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;
778 dev = __dev_getfirstbyhwtype(net, type);
784 EXPORT_SYMBOL(dev_getfirstbyhwtype);
787 * dev_get_by_flags - find any device with given flags
788 * @net: the applicable net namespace
789 * @if_flags: IFF_* values
790 * @mask: bitmask of bits in if_flags to check
792 * Search for any interface with the given flags. Returns NULL if a device
793 * is not found or a pointer to the device. The device returned has
794 * had a reference added and the pointer is safe until the user calls
795 * dev_put to indicate they have finished with it.
798 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
801 struct net_device *dev, *ret;
805 for_each_netdev_rcu(net, dev) {
806 if (((dev->flags ^ if_flags) & mask) == 0) {
815 EXPORT_SYMBOL(dev_get_by_flags);
818 * dev_valid_name - check if name is okay for network device
821 * Network device names need to be valid file names to
822 * to allow sysfs to work. We also disallow any kind of
825 int dev_valid_name(const char *name)
829 if (strlen(name) >= IFNAMSIZ)
831 if (!strcmp(name, ".") || !strcmp(name, ".."))
835 if (*name == '/' || isspace(*name))
841 EXPORT_SYMBOL(dev_valid_name);
844 * __dev_alloc_name - allocate a name for a device
845 * @net: network namespace to allocate the device name in
846 * @name: name format string
847 * @buf: scratch buffer and result name string
849 * Passed a format string - eg "lt%d" it will try and find a suitable
850 * id. It scans list of devices to build up a free map, then chooses
851 * the first empty slot. The caller must hold the dev_base or rtnl lock
852 * while allocating the name and adding the device in order to avoid
854 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
855 * Returns the number of the unit assigned or a negative errno code.
858 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
862 const int max_netdevices = 8*PAGE_SIZE;
863 unsigned long *inuse;
864 struct net_device *d;
866 p = strnchr(name, IFNAMSIZ-1, '%');
869 * Verify the string as this thing may have come from
870 * the user. There must be either one "%d" and no other "%"
873 if (p[1] != 'd' || strchr(p + 2, '%'))
876 /* Use one page as a bit array of possible slots */
877 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
881 for_each_netdev(net, d) {
882 if (!sscanf(d->name, name, &i))
884 if (i < 0 || i >= max_netdevices)
887 /* avoid cases where sscanf is not exact inverse of printf */
888 snprintf(buf, IFNAMSIZ, name, i);
889 if (!strncmp(buf, d->name, IFNAMSIZ))
893 i = find_first_zero_bit(inuse, max_netdevices);
894 free_page((unsigned long) inuse);
898 snprintf(buf, IFNAMSIZ, name, i);
899 if (!__dev_get_by_name(net, buf))
902 /* It is possible to run out of possible slots
903 * when the name is long and there isn't enough space left
904 * for the digits, or if all bits are used.
910 * dev_alloc_name - allocate a name for a device
912 * @name: name format string
914 * Passed a format string - eg "lt%d" it will try and find a suitable
915 * id. It scans list of devices to build up a free map, then chooses
916 * the first empty slot. The caller must hold the dev_base or rtnl lock
917 * while allocating the name and adding the device in order to avoid
919 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
920 * Returns the number of the unit assigned or a negative errno code.
923 int dev_alloc_name(struct net_device *dev, const char *name)
929 BUG_ON(!dev_net(dev));
931 ret = __dev_alloc_name(net, name, buf);
933 strlcpy(dev->name, buf, IFNAMSIZ);
936 EXPORT_SYMBOL(dev_alloc_name);
938 static int dev_get_valid_name(struct net *net, const char *name, char *buf,
941 if (!dev_valid_name(name))
944 if (fmt && strchr(name, '%'))
945 return __dev_alloc_name(net, name, buf);
946 else if (__dev_get_by_name(net, name))
948 else if (buf != name)
949 strlcpy(buf, name, IFNAMSIZ);
955 * dev_change_name - change name of a device
957 * @newname: name (or format string) must be at least IFNAMSIZ
959 * Change name of a device, can pass format strings "eth%d".
962 int dev_change_name(struct net_device *dev, const char *newname)
964 char oldname[IFNAMSIZ];
970 BUG_ON(!dev_net(dev));
973 if (dev->flags & IFF_UP)
976 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
979 memcpy(oldname, dev->name, IFNAMSIZ);
981 err = dev_get_valid_name(net, newname, dev->name, 1);
986 /* For now only devices in the initial network namespace
989 if (net_eq(net, &init_net)) {
990 ret = device_rename(&dev->dev, dev->name);
992 memcpy(dev->name, oldname, IFNAMSIZ);
997 write_lock_bh(&dev_base_lock);
998 hlist_del(&dev->name_hlist);
999 write_unlock_bh(&dev_base_lock);
1003 write_lock_bh(&dev_base_lock);
1004 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1005 write_unlock_bh(&dev_base_lock);
1007 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1008 ret = notifier_to_errno(ret);
1011 /* err >= 0 after dev_alloc_name() or stores the first errno */
1014 memcpy(dev->name, oldname, IFNAMSIZ);
1018 "%s: name change rollback failed: %d.\n",
1027 * dev_set_alias - change ifalias of a device
1029 * @alias: name up to IFALIASZ
1030 * @len: limit of bytes to copy from info
1032 * Set ifalias for a device,
1034 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1038 if (len >= IFALIASZ)
1043 kfree(dev->ifalias);
1044 dev->ifalias = NULL;
1049 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1053 strlcpy(dev->ifalias, alias, len+1);
1059 * netdev_features_change - device changes features
1060 * @dev: device to cause notification
1062 * Called to indicate a device has changed features.
1064 void netdev_features_change(struct net_device *dev)
1066 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1068 EXPORT_SYMBOL(netdev_features_change);
1071 * netdev_state_change - device changes state
1072 * @dev: device to cause notification
1074 * Called to indicate a device has changed state. This function calls
1075 * the notifier chains for netdev_chain and sends a NEWLINK message
1076 * to the routing socket.
1078 void netdev_state_change(struct net_device *dev)
1080 if (dev->flags & IFF_UP) {
1081 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1082 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1085 EXPORT_SYMBOL(netdev_state_change);
1087 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1089 return call_netdevice_notifiers(event, dev);
1091 EXPORT_SYMBOL(netdev_bonding_change);
1094 * dev_load - load a network module
1095 * @net: the applicable net namespace
1096 * @name: name of interface
1098 * If a network interface is not present and the process has suitable
1099 * privileges this function loads the module. If module loading is not
1100 * available in this kernel then it becomes a nop.
1103 void dev_load(struct net *net, const char *name)
1105 struct net_device *dev;
1108 dev = dev_get_by_name_rcu(net, name);
1111 if (!dev && capable(CAP_NET_ADMIN))
1112 request_module("%s", name);
1114 EXPORT_SYMBOL(dev_load);
1116 static int __dev_open(struct net_device *dev)
1118 const struct net_device_ops *ops = dev->netdev_ops;
1124 * Is it even present?
1126 if (!netif_device_present(dev))
1129 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1130 ret = notifier_to_errno(ret);
1135 * Call device private open method
1137 set_bit(__LINK_STATE_START, &dev->state);
1139 if (ops->ndo_validate_addr)
1140 ret = ops->ndo_validate_addr(dev);
1142 if (!ret && ops->ndo_open)
1143 ret = ops->ndo_open(dev);
1146 * If it went open OK then:
1150 clear_bit(__LINK_STATE_START, &dev->state);
1155 dev->flags |= IFF_UP;
1160 net_dmaengine_get();
1163 * Initialize multicasting status
1165 dev_set_rx_mode(dev);
1168 * Wakeup transmit queue engine
1177 * dev_open - prepare an interface for use.
1178 * @dev: device to open
1180 * Takes a device from down to up state. The device's private open
1181 * function is invoked and then the multicast lists are loaded. Finally
1182 * the device is moved into the up state and a %NETDEV_UP message is
1183 * sent to the netdev notifier chain.
1185 * Calling this function on an active interface is a nop. On a failure
1186 * a negative errno code is returned.
1188 int dev_open(struct net_device *dev)
1195 if (dev->flags & IFF_UP)
1201 ret = __dev_open(dev);
1206 * ... and announce new interface.
1208 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1209 call_netdevice_notifiers(NETDEV_UP, dev);
1213 EXPORT_SYMBOL(dev_open);
1215 static int __dev_close(struct net_device *dev)
1217 const struct net_device_ops *ops = dev->netdev_ops;
1223 * Tell people we are going down, so that they can
1224 * prepare to death, when device is still operating.
1226 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1228 clear_bit(__LINK_STATE_START, &dev->state);
1230 /* Synchronize to scheduled poll. We cannot touch poll list,
1231 * it can be even on different cpu. So just clear netif_running().
1233 * dev->stop() will invoke napi_disable() on all of it's
1234 * napi_struct instances on this device.
1236 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1238 dev_deactivate(dev);
1241 * Call the device specific close. This cannot fail.
1242 * Only if device is UP
1244 * We allow it to be called even after a DETACH hot-plug
1251 * Device is now down.
1254 dev->flags &= ~IFF_UP;
1259 net_dmaengine_put();
1265 * dev_close - shutdown an interface.
1266 * @dev: device to shutdown
1268 * This function moves an active device into down state. A
1269 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1270 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1273 int dev_close(struct net_device *dev)
1275 if (!(dev->flags & IFF_UP))
1281 * Tell people we are down
1283 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1284 call_netdevice_notifiers(NETDEV_DOWN, dev);
1288 EXPORT_SYMBOL(dev_close);
1292 * dev_disable_lro - disable Large Receive Offload on a device
1295 * Disable Large Receive Offload (LRO) on a net device. Must be
1296 * called under RTNL. This is needed if received packets may be
1297 * forwarded to another interface.
1299 void dev_disable_lro(struct net_device *dev)
1301 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1302 dev->ethtool_ops->set_flags) {
1303 u32 flags = dev->ethtool_ops->get_flags(dev);
1304 if (flags & ETH_FLAG_LRO) {
1305 flags &= ~ETH_FLAG_LRO;
1306 dev->ethtool_ops->set_flags(dev, flags);
1309 WARN_ON(dev->features & NETIF_F_LRO);
1311 EXPORT_SYMBOL(dev_disable_lro);
1314 static int dev_boot_phase = 1;
1317 * Device change register/unregister. These are not inline or static
1318 * as we export them to the world.
1322 * register_netdevice_notifier - register a network notifier block
1325 * Register a notifier to be called when network device events occur.
1326 * The notifier passed is linked into the kernel structures and must
1327 * not be reused until it has been unregistered. A negative errno code
1328 * is returned on a failure.
1330 * When registered all registration and up events are replayed
1331 * to the new notifier to allow device to have a race free
1332 * view of the network device list.
1335 int register_netdevice_notifier(struct notifier_block *nb)
1337 struct net_device *dev;
1338 struct net_device *last;
1343 err = raw_notifier_chain_register(&netdev_chain, nb);
1349 for_each_netdev(net, dev) {
1350 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1351 err = notifier_to_errno(err);
1355 if (!(dev->flags & IFF_UP))
1358 nb->notifier_call(nb, NETDEV_UP, dev);
1369 for_each_netdev(net, dev) {
1373 if (dev->flags & IFF_UP) {
1374 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1375 nb->notifier_call(nb, NETDEV_DOWN, dev);
1377 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1378 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1382 raw_notifier_chain_unregister(&netdev_chain, nb);
1385 EXPORT_SYMBOL(register_netdevice_notifier);
1388 * unregister_netdevice_notifier - unregister a network notifier block
1391 * Unregister a notifier previously registered by
1392 * register_netdevice_notifier(). The notifier is unlinked into the
1393 * kernel structures and may then be reused. A negative errno code
1394 * is returned on a failure.
1397 int unregister_netdevice_notifier(struct notifier_block *nb)
1402 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1406 EXPORT_SYMBOL(unregister_netdevice_notifier);
1409 * call_netdevice_notifiers - call all network notifier blocks
1410 * @val: value passed unmodified to notifier function
1411 * @dev: net_device pointer passed unmodified to notifier function
1413 * Call all network notifier blocks. Parameters and return value
1414 * are as for raw_notifier_call_chain().
1417 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1419 return raw_notifier_call_chain(&netdev_chain, val, dev);
1422 /* When > 0 there are consumers of rx skb time stamps */
1423 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1425 void net_enable_timestamp(void)
1427 atomic_inc(&netstamp_needed);
1429 EXPORT_SYMBOL(net_enable_timestamp);
1431 void net_disable_timestamp(void)
1433 atomic_dec(&netstamp_needed);
1435 EXPORT_SYMBOL(net_disable_timestamp);
1437 static inline void net_timestamp(struct sk_buff *skb)
1439 if (atomic_read(&netstamp_needed))
1440 __net_timestamp(skb);
1442 skb->tstamp.tv64 = 0;
1446 * dev_forward_skb - loopback an skb to another netif
1448 * @dev: destination network device
1449 * @skb: buffer to forward
1452 * NET_RX_SUCCESS (no congestion)
1453 * NET_RX_DROP (packet was dropped)
1455 * dev_forward_skb can be used for injecting an skb from the
1456 * start_xmit function of one device into the receive queue
1457 * of another device.
1459 * The receiving device may be in another namespace, so
1460 * we have to clear all information in the skb that could
1461 * impact namespace isolation.
1463 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1467 if (!(dev->flags & IFF_UP))
1470 if (skb->len > (dev->mtu + dev->hard_header_len))
1473 skb_set_dev(skb, dev);
1474 skb->tstamp.tv64 = 0;
1475 skb->pkt_type = PACKET_HOST;
1476 skb->protocol = eth_type_trans(skb, dev);
1477 return netif_rx(skb);
1479 EXPORT_SYMBOL_GPL(dev_forward_skb);
1482 * Support routine. Sends outgoing frames to any network
1483 * taps currently in use.
1486 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1488 struct packet_type *ptype;
1490 #ifdef CONFIG_NET_CLS_ACT
1491 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1498 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1499 /* Never send packets back to the socket
1500 * they originated from - MvS (miquels@drinkel.ow.org)
1502 if ((ptype->dev == dev || !ptype->dev) &&
1503 (ptype->af_packet_priv == NULL ||
1504 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1505 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1509 /* skb->nh should be correctly
1510 set by sender, so that the second statement is
1511 just protection against buggy protocols.
1513 skb_reset_mac_header(skb2);
1515 if (skb_network_header(skb2) < skb2->data ||
1516 skb2->network_header > skb2->tail) {
1517 if (net_ratelimit())
1518 printk(KERN_CRIT "protocol %04x is "
1520 skb2->protocol, dev->name);
1521 skb_reset_network_header(skb2);
1524 skb2->transport_header = skb2->network_header;
1525 skb2->pkt_type = PACKET_OUTGOING;
1526 ptype->func(skb2, skb->dev, ptype, skb->dev);
1533 static inline void __netif_reschedule(struct Qdisc *q)
1535 struct softnet_data *sd;
1536 unsigned long flags;
1538 local_irq_save(flags);
1539 sd = &__get_cpu_var(softnet_data);
1540 q->next_sched = sd->output_queue;
1541 sd->output_queue = q;
1542 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1543 local_irq_restore(flags);
1546 void __netif_schedule(struct Qdisc *q)
1548 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1549 __netif_reschedule(q);
1551 EXPORT_SYMBOL(__netif_schedule);
1553 void dev_kfree_skb_irq(struct sk_buff *skb)
1555 if (atomic_dec_and_test(&skb->users)) {
1556 struct softnet_data *sd;
1557 unsigned long flags;
1559 local_irq_save(flags);
1560 sd = &__get_cpu_var(softnet_data);
1561 skb->next = sd->completion_queue;
1562 sd->completion_queue = skb;
1563 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1564 local_irq_restore(flags);
1567 EXPORT_SYMBOL(dev_kfree_skb_irq);
1569 void dev_kfree_skb_any(struct sk_buff *skb)
1571 if (in_irq() || irqs_disabled())
1572 dev_kfree_skb_irq(skb);
1576 EXPORT_SYMBOL(dev_kfree_skb_any);
1580 * netif_device_detach - mark device as removed
1581 * @dev: network device
1583 * Mark device as removed from system and therefore no longer available.
1585 void netif_device_detach(struct net_device *dev)
1587 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1588 netif_running(dev)) {
1589 netif_tx_stop_all_queues(dev);
1592 EXPORT_SYMBOL(netif_device_detach);
1595 * netif_device_attach - mark device as attached
1596 * @dev: network device
1598 * Mark device as attached from system and restart if needed.
1600 void netif_device_attach(struct net_device *dev)
1602 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1603 netif_running(dev)) {
1604 netif_tx_wake_all_queues(dev);
1605 __netdev_watchdog_up(dev);
1608 EXPORT_SYMBOL(netif_device_attach);
1610 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1612 return ((features & NETIF_F_GEN_CSUM) ||
1613 ((features & NETIF_F_IP_CSUM) &&
1614 protocol == htons(ETH_P_IP)) ||
1615 ((features & NETIF_F_IPV6_CSUM) &&
1616 protocol == htons(ETH_P_IPV6)) ||
1617 ((features & NETIF_F_FCOE_CRC) &&
1618 protocol == htons(ETH_P_FCOE)));
1621 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1623 if (can_checksum_protocol(dev->features, skb->protocol))
1626 if (skb->protocol == htons(ETH_P_8021Q)) {
1627 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1628 if (can_checksum_protocol(dev->features & dev->vlan_features,
1629 veh->h_vlan_encapsulated_proto))
1637 * skb_dev_set -- assign a new device to a buffer
1638 * @skb: buffer for the new device
1639 * @dev: network device
1641 * If an skb is owned by a device already, we have to reset
1642 * all data private to the namespace a device belongs to
1643 * before assigning it a new device.
1645 #ifdef CONFIG_NET_NS
1646 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1649 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1652 skb_init_secmark(skb);
1656 skb->ipvs_property = 0;
1657 #ifdef CONFIG_NET_SCHED
1663 EXPORT_SYMBOL(skb_set_dev);
1664 #endif /* CONFIG_NET_NS */
1667 * Invalidate hardware checksum when packet is to be mangled, and
1668 * complete checksum manually on outgoing path.
1670 int skb_checksum_help(struct sk_buff *skb)
1673 int ret = 0, offset;
1675 if (skb->ip_summed == CHECKSUM_COMPLETE)
1676 goto out_set_summed;
1678 if (unlikely(skb_shinfo(skb)->gso_size)) {
1679 /* Let GSO fix up the checksum. */
1680 goto out_set_summed;
1683 offset = skb->csum_start - skb_headroom(skb);
1684 BUG_ON(offset >= skb_headlen(skb));
1685 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1687 offset += skb->csum_offset;
1688 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1690 if (skb_cloned(skb) &&
1691 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1692 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1697 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1699 skb->ip_summed = CHECKSUM_NONE;
1703 EXPORT_SYMBOL(skb_checksum_help);
1706 * skb_gso_segment - Perform segmentation on skb.
1707 * @skb: buffer to segment
1708 * @features: features for the output path (see dev->features)
1710 * This function segments the given skb and returns a list of segments.
1712 * It may return NULL if the skb requires no segmentation. This is
1713 * only possible when GSO is used for verifying header integrity.
1715 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1717 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1718 struct packet_type *ptype;
1719 __be16 type = skb->protocol;
1722 skb_reset_mac_header(skb);
1723 skb->mac_len = skb->network_header - skb->mac_header;
1724 __skb_pull(skb, skb->mac_len);
1726 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1727 struct net_device *dev = skb->dev;
1728 struct ethtool_drvinfo info = {};
1730 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1731 dev->ethtool_ops->get_drvinfo(dev, &info);
1733 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1735 info.driver, dev ? dev->features : 0L,
1736 skb->sk ? skb->sk->sk_route_caps : 0L,
1737 skb->len, skb->data_len, skb->ip_summed);
1739 if (skb_header_cloned(skb) &&
1740 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1741 return ERR_PTR(err);
1745 list_for_each_entry_rcu(ptype,
1746 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1747 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1748 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1749 err = ptype->gso_send_check(skb);
1750 segs = ERR_PTR(err);
1751 if (err || skb_gso_ok(skb, features))
1753 __skb_push(skb, (skb->data -
1754 skb_network_header(skb)));
1756 segs = ptype->gso_segment(skb, features);
1762 __skb_push(skb, skb->data - skb_mac_header(skb));
1766 EXPORT_SYMBOL(skb_gso_segment);
1768 /* Take action when hardware reception checksum errors are detected. */
1770 void netdev_rx_csum_fault(struct net_device *dev)
1772 if (net_ratelimit()) {
1773 printk(KERN_ERR "%s: hw csum failure.\n",
1774 dev ? dev->name : "<unknown>");
1778 EXPORT_SYMBOL(netdev_rx_csum_fault);
1781 /* Actually, we should eliminate this check as soon as we know, that:
1782 * 1. IOMMU is present and allows to map all the memory.
1783 * 2. No high memory really exists on this machine.
1786 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1788 #ifdef CONFIG_HIGHMEM
1791 if (dev->features & NETIF_F_HIGHDMA)
1794 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1795 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1803 void (*destructor)(struct sk_buff *skb);
1806 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1808 static void dev_gso_skb_destructor(struct sk_buff *skb)
1810 struct dev_gso_cb *cb;
1813 struct sk_buff *nskb = skb->next;
1815 skb->next = nskb->next;
1818 } while (skb->next);
1820 cb = DEV_GSO_CB(skb);
1822 cb->destructor(skb);
1826 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1827 * @skb: buffer to segment
1829 * This function segments the given skb and stores the list of segments
1832 static int dev_gso_segment(struct sk_buff *skb)
1834 struct net_device *dev = skb->dev;
1835 struct sk_buff *segs;
1836 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1839 segs = skb_gso_segment(skb, features);
1841 /* Verifying header integrity only. */
1846 return PTR_ERR(segs);
1849 DEV_GSO_CB(skb)->destructor = skb->destructor;
1850 skb->destructor = dev_gso_skb_destructor;
1855 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1856 struct netdev_queue *txq)
1858 const struct net_device_ops *ops = dev->netdev_ops;
1859 int rc = NETDEV_TX_OK;
1861 if (likely(!skb->next)) {
1862 if (!list_empty(&ptype_all))
1863 dev_queue_xmit_nit(skb, dev);
1865 if (netif_needs_gso(dev, skb)) {
1866 if (unlikely(dev_gso_segment(skb)))
1873 * If device doesnt need skb->dst, release it right now while
1874 * its hot in this cpu cache
1876 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1879 rc = ops->ndo_start_xmit(skb, dev);
1880 if (rc == NETDEV_TX_OK)
1881 txq_trans_update(txq);
1883 * TODO: if skb_orphan() was called by
1884 * dev->hard_start_xmit() (for example, the unmodified
1885 * igb driver does that; bnx2 doesn't), then
1886 * skb_tx_software_timestamp() will be unable to send
1887 * back the time stamp.
1889 * How can this be prevented? Always create another
1890 * reference to the socket before calling
1891 * dev->hard_start_xmit()? Prevent that skb_orphan()
1892 * does anything in dev->hard_start_xmit() by clearing
1893 * the skb destructor before the call and restoring it
1894 * afterwards, then doing the skb_orphan() ourselves?
1901 struct sk_buff *nskb = skb->next;
1903 skb->next = nskb->next;
1907 * If device doesnt need nskb->dst, release it right now while
1908 * its hot in this cpu cache
1910 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1913 rc = ops->ndo_start_xmit(nskb, dev);
1914 if (unlikely(rc != NETDEV_TX_OK)) {
1915 if (rc & ~NETDEV_TX_MASK)
1916 goto out_kfree_gso_skb;
1917 nskb->next = skb->next;
1921 txq_trans_update(txq);
1922 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1923 return NETDEV_TX_BUSY;
1924 } while (skb->next);
1927 if (likely(skb->next == NULL))
1928 skb->destructor = DEV_GSO_CB(skb)->destructor;
1934 static u32 hashrnd __read_mostly;
1936 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1940 if (skb_rx_queue_recorded(skb)) {
1941 hash = skb_get_rx_queue(skb);
1942 while (unlikely(hash >= dev->real_num_tx_queues))
1943 hash -= dev->real_num_tx_queues;
1947 if (skb->sk && skb->sk->sk_hash)
1948 hash = skb->sk->sk_hash;
1950 hash = skb->protocol;
1952 hash = jhash_1word(hash, hashrnd);
1954 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1956 EXPORT_SYMBOL(skb_tx_hash);
1958 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1960 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1961 if (net_ratelimit()) {
1962 netdev_warn(dev, "selects TX queue %d, but "
1963 "real number of TX queues is %d\n",
1964 queue_index, dev->real_num_tx_queues);
1971 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1972 struct sk_buff *skb)
1975 struct sock *sk = skb->sk;
1977 if (sk_tx_queue_recorded(sk)) {
1978 queue_index = sk_tx_queue_get(sk);
1980 const struct net_device_ops *ops = dev->netdev_ops;
1982 if (ops->ndo_select_queue) {
1983 queue_index = ops->ndo_select_queue(dev, skb);
1984 queue_index = dev_cap_txqueue(dev, queue_index);
1987 if (dev->real_num_tx_queues > 1)
1988 queue_index = skb_tx_hash(dev, skb);
1990 if (sk && sk->sk_dst_cache)
1991 sk_tx_queue_set(sk, queue_index);
1995 skb_set_queue_mapping(skb, queue_index);
1996 return netdev_get_tx_queue(dev, queue_index);
1999 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2000 struct net_device *dev,
2001 struct netdev_queue *txq)
2003 spinlock_t *root_lock = qdisc_lock(q);
2006 spin_lock(root_lock);
2007 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2010 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2011 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
2013 * This is a work-conserving queue; there are no old skbs
2014 * waiting to be sent out; and the qdisc is not running -
2015 * xmit the skb directly.
2017 __qdisc_update_bstats(q, skb->len);
2018 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
2021 clear_bit(__QDISC_STATE_RUNNING, &q->state);
2023 rc = NET_XMIT_SUCCESS;
2025 rc = qdisc_enqueue_root(skb, q);
2028 spin_unlock(root_lock);
2034 * Returns true if either:
2035 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2036 * 2. skb is fragmented and the device does not support SG, or if
2037 * at least one of fragments is in highmem and device does not
2038 * support DMA from it.
2040 static inline int skb_needs_linearize(struct sk_buff *skb,
2041 struct net_device *dev)
2043 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
2044 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
2045 illegal_highdma(dev, skb)));
2049 * dev_queue_xmit - transmit a buffer
2050 * @skb: buffer to transmit
2052 * Queue a buffer for transmission to a network device. The caller must
2053 * have set the device and priority and built the buffer before calling
2054 * this function. The function can be called from an interrupt.
2056 * A negative errno code is returned on a failure. A success does not
2057 * guarantee the frame will be transmitted as it may be dropped due
2058 * to congestion or traffic shaping.
2060 * -----------------------------------------------------------------------------------
2061 * I notice this method can also return errors from the queue disciplines,
2062 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2065 * Regardless of the return value, the skb is consumed, so it is currently
2066 * difficult to retry a send to this method. (You can bump the ref count
2067 * before sending to hold a reference for retry if you are careful.)
2069 * When calling this method, interrupts MUST be enabled. This is because
2070 * the BH enable code must have IRQs enabled so that it will not deadlock.
2073 int dev_queue_xmit(struct sk_buff *skb)
2075 struct net_device *dev = skb->dev;
2076 struct netdev_queue *txq;
2080 /* GSO will handle the following emulations directly. */
2081 if (netif_needs_gso(dev, skb))
2084 /* Convert a paged skb to linear, if required */
2085 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2088 /* If packet is not checksummed and device does not support
2089 * checksumming for this protocol, complete checksumming here.
2091 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2092 skb_set_transport_header(skb, skb->csum_start -
2094 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2099 /* Disable soft irqs for various locks below. Also
2100 * stops preemption for RCU.
2104 txq = dev_pick_tx(dev, skb);
2105 q = rcu_dereference_bh(txq->qdisc);
2107 #ifdef CONFIG_NET_CLS_ACT
2108 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2111 rc = __dev_xmit_skb(skb, q, dev, txq);
2115 /* The device has no queue. Common case for software devices:
2116 loopback, all the sorts of tunnels...
2118 Really, it is unlikely that netif_tx_lock protection is necessary
2119 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2121 However, it is possible, that they rely on protection
2124 Check this and shot the lock. It is not prone from deadlocks.
2125 Either shot noqueue qdisc, it is even simpler 8)
2127 if (dev->flags & IFF_UP) {
2128 int cpu = smp_processor_id(); /* ok because BHs are off */
2130 if (txq->xmit_lock_owner != cpu) {
2132 HARD_TX_LOCK(dev, txq, cpu);
2134 if (!netif_tx_queue_stopped(txq)) {
2135 rc = dev_hard_start_xmit(skb, dev, txq);
2136 if (dev_xmit_complete(rc)) {
2137 HARD_TX_UNLOCK(dev, txq);
2141 HARD_TX_UNLOCK(dev, txq);
2142 if (net_ratelimit())
2143 printk(KERN_CRIT "Virtual device %s asks to "
2144 "queue packet!\n", dev->name);
2146 /* Recursion is detected! It is possible,
2148 if (net_ratelimit())
2149 printk(KERN_CRIT "Dead loop on virtual device "
2150 "%s, fix it urgently!\n", dev->name);
2155 rcu_read_unlock_bh();
2161 rcu_read_unlock_bh();
2164 EXPORT_SYMBOL(dev_queue_xmit);
2167 /*=======================================================================
2169 =======================================================================*/
2171 int netdev_max_backlog __read_mostly = 1000;
2172 int netdev_budget __read_mostly = 300;
2173 int weight_p __read_mostly = 64; /* old backlog weight */
2175 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2179 * get_rps_cpu is called from netif_receive_skb and returns the target
2180 * CPU from the RPS map of the receiving queue for a given skb.
2182 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb)
2184 struct ipv6hdr *ip6;
2186 struct netdev_rx_queue *rxqueue;
2187 struct rps_map *map;
2190 u32 addr1, addr2, ports, ihl;
2194 if (skb_rx_queue_recorded(skb)) {
2195 u16 index = skb_get_rx_queue(skb);
2196 if (unlikely(index >= dev->num_rx_queues)) {
2197 if (net_ratelimit()) {
2198 netdev_warn(dev, "received packet on queue "
2199 "%u, but number of RX queues is %u\n",
2200 index, dev->num_rx_queues);
2204 rxqueue = dev->_rx + index;
2208 if (!rxqueue->rps_map)
2212 goto got_hash; /* Skip hash computation on packet header */
2214 switch (skb->protocol) {
2215 case __constant_htons(ETH_P_IP):
2216 if (!pskb_may_pull(skb, sizeof(*ip)))
2219 ip = (struct iphdr *) skb->data;
2220 ip_proto = ip->protocol;
2225 case __constant_htons(ETH_P_IPV6):
2226 if (!pskb_may_pull(skb, sizeof(*ip6)))
2229 ip6 = (struct ipv6hdr *) skb->data;
2230 ip_proto = ip6->nexthdr;
2231 addr1 = ip6->saddr.s6_addr32[3];
2232 addr2 = ip6->daddr.s6_addr32[3];
2246 case IPPROTO_UDPLITE:
2247 if (pskb_may_pull(skb, (ihl * 4) + 4))
2248 ports = *((u32 *) (skb->data + (ihl * 4)));
2255 skb->rxhash = jhash_3words(addr1, addr2, ports, hashrnd);
2260 map = rcu_dereference(rxqueue->rps_map);
2262 u16 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2264 if (cpu_online(tcpu)) {
2276 * This structure holds the per-CPU mask of CPUs for which IPIs are scheduled
2277 * to be sent to kick remote softirq processing. There are two masks since
2278 * the sending of IPIs must be done with interrupts enabled. The select field
2279 * indicates the current mask that enqueue_backlog uses to schedule IPIs.
2280 * select is flipped before net_rps_action is called while still under lock,
2281 * net_rps_action then uses the non-selected mask to send the IPIs and clears
2282 * it without conflicting with enqueue_backlog operation.
2284 struct rps_remote_softirq_cpus {
2288 static DEFINE_PER_CPU(struct rps_remote_softirq_cpus, rps_remote_softirq_cpus);
2290 /* Called from hardirq (IPI) context */
2291 static void trigger_softirq(void *data)
2293 struct softnet_data *queue = data;
2294 __napi_schedule(&queue->backlog);
2295 __get_cpu_var(netdev_rx_stat).received_rps++;
2297 #endif /* CONFIG_SMP */
2300 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2301 * queue (may be a remote CPU queue).
2303 static int enqueue_to_backlog(struct sk_buff *skb, int cpu)
2305 struct softnet_data *queue;
2306 unsigned long flags;
2308 queue = &per_cpu(softnet_data, cpu);
2310 local_irq_save(flags);
2311 __get_cpu_var(netdev_rx_stat).total++;
2313 spin_lock(&queue->input_pkt_queue.lock);
2314 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2315 if (queue->input_pkt_queue.qlen) {
2317 __skb_queue_tail(&queue->input_pkt_queue, skb);
2318 spin_unlock_irqrestore(&queue->input_pkt_queue.lock,
2320 return NET_RX_SUCCESS;
2323 /* Schedule NAPI for backlog device */
2324 if (napi_schedule_prep(&queue->backlog)) {
2326 if (cpu != smp_processor_id()) {
2327 struct rps_remote_softirq_cpus *rcpus =
2328 &__get_cpu_var(rps_remote_softirq_cpus);
2330 cpu_set(cpu, rcpus->mask[rcpus->select]);
2331 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2333 __napi_schedule(&queue->backlog);
2335 __napi_schedule(&queue->backlog);
2341 spin_unlock(&queue->input_pkt_queue.lock);
2343 __get_cpu_var(netdev_rx_stat).dropped++;
2344 local_irq_restore(flags);
2351 * netif_rx - post buffer to the network code
2352 * @skb: buffer to post
2354 * This function receives a packet from a device driver and queues it for
2355 * the upper (protocol) levels to process. It always succeeds. The buffer
2356 * may be dropped during processing for congestion control or by the
2360 * NET_RX_SUCCESS (no congestion)
2361 * NET_RX_DROP (packet was dropped)
2365 int netif_rx(struct sk_buff *skb)
2369 /* if netpoll wants it, pretend we never saw it */
2370 if (netpoll_rx(skb))
2373 if (!skb->tstamp.tv64)
2377 cpu = get_rps_cpu(skb->dev, skb);
2379 cpu = smp_processor_id();
2381 cpu = smp_processor_id();
2384 return enqueue_to_backlog(skb, cpu);
2386 EXPORT_SYMBOL(netif_rx);
2388 int netif_rx_ni(struct sk_buff *skb)
2393 err = netif_rx(skb);
2394 if (local_softirq_pending())
2400 EXPORT_SYMBOL(netif_rx_ni);
2402 static void net_tx_action(struct softirq_action *h)
2404 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2406 if (sd->completion_queue) {
2407 struct sk_buff *clist;
2409 local_irq_disable();
2410 clist = sd->completion_queue;
2411 sd->completion_queue = NULL;
2415 struct sk_buff *skb = clist;
2416 clist = clist->next;
2418 WARN_ON(atomic_read(&skb->users));
2423 if (sd->output_queue) {
2426 local_irq_disable();
2427 head = sd->output_queue;
2428 sd->output_queue = NULL;
2432 struct Qdisc *q = head;
2433 spinlock_t *root_lock;
2435 head = head->next_sched;
2437 root_lock = qdisc_lock(q);
2438 if (spin_trylock(root_lock)) {
2439 smp_mb__before_clear_bit();
2440 clear_bit(__QDISC_STATE_SCHED,
2443 spin_unlock(root_lock);
2445 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2447 __netif_reschedule(q);
2449 smp_mb__before_clear_bit();
2450 clear_bit(__QDISC_STATE_SCHED,
2458 static inline int deliver_skb(struct sk_buff *skb,
2459 struct packet_type *pt_prev,
2460 struct net_device *orig_dev)
2462 atomic_inc(&skb->users);
2463 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2466 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2468 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2469 /* This hook is defined here for ATM LANE */
2470 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2471 unsigned char *addr) __read_mostly;
2472 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2476 * If bridge module is loaded call bridging hook.
2477 * returns NULL if packet was consumed.
2479 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2480 struct sk_buff *skb) __read_mostly;
2481 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2483 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2484 struct packet_type **pt_prev, int *ret,
2485 struct net_device *orig_dev)
2487 struct net_bridge_port *port;
2489 if (skb->pkt_type == PACKET_LOOPBACK ||
2490 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2494 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2498 return br_handle_frame_hook(port, skb);
2501 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2504 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2505 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2506 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2508 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2509 struct packet_type **pt_prev,
2511 struct net_device *orig_dev)
2513 if (skb->dev->macvlan_port == NULL)
2517 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2520 return macvlan_handle_frame_hook(skb);
2523 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2526 #ifdef CONFIG_NET_CLS_ACT
2527 /* TODO: Maybe we should just force sch_ingress to be compiled in
2528 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2529 * a compare and 2 stores extra right now if we dont have it on
2530 * but have CONFIG_NET_CLS_ACT
2531 * NOTE: This doesnt stop any functionality; if you dont have
2532 * the ingress scheduler, you just cant add policies on ingress.
2535 static int ing_filter(struct sk_buff *skb)
2537 struct net_device *dev = skb->dev;
2538 u32 ttl = G_TC_RTTL(skb->tc_verd);
2539 struct netdev_queue *rxq;
2540 int result = TC_ACT_OK;
2543 if (MAX_RED_LOOP < ttl++) {
2545 "Redir loop detected Dropping packet (%d->%d)\n",
2546 skb->skb_iif, dev->ifindex);
2550 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2551 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2553 rxq = &dev->rx_queue;
2556 if (q != &noop_qdisc) {
2557 spin_lock(qdisc_lock(q));
2558 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2559 result = qdisc_enqueue_root(skb, q);
2560 spin_unlock(qdisc_lock(q));
2566 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2567 struct packet_type **pt_prev,
2568 int *ret, struct net_device *orig_dev)
2570 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2574 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2577 /* Huh? Why does turning on AF_PACKET affect this? */
2578 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2581 switch (ing_filter(skb)) {
2595 * netif_nit_deliver - deliver received packets to network taps
2598 * This function is used to deliver incoming packets to network
2599 * taps. It should be used when the normal netif_receive_skb path
2600 * is bypassed, for example because of VLAN acceleration.
2602 void netif_nit_deliver(struct sk_buff *skb)
2604 struct packet_type *ptype;
2606 if (list_empty(&ptype_all))
2609 skb_reset_network_header(skb);
2610 skb_reset_transport_header(skb);
2611 skb->mac_len = skb->network_header - skb->mac_header;
2614 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2615 if (!ptype->dev || ptype->dev == skb->dev)
2616 deliver_skb(skb, ptype, skb->dev);
2621 int __netif_receive_skb(struct sk_buff *skb)
2623 struct packet_type *ptype, *pt_prev;
2624 struct net_device *orig_dev;
2625 struct net_device *master;
2626 struct net_device *null_or_orig;
2627 struct net_device *null_or_bond;
2628 int ret = NET_RX_DROP;
2631 if (!skb->tstamp.tv64)
2634 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2635 return NET_RX_SUCCESS;
2637 /* if we've gotten here through NAPI, check netpoll */
2638 if (netpoll_receive_skb(skb))
2642 skb->skb_iif = skb->dev->ifindex;
2644 null_or_orig = NULL;
2645 orig_dev = skb->dev;
2646 master = ACCESS_ONCE(orig_dev->master);
2648 if (skb_bond_should_drop(skb, master))
2649 null_or_orig = orig_dev; /* deliver only exact match */
2654 __get_cpu_var(netdev_rx_stat).total++;
2656 skb_reset_network_header(skb);
2657 skb_reset_transport_header(skb);
2658 skb->mac_len = skb->network_header - skb->mac_header;
2664 #ifdef CONFIG_NET_CLS_ACT
2665 if (skb->tc_verd & TC_NCLS) {
2666 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2671 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2672 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2673 ptype->dev == orig_dev) {
2675 ret = deliver_skb(skb, pt_prev, orig_dev);
2680 #ifdef CONFIG_NET_CLS_ACT
2681 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2687 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2690 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2695 * Make sure frames received on VLAN interfaces stacked on
2696 * bonding interfaces still make their way to any base bonding
2697 * device that may have registered for a specific ptype. The
2698 * handler may have to adjust skb->dev and orig_dev.
2700 null_or_bond = NULL;
2701 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2702 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2703 null_or_bond = vlan_dev_real_dev(skb->dev);
2706 type = skb->protocol;
2707 list_for_each_entry_rcu(ptype,
2708 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2709 if (ptype->type == type && (ptype->dev == null_or_orig ||
2710 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2711 ptype->dev == null_or_bond)) {
2713 ret = deliver_skb(skb, pt_prev, orig_dev);
2719 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2722 /* Jamal, now you will not able to escape explaining
2723 * me how you were going to use this. :-)
2734 * netif_receive_skb - process receive buffer from network
2735 * @skb: buffer to process
2737 * netif_receive_skb() is the main receive data processing function.
2738 * It always succeeds. The buffer may be dropped during processing
2739 * for congestion control or by the protocol layers.
2741 * This function may only be called from softirq context and interrupts
2742 * should be enabled.
2744 * Return values (usually ignored):
2745 * NET_RX_SUCCESS: no congestion
2746 * NET_RX_DROP: packet was dropped
2748 int netif_receive_skb(struct sk_buff *skb)
2753 cpu = get_rps_cpu(skb->dev, skb);
2756 return __netif_receive_skb(skb);
2758 return enqueue_to_backlog(skb, cpu);
2760 return __netif_receive_skb(skb);
2763 EXPORT_SYMBOL(netif_receive_skb);
2765 /* Network device is going away, flush any packets still pending */
2766 static void flush_backlog(void *arg)
2768 struct net_device *dev = arg;
2769 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2770 struct sk_buff *skb, *tmp;
2772 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2773 if (skb->dev == dev) {
2774 __skb_unlink(skb, &queue->input_pkt_queue);
2779 static int napi_gro_complete(struct sk_buff *skb)
2781 struct packet_type *ptype;
2782 __be16 type = skb->protocol;
2783 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2786 if (NAPI_GRO_CB(skb)->count == 1) {
2787 skb_shinfo(skb)->gso_size = 0;
2792 list_for_each_entry_rcu(ptype, head, list) {
2793 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2796 err = ptype->gro_complete(skb);
2802 WARN_ON(&ptype->list == head);
2804 return NET_RX_SUCCESS;
2808 return netif_receive_skb(skb);
2811 static void napi_gro_flush(struct napi_struct *napi)
2813 struct sk_buff *skb, *next;
2815 for (skb = napi->gro_list; skb; skb = next) {
2818 napi_gro_complete(skb);
2821 napi->gro_count = 0;
2822 napi->gro_list = NULL;
2825 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2827 struct sk_buff **pp = NULL;
2828 struct packet_type *ptype;
2829 __be16 type = skb->protocol;
2830 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2833 enum gro_result ret;
2835 if (!(skb->dev->features & NETIF_F_GRO))
2838 if (skb_is_gso(skb) || skb_has_frags(skb))
2842 list_for_each_entry_rcu(ptype, head, list) {
2843 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2846 skb_set_network_header(skb, skb_gro_offset(skb));
2847 mac_len = skb->network_header - skb->mac_header;
2848 skb->mac_len = mac_len;
2849 NAPI_GRO_CB(skb)->same_flow = 0;
2850 NAPI_GRO_CB(skb)->flush = 0;
2851 NAPI_GRO_CB(skb)->free = 0;
2853 pp = ptype->gro_receive(&napi->gro_list, skb);
2858 if (&ptype->list == head)
2861 same_flow = NAPI_GRO_CB(skb)->same_flow;
2862 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2865 struct sk_buff *nskb = *pp;
2869 napi_gro_complete(nskb);
2876 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2880 NAPI_GRO_CB(skb)->count = 1;
2881 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2882 skb->next = napi->gro_list;
2883 napi->gro_list = skb;
2887 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2888 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2890 BUG_ON(skb->end - skb->tail < grow);
2892 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2895 skb->data_len -= grow;
2897 skb_shinfo(skb)->frags[0].page_offset += grow;
2898 skb_shinfo(skb)->frags[0].size -= grow;
2900 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2901 put_page(skb_shinfo(skb)->frags[0].page);
2902 memmove(skb_shinfo(skb)->frags,
2903 skb_shinfo(skb)->frags + 1,
2904 --skb_shinfo(skb)->nr_frags);
2915 EXPORT_SYMBOL(dev_gro_receive);
2918 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2922 if (netpoll_rx_on(skb))
2925 for (p = napi->gro_list; p; p = p->next) {
2926 NAPI_GRO_CB(p)->same_flow =
2927 (p->dev == skb->dev) &&
2928 !compare_ether_header(skb_mac_header(p),
2929 skb_gro_mac_header(skb));
2930 NAPI_GRO_CB(p)->flush = 0;
2933 return dev_gro_receive(napi, skb);
2936 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
2940 if (netif_receive_skb(skb))
2945 case GRO_MERGED_FREE:
2956 EXPORT_SYMBOL(napi_skb_finish);
2958 void skb_gro_reset_offset(struct sk_buff *skb)
2960 NAPI_GRO_CB(skb)->data_offset = 0;
2961 NAPI_GRO_CB(skb)->frag0 = NULL;
2962 NAPI_GRO_CB(skb)->frag0_len = 0;
2964 if (skb->mac_header == skb->tail &&
2965 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2966 NAPI_GRO_CB(skb)->frag0 =
2967 page_address(skb_shinfo(skb)->frags[0].page) +
2968 skb_shinfo(skb)->frags[0].page_offset;
2969 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2972 EXPORT_SYMBOL(skb_gro_reset_offset);
2974 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2976 skb_gro_reset_offset(skb);
2978 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2980 EXPORT_SYMBOL(napi_gro_receive);
2982 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2984 __skb_pull(skb, skb_headlen(skb));
2985 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2989 EXPORT_SYMBOL(napi_reuse_skb);
2991 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2993 struct sk_buff *skb = napi->skb;
2996 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3002 EXPORT_SYMBOL(napi_get_frags);
3004 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3010 skb->protocol = eth_type_trans(skb, skb->dev);
3012 if (ret == GRO_HELD)
3013 skb_gro_pull(skb, -ETH_HLEN);
3014 else if (netif_receive_skb(skb))
3019 case GRO_MERGED_FREE:
3020 napi_reuse_skb(napi, skb);
3029 EXPORT_SYMBOL(napi_frags_finish);
3031 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3033 struct sk_buff *skb = napi->skb;
3040 skb_reset_mac_header(skb);
3041 skb_gro_reset_offset(skb);
3043 off = skb_gro_offset(skb);
3044 hlen = off + sizeof(*eth);
3045 eth = skb_gro_header_fast(skb, off);
3046 if (skb_gro_header_hard(skb, hlen)) {
3047 eth = skb_gro_header_slow(skb, hlen, off);
3048 if (unlikely(!eth)) {
3049 napi_reuse_skb(napi, skb);
3055 skb_gro_pull(skb, sizeof(*eth));
3058 * This works because the only protocols we care about don't require
3059 * special handling. We'll fix it up properly at the end.
3061 skb->protocol = eth->h_proto;
3066 EXPORT_SYMBOL(napi_frags_skb);
3068 gro_result_t napi_gro_frags(struct napi_struct *napi)
3070 struct sk_buff *skb = napi_frags_skb(napi);
3075 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3077 EXPORT_SYMBOL(napi_gro_frags);
3079 static int process_backlog(struct napi_struct *napi, int quota)
3082 struct softnet_data *queue = &__get_cpu_var(softnet_data);
3083 unsigned long start_time = jiffies;
3085 napi->weight = weight_p;
3087 struct sk_buff *skb;
3089 spin_lock_irq(&queue->input_pkt_queue.lock);
3090 skb = __skb_dequeue(&queue->input_pkt_queue);
3092 __napi_complete(napi);
3093 spin_unlock_irq(&queue->input_pkt_queue.lock);
3096 spin_unlock_irq(&queue->input_pkt_queue.lock);
3098 __netif_receive_skb(skb);
3099 } while (++work < quota && jiffies == start_time);
3105 * __napi_schedule - schedule for receive
3106 * @n: entry to schedule
3108 * The entry's receive function will be scheduled to run
3110 void __napi_schedule(struct napi_struct *n)
3112 unsigned long flags;
3114 local_irq_save(flags);
3115 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
3116 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3117 local_irq_restore(flags);
3119 EXPORT_SYMBOL(__napi_schedule);
3121 void __napi_complete(struct napi_struct *n)
3123 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3124 BUG_ON(n->gro_list);
3126 list_del(&n->poll_list);
3127 smp_mb__before_clear_bit();
3128 clear_bit(NAPI_STATE_SCHED, &n->state);
3130 EXPORT_SYMBOL(__napi_complete);
3132 void napi_complete(struct napi_struct *n)
3134 unsigned long flags;
3137 * don't let napi dequeue from the cpu poll list
3138 * just in case its running on a different cpu
3140 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3144 local_irq_save(flags);
3146 local_irq_restore(flags);
3148 EXPORT_SYMBOL(napi_complete);
3150 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3151 int (*poll)(struct napi_struct *, int), int weight)
3153 INIT_LIST_HEAD(&napi->poll_list);
3154 napi->gro_count = 0;
3155 napi->gro_list = NULL;
3158 napi->weight = weight;
3159 list_add(&napi->dev_list, &dev->napi_list);
3161 #ifdef CONFIG_NETPOLL
3162 spin_lock_init(&napi->poll_lock);
3163 napi->poll_owner = -1;
3165 set_bit(NAPI_STATE_SCHED, &napi->state);
3167 EXPORT_SYMBOL(netif_napi_add);
3169 void netif_napi_del(struct napi_struct *napi)
3171 struct sk_buff *skb, *next;
3173 list_del_init(&napi->dev_list);
3174 napi_free_frags(napi);
3176 for (skb = napi->gro_list; skb; skb = next) {
3182 napi->gro_list = NULL;
3183 napi->gro_count = 0;
3185 EXPORT_SYMBOL(netif_napi_del);
3189 * net_rps_action sends any pending IPI's for rps. This is only called from
3190 * softirq and interrupts must be enabled.
3192 static void net_rps_action(cpumask_t *mask)
3196 /* Send pending IPI's to kick RPS processing on remote cpus. */
3197 for_each_cpu_mask_nr(cpu, *mask) {
3198 struct softnet_data *queue = &per_cpu(softnet_data, cpu);
3199 if (cpu_online(cpu))
3200 __smp_call_function_single(cpu, &queue->csd, 0);
3206 static void net_rx_action(struct softirq_action *h)
3208 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
3209 unsigned long time_limit = jiffies + 2;
3210 int budget = netdev_budget;
3214 struct rps_remote_softirq_cpus *rcpus;
3217 local_irq_disable();
3219 while (!list_empty(list)) {
3220 struct napi_struct *n;
3223 /* If softirq window is exhuasted then punt.
3224 * Allow this to run for 2 jiffies since which will allow
3225 * an average latency of 1.5/HZ.
3227 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3232 /* Even though interrupts have been re-enabled, this
3233 * access is safe because interrupts can only add new
3234 * entries to the tail of this list, and only ->poll()
3235 * calls can remove this head entry from the list.
3237 n = list_first_entry(list, struct napi_struct, poll_list);
3239 have = netpoll_poll_lock(n);
3243 /* This NAPI_STATE_SCHED test is for avoiding a race
3244 * with netpoll's poll_napi(). Only the entity which
3245 * obtains the lock and sees NAPI_STATE_SCHED set will
3246 * actually make the ->poll() call. Therefore we avoid
3247 * accidently calling ->poll() when NAPI is not scheduled.
3250 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3251 work = n->poll(n, weight);
3255 WARN_ON_ONCE(work > weight);
3259 local_irq_disable();
3261 /* Drivers must not modify the NAPI state if they
3262 * consume the entire weight. In such cases this code
3263 * still "owns" the NAPI instance and therefore can
3264 * move the instance around on the list at-will.
3266 if (unlikely(work == weight)) {
3267 if (unlikely(napi_disable_pending(n))) {
3270 local_irq_disable();
3272 list_move_tail(&n->poll_list, list);
3275 netpoll_poll_unlock(have);
3279 rcpus = &__get_cpu_var(rps_remote_softirq_cpus);
3280 select = rcpus->select;
3285 net_rps_action(&rcpus->mask[select]);
3290 #ifdef CONFIG_NET_DMA
3292 * There may not be any more sk_buffs coming right now, so push
3293 * any pending DMA copies to hardware
3295 dma_issue_pending_all();
3301 __get_cpu_var(netdev_rx_stat).time_squeeze++;
3302 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3306 static gifconf_func_t *gifconf_list[NPROTO];
3309 * register_gifconf - register a SIOCGIF handler
3310 * @family: Address family
3311 * @gifconf: Function handler
3313 * Register protocol dependent address dumping routines. The handler
3314 * that is passed must not be freed or reused until it has been replaced
3315 * by another handler.
3317 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3319 if (family >= NPROTO)
3321 gifconf_list[family] = gifconf;
3324 EXPORT_SYMBOL(register_gifconf);
3328 * Map an interface index to its name (SIOCGIFNAME)
3332 * We need this ioctl for efficient implementation of the
3333 * if_indextoname() function required by the IPv6 API. Without
3334 * it, we would have to search all the interfaces to find a
3338 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3340 struct net_device *dev;
3344 * Fetch the caller's info block.
3347 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3351 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3357 strcpy(ifr.ifr_name, dev->name);
3360 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3366 * Perform a SIOCGIFCONF call. This structure will change
3367 * size eventually, and there is nothing I can do about it.
3368 * Thus we will need a 'compatibility mode'.
3371 static int dev_ifconf(struct net *net, char __user *arg)
3374 struct net_device *dev;
3381 * Fetch the caller's info block.
3384 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3391 * Loop over the interfaces, and write an info block for each.
3395 for_each_netdev(net, dev) {
3396 for (i = 0; i < NPROTO; i++) {
3397 if (gifconf_list[i]) {
3400 done = gifconf_list[i](dev, NULL, 0);
3402 done = gifconf_list[i](dev, pos + total,
3412 * All done. Write the updated control block back to the caller.
3414 ifc.ifc_len = total;
3417 * Both BSD and Solaris return 0 here, so we do too.
3419 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3422 #ifdef CONFIG_PROC_FS
3424 * This is invoked by the /proc filesystem handler to display a device
3427 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3430 struct net *net = seq_file_net(seq);
3432 struct net_device *dev;
3436 return SEQ_START_TOKEN;
3439 for_each_netdev_rcu(net, dev)
3446 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3448 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3449 first_net_device(seq_file_net(seq)) :
3450 next_net_device((struct net_device *)v);
3453 return rcu_dereference(dev);
3456 void dev_seq_stop(struct seq_file *seq, void *v)
3462 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3464 const struct net_device_stats *stats = dev_get_stats(dev);
3466 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3467 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3468 dev->name, stats->rx_bytes, stats->rx_packets,
3470 stats->rx_dropped + stats->rx_missed_errors,
3471 stats->rx_fifo_errors,
3472 stats->rx_length_errors + stats->rx_over_errors +
3473 stats->rx_crc_errors + stats->rx_frame_errors,
3474 stats->rx_compressed, stats->multicast,
3475 stats->tx_bytes, stats->tx_packets,
3476 stats->tx_errors, stats->tx_dropped,
3477 stats->tx_fifo_errors, stats->collisions,
3478 stats->tx_carrier_errors +
3479 stats->tx_aborted_errors +
3480 stats->tx_window_errors +
3481 stats->tx_heartbeat_errors,
3482 stats->tx_compressed);
3486 * Called from the PROCfs module. This now uses the new arbitrary sized
3487 * /proc/net interface to create /proc/net/dev
3489 static int dev_seq_show(struct seq_file *seq, void *v)
3491 if (v == SEQ_START_TOKEN)
3492 seq_puts(seq, "Inter-| Receive "
3494 " face |bytes packets errs drop fifo frame "
3495 "compressed multicast|bytes packets errs "
3496 "drop fifo colls carrier compressed\n");
3498 dev_seq_printf_stats(seq, v);
3502 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3504 struct netif_rx_stats *rc = NULL;
3506 while (*pos < nr_cpu_ids)
3507 if (cpu_online(*pos)) {
3508 rc = &per_cpu(netdev_rx_stat, *pos);
3515 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3517 return softnet_get_online(pos);
3520 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3523 return softnet_get_online(pos);
3526 static void softnet_seq_stop(struct seq_file *seq, void *v)
3530 static int softnet_seq_show(struct seq_file *seq, void *v)
3532 struct netif_rx_stats *s = v;
3534 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3535 s->total, s->dropped, s->time_squeeze, 0,
3536 0, 0, 0, 0, /* was fastroute */
3537 s->cpu_collision, s->received_rps);
3541 static const struct seq_operations dev_seq_ops = {
3542 .start = dev_seq_start,
3543 .next = dev_seq_next,
3544 .stop = dev_seq_stop,
3545 .show = dev_seq_show,
3548 static int dev_seq_open(struct inode *inode, struct file *file)
3550 return seq_open_net(inode, file, &dev_seq_ops,
3551 sizeof(struct seq_net_private));
3554 static const struct file_operations dev_seq_fops = {
3555 .owner = THIS_MODULE,
3556 .open = dev_seq_open,
3558 .llseek = seq_lseek,
3559 .release = seq_release_net,
3562 static const struct seq_operations softnet_seq_ops = {
3563 .start = softnet_seq_start,
3564 .next = softnet_seq_next,
3565 .stop = softnet_seq_stop,
3566 .show = softnet_seq_show,
3569 static int softnet_seq_open(struct inode *inode, struct file *file)
3571 return seq_open(file, &softnet_seq_ops);
3574 static const struct file_operations softnet_seq_fops = {
3575 .owner = THIS_MODULE,
3576 .open = softnet_seq_open,
3578 .llseek = seq_lseek,
3579 .release = seq_release,
3582 static void *ptype_get_idx(loff_t pos)
3584 struct packet_type *pt = NULL;
3588 list_for_each_entry_rcu(pt, &ptype_all, list) {
3594 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3595 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3604 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3608 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3611 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3613 struct packet_type *pt;
3614 struct list_head *nxt;
3618 if (v == SEQ_START_TOKEN)
3619 return ptype_get_idx(0);
3622 nxt = pt->list.next;
3623 if (pt->type == htons(ETH_P_ALL)) {
3624 if (nxt != &ptype_all)
3627 nxt = ptype_base[0].next;
3629 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3631 while (nxt == &ptype_base[hash]) {
3632 if (++hash >= PTYPE_HASH_SIZE)
3634 nxt = ptype_base[hash].next;
3637 return list_entry(nxt, struct packet_type, list);
3640 static void ptype_seq_stop(struct seq_file *seq, void *v)
3646 static int ptype_seq_show(struct seq_file *seq, void *v)
3648 struct packet_type *pt = v;
3650 if (v == SEQ_START_TOKEN)
3651 seq_puts(seq, "Type Device Function\n");
3652 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3653 if (pt->type == htons(ETH_P_ALL))
3654 seq_puts(seq, "ALL ");
3656 seq_printf(seq, "%04x", ntohs(pt->type));
3658 seq_printf(seq, " %-8s %pF\n",
3659 pt->dev ? pt->dev->name : "", pt->func);
3665 static const struct seq_operations ptype_seq_ops = {
3666 .start = ptype_seq_start,
3667 .next = ptype_seq_next,
3668 .stop = ptype_seq_stop,
3669 .show = ptype_seq_show,
3672 static int ptype_seq_open(struct inode *inode, struct file *file)
3674 return seq_open_net(inode, file, &ptype_seq_ops,
3675 sizeof(struct seq_net_private));
3678 static const struct file_operations ptype_seq_fops = {
3679 .owner = THIS_MODULE,
3680 .open = ptype_seq_open,
3682 .llseek = seq_lseek,
3683 .release = seq_release_net,
3687 static int __net_init dev_proc_net_init(struct net *net)
3691 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3693 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3695 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3698 if (wext_proc_init(net))
3704 proc_net_remove(net, "ptype");
3706 proc_net_remove(net, "softnet_stat");
3708 proc_net_remove(net, "dev");
3712 static void __net_exit dev_proc_net_exit(struct net *net)
3714 wext_proc_exit(net);
3716 proc_net_remove(net, "ptype");
3717 proc_net_remove(net, "softnet_stat");
3718 proc_net_remove(net, "dev");
3721 static struct pernet_operations __net_initdata dev_proc_ops = {
3722 .init = dev_proc_net_init,
3723 .exit = dev_proc_net_exit,
3726 static int __init dev_proc_init(void)
3728 return register_pernet_subsys(&dev_proc_ops);
3731 #define dev_proc_init() 0
3732 #endif /* CONFIG_PROC_FS */
3736 * netdev_set_master - set up master/slave pair
3737 * @slave: slave device
3738 * @master: new master device
3740 * Changes the master device of the slave. Pass %NULL to break the
3741 * bonding. The caller must hold the RTNL semaphore. On a failure
3742 * a negative errno code is returned. On success the reference counts
3743 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3744 * function returns zero.
3746 int netdev_set_master(struct net_device *slave, struct net_device *master)
3748 struct net_device *old = slave->master;
3758 slave->master = master;
3765 slave->flags |= IFF_SLAVE;
3767 slave->flags &= ~IFF_SLAVE;
3769 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3772 EXPORT_SYMBOL(netdev_set_master);
3774 static void dev_change_rx_flags(struct net_device *dev, int flags)
3776 const struct net_device_ops *ops = dev->netdev_ops;
3778 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3779 ops->ndo_change_rx_flags(dev, flags);
3782 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3784 unsigned short old_flags = dev->flags;
3790 dev->flags |= IFF_PROMISC;
3791 dev->promiscuity += inc;
3792 if (dev->promiscuity == 0) {
3795 * If inc causes overflow, untouch promisc and return error.
3798 dev->flags &= ~IFF_PROMISC;
3800 dev->promiscuity -= inc;
3801 printk(KERN_WARNING "%s: promiscuity touches roof, "
3802 "set promiscuity failed, promiscuity feature "
3803 "of device might be broken.\n", dev->name);
3807 if (dev->flags != old_flags) {
3808 printk(KERN_INFO "device %s %s promiscuous mode\n",
3809 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3811 if (audit_enabled) {
3812 current_uid_gid(&uid, &gid);
3813 audit_log(current->audit_context, GFP_ATOMIC,
3814 AUDIT_ANOM_PROMISCUOUS,
3815 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3816 dev->name, (dev->flags & IFF_PROMISC),
3817 (old_flags & IFF_PROMISC),
3818 audit_get_loginuid(current),
3820 audit_get_sessionid(current));
3823 dev_change_rx_flags(dev, IFF_PROMISC);
3829 * dev_set_promiscuity - update promiscuity count on a device
3833 * Add or remove promiscuity from a device. While the count in the device
3834 * remains above zero the interface remains promiscuous. Once it hits zero
3835 * the device reverts back to normal filtering operation. A negative inc
3836 * value is used to drop promiscuity on the device.
3837 * Return 0 if successful or a negative errno code on error.
3839 int dev_set_promiscuity(struct net_device *dev, int inc)
3841 unsigned short old_flags = dev->flags;
3844 err = __dev_set_promiscuity(dev, inc);
3847 if (dev->flags != old_flags)
3848 dev_set_rx_mode(dev);
3851 EXPORT_SYMBOL(dev_set_promiscuity);
3854 * dev_set_allmulti - update allmulti count on a device
3858 * Add or remove reception of all multicast frames to a device. While the
3859 * count in the device remains above zero the interface remains listening
3860 * to all interfaces. Once it hits zero the device reverts back to normal
3861 * filtering operation. A negative @inc value is used to drop the counter
3862 * when releasing a resource needing all multicasts.
3863 * Return 0 if successful or a negative errno code on error.
3866 int dev_set_allmulti(struct net_device *dev, int inc)
3868 unsigned short old_flags = dev->flags;
3872 dev->flags |= IFF_ALLMULTI;
3873 dev->allmulti += inc;
3874 if (dev->allmulti == 0) {
3877 * If inc causes overflow, untouch allmulti and return error.
3880 dev->flags &= ~IFF_ALLMULTI;
3882 dev->allmulti -= inc;
3883 printk(KERN_WARNING "%s: allmulti touches roof, "
3884 "set allmulti failed, allmulti feature of "
3885 "device might be broken.\n", dev->name);
3889 if (dev->flags ^ old_flags) {
3890 dev_change_rx_flags(dev, IFF_ALLMULTI);
3891 dev_set_rx_mode(dev);
3895 EXPORT_SYMBOL(dev_set_allmulti);
3898 * Upload unicast and multicast address lists to device and
3899 * configure RX filtering. When the device doesn't support unicast
3900 * filtering it is put in promiscuous mode while unicast addresses
3903 void __dev_set_rx_mode(struct net_device *dev)
3905 const struct net_device_ops *ops = dev->netdev_ops;
3907 /* dev_open will call this function so the list will stay sane. */
3908 if (!(dev->flags&IFF_UP))
3911 if (!netif_device_present(dev))
3914 if (ops->ndo_set_rx_mode)
3915 ops->ndo_set_rx_mode(dev);
3917 /* Unicast addresses changes may only happen under the rtnl,
3918 * therefore calling __dev_set_promiscuity here is safe.
3920 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
3921 __dev_set_promiscuity(dev, 1);
3922 dev->uc_promisc = 1;
3923 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
3924 __dev_set_promiscuity(dev, -1);
3925 dev->uc_promisc = 0;
3928 if (ops->ndo_set_multicast_list)
3929 ops->ndo_set_multicast_list(dev);
3933 void dev_set_rx_mode(struct net_device *dev)
3935 netif_addr_lock_bh(dev);
3936 __dev_set_rx_mode(dev);
3937 netif_addr_unlock_bh(dev);
3940 /* hw addresses list handling functions */
3942 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3943 int addr_len, unsigned char addr_type)
3945 struct netdev_hw_addr *ha;
3948 if (addr_len > MAX_ADDR_LEN)
3951 list_for_each_entry(ha, &list->list, list) {
3952 if (!memcmp(ha->addr, addr, addr_len) &&
3953 ha->type == addr_type) {
3960 alloc_size = sizeof(*ha);
3961 if (alloc_size < L1_CACHE_BYTES)
3962 alloc_size = L1_CACHE_BYTES;
3963 ha = kmalloc(alloc_size, GFP_ATOMIC);
3966 memcpy(ha->addr, addr, addr_len);
3967 ha->type = addr_type;
3970 list_add_tail_rcu(&ha->list, &list->list);
3975 static void ha_rcu_free(struct rcu_head *head)
3977 struct netdev_hw_addr *ha;
3979 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3983 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3984 int addr_len, unsigned char addr_type)
3986 struct netdev_hw_addr *ha;
3988 list_for_each_entry(ha, &list->list, list) {
3989 if (!memcmp(ha->addr, addr, addr_len) &&
3990 (ha->type == addr_type || !addr_type)) {
3993 list_del_rcu(&ha->list);
3994 call_rcu(&ha->rcu_head, ha_rcu_free);
4002 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
4003 struct netdev_hw_addr_list *from_list,
4005 unsigned char addr_type)
4008 struct netdev_hw_addr *ha, *ha2;
4011 list_for_each_entry(ha, &from_list->list, list) {
4012 type = addr_type ? addr_type : ha->type;
4013 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
4020 list_for_each_entry(ha2, &from_list->list, list) {
4023 type = addr_type ? addr_type : ha2->type;
4024 __hw_addr_del(to_list, ha2->addr, addr_len, type);
4029 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
4030 struct netdev_hw_addr_list *from_list,
4032 unsigned char addr_type)
4034 struct netdev_hw_addr *ha;
4037 list_for_each_entry(ha, &from_list->list, list) {
4038 type = addr_type ? addr_type : ha->type;
4039 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
4043 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
4044 struct netdev_hw_addr_list *from_list,
4048 struct netdev_hw_addr *ha, *tmp;
4050 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
4052 err = __hw_addr_add(to_list, ha->addr,
4053 addr_len, ha->type);
4058 } else if (ha->refcount == 1) {
4059 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
4060 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
4066 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
4067 struct netdev_hw_addr_list *from_list,
4070 struct netdev_hw_addr *ha, *tmp;
4072 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
4074 __hw_addr_del(to_list, ha->addr,
4075 addr_len, ha->type);
4077 __hw_addr_del(from_list, ha->addr,
4078 addr_len, ha->type);
4083 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
4085 struct netdev_hw_addr *ha, *tmp;
4087 list_for_each_entry_safe(ha, tmp, &list->list, list) {
4088 list_del_rcu(&ha->list);
4089 call_rcu(&ha->rcu_head, ha_rcu_free);
4094 static void __hw_addr_init(struct netdev_hw_addr_list *list)
4096 INIT_LIST_HEAD(&list->list);
4100 /* Device addresses handling functions */
4102 static void dev_addr_flush(struct net_device *dev)
4104 /* rtnl_mutex must be held here */
4106 __hw_addr_flush(&dev->dev_addrs);
4107 dev->dev_addr = NULL;
4110 static int dev_addr_init(struct net_device *dev)
4112 unsigned char addr[MAX_ADDR_LEN];
4113 struct netdev_hw_addr *ha;
4116 /* rtnl_mutex must be held here */
4118 __hw_addr_init(&dev->dev_addrs);
4119 memset(addr, 0, sizeof(addr));
4120 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
4121 NETDEV_HW_ADDR_T_LAN);
4124 * Get the first (previously created) address from the list
4125 * and set dev_addr pointer to this location.
4127 ha = list_first_entry(&dev->dev_addrs.list,
4128 struct netdev_hw_addr, list);
4129 dev->dev_addr = ha->addr;
4135 * dev_addr_add - Add a device address
4137 * @addr: address to add
4138 * @addr_type: address type
4140 * Add a device address to the device or increase the reference count if
4141 * it already exists.
4143 * The caller must hold the rtnl_mutex.
4145 int dev_addr_add(struct net_device *dev, unsigned char *addr,
4146 unsigned char addr_type)
4152 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
4154 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4157 EXPORT_SYMBOL(dev_addr_add);
4160 * dev_addr_del - Release a device address.
4162 * @addr: address to delete
4163 * @addr_type: address type
4165 * Release reference to a device address and remove it from the device
4166 * if the reference count drops to zero.
4168 * The caller must hold the rtnl_mutex.
4170 int dev_addr_del(struct net_device *dev, unsigned char *addr,
4171 unsigned char addr_type)
4174 struct netdev_hw_addr *ha;
4179 * We can not remove the first address from the list because
4180 * dev->dev_addr points to that.
4182 ha = list_first_entry(&dev->dev_addrs.list,
4183 struct netdev_hw_addr, list);
4184 if (ha->addr == dev->dev_addr && ha->refcount == 1)
4187 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
4190 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4193 EXPORT_SYMBOL(dev_addr_del);
4196 * dev_addr_add_multiple - Add device addresses from another device
4197 * @to_dev: device to which addresses will be added
4198 * @from_dev: device from which addresses will be added
4199 * @addr_type: address type - 0 means type will be used from from_dev
4201 * Add device addresses of the one device to another.
4203 * The caller must hold the rtnl_mutex.
4205 int dev_addr_add_multiple(struct net_device *to_dev,
4206 struct net_device *from_dev,
4207 unsigned char addr_type)
4213 if (from_dev->addr_len != to_dev->addr_len)
4215 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
4216 to_dev->addr_len, addr_type);
4218 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
4221 EXPORT_SYMBOL(dev_addr_add_multiple);
4224 * dev_addr_del_multiple - Delete device addresses by another device
4225 * @to_dev: device where the addresses will be deleted
4226 * @from_dev: device by which addresses the addresses will be deleted
4227 * @addr_type: address type - 0 means type will used from from_dev
4229 * Deletes addresses in to device by the list of addresses in from device.
4231 * The caller must hold the rtnl_mutex.
4233 int dev_addr_del_multiple(struct net_device *to_dev,
4234 struct net_device *from_dev,
4235 unsigned char addr_type)
4239 if (from_dev->addr_len != to_dev->addr_len)
4241 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
4242 to_dev->addr_len, addr_type);
4243 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
4246 EXPORT_SYMBOL(dev_addr_del_multiple);
4248 /* multicast addresses handling functions */
4250 int __dev_addr_delete(struct dev_addr_list **list, int *count,
4251 void *addr, int alen, int glbl)
4253 struct dev_addr_list *da;
4255 for (; (da = *list) != NULL; list = &da->next) {
4256 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
4257 alen == da->da_addrlen) {
4259 int old_glbl = da->da_gusers;
4276 int __dev_addr_add(struct dev_addr_list **list, int *count,
4277 void *addr, int alen, int glbl)
4279 struct dev_addr_list *da;
4281 for (da = *list; da != NULL; da = da->next) {
4282 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
4283 da->da_addrlen == alen) {
4285 int old_glbl = da->da_gusers;
4295 da = kzalloc(sizeof(*da), GFP_ATOMIC);
4298 memcpy(da->da_addr, addr, alen);
4299 da->da_addrlen = alen;
4301 da->da_gusers = glbl ? 1 : 0;
4309 * dev_unicast_delete - Release secondary unicast address.
4311 * @addr: address to delete
4313 * Release reference to a secondary unicast address and remove it
4314 * from the device if the reference count drops to zero.
4316 * The caller must hold the rtnl_mutex.
4318 int dev_unicast_delete(struct net_device *dev, void *addr)
4324 netif_addr_lock_bh(dev);
4325 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
4326 NETDEV_HW_ADDR_T_UNICAST);
4328 __dev_set_rx_mode(dev);
4329 netif_addr_unlock_bh(dev);
4332 EXPORT_SYMBOL(dev_unicast_delete);
4335 * dev_unicast_add - add a secondary unicast address
4337 * @addr: address to add
4339 * Add a secondary unicast address to the device or increase
4340 * the reference count if it already exists.
4342 * The caller must hold the rtnl_mutex.
4344 int dev_unicast_add(struct net_device *dev, void *addr)
4350 netif_addr_lock_bh(dev);
4351 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
4352 NETDEV_HW_ADDR_T_UNICAST);
4354 __dev_set_rx_mode(dev);
4355 netif_addr_unlock_bh(dev);
4358 EXPORT_SYMBOL(dev_unicast_add);
4360 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
4361 struct dev_addr_list **from, int *from_count)
4363 struct dev_addr_list *da, *next;
4367 while (da != NULL) {
4369 if (!da->da_synced) {
4370 err = __dev_addr_add(to, to_count,
4371 da->da_addr, da->da_addrlen, 0);
4376 } else if (da->da_users == 1) {
4377 __dev_addr_delete(to, to_count,
4378 da->da_addr, da->da_addrlen, 0);
4379 __dev_addr_delete(from, from_count,
4380 da->da_addr, da->da_addrlen, 0);
4386 EXPORT_SYMBOL_GPL(__dev_addr_sync);
4388 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4389 struct dev_addr_list **from, int *from_count)
4391 struct dev_addr_list *da, *next;
4394 while (da != NULL) {
4396 if (da->da_synced) {
4397 __dev_addr_delete(to, to_count,
4398 da->da_addr, da->da_addrlen, 0);
4400 __dev_addr_delete(from, from_count,
4401 da->da_addr, da->da_addrlen, 0);
4406 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4409 * dev_unicast_sync - Synchronize device's unicast list to another device
4410 * @to: destination device
4411 * @from: source device
4413 * Add newly added addresses to the destination device and release
4414 * addresses that have no users left. The source device must be
4415 * locked by netif_tx_lock_bh.
4417 * This function is intended to be called from the dev->set_rx_mode
4418 * function of layered software devices.
4420 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4424 if (to->addr_len != from->addr_len)
4427 netif_addr_lock_bh(to);
4428 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4430 __dev_set_rx_mode(to);
4431 netif_addr_unlock_bh(to);
4434 EXPORT_SYMBOL(dev_unicast_sync);
4437 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4438 * @to: destination device
4439 * @from: source device
4441 * Remove all addresses that were added to the destination device by
4442 * dev_unicast_sync(). This function is intended to be called from the
4443 * dev->stop function of layered software devices.
4445 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4447 if (to->addr_len != from->addr_len)
4450 netif_addr_lock_bh(from);
4451 netif_addr_lock(to);
4452 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4453 __dev_set_rx_mode(to);
4454 netif_addr_unlock(to);
4455 netif_addr_unlock_bh(from);
4457 EXPORT_SYMBOL(dev_unicast_unsync);
4459 void dev_unicast_flush(struct net_device *dev)
4461 netif_addr_lock_bh(dev);
4462 __hw_addr_flush(&dev->uc);
4463 netif_addr_unlock_bh(dev);
4465 EXPORT_SYMBOL(dev_unicast_flush);
4467 static void dev_unicast_init(struct net_device *dev)
4469 __hw_addr_init(&dev->uc);
4473 static void __dev_addr_discard(struct dev_addr_list **list)
4475 struct dev_addr_list *tmp;
4477 while (*list != NULL) {
4480 if (tmp->da_users > tmp->da_gusers)
4481 printk("__dev_addr_discard: address leakage! "
4482 "da_users=%d\n", tmp->da_users);
4487 void dev_addr_discard(struct net_device *dev)
4489 netif_addr_lock_bh(dev);
4491 __dev_addr_discard(&dev->mc_list);
4492 netdev_mc_count(dev) = 0;
4494 netif_addr_unlock_bh(dev);
4496 EXPORT_SYMBOL(dev_addr_discard);
4499 * dev_get_flags - get flags reported to userspace
4502 * Get the combination of flag bits exported through APIs to userspace.
4504 unsigned dev_get_flags(const struct net_device *dev)
4508 flags = (dev->flags & ~(IFF_PROMISC |
4513 (dev->gflags & (IFF_PROMISC |
4516 if (netif_running(dev)) {
4517 if (netif_oper_up(dev))
4518 flags |= IFF_RUNNING;
4519 if (netif_carrier_ok(dev))
4520 flags |= IFF_LOWER_UP;
4521 if (netif_dormant(dev))
4522 flags |= IFF_DORMANT;
4527 EXPORT_SYMBOL(dev_get_flags);
4529 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4531 int old_flags = dev->flags;
4537 * Set the flags on our device.
4540 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4541 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4543 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4547 * Load in the correct multicast list now the flags have changed.
4550 if ((old_flags ^ flags) & IFF_MULTICAST)
4551 dev_change_rx_flags(dev, IFF_MULTICAST);
4553 dev_set_rx_mode(dev);
4556 * Have we downed the interface. We handle IFF_UP ourselves
4557 * according to user attempts to set it, rather than blindly
4562 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4563 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4566 dev_set_rx_mode(dev);
4569 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4570 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4572 dev->gflags ^= IFF_PROMISC;
4573 dev_set_promiscuity(dev, inc);
4576 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4577 is important. Some (broken) drivers set IFF_PROMISC, when
4578 IFF_ALLMULTI is requested not asking us and not reporting.
4580 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4581 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4583 dev->gflags ^= IFF_ALLMULTI;
4584 dev_set_allmulti(dev, inc);
4590 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4592 unsigned int changes = dev->flags ^ old_flags;
4594 if (changes & IFF_UP) {
4595 if (dev->flags & IFF_UP)
4596 call_netdevice_notifiers(NETDEV_UP, dev);
4598 call_netdevice_notifiers(NETDEV_DOWN, dev);
4601 if (dev->flags & IFF_UP &&
4602 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4603 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4607 * dev_change_flags - change device settings
4609 * @flags: device state flags
4611 * Change settings on device based state flags. The flags are
4612 * in the userspace exported format.
4614 int dev_change_flags(struct net_device *dev, unsigned flags)
4617 int old_flags = dev->flags;
4619 ret = __dev_change_flags(dev, flags);
4623 changes = old_flags ^ dev->flags;
4625 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4627 __dev_notify_flags(dev, old_flags);
4630 EXPORT_SYMBOL(dev_change_flags);
4633 * dev_set_mtu - Change maximum transfer unit
4635 * @new_mtu: new transfer unit
4637 * Change the maximum transfer size of the network device.
4639 int dev_set_mtu(struct net_device *dev, int new_mtu)
4641 const struct net_device_ops *ops = dev->netdev_ops;
4644 if (new_mtu == dev->mtu)
4647 /* MTU must be positive. */
4651 if (!netif_device_present(dev))
4655 if (ops->ndo_change_mtu)
4656 err = ops->ndo_change_mtu(dev, new_mtu);
4660 if (!err && dev->flags & IFF_UP)
4661 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4664 EXPORT_SYMBOL(dev_set_mtu);
4667 * dev_set_mac_address - Change Media Access Control Address
4671 * Change the hardware (MAC) address of the device
4673 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4675 const struct net_device_ops *ops = dev->netdev_ops;
4678 if (!ops->ndo_set_mac_address)
4680 if (sa->sa_family != dev->type)
4682 if (!netif_device_present(dev))
4684 err = ops->ndo_set_mac_address(dev, sa);
4686 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4689 EXPORT_SYMBOL(dev_set_mac_address);
4692 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4694 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4697 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4703 case SIOCGIFFLAGS: /* Get interface flags */
4704 ifr->ifr_flags = (short) dev_get_flags(dev);
4707 case SIOCGIFMETRIC: /* Get the metric on the interface
4708 (currently unused) */
4709 ifr->ifr_metric = 0;
4712 case SIOCGIFMTU: /* Get the MTU of a device */
4713 ifr->ifr_mtu = dev->mtu;
4718 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4720 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4721 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4722 ifr->ifr_hwaddr.sa_family = dev->type;
4730 ifr->ifr_map.mem_start = dev->mem_start;
4731 ifr->ifr_map.mem_end = dev->mem_end;
4732 ifr->ifr_map.base_addr = dev->base_addr;
4733 ifr->ifr_map.irq = dev->irq;
4734 ifr->ifr_map.dma = dev->dma;
4735 ifr->ifr_map.port = dev->if_port;
4739 ifr->ifr_ifindex = dev->ifindex;
4743 ifr->ifr_qlen = dev->tx_queue_len;
4747 /* dev_ioctl() should ensure this case
4759 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4761 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4764 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4765 const struct net_device_ops *ops;
4770 ops = dev->netdev_ops;
4773 case SIOCSIFFLAGS: /* Set interface flags */
4774 return dev_change_flags(dev, ifr->ifr_flags);
4776 case SIOCSIFMETRIC: /* Set the metric on the interface
4777 (currently unused) */
4780 case SIOCSIFMTU: /* Set the MTU of a device */
4781 return dev_set_mtu(dev, ifr->ifr_mtu);
4784 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4786 case SIOCSIFHWBROADCAST:
4787 if (ifr->ifr_hwaddr.sa_family != dev->type)
4789 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4790 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4791 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4795 if (ops->ndo_set_config) {
4796 if (!netif_device_present(dev))
4798 return ops->ndo_set_config(dev, &ifr->ifr_map);
4803 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4804 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4806 if (!netif_device_present(dev))
4808 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4812 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4813 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4815 if (!netif_device_present(dev))
4817 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4821 if (ifr->ifr_qlen < 0)
4823 dev->tx_queue_len = ifr->ifr_qlen;
4827 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4828 return dev_change_name(dev, ifr->ifr_newname);
4831 * Unknown or private ioctl
4834 if ((cmd >= SIOCDEVPRIVATE &&
4835 cmd <= SIOCDEVPRIVATE + 15) ||
4836 cmd == SIOCBONDENSLAVE ||
4837 cmd == SIOCBONDRELEASE ||
4838 cmd == SIOCBONDSETHWADDR ||
4839 cmd == SIOCBONDSLAVEINFOQUERY ||
4840 cmd == SIOCBONDINFOQUERY ||
4841 cmd == SIOCBONDCHANGEACTIVE ||
4842 cmd == SIOCGMIIPHY ||
4843 cmd == SIOCGMIIREG ||
4844 cmd == SIOCSMIIREG ||
4845 cmd == SIOCBRADDIF ||
4846 cmd == SIOCBRDELIF ||
4847 cmd == SIOCSHWTSTAMP ||
4848 cmd == SIOCWANDEV) {
4850 if (ops->ndo_do_ioctl) {
4851 if (netif_device_present(dev))
4852 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4864 * This function handles all "interface"-type I/O control requests. The actual
4865 * 'doing' part of this is dev_ifsioc above.
4869 * dev_ioctl - network device ioctl
4870 * @net: the applicable net namespace
4871 * @cmd: command to issue
4872 * @arg: pointer to a struct ifreq in user space
4874 * Issue ioctl functions to devices. This is normally called by the
4875 * user space syscall interfaces but can sometimes be useful for
4876 * other purposes. The return value is the return from the syscall if
4877 * positive or a negative errno code on error.
4880 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4886 /* One special case: SIOCGIFCONF takes ifconf argument
4887 and requires shared lock, because it sleeps writing
4891 if (cmd == SIOCGIFCONF) {
4893 ret = dev_ifconf(net, (char __user *) arg);
4897 if (cmd == SIOCGIFNAME)
4898 return dev_ifname(net, (struct ifreq __user *)arg);
4900 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4903 ifr.ifr_name[IFNAMSIZ-1] = 0;
4905 colon = strchr(ifr.ifr_name, ':');
4910 * See which interface the caller is talking about.
4915 * These ioctl calls:
4916 * - can be done by all.
4917 * - atomic and do not require locking.
4928 dev_load(net, ifr.ifr_name);
4930 ret = dev_ifsioc_locked(net, &ifr, cmd);
4935 if (copy_to_user(arg, &ifr,
4936 sizeof(struct ifreq)))
4942 dev_load(net, ifr.ifr_name);
4944 ret = dev_ethtool(net, &ifr);
4949 if (copy_to_user(arg, &ifr,
4950 sizeof(struct ifreq)))
4956 * These ioctl calls:
4957 * - require superuser power.
4958 * - require strict serialization.
4964 if (!capable(CAP_NET_ADMIN))
4966 dev_load(net, ifr.ifr_name);
4968 ret = dev_ifsioc(net, &ifr, cmd);
4973 if (copy_to_user(arg, &ifr,
4974 sizeof(struct ifreq)))
4980 * These ioctl calls:
4981 * - require superuser power.
4982 * - require strict serialization.
4983 * - do not return a value
4993 case SIOCSIFHWBROADCAST:
4996 case SIOCBONDENSLAVE:
4997 case SIOCBONDRELEASE:
4998 case SIOCBONDSETHWADDR:
4999 case SIOCBONDCHANGEACTIVE:
5003 if (!capable(CAP_NET_ADMIN))
5006 case SIOCBONDSLAVEINFOQUERY:
5007 case SIOCBONDINFOQUERY:
5008 dev_load(net, ifr.ifr_name);
5010 ret = dev_ifsioc(net, &ifr, cmd);
5015 /* Get the per device memory space. We can add this but
5016 * currently do not support it */
5018 /* Set the per device memory buffer space.
5019 * Not applicable in our case */
5024 * Unknown or private ioctl.
5027 if (cmd == SIOCWANDEV ||
5028 (cmd >= SIOCDEVPRIVATE &&
5029 cmd <= SIOCDEVPRIVATE + 15)) {
5030 dev_load(net, ifr.ifr_name);
5032 ret = dev_ifsioc(net, &ifr, cmd);
5034 if (!ret && copy_to_user(arg, &ifr,
5035 sizeof(struct ifreq)))
5039 /* Take care of Wireless Extensions */
5040 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5041 return wext_handle_ioctl(net, &ifr, cmd, arg);
5048 * dev_new_index - allocate an ifindex
5049 * @net: the applicable net namespace
5051 * Returns a suitable unique value for a new device interface
5052 * number. The caller must hold the rtnl semaphore or the
5053 * dev_base_lock to be sure it remains unique.
5055 static int dev_new_index(struct net *net)
5061 if (!__dev_get_by_index(net, ifindex))
5066 /* Delayed registration/unregisteration */
5067 static LIST_HEAD(net_todo_list);
5069 static void net_set_todo(struct net_device *dev)
5071 list_add_tail(&dev->todo_list, &net_todo_list);
5074 static void rollback_registered_many(struct list_head *head)
5076 struct net_device *dev, *tmp;
5078 BUG_ON(dev_boot_phase);
5081 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5082 /* Some devices call without registering
5083 * for initialization unwind. Remove those
5084 * devices and proceed with the remaining.
5086 if (dev->reg_state == NETREG_UNINITIALIZED) {
5087 pr_debug("unregister_netdevice: device %s/%p never "
5088 "was registered\n", dev->name, dev);
5091 list_del(&dev->unreg_list);
5095 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5097 /* If device is running, close it first. */
5100 /* And unlink it from device chain. */
5101 unlist_netdevice(dev);
5103 dev->reg_state = NETREG_UNREGISTERING;
5108 list_for_each_entry(dev, head, unreg_list) {
5109 /* Shutdown queueing discipline. */
5113 /* Notify protocols, that we are about to destroy
5114 this device. They should clean all the things.
5116 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5118 if (!dev->rtnl_link_ops ||
5119 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5120 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5123 * Flush the unicast and multicast chains
5125 dev_unicast_flush(dev);
5126 dev_addr_discard(dev);
5128 if (dev->netdev_ops->ndo_uninit)
5129 dev->netdev_ops->ndo_uninit(dev);
5131 /* Notifier chain MUST detach us from master device. */
5132 WARN_ON(dev->master);
5134 /* Remove entries from kobject tree */
5135 netdev_unregister_kobject(dev);
5138 /* Process any work delayed until the end of the batch */
5139 dev = list_first_entry(head, struct net_device, unreg_list);
5140 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5144 list_for_each_entry(dev, head, unreg_list)
5148 static void rollback_registered(struct net_device *dev)
5152 list_add(&dev->unreg_list, &single);
5153 rollback_registered_many(&single);
5156 static void __netdev_init_queue_locks_one(struct net_device *dev,
5157 struct netdev_queue *dev_queue,
5160 spin_lock_init(&dev_queue->_xmit_lock);
5161 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
5162 dev_queue->xmit_lock_owner = -1;
5165 static void netdev_init_queue_locks(struct net_device *dev)
5167 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
5168 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
5171 unsigned long netdev_fix_features(unsigned long features, const char *name)
5173 /* Fix illegal SG+CSUM combinations. */
5174 if ((features & NETIF_F_SG) &&
5175 !(features & NETIF_F_ALL_CSUM)) {
5177 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
5178 "checksum feature.\n", name);
5179 features &= ~NETIF_F_SG;
5182 /* TSO requires that SG is present as well. */
5183 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
5185 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
5186 "SG feature.\n", name);
5187 features &= ~NETIF_F_TSO;
5190 if (features & NETIF_F_UFO) {
5191 if (!(features & NETIF_F_GEN_CSUM)) {
5193 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5194 "since no NETIF_F_HW_CSUM feature.\n",
5196 features &= ~NETIF_F_UFO;
5199 if (!(features & NETIF_F_SG)) {
5201 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5202 "since no NETIF_F_SG feature.\n", name);
5203 features &= ~NETIF_F_UFO;
5209 EXPORT_SYMBOL(netdev_fix_features);
5212 * netif_stacked_transfer_operstate - transfer operstate
5213 * @rootdev: the root or lower level device to transfer state from
5214 * @dev: the device to transfer operstate to
5216 * Transfer operational state from root to device. This is normally
5217 * called when a stacking relationship exists between the root
5218 * device and the device(a leaf device).
5220 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5221 struct net_device *dev)
5223 if (rootdev->operstate == IF_OPER_DORMANT)
5224 netif_dormant_on(dev);
5226 netif_dormant_off(dev);
5228 if (netif_carrier_ok(rootdev)) {
5229 if (!netif_carrier_ok(dev))
5230 netif_carrier_on(dev);
5232 if (netif_carrier_ok(dev))
5233 netif_carrier_off(dev);
5236 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5239 * register_netdevice - register a network device
5240 * @dev: device to register
5242 * Take a completed network device structure and add it to the kernel
5243 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5244 * chain. 0 is returned on success. A negative errno code is returned
5245 * on a failure to set up the device, or if the name is a duplicate.
5247 * Callers must hold the rtnl semaphore. You may want
5248 * register_netdev() instead of this.
5251 * The locking appears insufficient to guarantee two parallel registers
5252 * will not get the same name.
5255 int register_netdevice(struct net_device *dev)
5258 struct net *net = dev_net(dev);
5260 BUG_ON(dev_boot_phase);
5265 /* When net_device's are persistent, this will be fatal. */
5266 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5269 spin_lock_init(&dev->addr_list_lock);
5270 netdev_set_addr_lockdep_class(dev);
5271 netdev_init_queue_locks(dev);
5275 if (!dev->num_rx_queues) {
5277 * Allocate a single RX queue if driver never called
5281 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
5287 dev->_rx->first = dev->_rx;
5288 atomic_set(&dev->_rx->count, 1);
5289 dev->num_rx_queues = 1;
5292 /* Init, if this function is available */
5293 if (dev->netdev_ops->ndo_init) {
5294 ret = dev->netdev_ops->ndo_init(dev);
5302 ret = dev_get_valid_name(net, dev->name, dev->name, 0);
5306 dev->ifindex = dev_new_index(net);
5307 if (dev->iflink == -1)
5308 dev->iflink = dev->ifindex;
5310 /* Fix illegal checksum combinations */
5311 if ((dev->features & NETIF_F_HW_CSUM) &&
5312 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5313 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5315 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5318 if ((dev->features & NETIF_F_NO_CSUM) &&
5319 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5320 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5322 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5325 dev->features = netdev_fix_features(dev->features, dev->name);
5327 /* Enable software GSO if SG is supported. */
5328 if (dev->features & NETIF_F_SG)
5329 dev->features |= NETIF_F_GSO;
5331 netdev_initialize_kobject(dev);
5333 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5334 ret = notifier_to_errno(ret);
5338 ret = netdev_register_kobject(dev);
5341 dev->reg_state = NETREG_REGISTERED;
5344 * Default initial state at registry is that the
5345 * device is present.
5348 set_bit(__LINK_STATE_PRESENT, &dev->state);
5350 dev_init_scheduler(dev);
5352 list_netdevice(dev);
5354 /* Notify protocols, that a new device appeared. */
5355 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5356 ret = notifier_to_errno(ret);
5358 rollback_registered(dev);
5359 dev->reg_state = NETREG_UNREGISTERED;
5362 * Prevent userspace races by waiting until the network
5363 * device is fully setup before sending notifications.
5365 if (!dev->rtnl_link_ops ||
5366 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5367 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5373 if (dev->netdev_ops->ndo_uninit)
5374 dev->netdev_ops->ndo_uninit(dev);
5377 EXPORT_SYMBOL(register_netdevice);
5380 * init_dummy_netdev - init a dummy network device for NAPI
5381 * @dev: device to init
5383 * This takes a network device structure and initialize the minimum
5384 * amount of fields so it can be used to schedule NAPI polls without
5385 * registering a full blown interface. This is to be used by drivers
5386 * that need to tie several hardware interfaces to a single NAPI
5387 * poll scheduler due to HW limitations.
5389 int init_dummy_netdev(struct net_device *dev)
5391 /* Clear everything. Note we don't initialize spinlocks
5392 * are they aren't supposed to be taken by any of the
5393 * NAPI code and this dummy netdev is supposed to be
5394 * only ever used for NAPI polls
5396 memset(dev, 0, sizeof(struct net_device));
5398 /* make sure we BUG if trying to hit standard
5399 * register/unregister code path
5401 dev->reg_state = NETREG_DUMMY;
5403 /* initialize the ref count */
5404 atomic_set(&dev->refcnt, 1);
5406 /* NAPI wants this */
5407 INIT_LIST_HEAD(&dev->napi_list);
5409 /* a dummy interface is started by default */
5410 set_bit(__LINK_STATE_PRESENT, &dev->state);
5411 set_bit(__LINK_STATE_START, &dev->state);
5415 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5419 * register_netdev - register a network device
5420 * @dev: device to register
5422 * Take a completed network device structure and add it to the kernel
5423 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5424 * chain. 0 is returned on success. A negative errno code is returned
5425 * on a failure to set up the device, or if the name is a duplicate.
5427 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5428 * and expands the device name if you passed a format string to
5431 int register_netdev(struct net_device *dev)
5438 * If the name is a format string the caller wants us to do a
5441 if (strchr(dev->name, '%')) {
5442 err = dev_alloc_name(dev, dev->name);
5447 err = register_netdevice(dev);
5452 EXPORT_SYMBOL(register_netdev);
5455 * netdev_wait_allrefs - wait until all references are gone.
5457 * This is called when unregistering network devices.
5459 * Any protocol or device that holds a reference should register
5460 * for netdevice notification, and cleanup and put back the
5461 * reference if they receive an UNREGISTER event.
5462 * We can get stuck here if buggy protocols don't correctly
5465 static void netdev_wait_allrefs(struct net_device *dev)
5467 unsigned long rebroadcast_time, warning_time;
5469 linkwatch_forget_dev(dev);
5471 rebroadcast_time = warning_time = jiffies;
5472 while (atomic_read(&dev->refcnt) != 0) {
5473 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5476 /* Rebroadcast unregister notification */
5477 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5478 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5479 * should have already handle it the first time */
5481 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5483 /* We must not have linkwatch events
5484 * pending on unregister. If this
5485 * happens, we simply run the queue
5486 * unscheduled, resulting in a noop
5489 linkwatch_run_queue();
5494 rebroadcast_time = jiffies;
5499 if (time_after(jiffies, warning_time + 10 * HZ)) {
5500 printk(KERN_EMERG "unregister_netdevice: "
5501 "waiting for %s to become free. Usage "
5503 dev->name, atomic_read(&dev->refcnt));
5504 warning_time = jiffies;
5513 * register_netdevice(x1);
5514 * register_netdevice(x2);
5516 * unregister_netdevice(y1);
5517 * unregister_netdevice(y2);
5523 * We are invoked by rtnl_unlock().
5524 * This allows us to deal with problems:
5525 * 1) We can delete sysfs objects which invoke hotplug
5526 * without deadlocking with linkwatch via keventd.
5527 * 2) Since we run with the RTNL semaphore not held, we can sleep
5528 * safely in order to wait for the netdev refcnt to drop to zero.
5530 * We must not return until all unregister events added during
5531 * the interval the lock was held have been completed.
5533 void netdev_run_todo(void)
5535 struct list_head list;
5537 /* Snapshot list, allow later requests */
5538 list_replace_init(&net_todo_list, &list);
5542 while (!list_empty(&list)) {
5543 struct net_device *dev
5544 = list_first_entry(&list, struct net_device, todo_list);
5545 list_del(&dev->todo_list);
5547 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5548 printk(KERN_ERR "network todo '%s' but state %d\n",
5549 dev->name, dev->reg_state);
5554 dev->reg_state = NETREG_UNREGISTERED;
5556 on_each_cpu(flush_backlog, dev, 1);
5558 netdev_wait_allrefs(dev);
5561 BUG_ON(atomic_read(&dev->refcnt));
5562 WARN_ON(dev->ip_ptr);
5563 WARN_ON(dev->ip6_ptr);
5564 WARN_ON(dev->dn_ptr);
5566 if (dev->destructor)
5567 dev->destructor(dev);
5569 /* Free network device */
5570 kobject_put(&dev->dev.kobj);
5575 * dev_txq_stats_fold - fold tx_queues stats
5576 * @dev: device to get statistics from
5577 * @stats: struct net_device_stats to hold results
5579 void dev_txq_stats_fold(const struct net_device *dev,
5580 struct net_device_stats *stats)
5582 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5584 struct netdev_queue *txq;
5586 for (i = 0; i < dev->num_tx_queues; i++) {
5587 txq = netdev_get_tx_queue(dev, i);
5588 tx_bytes += txq->tx_bytes;
5589 tx_packets += txq->tx_packets;
5590 tx_dropped += txq->tx_dropped;
5592 if (tx_bytes || tx_packets || tx_dropped) {
5593 stats->tx_bytes = tx_bytes;
5594 stats->tx_packets = tx_packets;
5595 stats->tx_dropped = tx_dropped;
5598 EXPORT_SYMBOL(dev_txq_stats_fold);
5601 * dev_get_stats - get network device statistics
5602 * @dev: device to get statistics from
5604 * Get network statistics from device. The device driver may provide
5605 * its own method by setting dev->netdev_ops->get_stats; otherwise
5606 * the internal statistics structure is used.
5608 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5610 const struct net_device_ops *ops = dev->netdev_ops;
5612 if (ops->ndo_get_stats)
5613 return ops->ndo_get_stats(dev);
5615 dev_txq_stats_fold(dev, &dev->stats);
5618 EXPORT_SYMBOL(dev_get_stats);
5620 static void netdev_init_one_queue(struct net_device *dev,
5621 struct netdev_queue *queue,
5627 static void netdev_init_queues(struct net_device *dev)
5629 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5630 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5631 spin_lock_init(&dev->tx_global_lock);
5635 * alloc_netdev_mq - allocate network device
5636 * @sizeof_priv: size of private data to allocate space for
5637 * @name: device name format string
5638 * @setup: callback to initialize device
5639 * @queue_count: the number of subqueues to allocate
5641 * Allocates a struct net_device with private data area for driver use
5642 * and performs basic initialization. Also allocates subquue structs
5643 * for each queue on the device at the end of the netdevice.
5645 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5646 void (*setup)(struct net_device *), unsigned int queue_count)
5648 struct netdev_queue *tx;
5649 struct netdev_rx_queue *rx;
5650 struct net_device *dev;
5652 struct net_device *p;
5655 BUG_ON(strlen(name) >= sizeof(dev->name));
5657 alloc_size = sizeof(struct net_device);
5659 /* ensure 32-byte alignment of private area */
5660 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5661 alloc_size += sizeof_priv;
5663 /* ensure 32-byte alignment of whole construct */
5664 alloc_size += NETDEV_ALIGN - 1;
5666 p = kzalloc(alloc_size, GFP_KERNEL);
5668 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5672 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5674 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5679 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5681 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5686 atomic_set(&rx->count, queue_count);
5689 * Set a pointer to first element in the array which holds the
5692 for (i = 0; i < queue_count; i++)
5695 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5696 dev->padded = (char *)dev - (char *)p;
5698 if (dev_addr_init(dev))
5701 dev_unicast_init(dev);
5703 dev_net_set(dev, &init_net);
5706 dev->num_tx_queues = queue_count;
5707 dev->real_num_tx_queues = queue_count;
5710 dev->num_rx_queues = queue_count;
5712 dev->gso_max_size = GSO_MAX_SIZE;
5714 netdev_init_queues(dev);
5716 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5717 dev->ethtool_ntuple_list.count = 0;
5718 INIT_LIST_HEAD(&dev->napi_list);
5719 INIT_LIST_HEAD(&dev->unreg_list);
5720 INIT_LIST_HEAD(&dev->link_watch_list);
5721 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5723 strcpy(dev->name, name);
5734 EXPORT_SYMBOL(alloc_netdev_mq);
5737 * free_netdev - free network device
5740 * This function does the last stage of destroying an allocated device
5741 * interface. The reference to the device object is released.
5742 * If this is the last reference then it will be freed.
5744 void free_netdev(struct net_device *dev)
5746 struct napi_struct *p, *n;
5748 release_net(dev_net(dev));
5752 /* Flush device addresses */
5753 dev_addr_flush(dev);
5755 /* Clear ethtool n-tuple list */
5756 ethtool_ntuple_flush(dev);
5758 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5761 /* Compatibility with error handling in drivers */
5762 if (dev->reg_state == NETREG_UNINITIALIZED) {
5763 kfree((char *)dev - dev->padded);
5767 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5768 dev->reg_state = NETREG_RELEASED;
5770 /* will free via device release */
5771 put_device(&dev->dev);
5773 EXPORT_SYMBOL(free_netdev);
5776 * synchronize_net - Synchronize with packet receive processing
5778 * Wait for packets currently being received to be done.
5779 * Does not block later packets from starting.
5781 void synchronize_net(void)
5786 EXPORT_SYMBOL(synchronize_net);
5789 * unregister_netdevice_queue - remove device from the kernel
5793 * This function shuts down a device interface and removes it
5794 * from the kernel tables.
5795 * If head not NULL, device is queued to be unregistered later.
5797 * Callers must hold the rtnl semaphore. You may want
5798 * unregister_netdev() instead of this.
5801 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5806 list_move_tail(&dev->unreg_list, head);
5808 rollback_registered(dev);
5809 /* Finish processing unregister after unlock */
5813 EXPORT_SYMBOL(unregister_netdevice_queue);
5816 * unregister_netdevice_many - unregister many devices
5817 * @head: list of devices
5819 void unregister_netdevice_many(struct list_head *head)
5821 struct net_device *dev;
5823 if (!list_empty(head)) {
5824 rollback_registered_many(head);
5825 list_for_each_entry(dev, head, unreg_list)
5829 EXPORT_SYMBOL(unregister_netdevice_many);
5832 * unregister_netdev - remove device from the kernel
5835 * This function shuts down a device interface and removes it
5836 * from the kernel tables.
5838 * This is just a wrapper for unregister_netdevice that takes
5839 * the rtnl semaphore. In general you want to use this and not
5840 * unregister_netdevice.
5842 void unregister_netdev(struct net_device *dev)
5845 unregister_netdevice(dev);
5848 EXPORT_SYMBOL(unregister_netdev);
5851 * dev_change_net_namespace - move device to different nethost namespace
5853 * @net: network namespace
5854 * @pat: If not NULL name pattern to try if the current device name
5855 * is already taken in the destination network namespace.
5857 * This function shuts down a device interface and moves it
5858 * to a new network namespace. On success 0 is returned, on
5859 * a failure a netagive errno code is returned.
5861 * Callers must hold the rtnl semaphore.
5864 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5870 /* Don't allow namespace local devices to be moved. */
5872 if (dev->features & NETIF_F_NETNS_LOCAL)
5876 /* Don't allow real devices to be moved when sysfs
5880 if (dev->dev.parent)
5884 /* Ensure the device has been registrered */
5886 if (dev->reg_state != NETREG_REGISTERED)
5889 /* Get out if there is nothing todo */
5891 if (net_eq(dev_net(dev), net))
5894 /* Pick the destination device name, and ensure
5895 * we can use it in the destination network namespace.
5898 if (__dev_get_by_name(net, dev->name)) {
5899 /* We get here if we can't use the current device name */
5902 if (dev_get_valid_name(net, pat, dev->name, 1))
5907 * And now a mini version of register_netdevice unregister_netdevice.
5910 /* If device is running close it first. */
5913 /* And unlink it from device chain */
5915 unlist_netdevice(dev);
5919 /* Shutdown queueing discipline. */
5922 /* Notify protocols, that we are about to destroy
5923 this device. They should clean all the things.
5925 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5926 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5929 * Flush the unicast and multicast chains
5931 dev_unicast_flush(dev);
5932 dev_addr_discard(dev);
5934 netdev_unregister_kobject(dev);
5936 /* Actually switch the network namespace */
5937 dev_net_set(dev, net);
5939 /* If there is an ifindex conflict assign a new one */
5940 if (__dev_get_by_index(net, dev->ifindex)) {
5941 int iflink = (dev->iflink == dev->ifindex);
5942 dev->ifindex = dev_new_index(net);
5944 dev->iflink = dev->ifindex;
5947 /* Fixup kobjects */
5948 err = netdev_register_kobject(dev);
5951 /* Add the device back in the hashes */
5952 list_netdevice(dev);
5954 /* Notify protocols, that a new device appeared. */
5955 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5958 * Prevent userspace races by waiting until the network
5959 * device is fully setup before sending notifications.
5961 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5968 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5970 static int dev_cpu_callback(struct notifier_block *nfb,
5971 unsigned long action,
5974 struct sk_buff **list_skb;
5975 struct Qdisc **list_net;
5976 struct sk_buff *skb;
5977 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5978 struct softnet_data *sd, *oldsd;
5980 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5983 local_irq_disable();
5984 cpu = smp_processor_id();
5985 sd = &per_cpu(softnet_data, cpu);
5986 oldsd = &per_cpu(softnet_data, oldcpu);
5988 /* Find end of our completion_queue. */
5989 list_skb = &sd->completion_queue;
5991 list_skb = &(*list_skb)->next;
5992 /* Append completion queue from offline CPU. */
5993 *list_skb = oldsd->completion_queue;
5994 oldsd->completion_queue = NULL;
5996 /* Find end of our output_queue. */
5997 list_net = &sd->output_queue;
5999 list_net = &(*list_net)->next_sched;
6000 /* Append output queue from offline CPU. */
6001 *list_net = oldsd->output_queue;
6002 oldsd->output_queue = NULL;
6004 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6007 /* Process offline CPU's input_pkt_queue */
6008 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
6016 * netdev_increment_features - increment feature set by one
6017 * @all: current feature set
6018 * @one: new feature set
6019 * @mask: mask feature set
6021 * Computes a new feature set after adding a device with feature set
6022 * @one to the master device with current feature set @all. Will not
6023 * enable anything that is off in @mask. Returns the new feature set.
6025 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
6028 /* If device needs checksumming, downgrade to it. */
6029 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
6030 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
6031 else if (mask & NETIF_F_ALL_CSUM) {
6032 /* If one device supports v4/v6 checksumming, set for all. */
6033 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
6034 !(all & NETIF_F_GEN_CSUM)) {
6035 all &= ~NETIF_F_ALL_CSUM;
6036 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
6039 /* If one device supports hw checksumming, set for all. */
6040 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
6041 all &= ~NETIF_F_ALL_CSUM;
6042 all |= NETIF_F_HW_CSUM;
6046 one |= NETIF_F_ALL_CSUM;
6048 one |= all & NETIF_F_ONE_FOR_ALL;
6049 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
6050 all |= one & mask & NETIF_F_ONE_FOR_ALL;
6054 EXPORT_SYMBOL(netdev_increment_features);
6056 static struct hlist_head *netdev_create_hash(void)
6059 struct hlist_head *hash;
6061 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6063 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6064 INIT_HLIST_HEAD(&hash[i]);
6069 /* Initialize per network namespace state */
6070 static int __net_init netdev_init(struct net *net)
6072 INIT_LIST_HEAD(&net->dev_base_head);
6074 net->dev_name_head = netdev_create_hash();
6075 if (net->dev_name_head == NULL)
6078 net->dev_index_head = netdev_create_hash();
6079 if (net->dev_index_head == NULL)
6085 kfree(net->dev_name_head);
6091 * netdev_drivername - network driver for the device
6092 * @dev: network device
6093 * @buffer: buffer for resulting name
6094 * @len: size of buffer
6096 * Determine network driver for device.
6098 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6100 const struct device_driver *driver;
6101 const struct device *parent;
6103 if (len <= 0 || !buffer)
6107 parent = dev->dev.parent;
6112 driver = parent->driver;
6113 if (driver && driver->name)
6114 strlcpy(buffer, driver->name, len);
6118 static void __net_exit netdev_exit(struct net *net)
6120 kfree(net->dev_name_head);
6121 kfree(net->dev_index_head);
6124 static struct pernet_operations __net_initdata netdev_net_ops = {
6125 .init = netdev_init,
6126 .exit = netdev_exit,
6129 static void __net_exit default_device_exit(struct net *net)
6131 struct net_device *dev, *aux;
6133 * Push all migratable network devices back to the
6134 * initial network namespace
6137 for_each_netdev_safe(net, dev, aux) {
6139 char fb_name[IFNAMSIZ];
6141 /* Ignore unmoveable devices (i.e. loopback) */
6142 if (dev->features & NETIF_F_NETNS_LOCAL)
6145 /* Leave virtual devices for the generic cleanup */
6146 if (dev->rtnl_link_ops)
6149 /* Push remaing network devices to init_net */
6150 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6151 err = dev_change_net_namespace(dev, &init_net, fb_name);
6153 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6154 __func__, dev->name, err);
6161 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6163 /* At exit all network devices most be removed from a network
6164 * namespace. Do this in the reverse order of registeration.
6165 * Do this across as many network namespaces as possible to
6166 * improve batching efficiency.
6168 struct net_device *dev;
6170 LIST_HEAD(dev_kill_list);
6173 list_for_each_entry(net, net_list, exit_list) {
6174 for_each_netdev_reverse(net, dev) {
6175 if (dev->rtnl_link_ops)
6176 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6178 unregister_netdevice_queue(dev, &dev_kill_list);
6181 unregister_netdevice_many(&dev_kill_list);
6185 static struct pernet_operations __net_initdata default_device_ops = {
6186 .exit = default_device_exit,
6187 .exit_batch = default_device_exit_batch,
6191 * Initialize the DEV module. At boot time this walks the device list and
6192 * unhooks any devices that fail to initialise (normally hardware not
6193 * present) and leaves us with a valid list of present and active devices.
6198 * This is called single threaded during boot, so no need
6199 * to take the rtnl semaphore.
6201 static int __init net_dev_init(void)
6203 int i, rc = -ENOMEM;
6205 BUG_ON(!dev_boot_phase);
6207 if (dev_proc_init())
6210 if (netdev_kobject_init())
6213 INIT_LIST_HEAD(&ptype_all);
6214 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6215 INIT_LIST_HEAD(&ptype_base[i]);
6217 if (register_pernet_subsys(&netdev_net_ops))
6221 * Initialise the packet receive queues.
6224 for_each_possible_cpu(i) {
6225 struct softnet_data *queue;
6227 queue = &per_cpu(softnet_data, i);
6228 skb_queue_head_init(&queue->input_pkt_queue);
6229 queue->completion_queue = NULL;
6230 INIT_LIST_HEAD(&queue->poll_list);
6233 queue->csd.func = trigger_softirq;
6234 queue->csd.info = queue;
6235 queue->csd.flags = 0;
6238 queue->backlog.poll = process_backlog;
6239 queue->backlog.weight = weight_p;
6240 queue->backlog.gro_list = NULL;
6241 queue->backlog.gro_count = 0;
6246 /* The loopback device is special if any other network devices
6247 * is present in a network namespace the loopback device must
6248 * be present. Since we now dynamically allocate and free the
6249 * loopback device ensure this invariant is maintained by
6250 * keeping the loopback device as the first device on the
6251 * list of network devices. Ensuring the loopback devices
6252 * is the first device that appears and the last network device
6255 if (register_pernet_device(&loopback_net_ops))
6258 if (register_pernet_device(&default_device_ops))
6261 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6262 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6264 hotcpu_notifier(dev_cpu_callback, 0);
6272 subsys_initcall(net_dev_init);
6274 static int __init initialize_hashrnd(void)
6276 get_random_bytes(&hashrnd, sizeof(hashrnd));
6280 late_initcall_sync(initialize_hashrnd);