 call_netdevice_notifiers(NETDEV_DOWN, dev);
1291 EXPORT_SYMBOL(dev_close);
1295 * dev_disable_lro - disable Large Receive Offload on a device
1298 * Disable Large Receive Offload (LRO) on a net device. Must be
1299 * called under RTNL. This is needed if received packets may be
1300 * forwarded to another interface.
1302 void dev_disable_lro(struct net_device *dev)
1304 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1305 dev->ethtool_ops->set_flags) {
1306 u32 flags = dev->ethtool_ops->get_flags(dev);
1307 if (flags & ETH_FLAG_LRO) {
1308 flags &= ~ETH_FLAG_LRO;
1309 dev->ethtool_ops->set_flags(dev, flags);
1312 WARN_ON(dev->features & NETIF_F_LRO);
1314 EXPORT_SYMBOL(dev_disable_lro);
1317 static int dev_boot_phase = 1;
1320 * Device change register/unregister. These are not inline or static
1321 * as we export them to the world.
1325 * register_netdevice_notifier - register a network notifier block
1328 * Register a notifier to be called when network device events occur.
1329 * The notifier passed is linked into the kernel structures and must
1330 * not be reused until it has been unregistered. A negative errno code
1331 * is returned on a failure.
1333 * When registered all registration and up events are replayed
1334 * to the new notifier to allow device to have a race free
1335 * view of the network device list.
1338 int register_netdevice_notifier(struct notifier_block *nb)
1340 struct net_device *dev;
1341 struct net_device *last;
1346 err = raw_notifier_chain_register(&netdev_chain, nb);
1352 for_each_netdev(net, dev) {
1353 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1354 err = notifier_to_errno(err);
1358 if (!(dev->flags & IFF_UP))
1361 nb->notifier_call(nb, NETDEV_UP, dev);
1372 for_each_netdev(net, dev) {
1376 if (dev->flags & IFF_UP) {
1377 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1378 nb->notifier_call(nb, NETDEV_DOWN, dev);
1380 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1381 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1385 raw_notifier_chain_unregister(&netdev_chain, nb);
1388 EXPORT_SYMBOL(register_netdevice_notifier);
1391 * unregister_netdevice_notifier - unregister a network notifier block
1394 * Unregister a notifier previously registered by
1395 * register_netdevice_notifier(). The notifier is unlinked into the
1396 * kernel structures and may then be reused. A negative errno code
1397 * is returned on a failure.
1400 int unregister_netdevice_notifier(struct notifier_block *nb)
1405 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1409 EXPORT_SYMBOL(unregister_netdevice_notifier);
1412 * call_netdevice_notifiers - call all network notifier blocks
1413 * @val: value passed unmodified to notifier function
1414 * @dev: net_device pointer passed unmodified to notifier function
1416 * Call all network notifier blocks. Parameters and return value
1417 * are as for raw_notifier_call_chain().
1420 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1422 return raw_notifier_call_chain(&netdev_chain, val, dev);
1425 /* When > 0 there are consumers of rx skb time stamps */
1426 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1428 void net_enable_timestamp(void)
1430 atomic_inc(&netstamp_needed);
1432 EXPORT_SYMBOL(net_enable_timestamp);
1434 void net_disable_timestamp(void)
1436 atomic_dec(&netstamp_needed);
1438 EXPORT_SYMBOL(net_disable_timestamp);
1440 static inline void net_timestamp(struct sk_buff *skb)
1442 if (atomic_read(&netstamp_needed))
1443 __net_timestamp(skb);
1445 skb->tstamp.tv64 = 0;
1449 * dev_forward_skb - loopback an skb to another netif
1451 * @dev: destination network device
1452 * @skb: buffer to forward
1455 * NET_RX_SUCCESS (no congestion)
1456 * NET_RX_DROP (packet was dropped)
1458 * dev_forward_skb can be used for injecting an skb from the
1459 * start_xmit function of one device into the receive queue
1460 * of another device.
1462 * The receiving device may be in another namespace, so
1463 * we have to clear all information in the skb that could
1464 * impact namespace isolation.
1466 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1470 if (!(dev->flags & IFF_UP))
1473 if (skb->len > (dev->mtu + dev->hard_header_len))
1476 skb_set_dev(skb, dev);
1477 skb->tstamp.tv64 = 0;
1478 skb->pkt_type = PACKET_HOST;
1479 skb->protocol = eth_type_trans(skb, dev);
1480 return netif_rx(skb);
1482 EXPORT_SYMBOL_GPL(dev_forward_skb);
1485 * Support routine. Sends outgoing frames to any network
1486 * taps currently in use.
1489 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1491 struct packet_type *ptype;
1493 #ifdef CONFIG_NET_CLS_ACT
1494 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1501 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1502 /* Never send packets back to the socket
1503 * they originated from - MvS (miquels@drinkel.ow.org)
1505 if ((ptype->dev == dev || !ptype->dev) &&
1506 (ptype->af_packet_priv == NULL ||
1507 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1508 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1512 /* skb->nh should be correctly
1513 set by sender, so that the second statement is
1514 just protection against buggy protocols.
1516 skb_reset_mac_header(skb2);
1518 if (skb_network_header(skb2) < skb2->data ||
1519 skb2->network_header > skb2->tail) {
1520 if (net_ratelimit())
1521 printk(KERN_CRIT "protocol %04x is "
1523 skb2->protocol, dev->name);
1524 skb_reset_network_header(skb2);
1527 skb2->transport_header = skb2->network_header;
1528 skb2->pkt_type = PACKET_OUTGOING;
1529 ptype->func(skb2, skb->dev, ptype, skb->dev);
1536 static inline void __netif_reschedule(struct Qdisc *q)
1538 struct softnet_data *sd;
1539 unsigned long flags;
1541 local_irq_save(flags);
1542 sd = &__get_cpu_var(softnet_data);
1543 q->next_sched = sd->output_queue;
1544 sd->output_queue = q;
1545 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1546 local_irq_restore(flags);
1549 void __netif_schedule(struct Qdisc *q)
1551 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1552 __netif_reschedule(q);
1554 EXPORT_SYMBOL(__netif_schedule);
1556 void dev_kfree_skb_irq(struct sk_buff *skb)
1558 if (atomic_dec_and_test(&skb->users)) {
1559 struct softnet_data *sd;
1560 unsigned long flags;
1562 local_irq_save(flags);
1563 sd = &__get_cpu_var(softnet_data);
1564 skb->next = sd->completion_queue;
1565 sd->completion_queue = skb;
1566 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1567 local_irq_restore(flags);
1570 EXPORT_SYMBOL(dev_kfree_skb_irq);
1572 void dev_kfree_skb_any(struct sk_buff *skb)
1574 if (in_irq() || irqs_disabled())
1575 dev_kfree_skb_irq(skb);
1579 EXPORT_SYMBOL(dev_kfree_skb_any);
1583 * netif_device_detach - mark device as removed
1584 * @dev: network device
1586 * Mark device as removed from system and therefore no longer available.
1588 void netif_device_detach(struct net_device *dev)
1590 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1591 netif_running(dev)) {
1592 netif_tx_stop_all_queues(dev);
1595 EXPORT_SYMBOL(netif_device_detach);
1598 * netif_device_attach - mark device as attached
1599 * @dev: network device
1601 * Mark device as attached from system and restart if needed.
1603 void netif_device_attach(struct net_device *dev)
1605 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1606 netif_running(dev)) {
1607 netif_tx_wake_all_queues(dev);
1608 __netdev_watchdog_up(dev);
1611 EXPORT_SYMBOL(netif_device_attach);
1613 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1615 return ((features & NETIF_F_GEN_CSUM) ||
1616 ((features & NETIF_F_IP_CSUM) &&
1617 protocol == htons(ETH_P_IP)) ||
1618 ((features & NETIF_F_IPV6_CSUM) &&
1619 protocol == htons(ETH_P_IPV6)) ||
1620 ((features & NETIF_F_FCOE_CRC) &&
1621 protocol == htons(ETH_P_FCOE)));
1624 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1626 if (can_checksum_protocol(dev->features, skb->protocol))
1629 if (skb->protocol == htons(ETH_P_8021Q)) {
1630 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1631 if (can_checksum_protocol(dev->features & dev->vlan_features,
1632 veh->h_vlan_encapsulated_proto))
1640 * skb_dev_set -- assign a new device to a buffer
1641 * @skb: buffer for the new device
1642 * @dev: network device
1644 * If an skb is owned by a device already, we have to reset
1645 * all data private to the namespace a device belongs to
1646 * before assigning it a new device.
1648 #ifdef CONFIG_NET_NS
1649 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1652 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1655 skb_init_secmark(skb);
1659 skb->ipvs_property = 0;
1660 #ifdef CONFIG_NET_SCHED
1666 EXPORT_SYMBOL(skb_set_dev);
1667 #endif /* CONFIG_NET_NS */
1670 * Invalidate hardware checksum when packet is to be mangled, and
1671 * complete checksum manually on outgoing path.
1673 int skb_checksum_help(struct sk_buff *skb)
1676 int ret = 0, offset;
1678 if (skb->ip_summed == CHECKSUM_COMPLETE)
1679 goto out_set_summed;
1681 if (unlikely(skb_shinfo(skb)->gso_size)) {
1682 /* Let GSO fix up the checksum. */
1683 goto out_set_summed;
1686 offset = skb->csum_start - skb_headroom(skb);
1687 BUG_ON(offset >= skb_headlen(skb));
1688 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1690 offset += skb->csum_offset;
1691 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1693 if (skb_cloned(skb) &&
1694 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1695 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1700 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1702 skb->ip_summed = CHECKSUM_NONE;
1706 EXPORT_SYMBOL(skb_checksum_help);
1709 * skb_gso_segment - Perform segmentation on skb.
1710 * @skb: buffer to segment
1711 * @features: features for the output path (see dev->features)
1713 * This function segments the given skb and returns a list of segments.
1715 * It may return NULL if the skb requires no segmentation. This is
1716 * only possible when GSO is used for verifying header integrity.
1718 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1720 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1721 struct packet_type *ptype;
1722 __be16 type = skb->protocol;
1725 skb_reset_mac_header(skb);
1726 skb->mac_len = skb->network_header - skb->mac_header;
1727 __skb_pull(skb, skb->mac_len);
1729 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1730 struct net_device *dev = skb->dev;
1731 struct ethtool_drvinfo info = {};
1733 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1734 dev->ethtool_ops->get_drvinfo(dev, &info);
1736 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1738 info.driver, dev ? dev->features : 0L,
1739 skb->sk ? skb->sk->sk_route_caps : 0L,
1740 skb->len, skb->data_len, skb->ip_summed);
1742 if (skb_header_cloned(skb) &&
1743 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1744 return ERR_PTR(err);
1748 list_for_each_entry_rcu(ptype,
1749 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1750 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1751 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1752 err = ptype->gso_send_check(skb);
1753 segs = ERR_PTR(err);
1754 if (err || skb_gso_ok(skb, features))
1756 __skb_push(skb, (skb->data -
1757 skb_network_header(skb)));
1759 segs = ptype->gso_segment(skb, features);
1765 __skb_push(skb, skb->data - skb_mac_header(skb));
1769 EXPORT_SYMBOL(skb_gso_segment);
1771 /* Take action when hardware reception checksum errors are detected. */
1773 void netdev_rx_csum_fault(struct net_device *dev)
1775 if (net_ratelimit()) {
1776 printk(KERN_ERR "%s: hw csum failure.\n",
1777 dev ? dev->name : "<unknown>");
1781 EXPORT_SYMBOL(netdev_rx_csum_fault);
1784 /* Actually, we should eliminate this check as soon as we know, that:
1785 * 1. IOMMU is present and allows to map all the memory.
1786 * 2. No high memory really exists on this machine.
1789 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1791 #ifdef CONFIG_HIGHMEM
1794 if (dev->features & NETIF_F_HIGHDMA)
1797 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1798 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1806 void (*destructor)(struct sk_buff *skb);
1809 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1811 static void dev_gso_skb_destructor(struct sk_buff *skb)
1813 struct dev_gso_cb *cb;
1816 struct sk_buff *nskb = skb->next;
1818 skb->next = nskb->next;
1821 } while (skb->next);
1823 cb = DEV_GSO_CB(skb);
1825 cb->destructor(skb);
1829 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1830 * @skb: buffer to segment
1832 * This function segments the given skb and stores the list of segments
1835 static int dev_gso_segment(struct sk_buff *skb)
1837 struct net_device *dev = skb->dev;
1838 struct sk_buff *segs;
1839 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1842 segs = skb_gso_segment(skb, features);
1844 /* Verifying header integrity only. */
1849 return PTR_ERR(segs);
1852 DEV_GSO_CB(skb)->destructor = skb->destructor;
1853 skb->destructor = dev_gso_skb_destructor;
1858 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1859 struct netdev_queue *txq)
1861 const struct net_device_ops *ops = dev->netdev_ops;
1862 int rc = NETDEV_TX_OK;
1864 if (likely(!skb->next)) {
1865 if (!list_empty(&ptype_all))
1866 dev_queue_xmit_nit(skb, dev);
1868 if (netif_needs_gso(dev, skb)) {
1869 if (unlikely(dev_gso_segment(skb)))
1876 * If device doesnt need skb->dst, release it right now while
1877 * its hot in this cpu cache
1879 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1882 rc = ops->ndo_start_xmit(skb, dev);
1883 if (rc == NETDEV_TX_OK)
1884 txq_trans_update(txq);
1886 * TODO: if skb_orphan() was called by
1887 * dev->hard_start_xmit() (for example, the unmodified
1888 * igb driver does that; bnx2 doesn't), then
1889 * skb_tx_software_timestamp() will be unable to send
1890 * back the time stamp.
1892 * How can this be prevented? Always create another
1893 * reference to the socket before calling
1894 * dev->hard_start_xmit()? Prevent that skb_orphan()
1895 * does anything in dev->hard_start_xmit() by clearing
1896 * the skb destructor before the call and restoring it
1897 * afterwards, then doing the skb_orphan() ourselves?
1904 struct sk_buff *nskb = skb->next;
1906 skb->next = nskb->next;
1910 * If device doesnt need nskb->dst, release it right now while
1911 * its hot in this cpu cache
1913 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1916 rc = ops->ndo_start_xmit(nskb, dev);
1917 if (unlikely(rc != NETDEV_TX_OK)) {
1918 if (rc & ~NETDEV_TX_MASK)
1919 goto out_kfree_gso_skb;
1920 nskb->next = skb->next;
1924 txq_trans_update(txq);
1925 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1926 return NETDEV_TX_BUSY;
1927 } while (skb->next);
1930 if (likely(skb->next == NULL))
1931 skb->destructor = DEV_GSO_CB(skb)->destructor;
1937 static u32 hashrnd __read_mostly;
1939 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1943 if (skb_rx_queue_recorded(skb)) {
1944 hash = skb_get_rx_queue(skb);
1945 while (unlikely(hash >= dev->real_num_tx_queues))
1946 hash -= dev->real_num_tx_queues;
1950 if (skb->sk && skb->sk->sk_hash)
1951 hash = skb->sk->sk_hash;
1953 hash = skb->protocol;
1955 hash = jhash_1word(hash, hashrnd);
1957 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1959 EXPORT_SYMBOL(skb_tx_hash);
1961 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1963 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1964 if (net_ratelimit()) {
1965 netdev_warn(dev, "selects TX queue %d, but "
1966 "real number of TX queues is %d\n",
1967 queue_index, dev->real_num_tx_queues);
1974 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1975 struct sk_buff *skb)
1978 struct sock *sk = skb->sk;
1980 if (sk_tx_queue_recorded(sk)) {
1981 queue_index = sk_tx_queue_get(sk);
1983 const struct net_device_ops *ops = dev->netdev_ops;
1985 if (ops->ndo_select_queue) {
1986 queue_index = ops->ndo_select_queue(dev, skb);
1987 queue_index = dev_cap_txqueue(dev, queue_index);
1990 if (dev->real_num_tx_queues > 1)
1991 queue_index = skb_tx_hash(dev, skb);
1993 if (sk && sk->sk_dst_cache)
1994 sk_tx_queue_set(sk, queue_index);
1998 skb_set_queue_mapping(skb, queue_index);
1999 return netdev_get_tx_queue(dev, queue_index);
2002 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2003 struct net_device *dev,
2004 struct netdev_queue *txq)
2006 spinlock_t *root_lock = qdisc_lock(q);
2009 spin_lock(root_lock);
2010 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2013 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2014 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
2016 * This is a work-conserving queue; there are no old skbs
2017 * waiting to be sent out; and the qdisc is not running -
2018 * xmit the skb directly.
2020 __qdisc_update_bstats(q, skb->len);
2021 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
2024 clear_bit(__QDISC_STATE_RUNNING, &q->state);
2026 rc = NET_XMIT_SUCCESS;
2028 rc = qdisc_enqueue_root(skb, q);
2031 spin_unlock(root_lock);
2037 * Returns true if either:
2038 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2039 * 2. skb is fragmented and the device does not support SG, or if
2040 * at least one of fragments is in highmem and device does not
2041 * support DMA from it.
2043 static inline int skb_needs_linearize(struct sk_buff *skb,
2044 struct net_device *dev)
2046 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
2047 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
2048 illegal_highdma(dev, skb)));
2052 * dev_queue_xmit - transmit a buffer
2053 * @skb: buffer to transmit
2055 * Queue a buffer for transmission to a network device. The caller must
2056 * have set the device and priority and built the buffer before calling
2057 * this function. The function can be called from an interrupt.
2059 * A negative errno code is returned on a failure. A success does not
2060 * guarantee the frame will be transmitted as it may be dropped due
2061 * to congestion or traffic shaping.
2063 * -----------------------------------------------------------------------------------
2064 * I notice this method can also return errors from the queue disciplines,
2065 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2068 * Regardless of the return value, the skb is consumed, so it is currently
2069 * difficult to retry a send to this method. (You can bump the ref count
2070 * before sending to hold a reference for retry if you are careful.)
2072 * When calling this method, interrupts MUST be enabled. This is because
2073 * the BH enable code must have IRQs enabled so that it will not deadlock.
2076 int dev_queue_xmit(struct sk_buff *skb)
2078 struct net_device *dev = skb->dev;
2079 struct netdev_queue *txq;
2083 /* GSO will handle the following emulations directly. */
2084 if (netif_needs_gso(dev, skb))
2087 /* Convert a paged skb to linear, if required */
2088 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2091 /* If packet is not checksummed and device does not support
2092 * checksumming for this protocol, complete checksumming here.
2094 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2095 skb_set_transport_header(skb, skb->csum_start -
2097 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2102 /* Disable soft irqs for various locks below. Also
2103 * stops preemption for RCU.
2107 txq = dev_pick_tx(dev, skb);
2108 q = rcu_dereference_bh(txq->qdisc);
2110 #ifdef CONFIG_NET_CLS_ACT
2111 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2114 rc = __dev_xmit_skb(skb, q, dev, txq);
2118 /* The device has no queue. Common case for software devices:
2119 loopback, all the sorts of tunnels...
2121 Really, it is unlikely that netif_tx_lock protection is necessary
2122 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2124 However, it is possible, that they rely on protection
2127 Check this and shot the lock. It is not prone from deadlocks.
2128 Either shot noqueue qdisc, it is even simpler 8)
2130 if (dev->flags & IFF_UP) {
2131 int cpu = smp_processor_id(); /* ok because BHs are off */
2133 if (txq->xmit_lock_owner != cpu) {
2135 HARD_TX_LOCK(dev, txq, cpu);
2137 if (!netif_tx_queue_stopped(txq)) {
2138 rc = dev_hard_start_xmit(skb, dev, txq);
2139 if (dev_xmit_complete(rc)) {
2140 HARD_TX_UNLOCK(dev, txq);
2144 HARD_TX_UNLOCK(dev, txq);
2145 if (net_ratelimit())
2146 printk(KERN_CRIT "Virtual device %s asks to "
2147 "queue packet!\n", dev->name);
2149 /* Recursion is detected! It is possible,
2151 if (net_ratelimit())
2152 printk(KERN_CRIT "Dead loop on virtual device "
2153 "%s, fix it urgently!\n", dev->name);
2158 rcu_read_unlock_bh();
2164 rcu_read_unlock_bh();
2167 EXPORT_SYMBOL(dev_queue_xmit);
2170 /*=======================================================================
2172 =======================================================================*/
2174 int netdev_max_backlog __read_mostly = 1000;
2175 int netdev_budget __read_mostly = 300;
2176 int weight_p __read_mostly = 64; /* old backlog weight */
2178 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2182 * get_rps_cpu is called from netif_receive_skb and returns the target
2183 * CPU from the RPS map of the receiving queue for a given skb.
2185 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb)
2187 struct ipv6hdr *ip6;
2189 struct netdev_rx_queue *rxqueue;
2190 struct rps_map *map;
2193 u32 addr1, addr2, ports, ihl;
2197 if (skb_rx_queue_recorded(skb)) {
2198 u16 index = skb_get_rx_queue(skb);
2199 if (unlikely(index >= dev->num_rx_queues)) {
2200 if (net_ratelimit()) {
2201 netdev_warn(dev, "received packet on queue "
2202 "%u, but number of RX queues is %u\n",
2203 index, dev->num_rx_queues);
2207 rxqueue = dev->_rx + index;
2211 if (!rxqueue->rps_map)
2215 goto got_hash; /* Skip hash computation on packet header */
2217 switch (skb->protocol) {
2218 case __constant_htons(ETH_P_IP):
2219 if (!pskb_may_pull(skb, sizeof(*ip)))
2222 ip = (struct iphdr *) skb->data;
2223 ip_proto = ip->protocol;
2228 case __constant_htons(ETH_P_IPV6):
2229 if (!pskb_may_pull(skb, sizeof(*ip6)))
2232 ip6 = (struct ipv6hdr *) skb->data;
2233 ip_proto = ip6->nexthdr;
2234 addr1 = ip6->saddr.s6_addr32[3];
2235 addr2 = ip6->daddr.s6_addr32[3];
2249 case IPPROTO_UDPLITE:
2250 if (pskb_may_pull(skb, (ihl * 4) + 4))
2251 ports = *((u32 *) (skb->data + (ihl * 4)));
2258 skb->rxhash = jhash_3words(addr1, addr2, ports, hashrnd);
2263 map = rcu_dereference(rxqueue->rps_map);
2265 u16 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2267 if (cpu_online(tcpu)) {
2279 * This structure holds the per-CPU mask of CPUs for which IPIs are scheduled
2280 * to be sent to kick remote softirq processing. There are two masks since
2281 * the sending of IPIs must be done with interrupts enabled. The select field
2282 * indicates the current mask that enqueue_backlog uses to schedule IPIs.
2283 * select is flipped before net_rps_action is called while still under lock,
2284 * net_rps_action then uses the non-selected mask to send the IPIs and clears
2285 * it without conflicting with enqueue_backlog operation.
2287 struct rps_remote_softirq_cpus {
2291 static DEFINE_PER_CPU(struct rps_remote_softirq_cpus, rps_remote_softirq_cpus);
2293 /* Called from hardirq (IPI) context */
2294 static void trigger_softirq(void *data)
2296 struct softnet_data *queue = data;
2297 __napi_schedule(&queue->backlog);
2298 __get_cpu_var(netdev_rx_stat).received_rps++;
2300 #endif /* CONFIG_SMP */
2303 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2304 * queue (may be a remote CPU queue).
2306 static int enqueue_to_backlog(struct sk_buff *skb, int cpu)
2308 struct softnet_data *queue;
2309 unsigned long flags;
2311 queue = &per_cpu(softnet_data, cpu);
2313 local_irq_save(flags);
2314 __get_cpu_var(netdev_rx_stat).total++;
2316 spin_lock(&queue->input_pkt_queue.lock);
2317 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2318 if (queue->input_pkt_queue.qlen) {
2320 __skb_queue_tail(&queue->input_pkt_queue, skb);
2321 spin_unlock_irqrestore(&queue->input_pkt_queue.lock,
2323 return NET_RX_SUCCESS;
2326 /* Schedule NAPI for backlog device */
2327 if (napi_schedule_prep(&queue->backlog)) {
2329 if (cpu != smp_processor_id()) {
2330 struct rps_remote_softirq_cpus *rcpus =
2331 &__get_cpu_var(rps_remote_softirq_cpus);
2333 cpu_set(cpu, rcpus->mask[rcpus->select]);
2334 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2336 __napi_schedule(&queue->backlog);
2338 __napi_schedule(&queue->backlog);
2344 spin_unlock(&queue->input_pkt_queue.lock);
2346 __get_cpu_var(netdev_rx_stat).dropped++;
2347 local_irq_restore(flags);
2354 * netif_rx - post buffer to the network code
2355 * @skb: buffer to post
2357 * This function receives a packet from a device driver and queues it for
2358 * the upper (protocol) levels to process. It always succeeds. The buffer
2359 * may be dropped during processing for congestion control or by the
2363 * NET_RX_SUCCESS (no congestion)
2364 * NET_RX_DROP (packet was dropped)
2368 int netif_rx(struct sk_buff *skb)
2372 /* if netpoll wants it, pretend we never saw it */
2373 if (netpoll_rx(skb))
2376 if (!skb->tstamp.tv64)
2380 cpu = get_rps_cpu(skb->dev, skb);
2382 cpu = smp_processor_id();
2384 cpu = smp_processor_id();
2387 return enqueue_to_backlog(skb, cpu);
2389 EXPORT_SYMBOL(netif_rx);
2391 int netif_rx_ni(struct sk_buff *skb)
2396 err = netif_rx(skb);
2397 if (local_softirq_pending())
2403 EXPORT_SYMBOL(netif_rx_ni);
2405 static void net_tx_action(struct softirq_action *h)
2407 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2409 if (sd->completion_queue) {
2410 struct sk_buff *clist;
2412 local_irq_disable();
2413 clist = sd->completion_queue;
2414 sd->completion_queue = NULL;
2418 struct sk_buff *skb = clist;
2419 clist = clist->next;
2421 WARN_ON(atomic_read(&skb->users));
2426 if (sd->output_queue) {
2429 local_irq_disable();
2430 head = sd->output_queue;
2431 sd->output_queue = NULL;
2435 struct Qdisc *q = head;
2436 spinlock_t *root_lock;
2438 head = head->next_sched;
2440 root_lock = qdisc_lock(q);
2441 if (spin_trylock(root_lock)) {
2442 smp_mb__before_clear_bit();
2443 clear_bit(__QDISC_STATE_SCHED,
2446 spin_unlock(root_lock);
2448 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2450 __netif_reschedule(q);
2452 smp_mb__before_clear_bit();
2453 clear_bit(__QDISC_STATE_SCHED,
2461 static inline int deliver_skb(struct sk_buff *skb,
2462 struct packet_type *pt_prev,
2463 struct net_device *orig_dev)
2465 atomic_inc(&skb->users);
2466 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2469 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2471 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2472 /* This hook is defined here for ATM LANE */
2473 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2474 unsigned char *addr) __read_mostly;
2475 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2479 * If bridge module is loaded call bridging hook.
2480 * returns NULL if packet was consumed.
2482 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2483 struct sk_buff *skb) __read_mostly;
2484 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2486 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2487 struct packet_type **pt_prev, int *ret,
2488 struct net_device *orig_dev)
2490 struct net_bridge_port *port;
2492 if (skb->pkt_type == PACKET_LOOPBACK ||
2493 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2497 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2501 return br_handle_frame_hook(port, skb);
2504 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2507 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2508 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2509 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2511 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2512 struct packet_type **pt_prev,
2514 struct net_device *orig_dev)
2516 if (skb->dev->macvlan_port == NULL)
2520 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2523 return macvlan_handle_frame_hook(skb);
2526 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2529 #ifdef CONFIG_NET_CLS_ACT
2530 /* TODO: Maybe we should just force sch_ingress to be compiled in
2531 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2532 * a compare and 2 stores extra right now if we dont have it on
2533 * but have CONFIG_NET_CLS_ACT
2534 * NOTE: This doesnt stop any functionality; if you dont have
2535 * the ingress scheduler, you just cant add policies on ingress.
2538 static int ing_filter(struct sk_buff *skb)
2540 struct net_device *dev = skb->dev;
2541 u32 ttl = G_TC_RTTL(skb->tc_verd);
2542 struct netdev_queue *rxq;
2543 int result = TC_ACT_OK;
2546 if (MAX_RED_LOOP < ttl++) {
2548 "Redir loop detected Dropping packet (%d->%d)\n",
2549 skb->skb_iif, dev->ifindex);
2553 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2554 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2556 rxq = &dev->rx_queue;
2559 if (q != &noop_qdisc) {
2560 spin_lock(qdisc_lock(q));
2561 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2562 result = qdisc_enqueue_root(skb, q);
2563 spin_unlock(qdisc_lock(q));
2569 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2570 struct packet_type **pt_prev,
2571 int *ret, struct net_device *orig_dev)
2573 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2577 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2580 /* Huh? Why does turning on AF_PACKET affect this? */
2581 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2584 switch (ing_filter(skb)) {
2598 * netif_nit_deliver - deliver received packets to network taps
2601 * This function is used to deliver incoming packets to network
2602 * taps. It should be used when the normal netif_receive_skb path
2603 * is bypassed, for example because of VLAN acceleration.
2605 void netif_nit_deliver(struct sk_buff *skb)
2607 struct packet_type *ptype;
2609 if (list_empty(&ptype_all))
2612 skb_reset_network_header(skb);
2613 skb_reset_transport_header(skb);
2614 skb->mac_len = skb->network_header - skb->mac_header;
2617 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2618 if (!ptype->dev || ptype->dev == skb->dev)
2619 deliver_skb(skb, ptype, skb->dev);
2624 int __netif_receive_skb(struct sk_buff *skb)
2626 struct packet_type *ptype, *pt_prev;
2627 struct net_device *orig_dev;
2628 struct net_device *master;
2629 struct net_device *null_or_orig;
2630 struct net_device *null_or_bond;
2631 int ret = NET_RX_DROP;
2634 if (!skb->tstamp.tv64)
2637 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2638 return NET_RX_SUCCESS;
2640 /* if we've gotten here through NAPI, check netpoll */
2641 if (netpoll_receive_skb(skb))
2645 skb->skb_iif = skb->dev->ifindex;
2647 null_or_orig = NULL;
2648 orig_dev = skb->dev;
2649 master = ACCESS_ONCE(orig_dev->master);
2651 if (skb_bond_should_drop(skb, master))
2652 null_or_orig = orig_dev; /* deliver only exact match */
2657 __get_cpu_var(netdev_rx_stat).total++;
2659 skb_reset_network_header(skb);
2660 skb_reset_transport_header(skb);
2661 skb->mac_len = skb->network_header - skb->mac_header;
2667 #ifdef CONFIG_NET_CLS_ACT
2668 if (skb->tc_verd & TC_NCLS) {
2669 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2674 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2675 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2676 ptype->dev == orig_dev) {
2678 ret = deliver_skb(skb, pt_prev, orig_dev);
2683 #ifdef CONFIG_NET_CLS_ACT
2684 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2690 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2693 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2698 * Make sure frames received on VLAN interfaces stacked on
2699 * bonding interfaces still make their way to any base bonding
2700 * device that may have registered for a specific ptype. The
2701 * handler may have to adjust skb->dev and orig_dev.
2703 null_or_bond = NULL;
2704 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2705 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2706 null_or_bond = vlan_dev_real_dev(skb->dev);
2709 type = skb->protocol;
2710 list_for_each_entry_rcu(ptype,
2711 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2712 if (ptype->type == type && (ptype->dev == null_or_orig ||
2713 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2714 ptype->dev == null_or_bond)) {
2716 ret = deliver_skb(skb, pt_prev, orig_dev);
2722 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2725 /* Jamal, now you will not able to escape explaining
2726 * me how you were going to use this. :-)
2737 * netif_receive_skb - process receive buffer from network
2738 * @skb: buffer to process
2740 * netif_receive_skb() is the main receive data processing function.
2741 * It always succeeds. The buffer may be dropped during processing
2742 * for congestion control or by the protocol layers.
2744 * This function may only be called from softirq context and interrupts
2745 * should be enabled.
2747 * Return values (usually ignored):
2748 * NET_RX_SUCCESS: no congestion
2749 * NET_RX_DROP: packet was dropped
2751 int netif_receive_skb(struct sk_buff *skb)
2756 cpu = get_rps_cpu(skb->dev, skb);
2759 return __netif_receive_skb(skb);
2761 return enqueue_to_backlog(skb, cpu);
2763 return __netif_receive_skb(skb);
2766 EXPORT_SYMBOL(netif_receive_skb);
2768 /* Network device is going away, flush any packets still pending */
2769 static void flush_backlog(void *arg)
2771 struct net_device *dev = arg;
2772 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2773 struct sk_buff *skb, *tmp;
2775 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2776 if (skb->dev == dev) {
2777 __skb_unlink(skb, &queue->input_pkt_queue);
2782 static int napi_gro_complete(struct sk_buff *skb)
2784 struct packet_type *ptype;
2785 __be16 type = skb->protocol;
2786 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2789 if (NAPI_GRO_CB(skb)->count == 1) {
2790 skb_shinfo(skb)->gso_size = 0;
2795 list_for_each_entry_rcu(ptype, head, list) {
2796 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2799 err = ptype->gro_complete(skb);
2805 WARN_ON(&ptype->list == head);
2807 return NET_RX_SUCCESS;
2811 return netif_receive_skb(skb);
2814 static void napi_gro_flush(struct napi_struct *napi)
2816 struct sk_buff *skb, *next;
2818 for (skb = napi->gro_list; skb; skb = next) {
2821 napi_gro_complete(skb);
2824 napi->gro_count = 0;
2825 napi->gro_list = NULL;
2828 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2830 struct sk_buff **pp = NULL;
2831 struct packet_type *ptype;
2832 __be16 type = skb->protocol;
2833 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2836 enum gro_result ret;
2838 if (!(skb->dev->features & NETIF_F_GRO))
2841 if (skb_is_gso(skb) || skb_has_frags(skb))
2845 list_for_each_entry_rcu(ptype, head, list) {
2846 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2849 skb_set_network_header(skb, skb_gro_offset(skb));
2850 mac_len = skb->network_header - skb->mac_header;
2851 skb->mac_len = mac_len;
2852 NAPI_GRO_CB(skb)->same_flow = 0;
2853 NAPI_GRO_CB(skb)->flush = 0;
2854 NAPI_GRO_CB(skb)->free = 0;
2856 pp = ptype->gro_receive(&napi->gro_list, skb);
2861 if (&ptype->list == head)
2864 same_flow = NAPI_GRO_CB(skb)->same_flow;
2865 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2868 struct sk_buff *nskb = *pp;
2872 napi_gro_complete(nskb);
2879 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2883 NAPI_GRO_CB(skb)->count = 1;
2884 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2885 skb->next = napi->gro_list;
2886 napi->gro_list = skb;
2890 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2891 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2893 BUG_ON(skb->end - skb->tail < grow);
2895 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2898 skb->data_len -= grow;
2900 skb_shinfo(skb)->frags[0].page_offset += grow;
2901 skb_shinfo(skb)->frags[0].size -= grow;
2903 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2904 put_page(skb_shinfo(skb)->frags[0].page);
2905 memmove(skb_shinfo(skb)->frags,
2906 skb_shinfo(skb)->frags + 1,
2907 --skb_shinfo(skb)->nr_frags);
2918 EXPORT_SYMBOL(dev_gro_receive);
2921 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2925 if (netpoll_rx_on(skb))
2928 for (p = napi->gro_list; p; p = p->next) {
2929 NAPI_GRO_CB(p)->same_flow =
2930 (p->dev == skb->dev) &&
2931 !compare_ether_header(skb_mac_header(p),
2932 skb_gro_mac_header(skb));
2933 NAPI_GRO_CB(p)->flush = 0;
2936 return dev_gro_receive(napi, skb);
2939 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
2943 if (netif_receive_skb(skb))
2948 case GRO_MERGED_FREE:
2959 EXPORT_SYMBOL(napi_skb_finish);
2961 void skb_gro_reset_offset(struct sk_buff *skb)
2963 NAPI_GRO_CB(skb)->data_offset = 0;
2964 NAPI_GRO_CB(skb)->frag0 = NULL;
2965 NAPI_GRO_CB(skb)->frag0_len = 0;
2967 if (skb->mac_header == skb->tail &&
2968 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2969 NAPI_GRO_CB(skb)->frag0 =
2970 page_address(skb_shinfo(skb)->frags[0].page) +
2971 skb_shinfo(skb)->frags[0].page_offset;
2972 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2975 EXPORT_SYMBOL(skb_gro_reset_offset);
2977 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2979 skb_gro_reset_offset(skb);
2981 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2983 EXPORT_SYMBOL(napi_gro_receive);
2985 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2987 __skb_pull(skb, skb_headlen(skb));
2988 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2992 EXPORT_SYMBOL(napi_reuse_skb);
2994 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2996 struct sk_buff *skb = napi->skb;
2999 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3005 EXPORT_SYMBOL(napi_get_frags);
3007 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3013 skb->protocol = eth_type_trans(skb, skb->dev);
3015 if (ret == GRO_HELD)
3016 skb_gro_pull(skb, -ETH_HLEN);
3017 else if (netif_receive_skb(skb))
3022 case GRO_MERGED_FREE:
3023 napi_reuse_skb(napi, skb);
3032 EXPORT_SYMBOL(napi_frags_finish);
3034 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3036 struct sk_buff *skb = napi->skb;
3043 skb_reset_mac_header(skb);
3044 skb_gro_reset_offset(skb);
3046 off = skb_gro_offset(skb);
3047 hlen = off + sizeof(*eth);
3048 eth = skb_gro_header_fast(skb, off);
3049 if (skb_gro_header_hard(skb, hlen)) {
3050 eth = skb_gro_header_slow(skb, hlen, off);
3051 if (unlikely(!eth)) {
3052 napi_reuse_skb(napi, skb);
3058 skb_gro_pull(skb, sizeof(*eth));
3061 * This works because the only protocols we care about don't require
3062 * special handling. We'll fix it up properly at the end.
3064 skb->protocol = eth->h_proto;
3069 EXPORT_SYMBOL(napi_frags_skb);
3071 gro_result_t napi_gro_frags(struct napi_struct *napi)
3073 struct sk_buff *skb = napi_frags_skb(napi);
3078 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3080 EXPORT_SYMBOL(napi_gro_frags);
3082 static int process_backlog(struct napi_struct *napi, int quota)
3085 struct softnet_data *queue = &__get_cpu_var(softnet_data);
3086 unsigned long start_time = jiffies;
3088 napi->weight = weight_p;
3090 struct sk_buff *skb;
3092 spin_lock_irq(&queue->input_pkt_queue.lock);
3093 skb = __skb_dequeue(&queue->input_pkt_queue);
3095 __napi_complete(napi);
3096 spin_unlock_irq(&queue->input_pkt_queue.lock);
3099 spin_unlock_irq(&queue->input_pkt_queue.lock);
3101 __netif_receive_skb(skb);
3102 } while (++work < quota && jiffies == start_time);
3108 * __napi_schedule - schedule for receive
3109 * @n: entry to schedule
3111 * The entry's receive function will be scheduled to run
3113 void __napi_schedule(struct napi_struct *n)
3115 unsigned long flags;
3117 local_irq_save(flags);
3118 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
3119 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3120 local_irq_restore(flags);
3122 EXPORT_SYMBOL(__napi_schedule);
3124 void __napi_complete(struct napi_struct *n)
3126 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3127 BUG_ON(n->gro_list);
3129 list_del(&n->poll_list);
3130 smp_mb__before_clear_bit();
3131 clear_bit(NAPI_STATE_SCHED, &n->state);
3133 EXPORT_SYMBOL(__napi_complete);
3135 void napi_complete(struct napi_struct *n)
3137 unsigned long flags;
3140 * don't let napi dequeue from the cpu poll list
3141 * just in case its running on a different cpu
3143 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3147 local_irq_save(flags);
3149 local_irq_restore(flags);
3151 EXPORT_SYMBOL(napi_complete);
3153 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3154 int (*poll)(struct napi_struct *, int), int weight)
3156 INIT_LIST_HEAD(&napi->poll_list);
3157 napi->gro_count = 0;
3158 napi->gro_list = NULL;
3161 napi->weight = weight;
3162 list_add(&napi->dev_list, &dev->napi_list);
3164 #ifdef CONFIG_NETPOLL
3165 spin_lock_init(&napi->poll_lock);
3166 napi->poll_owner = -1;
3168 set_bit(NAPI_STATE_SCHED, &napi->state);
3170 EXPORT_SYMBOL(netif_napi_add);
3172 void netif_napi_del(struct napi_struct *napi)
3174 struct sk_buff *skb, *next;
3176 list_del_init(&napi->dev_list);
3177 napi_free_frags(napi);
3179 for (skb = napi->gro_list; skb; skb = next) {
3185 napi->gro_list = NULL;
3186 napi->gro_count = 0;
3188 EXPORT_SYMBOL(netif_napi_del);
3192 * net_rps_action sends any pending IPI's for rps. This is only called from
3193 * softirq and interrupts must be enabled.
3195 static void net_rps_action(cpumask_t *mask)
3199 /* Send pending IPI's to kick RPS processing on remote cpus. */
3200 for_each_cpu_mask_nr(cpu, *mask) {
3201 struct softnet_data *queue = &per_cpu(softnet_data, cpu);
3202 if (cpu_online(cpu))
3203 __smp_call_function_single(cpu, &queue->csd, 0);
3209 static void net_rx_action(struct softirq_action *h)
3211 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
3212 unsigned long time_limit = jiffies + 2;
3213 int budget = netdev_budget;
3217 struct rps_remote_softirq_cpus *rcpus;
3220 local_irq_disable();
3222 while (!list_empty(list)) {
3223 struct napi_struct *n;
3226 /* If softirq window is exhuasted then punt.
3227 * Allow this to run for 2 jiffies since which will allow
3228 * an average latency of 1.5/HZ.
3230 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3235 /* Even though interrupts have been re-enabled, this
3236 * access is safe because interrupts can only add new
3237 * entries to the tail of this list, and only ->poll()
3238 * calls can remove this head entry from the list.
3240 n = list_first_entry(list, struct napi_struct, poll_list);
3242 have = netpoll_poll_lock(n);
3246 /* This NAPI_STATE_SCHED test is for avoiding a race
3247 * with netpoll's poll_napi(). Only the entity which
3248 * obtains the lock and sees NAPI_STATE_SCHED set will
3249 * actually make the ->poll() call. Therefore we avoid
3250 * accidently calling ->poll() when NAPI is not scheduled.
3253 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3254 work = n->poll(n, weight);
3258 WARN_ON_ONCE(work > weight);
3262 local_irq_disable();
3264 /* Drivers must not modify the NAPI state if they
3265 * consume the entire weight. In such cases this code
3266 * still "owns" the NAPI instance and therefore can
3267 * move the instance around on the list at-will.
3269 if (unlikely(work == weight)) {
3270 if (unlikely(napi_disable_pending(n))) {
3273 local_irq_disable();
3275 list_move_tail(&n->poll_list, list);
3278 netpoll_poll_unlock(have);
3282 rcpus = &__get_cpu_var(rps_remote_softirq_cpus);
3283 select = rcpus->select;
3288 net_rps_action(&rcpus->mask[select]);
3293 #ifdef CONFIG_NET_DMA
3295 * There may not be any more sk_buffs coming right now, so push
3296 * any pending DMA copies to hardware
3298 dma_issue_pending_all();
3304 __get_cpu_var(netdev_rx_stat).time_squeeze++;
3305 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3309 static gifconf_func_t *gifconf_list[NPROTO];
3312 * register_gifconf - register a SIOCGIF handler
3313 * @family: Address family
3314 * @gifconf: Function handler
3316 * Register protocol dependent address dumping routines. The handler
3317 * that is passed must not be freed or reused until it has been replaced
3318 * by another handler.
3320 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3322 if (family >= NPROTO)
3324 gifconf_list[family] = gifconf;
3327 EXPORT_SYMBOL(register_gifconf);
3331 * Map an interface index to its name (SIOCGIFNAME)
3335 * We need this ioctl for efficient implementation of the
3336 * if_indextoname() function required by the IPv6 API. Without
3337 * it, we would have to search all the interfaces to find a
3341 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3343 struct net_device *dev;
3347 * Fetch the caller's info block.
3350 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3354 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3360 strcpy(ifr.ifr_name, dev->name);
3363 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3369 * Perform a SIOCGIFCONF call. This structure will change
3370 * size eventually, and there is nothing I can do about it.
3371 * Thus we will need a 'compatibility mode'.
3374 static int dev_ifconf(struct net *net, char __user *arg)
3377 struct net_device *dev;
3384 * Fetch the caller's info block.
3387 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3394 * Loop over the interfaces, and write an info block for each.
3398 for_each_netdev(net, dev) {
3399 for (i = 0; i < NPROTO; i++) {
3400 if (gifconf_list[i]) {
3403 done = gifconf_list[i](dev, NULL, 0);
3405 done = gifconf_list[i](dev, pos + total,
3415 * All done. Write the updated control block back to the caller.
3417 ifc.ifc_len = total;
3420 * Both BSD and Solaris return 0 here, so we do too.
3422 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3425 #ifdef CONFIG_PROC_FS
3427 * This is invoked by the /proc filesystem handler to display a device
3430 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3433 struct net *net = seq_file_net(seq);
3435 struct net_device *dev;
3439 return SEQ_START_TOKEN;
3442 for_each_netdev_rcu(net, dev)
3449 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3451 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3452 first_net_device(seq_file_net(seq)) :
3453 next_net_device((struct net_device *)v);
3456 return rcu_dereference(dev);
3459 void dev_seq_stop(struct seq_file *seq, void *v)
3465 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3467 const struct net_device_stats *stats = dev_get_stats(dev);
3469 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3470 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3471 dev->name, stats->rx_bytes, stats->rx_packets,
3473 stats->rx_dropped + stats->rx_missed_errors,
3474 stats->rx_fifo_errors,
3475 stats->rx_length_errors + stats->rx_over_errors +
3476 stats->rx_crc_errors + stats->rx_frame_errors,
3477 stats->rx_compressed, stats->multicast,
3478 stats->tx_bytes, stats->tx_packets,
3479 stats->tx_errors, stats->tx_dropped,
3480 stats->tx_fifo_errors, stats->collisions,
3481 stats->tx_carrier_errors +
3482 stats->tx_aborted_errors +
3483 stats->tx_window_errors +
3484 stats->tx_heartbeat_errors,
3485 stats->tx_compressed);
3489 * Called from the PROCfs module. This now uses the new arbitrary sized
3490 * /proc/net interface to create /proc/net/dev
3492 static int dev_seq_show(struct seq_file *seq, void *v)
3494 if (v == SEQ_START_TOKEN)
3495 seq_puts(seq, "Inter-| Receive "
3497 " face |bytes packets errs drop fifo frame "
3498 "compressed multicast|bytes packets errs "
3499 "drop fifo colls carrier compressed\n");
3501 dev_seq_printf_stats(seq, v);
3505 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3507 struct netif_rx_stats *rc = NULL;
3509 while (*pos < nr_cpu_ids)
3510 if (cpu_online(*pos)) {
3511 rc = &per_cpu(netdev_rx_stat, *pos);
3518 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3520 return softnet_get_online(pos);
3523 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3526 return softnet_get_online(pos);
3529 static void softnet_seq_stop(struct seq_file *seq, void *v)
3533 static int softnet_seq_show(struct seq_file *seq, void *v)
3535 struct netif_rx_stats *s = v;
3537 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3538 s->total, s->dropped, s->time_squeeze, 0,
3539 0, 0, 0, 0, /* was fastroute */
3540 s->cpu_collision, s->received_rps);
3544 static const struct seq_operations dev_seq_ops = {
3545 .start = dev_seq_start,
3546 .next = dev_seq_next,
3547 .stop = dev_seq_stop,
3548 .show = dev_seq_show,
3551 static int dev_seq_open(struct inode *inode, struct file *file)
3553 return seq_open_net(inode, file, &dev_seq_ops,
3554 sizeof(struct seq_net_private));
3557 static const struct file_operations dev_seq_fops = {
3558 .owner = THIS_MODULE,
3559 .open = dev_seq_open,
3561 .llseek = seq_lseek,
3562 .release = seq_release_net,
3565 static const struct seq_operations softnet_seq_ops = {
3566 .start = softnet_seq_start,
3567 .next = softnet_seq_next,
3568 .stop = softnet_seq_stop,
3569 .show = softnet_seq_show,
3572 static int softnet_seq_open(struct inode *inode, struct file *file)
3574 return seq_open(file, &softnet_seq_ops);
3577 static const struct file_operations softnet_seq_fops = {
3578 .owner = THIS_MODULE,
3579 .open = softnet_seq_open,
3581 .llseek = seq_lseek,
3582 .release = seq_release,
3585 static void *ptype_get_idx(loff_t pos)
3587 struct packet_type *pt = NULL;
3591 list_for_each_entry_rcu(pt, &ptype_all, list) {
3597 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3598 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3607 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3611 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3614 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3616 struct packet_type *pt;
3617 struct list_head *nxt;
3621 if (v == SEQ_START_TOKEN)
3622 return ptype_get_idx(0);
3625 nxt = pt->list.next;
3626 if (pt->type == htons(ETH_P_ALL)) {
3627 if (nxt != &ptype_all)
3630 nxt = ptype_base[0].next;
3632 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3634 while (nxt == &ptype_base[hash]) {
3635 if (++hash >= PTYPE_HASH_SIZE)
3637 nxt = ptype_base[hash].next;
3640 return list_entry(nxt, struct packet_type, list);
3643 static void ptype_seq_stop(struct seq_file *seq, void *v)
3649 static int ptype_seq_show(struct seq_file *seq, void *v)
3651 struct packet_type *pt = v;
3653 if (v == SEQ_START_TOKEN)
3654 seq_puts(seq, "Type Device Function\n");
3655 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3656 if (pt->type == htons(ETH_P_ALL))
3657 seq_puts(seq, "ALL ");
3659 seq_printf(seq, "%04x", ntohs(pt->type));
3661 seq_printf(seq, " %-8s %pF\n",
3662 pt->dev ? pt->dev->name : "", pt->func);
3668 static const struct seq_operations ptype_seq_ops = {
3669 .start = ptype_seq_start,
3670 .next = ptype_seq_next,
3671 .stop = ptype_seq_stop,
3672 .show = ptype_seq_show,
3675 static int ptype_seq_open(struct inode *inode, struct file *file)
3677 return seq_open_net(inode, file, &ptype_seq_ops,
3678 sizeof(struct seq_net_private));
3681 static const struct file_operations ptype_seq_fops = {
3682 .owner = THIS_MODULE,
3683 .open = ptype_seq_open,
3685 .llseek = seq_lseek,
3686 .release = seq_release_net,
3690 static int __net_init dev_proc_net_init(struct net *net)
3694 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3696 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3698 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3701 if (wext_proc_init(net))
3707 proc_net_remove(net, "ptype");
3709 proc_net_remove(net, "softnet_stat");
3711 proc_net_remove(net, "dev");
3715 static void __net_exit dev_proc_net_exit(struct net *net)
3717 wext_proc_exit(net);
3719 proc_net_remove(net, "ptype");
3720 proc_net_remove(net, "softnet_stat");
3721 proc_net_remove(net, "dev");
3724 static struct pernet_operations __net_initdata dev_proc_ops = {
3725 .init = dev_proc_net_init,
3726 .exit = dev_proc_net_exit,
3729 static int __init dev_proc_init(void)
3731 return register_pernet_subsys(&dev_proc_ops);
3734 #define dev_proc_init() 0
3735 #endif /* CONFIG_PROC_FS */
3739 * netdev_set_master - set up master/slave pair
3740 * @slave: slave device
3741 * @master: new master device
3743 * Changes the master device of the slave. Pass %NULL to break the
3744 * bonding. The caller must hold the RTNL semaphore. On a failure
3745 * a negative errno code is returned. On success the reference counts
3746 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3747 * function returns zero.
3749 int netdev_set_master(struct net_device *slave, struct net_device *master)
3751 struct net_device *old = slave->master;
3761 slave->master = master;
3768 slave->flags |= IFF_SLAVE;
3770 slave->flags &= ~IFF_SLAVE;
3772 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3775 EXPORT_SYMBOL(netdev_set_master);
3777 static void dev_change_rx_flags(struct net_device *dev, int flags)
3779 const struct net_device_ops *ops = dev->netdev_ops;
3781 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3782 ops->ndo_change_rx_flags(dev, flags);
3785 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3787 unsigned short old_flags = dev->flags;
3793 dev->flags |= IFF_PROMISC;
3794 dev->promiscuity += inc;
3795 if (dev->promiscuity == 0) {
3798 * If inc causes overflow, untouch promisc and return error.
3801 dev->flags &= ~IFF_PROMISC;
3803 dev->promiscuity -= inc;
3804 printk(KERN_WARNING "%s: promiscuity touches roof, "
3805 "set promiscuity failed, promiscuity feature "
3806 "of device might be broken.\n", dev->name);
3810 if (dev->flags != old_flags) {
3811 printk(KERN_INFO "device %s %s promiscuous mode\n",
3812 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3814 if (audit_enabled) {
3815 current_uid_gid(&uid, &gid);
3816 audit_log(current->audit_context, GFP_ATOMIC,
3817 AUDIT_ANOM_PROMISCUOUS,
3818 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3819 dev->name, (dev->flags & IFF_PROMISC),
3820 (old_flags & IFF_PROMISC),
3821 audit_get_loginuid(current),
3823 audit_get_sessionid(current));
3826 dev_change_rx_flags(dev, IFF_PROMISC);
3832 * dev_set_promiscuity - update promiscuity count on a device
3836 * Add or remove promiscuity from a device. While the count in the device
3837 * remains above zero the interface remains promiscuous. Once it hits zero
3838 * the device reverts back to normal filtering operation. A negative inc
3839 * value is used to drop promiscuity on the device.
3840 * Return 0 if successful or a negative errno code on error.
3842 int dev_set_promiscuity(struct net_device *dev, int inc)
3844 unsigned short old_flags = dev->flags;
3847 err = __dev_set_promiscuity(dev, inc);
3850 if (dev->flags != old_flags)
3851 dev_set_rx_mode(dev);
3854 EXPORT_SYMBOL(dev_set_promiscuity);
3857 * dev_set_allmulti - update allmulti count on a device
3861 * Add or remove reception of all multicast frames to a device. While the
3862 * count in the device remains above zero the interface remains listening
3863 * to all interfaces. Once it hits zero the device reverts back to normal
3864 * filtering operation. A negative @inc value is used to drop the counter
3865 * when releasing a resource needing all multicasts.
3866 * Return 0 if successful or a negative errno code on error.
3869 int dev_set_allmulti(struct net_device *dev, int inc)
3871 unsigned short old_flags = dev->flags;
3875 dev->flags |= IFF_ALLMULTI;
3876 dev->allmulti += inc;
3877 if (dev->allmulti == 0) {
3880 * If inc causes overflow, untouch allmulti and return error.
3883 dev->flags &= ~IFF_ALLMULTI;
3885 dev->allmulti -= inc;
3886 printk(KERN_WARNING "%s: allmulti touches roof, "
3887 "set allmulti failed, allmulti feature of "
3888 "device might be broken.\n", dev->name);
3892 if (dev->flags ^ old_flags) {
3893 dev_change_rx_flags(dev, IFF_ALLMULTI);
3894 dev_set_rx_mode(dev);
3898 EXPORT_SYMBOL(dev_set_allmulti);
3901 * Upload unicast and multicast address lists to device and
3902 * configure RX filtering. When the device doesn't support unicast
3903 * filtering it is put in promiscuous mode while unicast addresses
3906 void __dev_set_rx_mode(struct net_device *dev)
3908 const struct net_device_ops *ops = dev->netdev_ops;
3910 /* dev_open will call this function so the list will stay sane. */
3911 if (!(dev->flags&IFF_UP))
3914 if (!netif_device_present(dev))
3917 if (ops->ndo_set_rx_mode)
3918 ops->ndo_set_rx_mode(dev);
3920 /* Unicast addresses changes may only happen under the rtnl,
3921 * therefore calling __dev_set_promiscuity here is safe.
3923 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
3924 __dev_set_promiscuity(dev, 1);
3925 dev->uc_promisc = 1;
3926 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
3927 __dev_set_promiscuity(dev, -1);
3928 dev->uc_promisc = 0;
3931 if (ops->ndo_set_multicast_list)
3932 ops->ndo_set_multicast_list(dev);
3936 void dev_set_rx_mode(struct net_device *dev)
3938 netif_addr_lock_bh(dev);
3939 __dev_set_rx_mode(dev);
3940 netif_addr_unlock_bh(dev);
3943 /* hw addresses list handling functions */
3945 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3946 int addr_len, unsigned char addr_type)
3948 struct netdev_hw_addr *ha;
3951 if (addr_len > MAX_ADDR_LEN)
3954 list_for_each_entry(ha, &list->list, list) {
3955 if (!memcmp(ha->addr, addr, addr_len) &&
3956 ha->type == addr_type) {
3963 alloc_size = sizeof(*ha);
3964 if (alloc_size < L1_CACHE_BYTES)
3965 alloc_size = L1_CACHE_BYTES;
3966 ha = kmalloc(alloc_size, GFP_ATOMIC);
3969 memcpy(ha->addr, addr, addr_len);
3970 ha->type = addr_type;
3973 list_add_tail_rcu(&ha->list, &list->list);
3978 static void ha_rcu_free(struct rcu_head *head)
3980 struct netdev_hw_addr *ha;
3982 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3986 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3987 int addr_len, unsigned char addr_type)
3989 struct netdev_hw_addr *ha;
3991 list_for_each_entry(ha, &list->list, list) {
3992 if (!memcmp(ha->addr, addr, addr_len) &&
3993 (ha->type == addr_type || !addr_type)) {
3996 list_del_rcu(&ha->list);
3997 call_rcu(&ha->rcu_head, ha_rcu_free);
4005 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
4006 struct netdev_hw_addr_list *from_list,
4008 unsigned char addr_type)
4011 struct netdev_hw_addr *ha, *ha2;
4014 list_for_each_entry(ha, &from_list->list, list) {
4015 type = addr_type ? addr_type : ha->type;
4016 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
4023 list_for_each_entry(ha2, &from_list->list, list) {
4026 type = addr_type ? addr_type : ha2->type;
4027 __hw_addr_del(to_list, ha2->addr, addr_len, type);
4032 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
4033 struct netdev_hw_addr_list *from_list,
4035 unsigned char addr_type)
4037 struct netdev_hw_addr *ha;
4040 list_for_each_entry(ha, &from_list->list, list) {
4041 type = addr_type ? addr_type : ha->type;
4042 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
4046 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
4047 struct netdev_hw_addr_list *from_list,
4051 struct netdev_hw_addr *ha, *tmp;
4053 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
4055 err = __hw_addr_add(to_list, ha->addr,
4056 addr_len, ha->type);
4061 } else if (ha->refcount == 1) {
4062 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
4063 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
4069 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
4070 struct netdev_hw_addr_list *from_list,
4073 struct netdev_hw_addr *ha, *tmp;
4075 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
4077 __hw_addr_del(to_list, ha->addr,
4078 addr_len, ha->type);
4080 __hw_addr_del(from_list, ha->addr,
4081 addr_len, ha->type);
4086 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
4088 struct netdev_hw_addr *ha, *tmp;
4090 list_for_each_entry_safe(ha, tmp, &list->list, list) {
4091 list_del_rcu(&ha->list);
4092 call_rcu(&ha->rcu_head, ha_rcu_free);
4097 static void __hw_addr_init(struct netdev_hw_addr_list *list)
4099 INIT_LIST_HEAD(&list->list);
4103 /* Device addresses handling functions */
4105 static void dev_addr_flush(struct net_device *dev)
4107 /* rtnl_mutex must be held here */
4109 __hw_addr_flush(&dev->dev_addrs);
4110 dev->dev_addr = NULL;
4113 static int dev_addr_init(struct net_device *dev)
4115 unsigned char addr[MAX_ADDR_LEN];
4116 struct netdev_hw_addr *ha;
4119 /* rtnl_mutex must be held here */
4121 __hw_addr_init(&dev->dev_addrs);
4122 memset(addr, 0, sizeof(addr));
4123 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
4124 NETDEV_HW_ADDR_T_LAN);
4127 * Get the first (previously created) address from the list
4128 * and set dev_addr pointer to this location.
4130 ha = list_first_entry(&dev->dev_addrs.list,
4131 struct netdev_hw_addr, list);
4132 dev->dev_addr = ha->addr;
4138 * dev_addr_add - Add a device address
4140 * @addr: address to add
4141 * @addr_type: address type
4143 * Add a device address to the device or increase the reference count if
4144 * it already exists.
4146 * The caller must hold the rtnl_mutex.
4148 int dev_addr_add(struct net_device *dev, unsigned char *addr,
4149 unsigned char addr_type)
4155 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
4157 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4160 EXPORT_SYMBOL(dev_addr_add);
4163 * dev_addr_del - Release a device address.
4165 * @addr: address to delete
4166 * @addr_type: address type
4168 * Release reference to a device address and remove it from the device
4169 * if the reference count drops to zero.
4171 * The caller must hold the rtnl_mutex.
4173 int dev_addr_del(struct net_device *dev, unsigned char *addr,
4174 unsigned char addr_type)
4177 struct netdev_hw_addr *ha;
4182 * We can not remove the first address from the list because
4183 * dev->dev_addr points to that.
4185 ha = list_first_entry(&dev->dev_addrs.list,
4186 struct netdev_hw_addr, list);
4187 if (ha->addr == dev->dev_addr && ha->refcount == 1)
4190 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
4193 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4196 EXPORT_SYMBOL(dev_addr_del);
4199 * dev_addr_add_multiple - Add device addresses from another device
4200 * @to_dev: device to which addresses will be added
4201 * @from_dev: device from which addresses will be added
4202 * @addr_type: address type - 0 means type will be used from from_dev
4204 * Add device addresses of the one device to another.
4206 * The caller must hold the rtnl_mutex.
4208 int dev_addr_add_multiple(struct net_device *to_dev,
4209 struct net_device *from_dev,
4210 unsigned char addr_type)
4216 if (from_dev->addr_len != to_dev->addr_len)
4218 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
4219 to_dev->addr_len, addr_type);
4221 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
4224 EXPORT_SYMBOL(dev_addr_add_multiple);
4227 * dev_addr_del_multiple - Delete device addresses by another device
4228 * @to_dev: device where the addresses will be deleted
4229 * @from_dev: device by which addresses the addresses will be deleted
4230 * @addr_type: address type - 0 means type will used from from_dev
4232 * Deletes addresses in to device by the list of addresses in from device.
4234 * The caller must hold the rtnl_mutex.
4236 int dev_addr_del_multiple(struct net_device *to_dev,
4237 struct net_device *from_dev,
4238 unsigned char addr_type)
4242 if (from_dev->addr_len != to_dev->addr_len)
4244 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
4245 to_dev->addr_len, addr_type);
4246 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
4249 EXPORT_SYMBOL(dev_addr_del_multiple);
4251 /* multicast addresses handling functions */
4253 int __dev_addr_delete(struct dev_addr_list **list, int *count,
4254 void *addr, int alen, int glbl)
4256 struct dev_addr_list *da;
4258 for (; (da = *list) != NULL; list = &da->next) {
4259 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
4260 alen == da->da_addrlen) {
4262 int old_glbl = da->da_gusers;
4279 int __dev_addr_add(struct dev_addr_list **list, int *count,
4280 void *addr, int alen, int glbl)
4282 struct dev_addr_list *da;
4284 for (da = *list; da != NULL; da = da->next) {
4285 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
4286 da->da_addrlen == alen) {
4288 int old_glbl = da->da_gusers;
4298 da = kzalloc(sizeof(*da), GFP_ATOMIC);
4301 memcpy(da->da_addr, addr, alen);
4302 da->da_addrlen = alen;
4304 da->da_gusers = glbl ? 1 : 0;
4312 * dev_unicast_delete - Release secondary unicast address.
4314 * @addr: address to delete
4316 * Release reference to a secondary unicast address and remove it
4317 * from the device if the reference count drops to zero.
4319 * The caller must hold the rtnl_mutex.
4321 int dev_unicast_delete(struct net_device *dev, void *addr)
4327 netif_addr_lock_bh(dev);
4328 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
4329 NETDEV_HW_ADDR_T_UNICAST);
4331 __dev_set_rx_mode(dev);
4332 netif_addr_unlock_bh(dev);
4335 EXPORT_SYMBOL(dev_unicast_delete);
4338 * dev_unicast_add - add a secondary unicast address
4340 * @addr: address to add
4342 * Add a secondary unicast address to the device or increase
4343 * the reference count if it already exists.
4345 * The caller must hold the rtnl_mutex.
4347 int dev_unicast_add(struct net_device *dev, void *addr)
4353 netif_addr_lock_bh(dev);
4354 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
4355 NETDEV_HW_ADDR_T_UNICAST);
4357 __dev_set_rx_mode(dev);
4358 netif_addr_unlock_bh(dev);
4361 EXPORT_SYMBOL(dev_unicast_add);
4363 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
4364 struct dev_addr_list **from, int *from_count)
4366 struct dev_addr_list *da, *next;
4370 while (da != NULL) {
4372 if (!da->da_synced) {
4373 err = __dev_addr_add(to, to_count,
4374 da->da_addr, da->da_addrlen, 0);
4379 } else if (da->da_users == 1) {
4380 __dev_addr_delete(to, to_count,
4381 da->da_addr, da->da_addrlen, 0);
4382 __dev_addr_delete(from, from_count,
4383 da->da_addr, da->da_addrlen, 0);
4389 EXPORT_SYMBOL_GPL(__dev_addr_sync);
4391 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4392 struct dev_addr_list **from, int *from_count)
4394 struct dev_addr_list *da, *next;
4397 while (da != NULL) {
4399 if (da->da_synced) {
4400 __dev_addr_delete(to, to_count,
4401 da->da_addr, da->da_addrlen, 0);
4403 __dev_addr_delete(from, from_count,
4404 da->da_addr, da->da_addrlen, 0);
4409 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4412 * dev_unicast_sync - Synchronize device's unicast list to another device
4413 * @to: destination device
4414 * @from: source device
4416 * Add newly added addresses to the destination device and release
4417 * addresses that have no users left. The source device must be
4418 * locked by netif_tx_lock_bh.
4420 * This function is intended to be called from the dev->set_rx_mode
4421 * function of layered software devices.
4423 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4427 if (to->addr_len != from->addr_len)
4430 netif_addr_lock_bh(to);
4431 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4433 __dev_set_rx_mode(to);
4434 netif_addr_unlock_bh(to);
4437 EXPORT_SYMBOL(dev_unicast_sync);
4440 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4441 * @to: destination device
4442 * @from: source device
4444 * Remove all addresses that were added to the destination device by
4445 * dev_unicast_sync(). This function is intended to be called from the
4446 * dev->stop function of layered software devices.
4448 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4450 if (to->addr_len != from->addr_len)
4453 netif_addr_lock_bh(from);
4454 netif_addr_lock(to);
4455 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4456 __dev_set_rx_mode(to);
4457 netif_addr_unlock(to);
4458 netif_addr_unlock_bh(from);
4460 EXPORT_SYMBOL(dev_unicast_unsync);
4462 void dev_unicast_flush(struct net_device *dev)
4464 netif_addr_lock_bh(dev);
4465 __hw_addr_flush(&dev->uc);
4466 netif_addr_unlock_bh(dev);
4468 EXPORT_SYMBOL(dev_unicast_flush);
4470 static void dev_unicast_init(struct net_device *dev)
4472 __hw_addr_init(&dev->uc);
4476 static void __dev_addr_discard(struct dev_addr_list **list)
4478 struct dev_addr_list *tmp;
4480 while (*list != NULL) {
4483 if (tmp->da_users > tmp->da_gusers)
4484 printk("__dev_addr_discard: address leakage! "
4485 "da_users=%d\n", tmp->da_users);
4490 void dev_addr_discard(struct net_device *dev)
4492 netif_addr_lock_bh(dev);
4494 __dev_addr_discard(&dev->mc_list);
4495 netdev_mc_count(dev) = 0;
4497 netif_addr_unlock_bh(dev);
4499 EXPORT_SYMBOL(dev_addr_discard);
4502 * dev_get_flags - get flags reported to userspace
4505 * Get the combination of flag bits exported through APIs to userspace.
4507 unsigned dev_get_flags(const struct net_device *dev)
4511 flags = (dev->flags & ~(IFF_PROMISC |
4516 (dev->gflags & (IFF_PROMISC |
4519 if (netif_running(dev)) {
4520 if (netif_oper_up(dev))
4521 flags |= IFF_RUNNING;
4522 if (netif_carrier_ok(dev))
4523 flags |= IFF_LOWER_UP;
4524 if (netif_dormant(dev))
4525 flags |= IFF_DORMANT;
4530 EXPORT_SYMBOL(dev_get_flags);
4532 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4534 int old_flags = dev->flags;
4540 * Set the flags on our device.
4543 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4544 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4546 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4550 * Load in the correct multicast list now the flags have changed.
4553 if ((old_flags ^ flags) & IFF_MULTICAST)
4554 dev_change_rx_flags(dev, IFF_MULTICAST);
4556 dev_set_rx_mode(dev);
4559 * Have we downed the interface. We handle IFF_UP ourselves
4560 * according to user attempts to set it, rather than blindly
4565 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4566 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4569 dev_set_rx_mode(dev);
4572 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4573 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4575 dev->gflags ^= IFF_PROMISC;
4576 dev_set_promiscuity(dev, inc);
4579 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4580 is important. Some (broken) drivers set IFF_PROMISC, when
4581 IFF_ALLMULTI is requested not asking us and not reporting.
4583 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4584 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4586 dev->gflags ^= IFF_ALLMULTI;
4587 dev_set_allmulti(dev, inc);
4593 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4595 unsigned int changes = dev->flags ^ old_flags;
4597 if (changes & IFF_UP) {
4598 if (dev->flags & IFF_UP)
4599 call_netdevice_notifiers(NETDEV_UP, dev);
4601 call_netdevice_notifiers(NETDEV_DOWN, dev);
4604 if (dev->flags & IFF_UP &&
4605 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4606 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4610 * dev_change_flags - change device settings
4612 * @flags: device state flags
4614 * Change settings on device based state flags. The flags are
4615 * in the userspace exported format.
4617 int dev_change_flags(struct net_device *dev, unsigned flags)
4620 int old_flags = dev->flags;
4622 ret = __dev_change_flags(dev, flags);
4626 changes = old_flags ^ dev->flags;
4628 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4630 __dev_notify_flags(dev, old_flags);
4633 EXPORT_SYMBOL(dev_change_flags);
4636 * dev_set_mtu - Change maximum transfer unit
4638 * @new_mtu: new transfer unit
4640 * Change the maximum transfer size of the network device.
4642 int dev_set_mtu(struct net_device *dev, int new_mtu)
4644 const struct net_device_ops *ops = dev->netdev_ops;
4647 if (new_mtu == dev->mtu)
4650 /* MTU must be positive. */
4654 if (!netif_device_present(dev))
4658 if (ops->ndo_change_mtu)
4659 err = ops->ndo_change_mtu(dev, new_mtu);
4663 if (!err && dev->flags & IFF_UP)
4664 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4667 EXPORT_SYMBOL(dev_set_mtu);
4670 * dev_set_mac_address - Change Media Access Control Address
4674 * Change the hardware (MAC) address of the device
4676 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4678 const struct net_device_ops *ops = dev->netdev_ops;
4681 if (!ops->ndo_set_mac_address)
4683 if (sa->sa_family != dev->type)
4685 if (!netif_device_present(dev))
4687 err = ops->ndo_set_mac_address(dev, sa);
4689 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4692 EXPORT_SYMBOL(dev_set_mac_address);
4695 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4697 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4700 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4706 case SIOCGIFFLAGS: /* Get interface flags */
4707 ifr->ifr_flags = (short) dev_get_flags(dev);
4710 case SIOCGIFMETRIC: /* Get the metric on the interface
4711 (currently unused) */
4712 ifr->ifr_metric = 0;
4715 case SIOCGIFMTU: /* Get the MTU of a device */
4716 ifr->ifr_mtu = dev->mtu;
4721 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4723 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4724 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4725 ifr->ifr_hwaddr.sa_family = dev->type;
4733 ifr->ifr_map.mem_start = dev->mem_start;
4734 ifr->ifr_map.mem_end = dev->mem_end;
4735 ifr->ifr_map.base_addr = dev->base_addr;
4736 ifr->ifr_map.irq = dev->irq;
4737 ifr->ifr_map.dma = dev->dma;
4738 ifr->ifr_map.port = dev->if_port;
4742 ifr->ifr_ifindex = dev->ifindex;
4746 ifr->ifr_qlen = dev->tx_queue_len;
4750 /* dev_ioctl() should ensure this case
4762 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4764 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4767 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4768 const struct net_device_ops *ops;
4773 ops = dev->netdev_ops;
4776 case SIOCSIFFLAGS: /* Set interface flags */
4777 return dev_change_flags(dev, ifr->ifr_flags);
4779 case SIOCSIFMETRIC: /* Set the metric on the interface
4780 (currently unused) */
4783 case SIOCSIFMTU: /* Set the MTU of a device */
4784 return dev_set_mtu(dev, ifr->ifr_mtu);
4787 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4789 case SIOCSIFHWBROADCAST:
4790 if (ifr->ifr_hwaddr.sa_family != dev->type)
4792 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4793 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4794 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4798 if (ops->ndo_set_config) {
4799 if (!netif_device_present(dev))
4801 return ops->ndo_set_config(dev, &ifr->ifr_map);
4806 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4807 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4809 if (!netif_device_present(dev))
4811 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4815 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4816 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4818 if (!netif_device_present(dev))
4820 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4824 if (ifr->ifr_qlen < 0)
4826 dev->tx_queue_len = ifr->ifr_qlen;
4830 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4831 return dev_change_name(dev, ifr->ifr_newname);
4834 * Unknown or private ioctl
4837 if ((cmd >= SIOCDEVPRIVATE &&
4838 cmd <= SIOCDEVPRIVATE + 15) ||
4839 cmd == SIOCBONDENSLAVE ||
4840 cmd == SIOCBONDRELEASE ||
4841 cmd == SIOCBONDSETHWADDR ||
4842 cmd == SIOCBONDSLAVEINFOQUERY ||
4843 cmd == SIOCBONDINFOQUERY ||
4844 cmd == SIOCBONDCHANGEACTIVE ||
4845 cmd == SIOCGMIIPHY ||
4846 cmd == SIOCGMIIREG ||
4847 cmd == SIOCSMIIREG ||
4848 cmd == SIOCBRADDIF ||
4849 cmd == SIOCBRDELIF ||
4850 cmd == SIOCSHWTSTAMP ||
4851 cmd == SIOCWANDEV) {
4853 if (ops->ndo_do_ioctl) {
4854 if (netif_device_present(dev))
4855 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4867 * This function handles all "interface"-type I/O control requests. The actual
4868 * 'doing' part of this is dev_ifsioc above.
4872 * dev_ioctl - network device ioctl
4873 * @net: the applicable net namespace
4874 * @cmd: command to issue
4875 * @arg: pointer to a struct ifreq in user space
4877 * Issue ioctl functions to devices. This is normally called by the
4878 * user space syscall interfaces but can sometimes be useful for
4879 * other purposes. The return value is the return from the syscall if
4880 * positive or a negative errno code on error.
4883 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4889 /* One special case: SIOCGIFCONF takes ifconf argument
4890 and requires shared lock, because it sleeps writing
4894 if (cmd == SIOCGIFCONF) {
4896 ret = dev_ifconf(net, (char __user *) arg);
4900 if (cmd == SIOCGIFNAME)
4901 return dev_ifname(net, (struct ifreq __user *)arg);
4903 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4906 ifr.ifr_name[IFNAMSIZ-1] = 0;
4908 colon = strchr(ifr.ifr_name, ':');
4913 * See which interface the caller is talking about.
4918 * These ioctl calls:
4919 * - can be done by all.
4920 * - atomic and do not require locking.
4931 dev_load(net, ifr.ifr_name);
4933 ret = dev_ifsioc_locked(net, &ifr, cmd);
4938 if (copy_to_user(arg, &ifr,
4939 sizeof(struct ifreq)))
4945 dev_load(net, ifr.ifr_name);
4947 ret = dev_ethtool(net, &ifr);
4952 if (copy_to_user(arg, &ifr,
4953 sizeof(struct ifreq)))
4959 * These ioctl calls:
4960 * - require superuser power.
4961 * - require strict serialization.
4967 if (!capable(CAP_NET_ADMIN))
4969 dev_load(net, ifr.ifr_name);
4971 ret = dev_ifsioc(net, &ifr, cmd);
4976 if (copy_to_user(arg, &ifr,
4977 sizeof(struct ifreq)))
4983 * These ioctl calls:
4984 * - require superuser power.
4985 * - require strict serialization.
4986 * - do not return a value
4996 case SIOCSIFHWBROADCAST:
4999 case SIOCBONDENSLAVE:
5000 case SIOCBONDRELEASE:
5001 case SIOCBONDSETHWADDR:
5002 case SIOCBONDCHANGEACTIVE:
5006 if (!capable(CAP_NET_ADMIN))
5009 case SIOCBONDSLAVEINFOQUERY:
5010 case SIOCBONDINFOQUERY:
5011 dev_load(net, ifr.ifr_name);
5013 ret = dev_ifsioc(net, &ifr, cmd);
5018 /* Get the per device memory space. We can add this but
5019 * currently do not support it */
5021 /* Set the per device memory buffer space.
5022 * Not applicable in our case */
5027 * Unknown or private ioctl.
5030 if (cmd == SIOCWANDEV ||
5031 (cmd >= SIOCDEVPRIVATE &&
5032 cmd <= SIOCDEVPRIVATE + 15)) {
5033 dev_load(net, ifr.ifr_name);
5035 ret = dev_ifsioc(net, &ifr, cmd);
5037 if (!ret && copy_to_user(arg, &ifr,
5038 sizeof(struct ifreq)))
5042 /* Take care of Wireless Extensions */
5043 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5044 return wext_handle_ioctl(net, &ifr, cmd, arg);
5051 * dev_new_index - allocate an ifindex
5052 * @net: the applicable net namespace
5054 * Returns a suitable unique value for a new device interface
5055 * number. The caller must hold the rtnl semaphore or the
5056 * dev_base_lock to be sure it remains unique.
5058 static int dev_new_index(struct net *net)
5064 if (!__dev_get_by_index(net, ifindex))
5069 /* Delayed registration/unregisteration */
5070 static LIST_HEAD(net_todo_list);
5072 static void net_set_todo(struct net_device *dev)
5074 list_add_tail(&dev->todo_list, &net_todo_list);
5077 static void rollback_registered_many(struct list_head *head)
5079 struct net_device *dev, *tmp;
5081 BUG_ON(dev_boot_phase);
5084 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5085 /* Some devices call without registering
5086 * for initialization unwind. Remove those
5087 * devices and proceed with the remaining.
5089 if (dev->reg_state == NETREG_UNINITIALIZED) {
5090 pr_debug("unregister_netdevice: device %s/%p never "
5091 "was registered\n", dev->name, dev);
5094 list_del(&dev->unreg_list);
5098 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5100 /* If device is running, close it first. */
5103 /* And unlink it from device chain. */
5104 unlist_netdevice(dev);
5106 dev->reg_state = NETREG_UNREGISTERING;
5111 list_for_each_entry(dev, head, unreg_list) {
5112 /* Shutdown queueing discipline. */
5116 /* Notify protocols, that we are about to destroy
5117 this device. They should clean all the things.
5119 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5121 if (!dev->rtnl_link_ops ||
5122 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5123 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5126 * Flush the unicast and multicast chains
5128 dev_unicast_flush(dev);
5129 dev_addr_discard(dev);
5131 if (dev->netdev_ops->ndo_uninit)
5132 dev->netdev_ops->ndo_uninit(dev);
5134 /* Notifier chain MUST detach us from master device. */
5135 WARN_ON(dev->master);
5137 /* Remove entries from kobject tree */
5138 netdev_unregister_kobject(dev);
5141 /* Process any work delayed until the end of the batch */
5142 dev = list_first_entry(head, struct net_device, unreg_list);
5143 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5147 list_for_each_entry(dev, head, unreg_list)
5151 static void rollback_registered(struct net_device *dev)
5155 list_add(&dev->unreg_list, &single);
5156 rollback_registered_many(&single);
5159 static void __netdev_init_queue_locks_one(struct net_device *dev,
5160 struct netdev_queue *dev_queue,
5163 spin_lock_init(&dev_queue->_xmit_lock);
5164 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
5165 dev_queue->xmit_lock_owner = -1;
5168 static void netdev_init_queue_locks(struct net_device *dev)
5170 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
5171 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
5174 unsigned long netdev_fix_features(unsigned long features, const char *name)
5176 /* Fix illegal SG+CSUM combinations. */
5177 if ((features & NETIF_F_SG) &&
5178 !(features & NETIF_F_ALL_CSUM)) {
5180 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
5181 "checksum feature.\n", name);
5182 features &= ~NETIF_F_SG;
5185 /* TSO requires that SG is present as well. */
5186 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
5188 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
5189 "SG feature.\n", name);
5190 features &= ~NETIF_F_TSO;
5193 if (features & NETIF_F_UFO) {
5194 if (!(features & NETIF_F_GEN_CSUM)) {
5196 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5197 "since no NETIF_F_HW_CSUM feature.\n",
5199 features &= ~NETIF_F_UFO;
5202 if (!(features & NETIF_F_SG)) {
5204 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5205 "since no NETIF_F_SG feature.\n", name);
5206 features &= ~NETIF_F_UFO;
5212 EXPORT_SYMBOL(netdev_fix_features);
5215 * netif_stacked_transfer_operstate - transfer operstate
5216 * @rootdev: the root or lower level device to transfer state from
5217 * @dev: the device to transfer operstate to
5219 * Transfer operational state from root to device. This is normally
5220 * called when a stacking relationship exists between the root
5221 * device and the device(a leaf device).
5223 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5224 struct net_device *dev)
5226 if (rootdev->operstate == IF_OPER_DORMANT)
5227 netif_dormant_on(dev);
5229 netif_dormant_off(dev);
5231 if (netif_carrier_ok(rootdev)) {
5232 if (!netif_carrier_ok(dev))
5233 netif_carrier_on(dev);
5235 if (netif_carrier_ok(dev))
5236 netif_carrier_off(dev);
5239 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5242 * register_netdevice - register a network device
5243 * @dev: device to register
5245 * Take a completed network device structure and add it to the kernel
5246 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5247 * chain. 0 is returned on success. A negative errno code is returned
5248 * on a failure to set up the device, or if the name is a duplicate.
5250 * Callers must hold the rtnl semaphore. You may want
5251 * register_netdev() instead of this.
5254 * The locking appears insufficient to guarantee two parallel registers
5255 * will not get the same name.
5258 int register_netdevice(struct net_device *dev)
5261 struct net *net = dev_net(dev);
5263 BUG_ON(dev_boot_phase);
5268 /* When net_device's are persistent, this will be fatal. */
5269 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5272 spin_lock_init(&dev->addr_list_lock);
5273 netdev_set_addr_lockdep_class(dev);
5274 netdev_init_queue_locks(dev);
5278 if (!dev->num_rx_queues) {
5280 * Allocate a single RX queue if driver never called
5284 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
5290 dev->_rx->first = dev->_rx;
5291 atomic_set(&dev->_rx->count, 1);
5292 dev->num_rx_queues = 1;
5295 /* Init, if this function is available */
5296 if (dev->netdev_ops->ndo_init) {
5297 ret = dev->netdev_ops->ndo_init(dev);
5305 ret = dev_get_valid_name(net, dev->name, dev->name, 0);
5309 dev->ifindex = dev_new_index(net);
5310 if (dev->iflink == -1)
5311 dev->iflink = dev->ifindex;
5313 /* Fix illegal checksum combinations */
5314 if ((dev->features & NETIF_F_HW_CSUM) &&
5315 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5316 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5318 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5321 if ((dev->features & NETIF_F_NO_CSUM) &&
5322 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5323 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5325 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5328 dev->features = netdev_fix_features(dev->features, dev->name);
5330 /* Enable software GSO if SG is supported. */
5331 if (dev->features & NETIF_F_SG)
5332 dev->features |= NETIF_F_GSO;
5334 netdev_initialize_kobject(dev);
5336 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5337 ret = notifier_to_errno(ret);
5341 ret = netdev_register_kobject(dev);
5344 dev->reg_state = NETREG_REGISTERED;
5347 * Default initial state at registry is that the
5348 * device is present.
5351 set_bit(__LINK_STATE_PRESENT, &dev->state);
5353 dev_init_scheduler(dev);
5355 list_netdevice(dev);
5357 /* Notify protocols, that a new device appeared. */
5358 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5359 ret = notifier_to_errno(ret);
5361 rollback_registered(dev);
5362 dev->reg_state = NETREG_UNREGISTERED;
5365 * Prevent userspace races by waiting until the network
5366 * device is fully setup before sending notifications.
5368 if (!dev->rtnl_link_ops ||
5369 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5370 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5376 if (dev->netdev_ops->ndo_uninit)
5377 dev->netdev_ops->ndo_uninit(dev);
5380 EXPORT_SYMBOL(register_netdevice);
5383 * init_dummy_netdev - init a dummy network device for NAPI
5384 * @dev: device to init
5386 * This takes a network device structure and initialize the minimum
5387 * amount of fields so it can be used to schedule NAPI polls without
5388 * registering a full blown interface. This is to be used by drivers
5389 * that need to tie several hardware interfaces to a single NAPI
5390 * poll scheduler due to HW limitations.
5392 int init_dummy_netdev(struct net_device *dev)
5394 /* Clear everything. Note we don't initialize spinlocks
5395 * are they aren't supposed to be taken by any of the
5396 * NAPI code and this dummy netdev is supposed to be
5397 * only ever used for NAPI polls
5399 memset(dev, 0, sizeof(struct net_device));
5401 /* make sure we BUG if trying to hit standard
5402 * register/unregister code path
5404 dev->reg_state = NETREG_DUMMY;
5406 /* initialize the ref count */
5407 atomic_set(&dev->refcnt, 1);
5409 /* NAPI wants this */
5410 INIT_LIST_HEAD(&dev->napi_list);
5412 /* a dummy interface is started by default */
5413 set_bit(__LINK_STATE_PRESENT, &dev->state);
5414 set_bit(__LINK_STATE_START, &dev->state);
5418 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5422 * register_netdev - register a network device
5423 * @dev: device to register
5425 * Take a completed network device structure and add it to the kernel
5426 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5427 * chain. 0 is returned on success. A negative errno code is returned
5428 * on a failure to set up the device, or if the name is a duplicate.
5430 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5431 * and expands the device name if you passed a format string to
5434 int register_netdev(struct net_device *dev)
5441 * If the name is a format string the caller wants us to do a
5444 if (strchr(dev->name, '%')) {
5445 err = dev_alloc_name(dev, dev->name);
5450 err = register_netdevice(dev);
5455 EXPORT_SYMBOL(register_netdev);
5458 * netdev_wait_allrefs - wait until all references are gone.
5460 * This is called when unregistering network devices.
5462 * Any protocol or device that holds a reference should register
5463 * for netdevice notification, and cleanup and put back the
5464 * reference if they receive an UNREGISTER event.
5465 * We can get stuck here if buggy protocols don't correctly
5468 static void netdev_wait_allrefs(struct net_device *dev)
5470 unsigned long rebroadcast_time, warning_time;
5472 linkwatch_forget_dev(dev);
5474 rebroadcast_time = warning_time = jiffies;
5475 while (atomic_read(&dev->refcnt) != 0) {
5476 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5479 /* Rebroadcast unregister notification */
5480 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5481 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5482 * should have already handle it the first time */
5484 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5486 /* We must not have linkwatch events
5487 * pending on unregister. If this
5488 * happens, we simply run the queue
5489 * unscheduled, resulting in a noop
5492 linkwatch_run_queue();
5497 rebroadcast_time = jiffies;
5502 if (time_after(jiffies, warning_time + 10 * HZ)) {
5503 printk(KERN_EMERG "unregister_netdevice: "
5504 "waiting for %s to become free. Usage "
5506 dev->name, atomic_read(&dev->refcnt));
5507 warning_time = jiffies;
5516 * register_netdevice(x1);
5517 * register_netdevice(x2);
5519 * unregister_netdevice(y1);
5520 * unregister_netdevice(y2);
5526 * We are invoked by rtnl_unlock().
5527 * This allows us to deal with problems:
5528 * 1) We can delete sysfs objects which invoke hotplug
5529 * without deadlocking with linkwatch via keventd.
5530 * 2) Since we run with the RTNL semaphore not held, we can sleep
5531 * safely in order to wait for the netdev refcnt to drop to zero.
5533 * We must not return until all unregister events added during
5534 * the interval the lock was held have been completed.
5536 void netdev_run_todo(void)
5538 struct list_head list;
5540 /* Snapshot list, allow later requests */
5541 list_replace_init(&net_todo_list, &list);
5545 while (!list_empty(&list)) {
5546 struct net_device *dev
5547 = list_first_entry(&list, struct net_device, todo_list);
5548 list_del(&dev->todo_list);
5550 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5551 printk(KERN_ERR "network todo '%s' but state %d\n",
5552 dev->name, dev->reg_state);
5557 dev->reg_state = NETREG_UNREGISTERED;
5559 on_each_cpu(flush_backlog, dev, 1);
5561 netdev_wait_allrefs(dev);
5564 BUG_ON(atomic_read(&dev->refcnt));
5565 WARN_ON(dev->ip_ptr);
5566 WARN_ON(dev->ip6_ptr);
5567 WARN_ON(dev->dn_ptr);
5569 if (dev->destructor)
5570 dev->destructor(dev);
5572 /* Free network device */
5573 kobject_put(&dev->dev.kobj);
5578 * dev_txq_stats_fold - fold tx_queues stats
5579 * @dev: device to get statistics from
5580 * @stats: struct net_device_stats to hold results
5582 void dev_txq_stats_fold(const struct net_device *dev,
5583 struct net_device_stats *stats)
5585 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5587 struct netdev_queue *txq;
5589 for (i = 0; i < dev->num_tx_queues; i++) {
5590 txq = netdev_get_tx_queue(dev, i);
5591 tx_bytes += txq->tx_bytes;
5592 tx_packets += txq->tx_packets;
5593 tx_dropped += txq->tx_dropped;
5595 if (tx_bytes || tx_packets || tx_dropped) {
5596 stats->tx_bytes = tx_bytes;
5597 stats->tx_packets = tx_packets;
5598 stats->tx_dropped = tx_dropped;
5601 EXPORT_SYMBOL(dev_txq_stats_fold);
5604 * dev_get_stats - get network device statistics
5605 * @dev: device to get statistics from
5607 * Get network statistics from device. The device driver may provide
5608 * its own method by setting dev->netdev_ops->get_stats; otherwise
5609 * the internal statistics structure is used.
5611 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5613 const struct net_device_ops *ops = dev->netdev_ops;
5615 if (ops->ndo_get_stats)
5616 return ops->ndo_get_stats(dev);
5618 dev_txq_stats_fold(dev, &dev->stats);
5621 EXPORT_SYMBOL(dev_get_stats);
5623 static void netdev_init_one_queue(struct net_device *dev,
5624 struct netdev_queue *queue,
5630 static void netdev_init_queues(struct net_device *dev)
5632 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5633 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5634 spin_lock_init(&dev->tx_global_lock);
5638 * alloc_netdev_mq - allocate network device
5639 * @sizeof_priv: size of private data to allocate space for
5640 * @name: device name format string
5641 * @setup: callback to initialize device
5642 * @queue_count: the number of subqueues to allocate
5644 * Allocates a struct net_device with private data area for driver use
5645 * and performs basic initialization. Also allocates subquue structs
5646 * for each queue on the device at the end of the netdevice.
5648 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5649 void (*setup)(struct net_device *), unsigned int queue_count)
5651 struct netdev_queue *tx;
5652 struct netdev_rx_queue *rx;
5653 struct net_device *dev;
5655 struct net_device *p;
5658 BUG_ON(strlen(name) >= sizeof(dev->name));
5660 alloc_size = sizeof(struct net_device);
5662 /* ensure 32-byte alignment of private area */
5663 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5664 alloc_size += sizeof_priv;
5666 /* ensure 32-byte alignment of whole construct */
5667 alloc_size += NETDEV_ALIGN - 1;
5669 p = kzalloc(alloc_size, GFP_KERNEL);
5671 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5675 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5677 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5682 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5684 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5689 atomic_set(&rx->count, queue_count);
5692 * Set a pointer to first element in the array which holds the
5695 for (i = 0; i < queue_count; i++)
5698 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5699 dev->padded = (char *)dev - (char *)p;
5701 if (dev_addr_init(dev))
5704 dev_unicast_init(dev);
5706 dev_net_set(dev, &init_net);
5709 dev->num_tx_queues = queue_count;
5710 dev->real_num_tx_queues = queue_count;
5713 dev->num_rx_queues = queue_count;
5715 dev->gso_max_size = GSO_MAX_SIZE;
5717 netdev_init_queues(dev);
5719 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5720 dev->ethtool_ntuple_list.count = 0;
5721 INIT_LIST_HEAD(&dev->napi_list);
5722 INIT_LIST_HEAD(&dev->unreg_list);
5723 INIT_LIST_HEAD(&dev->link_watch_list);
5724 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5726 strcpy(dev->name, name);
5737 EXPORT_SYMBOL(alloc_netdev_mq);
5740 * free_netdev - free network device
5743 * This function does the last stage of destroying an allocated device
5744 * interface. The reference to the device object is released.
5745 * If this is the last reference then it will be freed.
5747 void free_netdev(struct net_device *dev)
5749 struct napi_struct *p, *n;
5751 release_net(dev_net(dev));
5755 /* Flush device addresses */
5756 dev_addr_flush(dev);
5758 /* Clear ethtool n-tuple list */
5759 ethtool_ntuple_flush(dev);
5761 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5764 /* Compatibility with error handling in drivers */
5765 if (dev->reg_state == NETREG_UNINITIALIZED) {
5766 kfree((char *)dev - dev->padded);
5770 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5771 dev->reg_state = NETREG_RELEASED;
5773 /* will free via device release */
5774 put_device(&dev->dev);
5776 EXPORT_SYMBOL(free_netdev);
5779 * synchronize_net - Synchronize with packet receive processing
5781 * Wait for packets currently being received to be done.
5782 * Does not block later packets from starting.
5784 void synchronize_net(void)
5789 EXPORT_SYMBOL(synchronize_net);
5792 * unregister_netdevice_queue - remove device from the kernel
5796 * This function shuts down a device interface and removes it
5797 * from the kernel tables.
5798 * If head not NULL, device is queued to be unregistered later.
5800 * Callers must hold the rtnl semaphore. You may want
5801 * unregister_netdev() instead of this.
5804 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5809 list_move_tail(&dev->unreg_list, head);
5811 rollback_registered(dev);
5812 /* Finish processing unregister after unlock */
5816 EXPORT_SYMBOL(unregister_netdevice_queue);
5819 * unregister_netdevice_many - unregister many devices
5820 * @head: list of devices
5822 void unregister_netdevice_many(struct list_head *head)
5824 struct net_device *dev;
5826 if (!list_empty(head)) {
5827 rollback_registered_many(head);
5828 list_for_each_entry(dev, head, unreg_list)
5832 EXPORT_SYMBOL(unregister_netdevice_many);
5835 * unregister_netdev - remove device from the kernel
5838 * This function shuts down a device interface and removes it
5839 * from the kernel tables.
5841 * This is just a wrapper for unregister_netdevice that takes
5842 * the rtnl semaphore. In general you want to use this and not
5843 * unregister_netdevice.
5845 void unregister_netdev(struct net_device *dev)
5848 unregister_netdevice(dev);
5851 EXPORT_SYMBOL(unregister_netdev);
5854 * dev_change_net_namespace - move device to different nethost namespace
5856 * @net: network namespace
5857 * @pat: If not NULL name pattern to try if the current device name
5858 * is already taken in the destination network namespace.
5860 * This function shuts down a device interface and moves it
5861 * to a new network namespace. On success 0 is returned, on
5862 * a failure a netagive errno code is returned.
5864 * Callers must hold the rtnl semaphore.
5867 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5873 /* Don't allow namespace local devices to be moved. */
5875 if (dev->features & NETIF_F_NETNS_LOCAL)
5879 /* Don't allow real devices to be moved when sysfs
5883 if (dev->dev.parent)
5887 /* Ensure the device has been registrered */
5889 if (dev->reg_state != NETREG_REGISTERED)
5892 /* Get out if there is nothing todo */
5894 if (net_eq(dev_net(dev), net))
5897 /* Pick the destination device name, and ensure
5898 * we can use it in the destination network namespace.
5901 if (__dev_get_by_name(net, dev->name)) {
5902 /* We get here if we can't use the current device name */
5905 if (dev_get_valid_name(net, pat, dev->name, 1))
5910 * And now a mini version of register_netdevice unregister_netdevice.
5913 /* If device is running close it first. */
5916 /* And unlink it from device chain */
5918 unlist_netdevice(dev);
5922 /* Shutdown queueing discipline. */
5925 /* Notify protocols, that we are about to destroy
5926 this device. They should clean all the things.
5928 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5929 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5932 * Flush the unicast and multicast chains
5934 dev_unicast_flush(dev);
5935 dev_addr_discard(dev);
5937 netdev_unregister_kobject(dev);
5939 /* Actually switch the network namespace */
5940 dev_net_set(dev, net);
5942 /* If there is an ifindex conflict assign a new one */
5943 if (__dev_get_by_index(net, dev->ifindex)) {
5944 int iflink = (dev->iflink == dev->ifindex);
5945 dev->ifindex = dev_new_index(net);
5947 dev->iflink = dev->ifindex;
5950 /* Fixup kobjects */
5951 err = netdev_register_kobject(dev);
5954 /* Add the device back in the hashes */
5955 list_netdevice(dev);
5957 /* Notify protocols, that a new device appeared. */
5958 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5961 * Prevent userspace races by waiting until the network
5962 * device is fully setup before sending notifications.
5964 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5971 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5973 static int dev_cpu_callback(struct notifier_block *nfb,
5974 unsigned long action,
5977 struct sk_buff **list_skb;
5978 struct Qdisc **list_net;
5979 struct sk_buff *skb;
5980 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5981 struct softnet_data *sd, *oldsd;
5983 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5986 local_irq_disable();
5987 cpu = smp_processor_id();
5988 sd = &per_cpu(softnet_data, cpu);
5989 oldsd = &per_cpu(softnet_data, oldcpu);
5991 /* Find end of our completion_queue. */
5992 list_skb = &sd->completion_queue;
5994 list_skb = &(*list_skb)->next;
5995 /* Append completion queue from offline CPU. */
5996 *list_skb = oldsd->completion_queue;
5997 oldsd->completion_queue = NULL;
5999 /* Find end of our output_queue. */
6000 list_net = &sd->output_queue;
6002 list_net = &(*list_net)->next_sched;
6003 /* Append output queue from offline CPU. */
6004 *list_net = oldsd->output_queue;
6005 oldsd->output_queue = NULL;
6007 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6010 /* Process offline CPU's input_pkt_queue */
6011 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
6019 * netdev_increment_features - increment feature set by one
6020 * @all: current feature set
6021 * @one: new feature set
6022 * @mask: mask feature set
6024 * Computes a new feature set after adding a device with feature set
6025 * @one to the master device with current feature set @all. Will not
6026 * enable anything that is off in @mask. Returns the new feature set.
6028 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
6031 /* If device needs checksumming, downgrade to it. */
6032 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
6033 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
6034 else if (mask & NETIF_F_ALL_CSUM) {
6035 /* If one device supports v4/v6 checksumming, set for all. */
6036 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
6037 !(all & NETIF_F_GEN_CSUM)) {
6038 all &= ~NETIF_F_ALL_CSUM;
6039 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
6042 /* If one device supports hw checksumming, set for all. */
6043 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
6044 all &= ~NETIF_F_ALL_CSUM;
6045 all |= NETIF_F_HW_CSUM;
6049 one |= NETIF_F_ALL_CSUM;
6051 one |= all & NETIF_F_ONE_FOR_ALL;
6052 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
6053 all |= one & mask & NETIF_F_ONE_FOR_ALL;
6057 EXPORT_SYMBOL(netdev_increment_features);
6059 static struct hlist_head *netdev_create_hash(void)
6062 struct hlist_head *hash;
6064 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6066 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6067 INIT_HLIST_HEAD(&hash[i]);
6072 /* Initialize per network namespace state */
6073 static int __net_init netdev_init(struct net *net)
6075 INIT_LIST_HEAD(&net->dev_base_head);
6077 net->dev_name_head = netdev_create_hash();
6078 if (net->dev_name_head == NULL)
6081 net->dev_index_head = netdev_create_hash();
6082 if (net->dev_index_head == NULL)
6088 kfree(net->dev_name_head);
6094 * netdev_drivername - network driver for the device
6095 * @dev: network device
6096 * @buffer: buffer for resulting name
6097 * @len: size of buffer
6099 * Determine network driver for device.
6101 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6103 const struct device_driver *driver;
6104 const struct device *parent;
6106 if (len <= 0 || !buffer)
6110 parent = dev->dev.parent;
6115 driver = parent->driver;
6116 if (driver && driver->name)
6117 strlcpy(buffer, driver->name, len);
6121 static void __net_exit netdev_exit(struct net *net)
6123 kfree(net->dev_name_head);
6124 kfree(net->dev_index_head);
6127 static struct pernet_operations __net_initdata netdev_net_ops = {
6128 .init = netdev_init,
6129 .exit = netdev_exit,
6132 static void __net_exit default_device_exit(struct net *net)
6134 struct net_device *dev, *aux;
6136 * Push all migratable network devices back to the
6137 * initial network namespace
6140 for_each_netdev_safe(net, dev, aux) {
6142 char fb_name[IFNAMSIZ];
6144 /* Ignore unmoveable devices (i.e. loopback) */
6145 if (dev->features & NETIF_F_NETNS_LOCAL)
6148 /* Leave virtual devices for the generic cleanup */
6149 if (dev->rtnl_link_ops)
6152 /* Push remaing network devices to init_net */
6153 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6154 err = dev_change_net_namespace(dev, &init_net, fb_name);
6156 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6157 __func__, dev->name, err);
6164 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6166 /* At exit all network devices most be removed from a network
6167 * namespace. Do this in the reverse order of registeration.
6168 * Do this across as many network namespaces as possible to
6169 * improve batching efficiency.
6171 struct net_device *dev;
6173 LIST_HEAD(dev_kill_list);
6176 list_for_each_entry(net, net_list, exit_list) {
6177 for_each_netdev_reverse(net, dev) {
6178 if (dev->rtnl_link_ops)
6179 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6181 unregister_netdevice_queue(dev, &dev_kill_list);
6184 unregister_netdevice_many(&dev_kill_list);
6188 static struct pernet_operations __net_initdata default_device_ops = {
6189 .exit = default_device_exit,
6190 .exit_batch = default_device_exit_batch,
6194 * Initialize the DEV module. At boot time this walks the device list and
6195 * unhooks any devices that fail to initialise (normally hardware not
6196 * present) and leaves us with a valid list of present and active devices.
6201 * This is called single threaded during boot, so no need
6202 * to take the rtnl semaphore.
6204 static int __init net_dev_init(void)
6206 int i, rc = -ENOMEM;
6208 BUG_ON(!dev_boot_phase);
6210 if (dev_proc_init())
6213 if (netdev_kobject_init())
6216 INIT_LIST_HEAD(&ptype_all);
6217 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6218 INIT_LIST_HEAD(&ptype_base[i]);
6220 if (register_pernet_subsys(&netdev_net_ops))
6224 * Initialise the packet receive queues.
6227 for_each_possible_cpu(i) {
6228 struct softnet_data *queue;
6230 queue = &per_cpu(softnet_data, i);
6231 skb_queue_head_init(&queue->input_pkt_queue);
6232 queue->completion_queue = NULL;
6233 INIT_LIST_HEAD(&queue->poll_list);
6236 queue->csd.func = trigger_softirq;
6237 queue->csd.info = queue;
6238 queue->csd.flags = 0;
6241 queue->backlog.poll = process_backlog;
6242 queue->backlog.weight = weight_p;
6243 queue->backlog.gro_list = NULL;
6244 queue->backlog.gro_count = 0;
6249 /* The loopback device is special if any other network devices
6250 * is present in a network namespace the loopback device must
6251 * be present. Since we now dynamically allocate and free the
6252 * loopback device ensure this invariant is maintained by
6253 * keeping the loopback device as the first device on the
6254 * list of network devices. Ensuring the loopback devices
6255 * is the first device that appears and the last network device
6258 if (register_pernet_device(&loopback_net_ops))
6261 if (register_pernet_device(&default_device_ops))
6264 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6265 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6267 hotcpu_notifier(dev_cpu_callback, 0);
6275 subsys_initcall(net_dev_init);
6277 static int __init initialize_hashrnd(void)
6279 get_random_bytes(&hashrnd, sizeof(hashrnd));
6283 late_initcall_sync(initialize_hashrnd);