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>
132 #include <linux/pci.h>
134 #include "net-sysfs.h"
136 /* Instead of increasing this, you should create a hash table. */
137 #define MAX_GRO_SKBS 8
139 /* This should be increased if a protocol with a bigger head is added. */
140 #define GRO_MAX_HEAD (MAX_HEADER + 128)
143 * The list of packet types we will receive (as opposed to discard)
144 * and the routines to invoke.
146 * Why 16. Because with 16 the only overlap we get on a hash of the
147 * low nibble of the protocol value is RARP/SNAP/X.25.
149 * NOTE: That is no longer true with the addition of VLAN tags. Not
150 * sure which should go first, but I bet it won't make much
151 * difference if we are running VLANs. The good news is that
152 * this protocol won't be in the list unless compiled in, so
153 * the average user (w/out VLANs) will not be adversely affected.
170 #define PTYPE_HASH_SIZE (16)
171 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
173 static DEFINE_SPINLOCK(ptype_lock);
174 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
175 static struct list_head ptype_all __read_mostly; /* Taps */
178 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
181 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
183 * Writers must hold the rtnl semaphore while they loop through the
184 * dev_base_head list, and hold dev_base_lock for writing when they do the
185 * actual updates. This allows pure readers to access the list even
186 * while a writer is preparing to update it.
188 * To put it another way, dev_base_lock is held for writing only to
189 * protect against pure readers; the rtnl semaphore provides the
190 * protection against other writers.
192 * See, for example usages, register_netdevice() and
193 * unregister_netdevice(), which must be called with the rtnl
196 DEFINE_RWLOCK(dev_base_lock);
197 EXPORT_SYMBOL(dev_base_lock);
199 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
201 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
202 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
205 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
207 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
210 static inline void rps_lock(struct softnet_data *queue)
213 spin_lock(&queue->input_pkt_queue.lock);
217 static inline void rps_unlock(struct softnet_data *queue)
220 spin_unlock(&queue->input_pkt_queue.lock);
224 /* Device list insertion */
225 static int list_netdevice(struct net_device *dev)
227 struct net *net = dev_net(dev);
231 write_lock_bh(&dev_base_lock);
232 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
233 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
234 hlist_add_head_rcu(&dev->index_hlist,
235 dev_index_hash(net, dev->ifindex));
236 write_unlock_bh(&dev_base_lock);
240 /* Device list removal
241 * caller must respect a RCU grace period before freeing/reusing dev
243 static void unlist_netdevice(struct net_device *dev)
247 /* Unlink dev from the device chain */
248 write_lock_bh(&dev_base_lock);
249 list_del_rcu(&dev->dev_list);
250 hlist_del_rcu(&dev->name_hlist);
251 hlist_del_rcu(&dev->index_hlist);
252 write_unlock_bh(&dev_base_lock);
259 static RAW_NOTIFIER_HEAD(netdev_chain);
262 * Device drivers call our routines to queue packets here. We empty the
263 * queue in the local softnet handler.
266 DEFINE_PER_CPU(struct softnet_data, softnet_data);
267 EXPORT_PER_CPU_SYMBOL(softnet_data);
269 #ifdef CONFIG_LOCKDEP
271 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
272 * according to dev->type
274 static const unsigned short netdev_lock_type[] =
275 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
276 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
277 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
278 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
279 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
280 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
281 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
282 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
283 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
284 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
285 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
286 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
287 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
288 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
289 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
290 ARPHRD_VOID, ARPHRD_NONE};
292 static const char *const netdev_lock_name[] =
293 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
294 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
295 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
296 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
297 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
298 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
299 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
300 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
301 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
302 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
303 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
304 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
305 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
306 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
307 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
308 "_xmit_VOID", "_xmit_NONE"};
310 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
311 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
313 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
317 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
318 if (netdev_lock_type[i] == dev_type)
320 /* the last key is used by default */
321 return ARRAY_SIZE(netdev_lock_type) - 1;
324 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
325 unsigned short dev_type)
329 i = netdev_lock_pos(dev_type);
330 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
331 netdev_lock_name[i]);
334 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
338 i = netdev_lock_pos(dev->type);
339 lockdep_set_class_and_name(&dev->addr_list_lock,
340 &netdev_addr_lock_key[i],
341 netdev_lock_name[i]);
344 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
345 unsigned short dev_type)
348 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
353 /*******************************************************************************
355 Protocol management and registration routines
357 *******************************************************************************/
360 * Add a protocol ID to the list. Now that the input handler is
361 * smarter we can dispense with all the messy stuff that used to be
364 * BEWARE!!! Protocol handlers, mangling input packets,
365 * MUST BE last in hash buckets and checking protocol handlers
366 * MUST start from promiscuous ptype_all chain in net_bh.
367 * It is true now, do not change it.
368 * Explanation follows: if protocol handler, mangling packet, will
369 * be the first on list, it is not able to sense, that packet
370 * is cloned and should be copied-on-write, so that it will
371 * change it and subsequent readers will get broken packet.
376 * dev_add_pack - add packet handler
377 * @pt: packet type declaration
379 * Add a protocol handler to the networking stack. The passed &packet_type
380 * is linked into kernel lists and may not be freed until it has been
381 * removed from the kernel lists.
383 * This call does not sleep therefore it can not
384 * guarantee all CPU's that are in middle of receiving packets
385 * will see the new packet type (until the next received packet).
388 void dev_add_pack(struct packet_type *pt)
392 spin_lock_bh(&ptype_lock);
393 if (pt->type == htons(ETH_P_ALL))
394 list_add_rcu(&pt->list, &ptype_all);
396 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
397 list_add_rcu(&pt->list, &ptype_base[hash]);
399 spin_unlock_bh(&ptype_lock);
401 EXPORT_SYMBOL(dev_add_pack);
404 * __dev_remove_pack - remove packet handler
405 * @pt: packet type declaration
407 * Remove a protocol handler that was previously added to the kernel
408 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
409 * from the kernel lists and can be freed or reused once this function
412 * The packet type might still be in use by receivers
413 * and must not be freed until after all the CPU's have gone
414 * through a quiescent state.
416 void __dev_remove_pack(struct packet_type *pt)
418 struct list_head *head;
419 struct packet_type *pt1;
421 spin_lock_bh(&ptype_lock);
423 if (pt->type == htons(ETH_P_ALL))
426 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
428 list_for_each_entry(pt1, head, list) {
430 list_del_rcu(&pt->list);
435 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
437 spin_unlock_bh(&ptype_lock);
439 EXPORT_SYMBOL(__dev_remove_pack);
442 * dev_remove_pack - remove packet handler
443 * @pt: packet type declaration
445 * Remove a protocol handler that was previously added to the kernel
446 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
447 * from the kernel lists and can be freed or reused once this function
450 * This call sleeps to guarantee that no CPU is looking at the packet
453 void dev_remove_pack(struct packet_type *pt)
455 __dev_remove_pack(pt);
459 EXPORT_SYMBOL(dev_remove_pack);
461 /******************************************************************************
463 Device Boot-time Settings Routines
465 *******************************************************************************/
467 /* Boot time configuration table */
468 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
471 * netdev_boot_setup_add - add new setup entry
472 * @name: name of the device
473 * @map: configured settings for the device
475 * Adds new setup entry to the dev_boot_setup list. The function
476 * returns 0 on error and 1 on success. This is a generic routine to
479 static int netdev_boot_setup_add(char *name, struct ifmap *map)
481 struct netdev_boot_setup *s;
485 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
486 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
487 memset(s[i].name, 0, sizeof(s[i].name));
488 strlcpy(s[i].name, name, IFNAMSIZ);
489 memcpy(&s[i].map, map, sizeof(s[i].map));
494 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
498 * netdev_boot_setup_check - check boot time settings
499 * @dev: the netdevice
501 * Check boot time settings for the device.
502 * The found settings are set for the device to be used
503 * later in the device probing.
504 * Returns 0 if no settings found, 1 if they are.
506 int netdev_boot_setup_check(struct net_device *dev)
508 struct netdev_boot_setup *s = dev_boot_setup;
511 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
512 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
513 !strcmp(dev->name, s[i].name)) {
514 dev->irq = s[i].map.irq;
515 dev->base_addr = s[i].map.base_addr;
516 dev->mem_start = s[i].map.mem_start;
517 dev->mem_end = s[i].map.mem_end;
523 EXPORT_SYMBOL(netdev_boot_setup_check);
527 * netdev_boot_base - get address from boot time settings
528 * @prefix: prefix for network device
529 * @unit: id for network device
531 * Check boot time settings for the base address of device.
532 * The found settings are set for the device to be used
533 * later in the device probing.
534 * Returns 0 if no settings found.
536 unsigned long netdev_boot_base(const char *prefix, int unit)
538 const struct netdev_boot_setup *s = dev_boot_setup;
542 sprintf(name, "%s%d", prefix, unit);
545 * If device already registered then return base of 1
546 * to indicate not to probe for this interface
548 if (__dev_get_by_name(&init_net, name))
551 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
552 if (!strcmp(name, s[i].name))
553 return s[i].map.base_addr;
558 * Saves at boot time configured settings for any netdevice.
560 int __init netdev_boot_setup(char *str)
565 str = get_options(str, ARRAY_SIZE(ints), ints);
570 memset(&map, 0, sizeof(map));
574 map.base_addr = ints[2];
576 map.mem_start = ints[3];
578 map.mem_end = ints[4];
580 /* Add new entry to the list */
581 return netdev_boot_setup_add(str, &map);
584 __setup("netdev=", netdev_boot_setup);
586 /*******************************************************************************
588 Device Interface Subroutines
590 *******************************************************************************/
593 * __dev_get_by_name - find a device by its name
594 * @net: the applicable net namespace
595 * @name: name to find
597 * Find an interface by name. Must be called under RTNL semaphore
598 * or @dev_base_lock. If the name is found a pointer to the device
599 * is returned. If the name is not found then %NULL is returned. The
600 * reference counters are not incremented so the caller must be
601 * careful with locks.
604 struct net_device *__dev_get_by_name(struct net *net, const char *name)
606 struct hlist_node *p;
607 struct net_device *dev;
608 struct hlist_head *head = dev_name_hash(net, name);
610 hlist_for_each_entry(dev, p, head, name_hlist)
611 if (!strncmp(dev->name, name, IFNAMSIZ))
616 EXPORT_SYMBOL(__dev_get_by_name);
619 * dev_get_by_name_rcu - find a device by its name
620 * @net: the applicable net namespace
621 * @name: name to find
623 * Find an interface by name.
624 * If the name is found a pointer to the device is returned.
625 * If the name is not found then %NULL is returned.
626 * The reference counters are not incremented so the caller must be
627 * careful with locks. The caller must hold RCU lock.
630 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
632 struct hlist_node *p;
633 struct net_device *dev;
634 struct hlist_head *head = dev_name_hash(net, name);
636 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
637 if (!strncmp(dev->name, name, IFNAMSIZ))
642 EXPORT_SYMBOL(dev_get_by_name_rcu);
645 * dev_get_by_name - find a device by its name
646 * @net: the applicable net namespace
647 * @name: name to find
649 * Find an interface by name. This can be called from any
650 * context and does its own locking. The returned handle has
651 * the usage count incremented and the caller must use dev_put() to
652 * release it when it is no longer needed. %NULL is returned if no
653 * matching device is found.
656 struct net_device *dev_get_by_name(struct net *net, const char *name)
658 struct net_device *dev;
661 dev = dev_get_by_name_rcu(net, name);
667 EXPORT_SYMBOL(dev_get_by_name);
670 * __dev_get_by_index - find a device by its ifindex
671 * @net: the applicable net namespace
672 * @ifindex: index of device
674 * Search for an interface by index. Returns %NULL if the device
675 * is not found or a pointer to the device. The device has not
676 * had its reference counter increased so the caller must be careful
677 * about locking. The caller must hold either the RTNL semaphore
681 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
683 struct hlist_node *p;
684 struct net_device *dev;
685 struct hlist_head *head = dev_index_hash(net, ifindex);
687 hlist_for_each_entry(dev, p, head, index_hlist)
688 if (dev->ifindex == ifindex)
693 EXPORT_SYMBOL(__dev_get_by_index);
696 * dev_get_by_index_rcu - find a device by its ifindex
697 * @net: the applicable net namespace
698 * @ifindex: index of device
700 * Search for an interface by index. Returns %NULL if the device
701 * is not found or a pointer to the device. The device has not
702 * had its reference counter increased so the caller must be careful
703 * about locking. The caller must hold RCU lock.
706 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
708 struct hlist_node *p;
709 struct net_device *dev;
710 struct hlist_head *head = dev_index_hash(net, ifindex);
712 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
713 if (dev->ifindex == ifindex)
718 EXPORT_SYMBOL(dev_get_by_index_rcu);
722 * dev_get_by_index - find a device by its ifindex
723 * @net: the applicable net namespace
724 * @ifindex: index of device
726 * Search for an interface by index. Returns NULL if the device
727 * is not found or a pointer to the device. The device returned has
728 * had a reference added and the pointer is safe until the user calls
729 * dev_put to indicate they have finished with it.
732 struct net_device *dev_get_by_index(struct net *net, int ifindex)
734 struct net_device *dev;
737 dev = dev_get_by_index_rcu(net, ifindex);
743 EXPORT_SYMBOL(dev_get_by_index);
746 * dev_getbyhwaddr - find a device by its hardware address
747 * @net: the applicable net namespace
748 * @type: media type of device
749 * @ha: hardware address
751 * Search for an interface by MAC address. Returns NULL if the device
752 * is not found or a pointer to the device. The caller must hold the
753 * rtnl semaphore. The returned device has not had its ref count increased
754 * and the caller must therefore be careful about locking
757 * If the API was consistent this would be __dev_get_by_hwaddr
760 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
762 struct net_device *dev;
766 for_each_netdev(net, dev)
767 if (dev->type == type &&
768 !memcmp(dev->dev_addr, ha, dev->addr_len))
773 EXPORT_SYMBOL(dev_getbyhwaddr);
775 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
777 struct net_device *dev;
780 for_each_netdev(net, dev)
781 if (dev->type == type)
786 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
788 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
790 struct net_device *dev, *ret = NULL;
793 for_each_netdev_rcu(net, dev)
794 if (dev->type == type) {
802 EXPORT_SYMBOL(dev_getfirstbyhwtype);
805 * dev_get_by_flags - find any device with given flags
806 * @net: the applicable net namespace
807 * @if_flags: IFF_* values
808 * @mask: bitmask of bits in if_flags to check
810 * Search for any interface with the given flags. Returns NULL if a device
811 * is not found or a pointer to the device. The device returned has
812 * had a reference added and the pointer is safe until the user calls
813 * dev_put to indicate they have finished with it.
816 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
819 struct net_device *dev, *ret;
823 for_each_netdev_rcu(net, dev) {
824 if (((dev->flags ^ if_flags) & mask) == 0) {
833 EXPORT_SYMBOL(dev_get_by_flags);
836 * dev_valid_name - check if name is okay for network device
839 * Network device names need to be valid file names to
840 * to allow sysfs to work. We also disallow any kind of
843 int dev_valid_name(const char *name)
847 if (strlen(name) >= IFNAMSIZ)
849 if (!strcmp(name, ".") || !strcmp(name, ".."))
853 if (*name == '/' || isspace(*name))
859 EXPORT_SYMBOL(dev_valid_name);
862 * __dev_alloc_name - allocate a name for a device
863 * @net: network namespace to allocate the device name in
864 * @name: name format string
865 * @buf: scratch buffer and result name string
867 * Passed a format string - eg "lt%d" it will try and find a suitable
868 * id. It scans list of devices to build up a free map, then chooses
869 * the first empty slot. The caller must hold the dev_base or rtnl lock
870 * while allocating the name and adding the device in order to avoid
872 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
873 * Returns the number of the unit assigned or a negative errno code.
876 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
880 const int max_netdevices = 8*PAGE_SIZE;
881 unsigned long *inuse;
882 struct net_device *d;
884 p = strnchr(name, IFNAMSIZ-1, '%');
887 * Verify the string as this thing may have come from
888 * the user. There must be either one "%d" and no other "%"
891 if (p[1] != 'd' || strchr(p + 2, '%'))
894 /* Use one page as a bit array of possible slots */
895 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
899 for_each_netdev(net, d) {
900 if (!sscanf(d->name, name, &i))
902 if (i < 0 || i >= max_netdevices)
905 /* avoid cases where sscanf is not exact inverse of printf */
906 snprintf(buf, IFNAMSIZ, name, i);
907 if (!strncmp(buf, d->name, IFNAMSIZ))
911 i = find_first_zero_bit(inuse, max_netdevices);
912 free_page((unsigned long) inuse);
916 snprintf(buf, IFNAMSIZ, name, i);
917 if (!__dev_get_by_name(net, buf))
920 /* It is possible to run out of possible slots
921 * when the name is long and there isn't enough space left
922 * for the digits, or if all bits are used.
928 * dev_alloc_name - allocate a name for a device
930 * @name: name format string
932 * Passed a format string - eg "lt%d" it will try and find a suitable
933 * id. It scans list of devices to build up a free map, then chooses
934 * the first empty slot. The caller must hold the dev_base or rtnl lock
935 * while allocating the name and adding the device in order to avoid
937 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
938 * Returns the number of the unit assigned or a negative errno code.
941 int dev_alloc_name(struct net_device *dev, const char *name)
947 BUG_ON(!dev_net(dev));
949 ret = __dev_alloc_name(net, name, buf);
951 strlcpy(dev->name, buf, IFNAMSIZ);
954 EXPORT_SYMBOL(dev_alloc_name);
956 static int dev_get_valid_name(struct net *net, const char *name, char *buf,
959 if (!dev_valid_name(name))
962 if (fmt && strchr(name, '%'))
963 return __dev_alloc_name(net, name, buf);
964 else if (__dev_get_by_name(net, name))
966 else if (buf != name)
967 strlcpy(buf, name, IFNAMSIZ);
973 * dev_change_name - change name of a device
975 * @newname: name (or format string) must be at least IFNAMSIZ
977 * Change name of a device, can pass format strings "eth%d".
980 int dev_change_name(struct net_device *dev, const char *newname)
982 char oldname[IFNAMSIZ];
988 BUG_ON(!dev_net(dev));
991 if (dev->flags & IFF_UP)
994 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
997 memcpy(oldname, dev->name, IFNAMSIZ);
999 err = dev_get_valid_name(net, newname, dev->name, 1);
1004 /* For now only devices in the initial network namespace
1007 if (net_eq(net, &init_net)) {
1008 ret = device_rename(&dev->dev, dev->name);
1010 memcpy(dev->name, oldname, IFNAMSIZ);
1015 write_lock_bh(&dev_base_lock);
1016 hlist_del(&dev->name_hlist);
1017 write_unlock_bh(&dev_base_lock);
1021 write_lock_bh(&dev_base_lock);
1022 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1023 write_unlock_bh(&dev_base_lock);
1025 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1026 ret = notifier_to_errno(ret);
1029 /* err >= 0 after dev_alloc_name() or stores the first errno */
1032 memcpy(dev->name, oldname, IFNAMSIZ);
1036 "%s: name change rollback failed: %d.\n",
1045 * dev_set_alias - change ifalias of a device
1047 * @alias: name up to IFALIASZ
1048 * @len: limit of bytes to copy from info
1050 * Set ifalias for a device,
1052 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1056 if (len >= IFALIASZ)
1061 kfree(dev->ifalias);
1062 dev->ifalias = NULL;
1067 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1071 strlcpy(dev->ifalias, alias, len+1);
1077 * netdev_features_change - device changes features
1078 * @dev: device to cause notification
1080 * Called to indicate a device has changed features.
1082 void netdev_features_change(struct net_device *dev)
1084 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1086 EXPORT_SYMBOL(netdev_features_change);
1089 * netdev_state_change - device changes state
1090 * @dev: device to cause notification
1092 * Called to indicate a device has changed state. This function calls
1093 * the notifier chains for netdev_chain and sends a NEWLINK message
1094 * to the routing socket.
1096 void netdev_state_change(struct net_device *dev)
1098 if (dev->flags & IFF_UP) {
1099 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1100 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1103 EXPORT_SYMBOL(netdev_state_change);
1105 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1107 return call_netdevice_notifiers(event, dev);
1109 EXPORT_SYMBOL(netdev_bonding_change);
1112 * dev_load - load a network module
1113 * @net: the applicable net namespace
1114 * @name: name of interface
1116 * If a network interface is not present and the process has suitable
1117 * privileges this function loads the module. If module loading is not
1118 * available in this kernel then it becomes a nop.
1121 void dev_load(struct net *net, const char *name)
1123 struct net_device *dev;
1126 dev = dev_get_by_name_rcu(net, name);
1129 if (!dev && capable(CAP_NET_ADMIN))
1130 request_module("%s", name);
1132 EXPORT_SYMBOL(dev_load);
1134 static int __dev_open(struct net_device *dev)
1136 const struct net_device_ops *ops = dev->netdev_ops;
1142 * Is it even present?
1144 if (!netif_device_present(dev))
1147 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1148 ret = notifier_to_errno(ret);
1153 * Call device private open method
1155 set_bit(__LINK_STATE_START, &dev->state);
1157 if (ops->ndo_validate_addr)
1158 ret = ops->ndo_validate_addr(dev);
1160 if (!ret && ops->ndo_open)
1161 ret = ops->ndo_open(dev);
1164 * If it went open OK then:
1168 clear_bit(__LINK_STATE_START, &dev->state);
1173 dev->flags |= IFF_UP;
1178 net_dmaengine_get();
1181 * Initialize multicasting status
1183 dev_set_rx_mode(dev);
1186 * Wakeup transmit queue engine
1195 * dev_open - prepare an interface for use.
1196 * @dev: device to open
1198 * Takes a device from down to up state. The device's private open
1199 * function is invoked and then the multicast lists are loaded. Finally
1200 * the device is moved into the up state and a %NETDEV_UP message is
1201 * sent to the netdev notifier chain.
1203 * Calling this function on an active interface is a nop. On a failure
1204 * a negative errno code is returned.
1206 int dev_open(struct net_device *dev)
1213 if (dev->flags & IFF_UP)
1219 ret = __dev_open(dev);
1224 * ... and announce new interface.
1226 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1227 call_netdevice_notifiers(NETDEV_UP, dev);
1231 EXPORT_SYMBOL(dev_open);
1233 static int __dev_close(struct net_device *dev)
1235 const struct net_device_ops *ops = dev->netdev_ops;
1241 * Tell people we are going down, so that they can
1242 * prepare to death, when device is still operating.
1244 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1246 clear_bit(__LINK_STATE_START, &dev->state);
1248 /* Synchronize to scheduled poll. We cannot touch poll list,
1249 * it can be even on different cpu. So just clear netif_running().
1251 * dev->stop() will invoke napi_disable() on all of it's
1252 * napi_struct instances on this device.
1254 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1256 dev_deactivate(dev);
1259 * Call the device specific close. This cannot fail.
1260 * Only if device is UP
1262 * We allow it to be called even after a DETACH hot-plug
1269 * Device is now down.
1272 dev->flags &= ~IFF_UP;
1277 net_dmaengine_put();
1283 * dev_close - shutdown an interface.
1284 * @dev: device to shutdown
1286 * This function moves an active device into down state. A
1287 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1288 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1291 int dev_close(struct net_device *dev)
1293 if (!(dev->flags & IFF_UP))
1299 * Tell people we are down
1301 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1302 call_netdevice_notifiers(NETDEV_DOWN, dev);
1306 EXPORT_SYMBOL(dev_close);
1310 * dev_disable_lro - disable Large Receive Offload on a device
1313 * Disable Large Receive Offload (LRO) on a net device. Must be
1314 * called under RTNL. This is needed if received packets may be
1315 * forwarded to another interface.
1317 void dev_disable_lro(struct net_device *dev)
1319 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1320 dev->ethtool_ops->set_flags) {
1321 u32 flags = dev->ethtool_ops->get_flags(dev);
1322 if (flags & ETH_FLAG_LRO) {
1323 flags &= ~ETH_FLAG_LRO;
1324 dev->ethtool_ops->set_flags(dev, flags);
1327 WARN_ON(dev->features & NETIF_F_LRO);
1329 EXPORT_SYMBOL(dev_disable_lro);
1332 static int dev_boot_phase = 1;
1335 * Device change register/unregister. These are not inline or static
1336 * as we export them to the world.
1340 * register_netdevice_notifier - register a network notifier block
1343 * Register a notifier to be called when network device events occur.
1344 * The notifier passed is linked into the kernel structures and must
1345 * not be reused until it has been unregistered. A negative errno code
1346 * is returned on a failure.
1348 * When registered all registration and up events are replayed
1349 * to the new notifier to allow device to have a race free
1350 * view of the network device list.
1353 int register_netdevice_notifier(struct notifier_block *nb)
1355 struct net_device *dev;
1356 struct net_device *last;
1361 err = raw_notifier_chain_register(&netdev_chain, nb);
1367 for_each_netdev(net, dev) {
1368 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1369 err = notifier_to_errno(err);
1373 if (!(dev->flags & IFF_UP))
1376 nb->notifier_call(nb, NETDEV_UP, dev);
1387 for_each_netdev(net, dev) {
1391 if (dev->flags & IFF_UP) {
1392 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1393 nb->notifier_call(nb, NETDEV_DOWN, dev);
1395 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1396 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1400 raw_notifier_chain_unregister(&netdev_chain, nb);
1403 EXPORT_SYMBOL(register_netdevice_notifier);
1406 * unregister_netdevice_notifier - unregister a network notifier block
1409 * Unregister a notifier previously registered by
1410 * register_netdevice_notifier(). The notifier is unlinked into the
1411 * kernel structures and may then be reused. A negative errno code
1412 * is returned on a failure.
1415 int unregister_netdevice_notifier(struct notifier_block *nb)
1420 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1424 EXPORT_SYMBOL(unregister_netdevice_notifier);
1427 * call_netdevice_notifiers - call all network notifier blocks
1428 * @val: value passed unmodified to notifier function
1429 * @dev: net_device pointer passed unmodified to notifier function
1431 * Call all network notifier blocks. Parameters and return value
1432 * are as for raw_notifier_call_chain().
1435 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1437 return raw_notifier_call_chain(&netdev_chain, val, dev);
1440 /* When > 0 there are consumers of rx skb time stamps */
1441 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1443 void net_enable_timestamp(void)
1445 atomic_inc(&netstamp_needed);
1447 EXPORT_SYMBOL(net_enable_timestamp);
1449 void net_disable_timestamp(void)
1451 atomic_dec(&netstamp_needed);
1453 EXPORT_SYMBOL(net_disable_timestamp);
1455 static inline void net_timestamp(struct sk_buff *skb)
1457 if (atomic_read(&netstamp_needed))
1458 __net_timestamp(skb);
1460 skb->tstamp.tv64 = 0;
1464 * dev_forward_skb - loopback an skb to another netif
1466 * @dev: destination network device
1467 * @skb: buffer to forward
1470 * NET_RX_SUCCESS (no congestion)
1471 * NET_RX_DROP (packet was dropped)
1473 * dev_forward_skb can be used for injecting an skb from the
1474 * start_xmit function of one device into the receive queue
1475 * of another device.
1477 * The receiving device may be in another namespace, so
1478 * we have to clear all information in the skb that could
1479 * impact namespace isolation.
1481 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1485 if (!(dev->flags & IFF_UP))
1488 if (skb->len > (dev->mtu + dev->hard_header_len))
1491 skb_set_dev(skb, dev);
1492 skb->tstamp.tv64 = 0;
1493 skb->pkt_type = PACKET_HOST;
1494 skb->protocol = eth_type_trans(skb, dev);
1495 return netif_rx(skb);
1497 EXPORT_SYMBOL_GPL(dev_forward_skb);
1500 * Support routine. Sends outgoing frames to any network
1501 * taps currently in use.
1504 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1506 struct packet_type *ptype;
1508 #ifdef CONFIG_NET_CLS_ACT
1509 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1516 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1517 /* Never send packets back to the socket
1518 * they originated from - MvS (miquels@drinkel.ow.org)
1520 if ((ptype->dev == dev || !ptype->dev) &&
1521 (ptype->af_packet_priv == NULL ||
1522 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1523 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1527 /* skb->nh should be correctly
1528 set by sender, so that the second statement is
1529 just protection against buggy protocols.
1531 skb_reset_mac_header(skb2);
1533 if (skb_network_header(skb2) < skb2->data ||
1534 skb2->network_header > skb2->tail) {
1535 if (net_ratelimit())
1536 printk(KERN_CRIT "protocol %04x is "
1538 skb2->protocol, dev->name);
1539 skb_reset_network_header(skb2);
1542 skb2->transport_header = skb2->network_header;
1543 skb2->pkt_type = PACKET_OUTGOING;
1544 ptype->func(skb2, skb->dev, ptype, skb->dev);
1551 static inline void __netif_reschedule(struct Qdisc *q)
1553 struct softnet_data *sd;
1554 unsigned long flags;
1556 local_irq_save(flags);
1557 sd = &__get_cpu_var(softnet_data);
1558 q->next_sched = sd->output_queue;
1559 sd->output_queue = q;
1560 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1561 local_irq_restore(flags);
1564 void __netif_schedule(struct Qdisc *q)
1566 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1567 __netif_reschedule(q);
1569 EXPORT_SYMBOL(__netif_schedule);
1571 void dev_kfree_skb_irq(struct sk_buff *skb)
1573 if (atomic_dec_and_test(&skb->users)) {
1574 struct softnet_data *sd;
1575 unsigned long flags;
1577 local_irq_save(flags);
1578 sd = &__get_cpu_var(softnet_data);
1579 skb->next = sd->completion_queue;
1580 sd->completion_queue = skb;
1581 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1582 local_irq_restore(flags);
1585 EXPORT_SYMBOL(dev_kfree_skb_irq);
1587 void dev_kfree_skb_any(struct sk_buff *skb)
1589 if (in_irq() || irqs_disabled())
1590 dev_kfree_skb_irq(skb);
1594 EXPORT_SYMBOL(dev_kfree_skb_any);
1598 * netif_device_detach - mark device as removed
1599 * @dev: network device
1601 * Mark device as removed from system and therefore no longer available.
1603 void netif_device_detach(struct net_device *dev)
1605 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1606 netif_running(dev)) {
1607 netif_tx_stop_all_queues(dev);
1610 EXPORT_SYMBOL(netif_device_detach);
1613 * netif_device_attach - mark device as attached
1614 * @dev: network device
1616 * Mark device as attached from system and restart if needed.
1618 void netif_device_attach(struct net_device *dev)
1620 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1621 netif_running(dev)) {
1622 netif_tx_wake_all_queues(dev);
1623 __netdev_watchdog_up(dev);
1626 EXPORT_SYMBOL(netif_device_attach);
1628 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1630 return ((features & NETIF_F_GEN_CSUM) ||
1631 ((features & NETIF_F_IP_CSUM) &&
1632 protocol == htons(ETH_P_IP)) ||
1633 ((features & NETIF_F_IPV6_CSUM) &&
1634 protocol == htons(ETH_P_IPV6)) ||
1635 ((features & NETIF_F_FCOE_CRC) &&
1636 protocol == htons(ETH_P_FCOE)));
1639 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1641 if (can_checksum_protocol(dev->features, skb->protocol))
1644 if (skb->protocol == htons(ETH_P_8021Q)) {
1645 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1646 if (can_checksum_protocol(dev->features & dev->vlan_features,
1647 veh->h_vlan_encapsulated_proto))
1655 * skb_dev_set -- assign a new device to a buffer
1656 * @skb: buffer for the new device
1657 * @dev: network device
1659 * If an skb is owned by a device already, we have to reset
1660 * all data private to the namespace a device belongs to
1661 * before assigning it a new device.
1663 #ifdef CONFIG_NET_NS
1664 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1667 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1670 skb_init_secmark(skb);
1674 skb->ipvs_property = 0;
1675 #ifdef CONFIG_NET_SCHED
1681 EXPORT_SYMBOL(skb_set_dev);
1682 #endif /* CONFIG_NET_NS */
1685 * Invalidate hardware checksum when packet is to be mangled, and
1686 * complete checksum manually on outgoing path.
1688 int skb_checksum_help(struct sk_buff *skb)
1691 int ret = 0, offset;
1693 if (skb->ip_summed == CHECKSUM_COMPLETE)
1694 goto out_set_summed;
1696 if (unlikely(skb_shinfo(skb)->gso_size)) {
1697 /* Let GSO fix up the checksum. */
1698 goto out_set_summed;
1701 offset = skb->csum_start - skb_headroom(skb);
1702 BUG_ON(offset >= skb_headlen(skb));
1703 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1705 offset += skb->csum_offset;
1706 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1708 if (skb_cloned(skb) &&
1709 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1710 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1715 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1717 skb->ip_summed = CHECKSUM_NONE;
1721 EXPORT_SYMBOL(skb_checksum_help);
1724 * skb_gso_segment - Perform segmentation on skb.
1725 * @skb: buffer to segment
1726 * @features: features for the output path (see dev->features)
1728 * This function segments the given skb and returns a list of segments.
1730 * It may return NULL if the skb requires no segmentation. This is
1731 * only possible when GSO is used for verifying header integrity.
1733 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1735 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1736 struct packet_type *ptype;
1737 __be16 type = skb->protocol;
1740 skb_reset_mac_header(skb);
1741 skb->mac_len = skb->network_header - skb->mac_header;
1742 __skb_pull(skb, skb->mac_len);
1744 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1745 struct net_device *dev = skb->dev;
1746 struct ethtool_drvinfo info = {};
1748 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1749 dev->ethtool_ops->get_drvinfo(dev, &info);
1751 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1753 info.driver, dev ? dev->features : 0L,
1754 skb->sk ? skb->sk->sk_route_caps : 0L,
1755 skb->len, skb->data_len, skb->ip_summed);
1757 if (skb_header_cloned(skb) &&
1758 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1759 return ERR_PTR(err);
1763 list_for_each_entry_rcu(ptype,
1764 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1765 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1766 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1767 err = ptype->gso_send_check(skb);
1768 segs = ERR_PTR(err);
1769 if (err || skb_gso_ok(skb, features))
1771 __skb_push(skb, (skb->data -
1772 skb_network_header(skb)));
1774 segs = ptype->gso_segment(skb, features);
1780 __skb_push(skb, skb->data - skb_mac_header(skb));
1784 EXPORT_SYMBOL(skb_gso_segment);
1786 /* Take action when hardware reception checksum errors are detected. */
1788 void netdev_rx_csum_fault(struct net_device *dev)
1790 if (net_ratelimit()) {
1791 printk(KERN_ERR "%s: hw csum failure.\n",
1792 dev ? dev->name : "<unknown>");
1796 EXPORT_SYMBOL(netdev_rx_csum_fault);
1799 /* Actually, we should eliminate this check as soon as we know, that:
1800 * 1. IOMMU is present and allows to map all the memory.
1801 * 2. No high memory really exists on this machine.
1804 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1806 #ifdef CONFIG_HIGHMEM
1808 if (!(dev->features & NETIF_F_HIGHDMA)) {
1809 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1810 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1814 if (PCI_DMA_BUS_IS_PHYS) {
1815 struct device *pdev = dev->dev.parent;
1817 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1818 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1819 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1828 void (*destructor)(struct sk_buff *skb);
1831 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1833 static void dev_gso_skb_destructor(struct sk_buff *skb)
1835 struct dev_gso_cb *cb;
1838 struct sk_buff *nskb = skb->next;
1840 skb->next = nskb->next;
1843 } while (skb->next);
1845 cb = DEV_GSO_CB(skb);
1847 cb->destructor(skb);
1851 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1852 * @skb: buffer to segment
1854 * This function segments the given skb and stores the list of segments
1857 static int dev_gso_segment(struct sk_buff *skb)
1859 struct net_device *dev = skb->dev;
1860 struct sk_buff *segs;
1861 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1864 segs = skb_gso_segment(skb, features);
1866 /* Verifying header integrity only. */
1871 return PTR_ERR(segs);
1874 DEV_GSO_CB(skb)->destructor = skb->destructor;
1875 skb->destructor = dev_gso_skb_destructor;
1880 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1881 struct netdev_queue *txq)
1883 const struct net_device_ops *ops = dev->netdev_ops;
1884 int rc = NETDEV_TX_OK;
1886 if (likely(!skb->next)) {
1887 if (!list_empty(&ptype_all))
1888 dev_queue_xmit_nit(skb, dev);
1890 if (netif_needs_gso(dev, skb)) {
1891 if (unlikely(dev_gso_segment(skb)))
1898 * If device doesnt need skb->dst, release it right now while
1899 * its hot in this cpu cache
1901 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1904 rc = ops->ndo_start_xmit(skb, dev);
1905 if (rc == NETDEV_TX_OK)
1906 txq_trans_update(txq);
1908 * TODO: if skb_orphan() was called by
1909 * dev->hard_start_xmit() (for example, the unmodified
1910 * igb driver does that; bnx2 doesn't), then
1911 * skb_tx_software_timestamp() will be unable to send
1912 * back the time stamp.
1914 * How can this be prevented? Always create another
1915 * reference to the socket before calling
1916 * dev->hard_start_xmit()? Prevent that skb_orphan()
1917 * does anything in dev->hard_start_xmit() by clearing
1918 * the skb destructor before the call and restoring it
1919 * afterwards, then doing the skb_orphan() ourselves?
1926 struct sk_buff *nskb = skb->next;
1928 skb->next = nskb->next;
1932 * If device doesnt need nskb->dst, release it right now while
1933 * its hot in this cpu cache
1935 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1938 rc = ops->ndo_start_xmit(nskb, dev);
1939 if (unlikely(rc != NETDEV_TX_OK)) {
1940 if (rc & ~NETDEV_TX_MASK)
1941 goto out_kfree_gso_skb;
1942 nskb->next = skb->next;
1946 txq_trans_update(txq);
1947 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1948 return NETDEV_TX_BUSY;
1949 } while (skb->next);
1952 if (likely(skb->next == NULL))
1953 skb->destructor = DEV_GSO_CB(skb)->destructor;
1959 static u32 hashrnd __read_mostly;
1961 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1965 if (skb_rx_queue_recorded(skb)) {
1966 hash = skb_get_rx_queue(skb);
1967 while (unlikely(hash >= dev->real_num_tx_queues))
1968 hash -= dev->real_num_tx_queues;
1972 if (skb->sk && skb->sk->sk_hash)
1973 hash = skb->sk->sk_hash;
1975 hash = skb->protocol;
1977 hash = jhash_1word(hash, hashrnd);
1979 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1981 EXPORT_SYMBOL(skb_tx_hash);
1983 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1985 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1986 if (net_ratelimit()) {
1987 netdev_warn(dev, "selects TX queue %d, but "
1988 "real number of TX queues is %d\n",
1989 queue_index, dev->real_num_tx_queues);
1996 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1997 struct sk_buff *skb)
2000 struct sock *sk = skb->sk;
2002 if (sk_tx_queue_recorded(sk)) {
2003 queue_index = sk_tx_queue_get(sk);
2005 const struct net_device_ops *ops = dev->netdev_ops;
2007 if (ops->ndo_select_queue) {
2008 queue_index = ops->ndo_select_queue(dev, skb);
2009 queue_index = dev_cap_txqueue(dev, queue_index);
2012 if (dev->real_num_tx_queues > 1)
2013 queue_index = skb_tx_hash(dev, skb);
2015 if (sk && sk->sk_dst_cache)
2016 sk_tx_queue_set(sk, queue_index);
2020 skb_set_queue_mapping(skb, queue_index);
2021 return netdev_get_tx_queue(dev, queue_index);
2024 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2025 struct net_device *dev,
2026 struct netdev_queue *txq)
2028 spinlock_t *root_lock = qdisc_lock(q);
2031 spin_lock(root_lock);
2032 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2035 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2036 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
2038 * This is a work-conserving queue; there are no old skbs
2039 * waiting to be sent out; and the qdisc is not running -
2040 * xmit the skb directly.
2042 __qdisc_update_bstats(q, skb->len);
2043 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
2046 clear_bit(__QDISC_STATE_RUNNING, &q->state);
2048 rc = NET_XMIT_SUCCESS;
2050 rc = qdisc_enqueue_root(skb, q);
2053 spin_unlock(root_lock);
2059 * Returns true if either:
2060 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2061 * 2. skb is fragmented and the device does not support SG, or if
2062 * at least one of fragments is in highmem and device does not
2063 * support DMA from it.
2065 static inline int skb_needs_linearize(struct sk_buff *skb,
2066 struct net_device *dev)
2068 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
2069 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
2070 illegal_highdma(dev, skb)));
2074 * dev_queue_xmit - transmit a buffer
2075 * @skb: buffer to transmit
2077 * Queue a buffer for transmission to a network device. The caller must
2078 * have set the device and priority and built the buffer before calling
2079 * this function. The function can be called from an interrupt.
2081 * A negative errno code is returned on a failure. A success does not
2082 * guarantee the frame will be transmitted as it may be dropped due
2083 * to congestion or traffic shaping.
2085 * -----------------------------------------------------------------------------------
2086 * I notice this method can also return errors from the queue disciplines,
2087 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2090 * Regardless of the return value, the skb is consumed, so it is currently
2091 * difficult to retry a send to this method. (You can bump the ref count
2092 * before sending to hold a reference for retry if you are careful.)
2094 * When calling this method, interrupts MUST be enabled. This is because
2095 * the BH enable code must have IRQs enabled so that it will not deadlock.
2098 int dev_queue_xmit(struct sk_buff *skb)
2100 struct net_device *dev = skb->dev;
2101 struct netdev_queue *txq;
2105 /* GSO will handle the following emulations directly. */
2106 if (netif_needs_gso(dev, skb))
2109 /* Convert a paged skb to linear, if required */
2110 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2113 /* If packet is not checksummed and device does not support
2114 * checksumming for this protocol, complete checksumming here.
2116 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2117 skb_set_transport_header(skb, skb->csum_start -
2119 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2124 /* Disable soft irqs for various locks below. Also
2125 * stops preemption for RCU.
2129 txq = dev_pick_tx(dev, skb);
2130 q = rcu_dereference_bh(txq->qdisc);
2132 #ifdef CONFIG_NET_CLS_ACT
2133 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2136 rc = __dev_xmit_skb(skb, q, dev, txq);
2140 /* The device has no queue. Common case for software devices:
2141 loopback, all the sorts of tunnels...
2143 Really, it is unlikely that netif_tx_lock protection is necessary
2144 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2146 However, it is possible, that they rely on protection
2149 Check this and shot the lock. It is not prone from deadlocks.
2150 Either shot noqueue qdisc, it is even simpler 8)
2152 if (dev->flags & IFF_UP) {
2153 int cpu = smp_processor_id(); /* ok because BHs are off */
2155 if (txq->xmit_lock_owner != cpu) {
2157 HARD_TX_LOCK(dev, txq, cpu);
2159 if (!netif_tx_queue_stopped(txq)) {
2160 rc = dev_hard_start_xmit(skb, dev, txq);
2161 if (dev_xmit_complete(rc)) {
2162 HARD_TX_UNLOCK(dev, txq);
2166 HARD_TX_UNLOCK(dev, txq);
2167 if (net_ratelimit())
2168 printk(KERN_CRIT "Virtual device %s asks to "
2169 "queue packet!\n", dev->name);
2171 /* Recursion is detected! It is possible,
2173 if (net_ratelimit())
2174 printk(KERN_CRIT "Dead loop on virtual device "
2175 "%s, fix it urgently!\n", dev->name);
2180 rcu_read_unlock_bh();
2186 rcu_read_unlock_bh();
2189 EXPORT_SYMBOL(dev_queue_xmit);
2192 /*=======================================================================
2194 =======================================================================*/
2196 int netdev_max_backlog __read_mostly = 1000;
2197 int netdev_budget __read_mostly = 300;
2198 int weight_p __read_mostly = 64; /* old backlog weight */
2200 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2204 * get_rps_cpu is called from netif_receive_skb and returns the target
2205 * CPU from the RPS map of the receiving queue for a given skb.
2207 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb)
2209 struct ipv6hdr *ip6;
2211 struct netdev_rx_queue *rxqueue;
2212 struct rps_map *map;
2215 u32 addr1, addr2, ports, ihl;
2219 if (skb_rx_queue_recorded(skb)) {
2220 u16 index = skb_get_rx_queue(skb);
2221 if (unlikely(index >= dev->num_rx_queues)) {
2222 if (net_ratelimit()) {
2223 netdev_warn(dev, "received packet on queue "
2224 "%u, but number of RX queues is %u\n",
2225 index, dev->num_rx_queues);
2229 rxqueue = dev->_rx + index;
2233 if (!rxqueue->rps_map)
2237 goto got_hash; /* Skip hash computation on packet header */
2239 switch (skb->protocol) {
2240 case __constant_htons(ETH_P_IP):
2241 if (!pskb_may_pull(skb, sizeof(*ip)))
2244 ip = (struct iphdr *) skb->data;
2245 ip_proto = ip->protocol;
2250 case __constant_htons(ETH_P_IPV6):
2251 if (!pskb_may_pull(skb, sizeof(*ip6)))
2254 ip6 = (struct ipv6hdr *) skb->data;
2255 ip_proto = ip6->nexthdr;
2256 addr1 = ip6->saddr.s6_addr32[3];
2257 addr2 = ip6->daddr.s6_addr32[3];
2271 case IPPROTO_UDPLITE:
2272 if (pskb_may_pull(skb, (ihl * 4) + 4))
2273 ports = *((u32 *) (skb->data + (ihl * 4)));
2280 skb->rxhash = jhash_3words(addr1, addr2, ports, hashrnd);
2285 map = rcu_dereference(rxqueue->rps_map);
2287 u16 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2289 if (cpu_online(tcpu)) {
2301 * This structure holds the per-CPU mask of CPUs for which IPIs are scheduled
2302 * to be sent to kick remote softirq processing. There are two masks since
2303 * the sending of IPIs must be done with interrupts enabled. The select field
2304 * indicates the current mask that enqueue_backlog uses to schedule IPIs.
2305 * select is flipped before net_rps_action is called while still under lock,
2306 * net_rps_action then uses the non-selected mask to send the IPIs and clears
2307 * it without conflicting with enqueue_backlog operation.
2309 struct rps_remote_softirq_cpus {
2313 static DEFINE_PER_CPU(struct rps_remote_softirq_cpus, rps_remote_softirq_cpus);
2315 /* Called from hardirq (IPI) context */
2316 static void trigger_softirq(void *data)
2318 struct softnet_data *queue = data;
2319 __napi_schedule(&queue->backlog);
2320 __get_cpu_var(netdev_rx_stat).received_rps++;
2322 #endif /* CONFIG_SMP */
2325 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2326 * queue (may be a remote CPU queue).
2328 static int enqueue_to_backlog(struct sk_buff *skb, int cpu)
2330 struct softnet_data *queue;
2331 unsigned long flags;
2333 queue = &per_cpu(softnet_data, cpu);
2335 local_irq_save(flags);
2336 __get_cpu_var(netdev_rx_stat).total++;
2339 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2340 if (queue->input_pkt_queue.qlen) {
2342 __skb_queue_tail(&queue->input_pkt_queue, skb);
2344 local_irq_restore(flags);
2345 return NET_RX_SUCCESS;
2348 /* Schedule NAPI for backlog device */
2349 if (napi_schedule_prep(&queue->backlog)) {
2351 if (cpu != smp_processor_id()) {
2352 struct rps_remote_softirq_cpus *rcpus =
2353 &__get_cpu_var(rps_remote_softirq_cpus);
2355 cpu_set(cpu, rcpus->mask[rcpus->select]);
2356 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2358 __napi_schedule(&queue->backlog);
2360 __napi_schedule(&queue->backlog);
2368 __get_cpu_var(netdev_rx_stat).dropped++;
2369 local_irq_restore(flags);
2376 * netif_rx - post buffer to the network code
2377 * @skb: buffer to post
2379 * This function receives a packet from a device driver and queues it for
2380 * the upper (protocol) levels to process. It always succeeds. The buffer
2381 * may be dropped during processing for congestion control or by the
2385 * NET_RX_SUCCESS (no congestion)
2386 * NET_RX_DROP (packet was dropped)
2390 int netif_rx(struct sk_buff *skb)
2394 /* if netpoll wants it, pretend we never saw it */
2395 if (netpoll_rx(skb))
2398 if (!skb->tstamp.tv64)
2402 cpu = get_rps_cpu(skb->dev, skb);
2404 cpu = smp_processor_id();
2406 cpu = smp_processor_id();
2409 return enqueue_to_backlog(skb, cpu);
2411 EXPORT_SYMBOL(netif_rx);
2413 int netif_rx_ni(struct sk_buff *skb)
2418 err = netif_rx(skb);
2419 if (local_softirq_pending())
2425 EXPORT_SYMBOL(netif_rx_ni);
2427 static void net_tx_action(struct softirq_action *h)
2429 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2431 if (sd->completion_queue) {
2432 struct sk_buff *clist;
2434 local_irq_disable();
2435 clist = sd->completion_queue;
2436 sd->completion_queue = NULL;
2440 struct sk_buff *skb = clist;
2441 clist = clist->next;
2443 WARN_ON(atomic_read(&skb->users));
2448 if (sd->output_queue) {
2451 local_irq_disable();
2452 head = sd->output_queue;
2453 sd->output_queue = NULL;
2457 struct Qdisc *q = head;
2458 spinlock_t *root_lock;
2460 head = head->next_sched;
2462 root_lock = qdisc_lock(q);
2463 if (spin_trylock(root_lock)) {
2464 smp_mb__before_clear_bit();
2465 clear_bit(__QDISC_STATE_SCHED,
2468 spin_unlock(root_lock);
2470 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2472 __netif_reschedule(q);
2474 smp_mb__before_clear_bit();
2475 clear_bit(__QDISC_STATE_SCHED,
2483 static inline int deliver_skb(struct sk_buff *skb,
2484 struct packet_type *pt_prev,
2485 struct net_device *orig_dev)
2487 atomic_inc(&skb->users);
2488 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2491 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2493 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2494 /* This hook is defined here for ATM LANE */
2495 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2496 unsigned char *addr) __read_mostly;
2497 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2501 * If bridge module is loaded call bridging hook.
2502 * returns NULL if packet was consumed.
2504 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2505 struct sk_buff *skb) __read_mostly;
2506 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2508 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2509 struct packet_type **pt_prev, int *ret,
2510 struct net_device *orig_dev)
2512 struct net_bridge_port *port;
2514 if (skb->pkt_type == PACKET_LOOPBACK ||
2515 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2519 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2523 return br_handle_frame_hook(port, skb);
2526 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2529 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2530 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2531 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2533 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2534 struct packet_type **pt_prev,
2536 struct net_device *orig_dev)
2538 if (skb->dev->macvlan_port == NULL)
2542 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2545 return macvlan_handle_frame_hook(skb);
2548 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2551 #ifdef CONFIG_NET_CLS_ACT
2552 /* TODO: Maybe we should just force sch_ingress to be compiled in
2553 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2554 * a compare and 2 stores extra right now if we dont have it on
2555 * but have CONFIG_NET_CLS_ACT
2556 * NOTE: This doesnt stop any functionality; if you dont have
2557 * the ingress scheduler, you just cant add policies on ingress.
2560 static int ing_filter(struct sk_buff *skb)
2562 struct net_device *dev = skb->dev;
2563 u32 ttl = G_TC_RTTL(skb->tc_verd);
2564 struct netdev_queue *rxq;
2565 int result = TC_ACT_OK;
2568 if (MAX_RED_LOOP < ttl++) {
2570 "Redir loop detected Dropping packet (%d->%d)\n",
2571 skb->skb_iif, dev->ifindex);
2575 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2576 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2578 rxq = &dev->rx_queue;
2581 if (q != &noop_qdisc) {
2582 spin_lock(qdisc_lock(q));
2583 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2584 result = qdisc_enqueue_root(skb, q);
2585 spin_unlock(qdisc_lock(q));
2591 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2592 struct packet_type **pt_prev,
2593 int *ret, struct net_device *orig_dev)
2595 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2599 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2602 /* Huh? Why does turning on AF_PACKET affect this? */
2603 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2606 switch (ing_filter(skb)) {
2620 * netif_nit_deliver - deliver received packets to network taps
2623 * This function is used to deliver incoming packets to network
2624 * taps. It should be used when the normal netif_receive_skb path
2625 * is bypassed, for example because of VLAN acceleration.
2627 void netif_nit_deliver(struct sk_buff *skb)
2629 struct packet_type *ptype;
2631 if (list_empty(&ptype_all))
2634 skb_reset_network_header(skb);
2635 skb_reset_transport_header(skb);
2636 skb->mac_len = skb->network_header - skb->mac_header;
2639 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2640 if (!ptype->dev || ptype->dev == skb->dev)
2641 deliver_skb(skb, ptype, skb->dev);
2646 static int __netif_receive_skb(struct sk_buff *skb)
2648 struct packet_type *ptype, *pt_prev;
2649 struct net_device *orig_dev;
2650 struct net_device *master;
2651 struct net_device *null_or_orig;
2652 struct net_device *null_or_bond;
2653 int ret = NET_RX_DROP;
2656 if (!skb->tstamp.tv64)
2659 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2660 return NET_RX_SUCCESS;
2662 /* if we've gotten here through NAPI, check netpoll */
2663 if (netpoll_receive_skb(skb))
2667 skb->skb_iif = skb->dev->ifindex;
2669 null_or_orig = NULL;
2670 orig_dev = skb->dev;
2671 master = ACCESS_ONCE(orig_dev->master);
2673 if (skb_bond_should_drop(skb, master))
2674 null_or_orig = orig_dev; /* deliver only exact match */
2679 __get_cpu_var(netdev_rx_stat).total++;
2681 skb_reset_network_header(skb);
2682 skb_reset_transport_header(skb);
2683 skb->mac_len = skb->network_header - skb->mac_header;
2689 #ifdef CONFIG_NET_CLS_ACT
2690 if (skb->tc_verd & TC_NCLS) {
2691 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2696 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2697 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2698 ptype->dev == orig_dev) {
2700 ret = deliver_skb(skb, pt_prev, orig_dev);
2705 #ifdef CONFIG_NET_CLS_ACT
2706 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2712 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2715 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2720 * Make sure frames received on VLAN interfaces stacked on
2721 * bonding interfaces still make their way to any base bonding
2722 * device that may have registered for a specific ptype. The
2723 * handler may have to adjust skb->dev and orig_dev.
2725 null_or_bond = NULL;
2726 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2727 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2728 null_or_bond = vlan_dev_real_dev(skb->dev);
2731 type = skb->protocol;
2732 list_for_each_entry_rcu(ptype,
2733 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2734 if (ptype->type == type && (ptype->dev == null_or_orig ||
2735 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2736 ptype->dev == null_or_bond)) {
2738 ret = deliver_skb(skb, pt_prev, orig_dev);
2744 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2747 /* Jamal, now you will not able to escape explaining
2748 * me how you were going to use this. :-)
2759 * netif_receive_skb - process receive buffer from network
2760 * @skb: buffer to process
2762 * netif_receive_skb() is the main receive data processing function.
2763 * It always succeeds. The buffer may be dropped during processing
2764 * for congestion control or by the protocol layers.
2766 * This function may only be called from softirq context and interrupts
2767 * should be enabled.
2769 * Return values (usually ignored):
2770 * NET_RX_SUCCESS: no congestion
2771 * NET_RX_DROP: packet was dropped
2773 int netif_receive_skb(struct sk_buff *skb)
2778 cpu = get_rps_cpu(skb->dev, skb);
2781 return __netif_receive_skb(skb);
2783 return enqueue_to_backlog(skb, cpu);
2785 return __netif_receive_skb(skb);
2788 EXPORT_SYMBOL(netif_receive_skb);
2790 /* Network device is going away, flush any packets still pending */
2791 static void flush_backlog(void *arg)
2793 struct net_device *dev = arg;
2794 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2795 struct sk_buff *skb, *tmp;
2798 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2799 if (skb->dev == dev) {
2800 __skb_unlink(skb, &queue->input_pkt_queue);
2806 static int napi_gro_complete(struct sk_buff *skb)
2808 struct packet_type *ptype;
2809 __be16 type = skb->protocol;
2810 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2813 if (NAPI_GRO_CB(skb)->count == 1) {
2814 skb_shinfo(skb)->gso_size = 0;
2819 list_for_each_entry_rcu(ptype, head, list) {
2820 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2823 err = ptype->gro_complete(skb);
2829 WARN_ON(&ptype->list == head);
2831 return NET_RX_SUCCESS;
2835 return netif_receive_skb(skb);
2838 static void napi_gro_flush(struct napi_struct *napi)
2840 struct sk_buff *skb, *next;
2842 for (skb = napi->gro_list; skb; skb = next) {
2845 napi_gro_complete(skb);
2848 napi->gro_count = 0;
2849 napi->gro_list = NULL;
2852 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2854 struct sk_buff **pp = NULL;
2855 struct packet_type *ptype;
2856 __be16 type = skb->protocol;
2857 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2860 enum gro_result ret;
2862 if (!(skb->dev->features & NETIF_F_GRO))
2865 if (skb_is_gso(skb) || skb_has_frags(skb))
2869 list_for_each_entry_rcu(ptype, head, list) {
2870 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2873 skb_set_network_header(skb, skb_gro_offset(skb));
2874 mac_len = skb->network_header - skb->mac_header;
2875 skb->mac_len = mac_len;
2876 NAPI_GRO_CB(skb)->same_flow = 0;
2877 NAPI_GRO_CB(skb)->flush = 0;
2878 NAPI_GRO_CB(skb)->free = 0;
2880 pp = ptype->gro_receive(&napi->gro_list, skb);
2885 if (&ptype->list == head)
2888 same_flow = NAPI_GRO_CB(skb)->same_flow;
2889 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2892 struct sk_buff *nskb = *pp;
2896 napi_gro_complete(nskb);
2903 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2907 NAPI_GRO_CB(skb)->count = 1;
2908 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2909 skb->next = napi->gro_list;
2910 napi->gro_list = skb;
2914 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2915 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2917 BUG_ON(skb->end - skb->tail < grow);
2919 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2922 skb->data_len -= grow;
2924 skb_shinfo(skb)->frags[0].page_offset += grow;
2925 skb_shinfo(skb)->frags[0].size -= grow;
2927 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2928 put_page(skb_shinfo(skb)->frags[0].page);
2929 memmove(skb_shinfo(skb)->frags,
2930 skb_shinfo(skb)->frags + 1,
2931 --skb_shinfo(skb)->nr_frags);
2942 EXPORT_SYMBOL(dev_gro_receive);
2945 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2949 if (netpoll_rx_on(skb))
2952 for (p = napi->gro_list; p; p = p->next) {
2953 NAPI_GRO_CB(p)->same_flow =
2954 (p->dev == skb->dev) &&
2955 !compare_ether_header(skb_mac_header(p),
2956 skb_gro_mac_header(skb));
2957 NAPI_GRO_CB(p)->flush = 0;
2960 return dev_gro_receive(napi, skb);
2963 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
2967 if (netif_receive_skb(skb))
2972 case GRO_MERGED_FREE:
2983 EXPORT_SYMBOL(napi_skb_finish);
2985 void skb_gro_reset_offset(struct sk_buff *skb)
2987 NAPI_GRO_CB(skb)->data_offset = 0;
2988 NAPI_GRO_CB(skb)->frag0 = NULL;
2989 NAPI_GRO_CB(skb)->frag0_len = 0;
2991 if (skb->mac_header == skb->tail &&
2992 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2993 NAPI_GRO_CB(skb)->frag0 =
2994 page_address(skb_shinfo(skb)->frags[0].page) +
2995 skb_shinfo(skb)->frags[0].page_offset;
2996 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2999 EXPORT_SYMBOL(skb_gro_reset_offset);
3001 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3003 skb_gro_reset_offset(skb);
3005 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3007 EXPORT_SYMBOL(napi_gro_receive);
3009 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3011 __skb_pull(skb, skb_headlen(skb));
3012 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3016 EXPORT_SYMBOL(napi_reuse_skb);
3018 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3020 struct sk_buff *skb = napi->skb;
3023 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3029 EXPORT_SYMBOL(napi_get_frags);
3031 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3037 skb->protocol = eth_type_trans(skb, skb->dev);
3039 if (ret == GRO_HELD)
3040 skb_gro_pull(skb, -ETH_HLEN);
3041 else if (netif_receive_skb(skb))
3046 case GRO_MERGED_FREE:
3047 napi_reuse_skb(napi, skb);
3056 EXPORT_SYMBOL(napi_frags_finish);
3058 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3060 struct sk_buff *skb = napi->skb;
3067 skb_reset_mac_header(skb);
3068 skb_gro_reset_offset(skb);
3070 off = skb_gro_offset(skb);
3071 hlen = off + sizeof(*eth);
3072 eth = skb_gro_header_fast(skb, off);
3073 if (skb_gro_header_hard(skb, hlen)) {
3074 eth = skb_gro_header_slow(skb, hlen, off);
3075 if (unlikely(!eth)) {
3076 napi_reuse_skb(napi, skb);
3082 skb_gro_pull(skb, sizeof(*eth));
3085 * This works because the only protocols we care about don't require
3086 * special handling. We'll fix it up properly at the end.
3088 skb->protocol = eth->h_proto;
3093 EXPORT_SYMBOL(napi_frags_skb);
3095 gro_result_t napi_gro_frags(struct napi_struct *napi)
3097 struct sk_buff *skb = napi_frags_skb(napi);
3102 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3104 EXPORT_SYMBOL(napi_gro_frags);
3106 static int process_backlog(struct napi_struct *napi, int quota)
3109 struct softnet_data *queue = &__get_cpu_var(softnet_data);
3110 unsigned long start_time = jiffies;
3112 napi->weight = weight_p;
3114 struct sk_buff *skb;
3116 local_irq_disable();
3118 skb = __skb_dequeue(&queue->input_pkt_queue);
3120 __napi_complete(napi);
3121 spin_unlock_irq(&queue->input_pkt_queue.lock);
3127 __netif_receive_skb(skb);
3128 } while (++work < quota && jiffies == start_time);
3134 * __napi_schedule - schedule for receive
3135 * @n: entry to schedule
3137 * The entry's receive function will be scheduled to run
3139 void __napi_schedule(struct napi_struct *n)
3141 unsigned long flags;
3143 local_irq_save(flags);
3144 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
3145 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3146 local_irq_restore(flags);
3148 EXPORT_SYMBOL(__napi_schedule);
3150 void __napi_complete(struct napi_struct *n)
3152 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3153 BUG_ON(n->gro_list);
3155 list_del(&n->poll_list);
3156 smp_mb__before_clear_bit();
3157 clear_bit(NAPI_STATE_SCHED, &n->state);
3159 EXPORT_SYMBOL(__napi_complete);
3161 void napi_complete(struct napi_struct *n)
3163 unsigned long flags;
3166 * don't let napi dequeue from the cpu poll list
3167 * just in case its running on a different cpu
3169 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3173 local_irq_save(flags);
3175 local_irq_restore(flags);
3177 EXPORT_SYMBOL(napi_complete);
3179 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3180 int (*poll)(struct napi_struct *, int), int weight)
3182 INIT_LIST_HEAD(&napi->poll_list);
3183 napi->gro_count = 0;
3184 napi->gro_list = NULL;
3187 napi->weight = weight;
3188 list_add(&napi->dev_list, &dev->napi_list);
3190 #ifdef CONFIG_NETPOLL
3191 spin_lock_init(&napi->poll_lock);
3192 napi->poll_owner = -1;
3194 set_bit(NAPI_STATE_SCHED, &napi->state);
3196 EXPORT_SYMBOL(netif_napi_add);
3198 void netif_napi_del(struct napi_struct *napi)
3200 struct sk_buff *skb, *next;
3202 list_del_init(&napi->dev_list);
3203 napi_free_frags(napi);
3205 for (skb = napi->gro_list; skb; skb = next) {
3211 napi->gro_list = NULL;
3212 napi->gro_count = 0;
3214 EXPORT_SYMBOL(netif_napi_del);
3218 * net_rps_action sends any pending IPI's for rps. This is only called from
3219 * softirq and interrupts must be enabled.
3221 static void net_rps_action(cpumask_t *mask)
3225 /* Send pending IPI's to kick RPS processing on remote cpus. */
3226 for_each_cpu_mask_nr(cpu, *mask) {
3227 struct softnet_data *queue = &per_cpu(softnet_data, cpu);
3228 if (cpu_online(cpu))
3229 __smp_call_function_single(cpu, &queue->csd, 0);
3235 static void net_rx_action(struct softirq_action *h)
3237 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
3238 unsigned long time_limit = jiffies + 2;
3239 int budget = netdev_budget;
3243 struct rps_remote_softirq_cpus *rcpus;
3246 local_irq_disable();
3248 while (!list_empty(list)) {
3249 struct napi_struct *n;
3252 /* If softirq window is exhuasted then punt.
3253 * Allow this to run for 2 jiffies since which will allow
3254 * an average latency of 1.5/HZ.
3256 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3261 /* Even though interrupts have been re-enabled, this
3262 * access is safe because interrupts can only add new
3263 * entries to the tail of this list, and only ->poll()
3264 * calls can remove this head entry from the list.
3266 n = list_first_entry(list, struct napi_struct, poll_list);
3268 have = netpoll_poll_lock(n);
3272 /* This NAPI_STATE_SCHED test is for avoiding a race
3273 * with netpoll's poll_napi(). Only the entity which
3274 * obtains the lock and sees NAPI_STATE_SCHED set will
3275 * actually make the ->poll() call. Therefore we avoid
3276 * accidently calling ->poll() when NAPI is not scheduled.
3279 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3280 work = n->poll(n, weight);
3284 WARN_ON_ONCE(work > weight);
3288 local_irq_disable();
3290 /* Drivers must not modify the NAPI state if they
3291 * consume the entire weight. In such cases this code
3292 * still "owns" the NAPI instance and therefore can
3293 * move the instance around on the list at-will.
3295 if (unlikely(work == weight)) {
3296 if (unlikely(napi_disable_pending(n))) {
3299 local_irq_disable();
3301 list_move_tail(&n->poll_list, list);
3304 netpoll_poll_unlock(have);
3308 rcpus = &__get_cpu_var(rps_remote_softirq_cpus);
3309 select = rcpus->select;
3314 net_rps_action(&rcpus->mask[select]);
3319 #ifdef CONFIG_NET_DMA
3321 * There may not be any more sk_buffs coming right now, so push
3322 * any pending DMA copies to hardware
3324 dma_issue_pending_all();
3330 __get_cpu_var(netdev_rx_stat).time_squeeze++;
3331 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3335 static gifconf_func_t *gifconf_list[NPROTO];
3338 * register_gifconf - register a SIOCGIF handler
3339 * @family: Address family
3340 * @gifconf: Function handler
3342 * Register protocol dependent address dumping routines. The handler
3343 * that is passed must not be freed or reused until it has been replaced
3344 * by another handler.
3346 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3348 if (family >= NPROTO)
3350 gifconf_list[family] = gifconf;
3353 EXPORT_SYMBOL(register_gifconf);
3357 * Map an interface index to its name (SIOCGIFNAME)
3361 * We need this ioctl for efficient implementation of the
3362 * if_indextoname() function required by the IPv6 API. Without
3363 * it, we would have to search all the interfaces to find a
3367 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3369 struct net_device *dev;
3373 * Fetch the caller's info block.
3376 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3380 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3386 strcpy(ifr.ifr_name, dev->name);
3389 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3395 * Perform a SIOCGIFCONF call. This structure will change
3396 * size eventually, and there is nothing I can do about it.
3397 * Thus we will need a 'compatibility mode'.
3400 static int dev_ifconf(struct net *net, char __user *arg)
3403 struct net_device *dev;
3410 * Fetch the caller's info block.
3413 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3420 * Loop over the interfaces, and write an info block for each.
3424 for_each_netdev(net, dev) {
3425 for (i = 0; i < NPROTO; i++) {
3426 if (gifconf_list[i]) {
3429 done = gifconf_list[i](dev, NULL, 0);
3431 done = gifconf_list[i](dev, pos + total,
3441 * All done. Write the updated control block back to the caller.
3443 ifc.ifc_len = total;
3446 * Both BSD and Solaris return 0 here, so we do too.
3448 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3451 #ifdef CONFIG_PROC_FS
3453 * This is invoked by the /proc filesystem handler to display a device
3456 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3459 struct net *net = seq_file_net(seq);
3461 struct net_device *dev;
3465 return SEQ_START_TOKEN;
3468 for_each_netdev_rcu(net, dev)
3475 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3477 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3478 first_net_device(seq_file_net(seq)) :
3479 next_net_device((struct net_device *)v);
3482 return rcu_dereference(dev);
3485 void dev_seq_stop(struct seq_file *seq, void *v)
3491 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3493 const struct net_device_stats *stats = dev_get_stats(dev);
3495 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3496 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3497 dev->name, stats->rx_bytes, stats->rx_packets,
3499 stats->rx_dropped + stats->rx_missed_errors,
3500 stats->rx_fifo_errors,
3501 stats->rx_length_errors + stats->rx_over_errors +
3502 stats->rx_crc_errors + stats->rx_frame_errors,
3503 stats->rx_compressed, stats->multicast,
3504 stats->tx_bytes, stats->tx_packets,
3505 stats->tx_errors, stats->tx_dropped,
3506 stats->tx_fifo_errors, stats->collisions,
3507 stats->tx_carrier_errors +
3508 stats->tx_aborted_errors +
3509 stats->tx_window_errors +
3510 stats->tx_heartbeat_errors,
3511 stats->tx_compressed);
3515 * Called from the PROCfs module. This now uses the new arbitrary sized
3516 * /proc/net interface to create /proc/net/dev
3518 static int dev_seq_show(struct seq_file *seq, void *v)
3520 if (v == SEQ_START_TOKEN)
3521 seq_puts(seq, "Inter-| Receive "
3523 " face |bytes packets errs drop fifo frame "
3524 "compressed multicast|bytes packets errs "
3525 "drop fifo colls carrier compressed\n");
3527 dev_seq_printf_stats(seq, v);
3531 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3533 struct netif_rx_stats *rc = NULL;
3535 while (*pos < nr_cpu_ids)
3536 if (cpu_online(*pos)) {
3537 rc = &per_cpu(netdev_rx_stat, *pos);
3544 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3546 return softnet_get_online(pos);
3549 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3552 return softnet_get_online(pos);
3555 static void softnet_seq_stop(struct seq_file *seq, void *v)
3559 static int softnet_seq_show(struct seq_file *seq, void *v)
3561 struct netif_rx_stats *s = v;
3563 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3564 s->total, s->dropped, s->time_squeeze, 0,
3565 0, 0, 0, 0, /* was fastroute */
3566 s->cpu_collision, s->received_rps);
3570 static const struct seq_operations dev_seq_ops = {
3571 .start = dev_seq_start,
3572 .next = dev_seq_next,
3573 .stop = dev_seq_stop,
3574 .show = dev_seq_show,
3577 static int dev_seq_open(struct inode *inode, struct file *file)
3579 return seq_open_net(inode, file, &dev_seq_ops,
3580 sizeof(struct seq_net_private));
3583 static const struct file_operations dev_seq_fops = {
3584 .owner = THIS_MODULE,
3585 .open = dev_seq_open,
3587 .llseek = seq_lseek,
3588 .release = seq_release_net,
3591 static const struct seq_operations softnet_seq_ops = {
3592 .start = softnet_seq_start,
3593 .next = softnet_seq_next,
3594 .stop = softnet_seq_stop,
3595 .show = softnet_seq_show,
3598 static int softnet_seq_open(struct inode *inode, struct file *file)
3600 return seq_open(file, &softnet_seq_ops);
3603 static const struct file_operations softnet_seq_fops = {
3604 .owner = THIS_MODULE,
3605 .open = softnet_seq_open,
3607 .llseek = seq_lseek,
3608 .release = seq_release,
3611 static void *ptype_get_idx(loff_t pos)
3613 struct packet_type *pt = NULL;
3617 list_for_each_entry_rcu(pt, &ptype_all, list) {
3623 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3624 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3633 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3637 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3640 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3642 struct packet_type *pt;
3643 struct list_head *nxt;
3647 if (v == SEQ_START_TOKEN)
3648 return ptype_get_idx(0);
3651 nxt = pt->list.next;
3652 if (pt->type == htons(ETH_P_ALL)) {
3653 if (nxt != &ptype_all)
3656 nxt = ptype_base[0].next;
3658 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3660 while (nxt == &ptype_base[hash]) {
3661 if (++hash >= PTYPE_HASH_SIZE)
3663 nxt = ptype_base[hash].next;
3666 return list_entry(nxt, struct packet_type, list);
3669 static void ptype_seq_stop(struct seq_file *seq, void *v)
3675 static int ptype_seq_show(struct seq_file *seq, void *v)
3677 struct packet_type *pt = v;
3679 if (v == SEQ_START_TOKEN)
3680 seq_puts(seq, "Type Device Function\n");
3681 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3682 if (pt->type == htons(ETH_P_ALL))
3683 seq_puts(seq, "ALL ");
3685 seq_printf(seq, "%04x", ntohs(pt->type));
3687 seq_printf(seq, " %-8s %pF\n",
3688 pt->dev ? pt->dev->name : "", pt->func);
3694 static const struct seq_operations ptype_seq_ops = {
3695 .start = ptype_seq_start,
3696 .next = ptype_seq_next,
3697 .stop = ptype_seq_stop,
3698 .show = ptype_seq_show,
3701 static int ptype_seq_open(struct inode *inode, struct file *file)
3703 return seq_open_net(inode, file, &ptype_seq_ops,
3704 sizeof(struct seq_net_private));
3707 static const struct file_operations ptype_seq_fops = {
3708 .owner = THIS_MODULE,
3709 .open = ptype_seq_open,
3711 .llseek = seq_lseek,
3712 .release = seq_release_net,
3716 static int __net_init dev_proc_net_init(struct net *net)
3720 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3722 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3724 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3727 if (wext_proc_init(net))
3733 proc_net_remove(net, "ptype");
3735 proc_net_remove(net, "softnet_stat");
3737 proc_net_remove(net, "dev");
3741 static void __net_exit dev_proc_net_exit(struct net *net)
3743 wext_proc_exit(net);
3745 proc_net_remove(net, "ptype");
3746 proc_net_remove(net, "softnet_stat");
3747 proc_net_remove(net, "dev");
3750 static struct pernet_operations __net_initdata dev_proc_ops = {
3751 .init = dev_proc_net_init,
3752 .exit = dev_proc_net_exit,
3755 static int __init dev_proc_init(void)
3757 return register_pernet_subsys(&dev_proc_ops);
3760 #define dev_proc_init() 0
3761 #endif /* CONFIG_PROC_FS */
3765 * netdev_set_master - set up master/slave pair
3766 * @slave: slave device
3767 * @master: new master device
3769 * Changes the master device of the slave. Pass %NULL to break the
3770 * bonding. The caller must hold the RTNL semaphore. On a failure
3771 * a negative errno code is returned. On success the reference counts
3772 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3773 * function returns zero.
3775 int netdev_set_master(struct net_device *slave, struct net_device *master)
3777 struct net_device *old = slave->master;
3787 slave->master = master;
3794 slave->flags |= IFF_SLAVE;
3796 slave->flags &= ~IFF_SLAVE;
3798 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3801 EXPORT_SYMBOL(netdev_set_master);
3803 static void dev_change_rx_flags(struct net_device *dev, int flags)
3805 const struct net_device_ops *ops = dev->netdev_ops;
3807 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3808 ops->ndo_change_rx_flags(dev, flags);
3811 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3813 unsigned short old_flags = dev->flags;
3819 dev->flags |= IFF_PROMISC;
3820 dev->promiscuity += inc;
3821 if (dev->promiscuity == 0) {
3824 * If inc causes overflow, untouch promisc and return error.
3827 dev->flags &= ~IFF_PROMISC;
3829 dev->promiscuity -= inc;
3830 printk(KERN_WARNING "%s: promiscuity touches roof, "
3831 "set promiscuity failed, promiscuity feature "
3832 "of device might be broken.\n", dev->name);
3836 if (dev->flags != old_flags) {
3837 printk(KERN_INFO "device %s %s promiscuous mode\n",
3838 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3840 if (audit_enabled) {
3841 current_uid_gid(&uid, &gid);
3842 audit_log(current->audit_context, GFP_ATOMIC,
3843 AUDIT_ANOM_PROMISCUOUS,
3844 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3845 dev->name, (dev->flags & IFF_PROMISC),
3846 (old_flags & IFF_PROMISC),
3847 audit_get_loginuid(current),
3849 audit_get_sessionid(current));
3852 dev_change_rx_flags(dev, IFF_PROMISC);
3858 * dev_set_promiscuity - update promiscuity count on a device
3862 * Add or remove promiscuity from a device. While the count in the device
3863 * remains above zero the interface remains promiscuous. Once it hits zero
3864 * the device reverts back to normal filtering operation. A negative inc
3865 * value is used to drop promiscuity on the device.
3866 * Return 0 if successful or a negative errno code on error.
3868 int dev_set_promiscuity(struct net_device *dev, int inc)
3870 unsigned short old_flags = dev->flags;
3873 err = __dev_set_promiscuity(dev, inc);
3876 if (dev->flags != old_flags)
3877 dev_set_rx_mode(dev);
3880 EXPORT_SYMBOL(dev_set_promiscuity);
3883 * dev_set_allmulti - update allmulti count on a device
3887 * Add or remove reception of all multicast frames to a device. While the
3888 * count in the device remains above zero the interface remains listening
3889 * to all interfaces. Once it hits zero the device reverts back to normal
3890 * filtering operation. A negative @inc value is used to drop the counter
3891 * when releasing a resource needing all multicasts.
3892 * Return 0 if successful or a negative errno code on error.
3895 int dev_set_allmulti(struct net_device *dev, int inc)
3897 unsigned short old_flags = dev->flags;
3901 dev->flags |= IFF_ALLMULTI;
3902 dev->allmulti += inc;
3903 if (dev->allmulti == 0) {
3906 * If inc causes overflow, untouch allmulti and return error.
3909 dev->flags &= ~IFF_ALLMULTI;
3911 dev->allmulti -= inc;
3912 printk(KERN_WARNING "%s: allmulti touches roof, "
3913 "set allmulti failed, allmulti feature of "
3914 "device might be broken.\n", dev->name);
3918 if (dev->flags ^ old_flags) {
3919 dev_change_rx_flags(dev, IFF_ALLMULTI);
3920 dev_set_rx_mode(dev);
3924 EXPORT_SYMBOL(dev_set_allmulti);
3927 * Upload unicast and multicast address lists to device and
3928 * configure RX filtering. When the device doesn't support unicast
3929 * filtering it is put in promiscuous mode while unicast addresses
3932 void __dev_set_rx_mode(struct net_device *dev)
3934 const struct net_device_ops *ops = dev->netdev_ops;
3936 /* dev_open will call this function so the list will stay sane. */
3937 if (!(dev->flags&IFF_UP))
3940 if (!netif_device_present(dev))
3943 if (ops->ndo_set_rx_mode)
3944 ops->ndo_set_rx_mode(dev);
3946 /* Unicast addresses changes may only happen under the rtnl,
3947 * therefore calling __dev_set_promiscuity here is safe.
3949 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
3950 __dev_set_promiscuity(dev, 1);
3951 dev->uc_promisc = 1;
3952 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
3953 __dev_set_promiscuity(dev, -1);
3954 dev->uc_promisc = 0;
3957 if (ops->ndo_set_multicast_list)
3958 ops->ndo_set_multicast_list(dev);
3962 void dev_set_rx_mode(struct net_device *dev)
3964 netif_addr_lock_bh(dev);
3965 __dev_set_rx_mode(dev);
3966 netif_addr_unlock_bh(dev);
3969 /* hw addresses list handling functions */
3971 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3972 int addr_len, unsigned char addr_type)
3974 struct netdev_hw_addr *ha;
3977 if (addr_len > MAX_ADDR_LEN)
3980 list_for_each_entry(ha, &list->list, list) {
3981 if (!memcmp(ha->addr, addr, addr_len) &&
3982 ha->type == addr_type) {
3989 alloc_size = sizeof(*ha);
3990 if (alloc_size < L1_CACHE_BYTES)
3991 alloc_size = L1_CACHE_BYTES;
3992 ha = kmalloc(alloc_size, GFP_ATOMIC);
3995 memcpy(ha->addr, addr, addr_len);
3996 ha->type = addr_type;
3999 list_add_tail_rcu(&ha->list, &list->list);
4004 static void ha_rcu_free(struct rcu_head *head)
4006 struct netdev_hw_addr *ha;
4008 ha = container_of(head, struct netdev_hw_addr, rcu_head);
4012 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
4013 int addr_len, unsigned char addr_type)
4015 struct netdev_hw_addr *ha;
4017 list_for_each_entry(ha, &list->list, list) {
4018 if (!memcmp(ha->addr, addr, addr_len) &&
4019 (ha->type == addr_type || !addr_type)) {
4022 list_del_rcu(&ha->list);
4023 call_rcu(&ha->rcu_head, ha_rcu_free);
4031 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
4032 struct netdev_hw_addr_list *from_list,
4034 unsigned char addr_type)
4037 struct netdev_hw_addr *ha, *ha2;
4040 list_for_each_entry(ha, &from_list->list, list) {
4041 type = addr_type ? addr_type : ha->type;
4042 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
4049 list_for_each_entry(ha2, &from_list->list, list) {
4052 type = addr_type ? addr_type : ha2->type;
4053 __hw_addr_del(to_list, ha2->addr, addr_len, type);
4058 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
4059 struct netdev_hw_addr_list *from_list,
4061 unsigned char addr_type)
4063 struct netdev_hw_addr *ha;
4066 list_for_each_entry(ha, &from_list->list, list) {
4067 type = addr_type ? addr_type : ha->type;
4068 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
4072 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
4073 struct netdev_hw_addr_list *from_list,
4077 struct netdev_hw_addr *ha, *tmp;
4079 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
4081 err = __hw_addr_add(to_list, ha->addr,
4082 addr_len, ha->type);
4087 } else if (ha->refcount == 1) {
4088 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
4089 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
4095 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
4096 struct netdev_hw_addr_list *from_list,
4099 struct netdev_hw_addr *ha, *tmp;
4101 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
4103 __hw_addr_del(to_list, ha->addr,
4104 addr_len, ha->type);
4106 __hw_addr_del(from_list, ha->addr,
4107 addr_len, ha->type);
4112 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
4114 struct netdev_hw_addr *ha, *tmp;
4116 list_for_each_entry_safe(ha, tmp, &list->list, list) {
4117 list_del_rcu(&ha->list);
4118 call_rcu(&ha->rcu_head, ha_rcu_free);
4123 static void __hw_addr_init(struct netdev_hw_addr_list *list)
4125 INIT_LIST_HEAD(&list->list);
4129 /* Device addresses handling functions */
4131 static void dev_addr_flush(struct net_device *dev)
4133 /* rtnl_mutex must be held here */
4135 __hw_addr_flush(&dev->dev_addrs);
4136 dev->dev_addr = NULL;
4139 static int dev_addr_init(struct net_device *dev)
4141 unsigned char addr[MAX_ADDR_LEN];
4142 struct netdev_hw_addr *ha;
4145 /* rtnl_mutex must be held here */
4147 __hw_addr_init(&dev->dev_addrs);
4148 memset(addr, 0, sizeof(addr));
4149 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
4150 NETDEV_HW_ADDR_T_LAN);
4153 * Get the first (previously created) address from the list
4154 * and set dev_addr pointer to this location.
4156 ha = list_first_entry(&dev->dev_addrs.list,
4157 struct netdev_hw_addr, list);
4158 dev->dev_addr = ha->addr;
4164 * dev_addr_add - Add a device address
4166 * @addr: address to add
4167 * @addr_type: address type
4169 * Add a device address to the device or increase the reference count if
4170 * it already exists.
4172 * The caller must hold the rtnl_mutex.
4174 int dev_addr_add(struct net_device *dev, unsigned char *addr,
4175 unsigned char addr_type)
4181 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
4183 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4186 EXPORT_SYMBOL(dev_addr_add);
4189 * dev_addr_del - Release a device address.
4191 * @addr: address to delete
4192 * @addr_type: address type
4194 * Release reference to a device address and remove it from the device
4195 * if the reference count drops to zero.
4197 * The caller must hold the rtnl_mutex.
4199 int dev_addr_del(struct net_device *dev, unsigned char *addr,
4200 unsigned char addr_type)
4203 struct netdev_hw_addr *ha;
4208 * We can not remove the first address from the list because
4209 * dev->dev_addr points to that.
4211 ha = list_first_entry(&dev->dev_addrs.list,
4212 struct netdev_hw_addr, list);
4213 if (ha->addr == dev->dev_addr && ha->refcount == 1)
4216 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
4219 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4222 EXPORT_SYMBOL(dev_addr_del);
4225 * dev_addr_add_multiple - Add device addresses from another device
4226 * @to_dev: device to which addresses will be added
4227 * @from_dev: device from which addresses will be added
4228 * @addr_type: address type - 0 means type will be used from from_dev
4230 * Add device addresses of the one device to another.
4232 * The caller must hold the rtnl_mutex.
4234 int dev_addr_add_multiple(struct net_device *to_dev,
4235 struct net_device *from_dev,
4236 unsigned char addr_type)
4242 if (from_dev->addr_len != to_dev->addr_len)
4244 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
4245 to_dev->addr_len, addr_type);
4247 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
4250 EXPORT_SYMBOL(dev_addr_add_multiple);
4253 * dev_addr_del_multiple - Delete device addresses by another device
4254 * @to_dev: device where the addresses will be deleted
4255 * @from_dev: device by which addresses the addresses will be deleted
4256 * @addr_type: address type - 0 means type will used from from_dev
4258 * Deletes addresses in to device by the list of addresses in from device.
4260 * The caller must hold the rtnl_mutex.
4262 int dev_addr_del_multiple(struct net_device *to_dev,
4263 struct net_device *from_dev,
4264 unsigned char addr_type)
4268 if (from_dev->addr_len != to_dev->addr_len)
4270 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
4271 to_dev->addr_len, addr_type);
4272 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
4275 EXPORT_SYMBOL(dev_addr_del_multiple);
4277 /* multicast addresses handling functions */
4279 int __dev_addr_delete(struct dev_addr_list **list, int *count,
4280 void *addr, int alen, int glbl)
4282 struct dev_addr_list *da;
4284 for (; (da = *list) != NULL; list = &da->next) {
4285 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
4286 alen == da->da_addrlen) {
4288 int old_glbl = da->da_gusers;
4305 int __dev_addr_add(struct dev_addr_list **list, int *count,
4306 void *addr, int alen, int glbl)
4308 struct dev_addr_list *da;
4310 for (da = *list; da != NULL; da = da->next) {
4311 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
4312 da->da_addrlen == alen) {
4314 int old_glbl = da->da_gusers;
4324 da = kzalloc(sizeof(*da), GFP_ATOMIC);
4327 memcpy(da->da_addr, addr, alen);
4328 da->da_addrlen = alen;
4330 da->da_gusers = glbl ? 1 : 0;
4338 * dev_unicast_delete - Release secondary unicast address.
4340 * @addr: address to delete
4342 * Release reference to a secondary unicast address and remove it
4343 * from the device if the reference count drops to zero.
4345 * The caller must hold the rtnl_mutex.
4347 int dev_unicast_delete(struct net_device *dev, void *addr)
4353 netif_addr_lock_bh(dev);
4354 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
4355 NETDEV_HW_ADDR_T_UNICAST);
4357 __dev_set_rx_mode(dev);
4358 netif_addr_unlock_bh(dev);
4361 EXPORT_SYMBOL(dev_unicast_delete);
4364 * dev_unicast_add - add a secondary unicast address
4366 * @addr: address to add
4368 * Add a secondary unicast address to the device or increase
4369 * the reference count if it already exists.
4371 * The caller must hold the rtnl_mutex.
4373 int dev_unicast_add(struct net_device *dev, void *addr)
4379 netif_addr_lock_bh(dev);
4380 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
4381 NETDEV_HW_ADDR_T_UNICAST);
4383 __dev_set_rx_mode(dev);
4384 netif_addr_unlock_bh(dev);
4387 EXPORT_SYMBOL(dev_unicast_add);
4389 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
4390 struct dev_addr_list **from, int *from_count)
4392 struct dev_addr_list *da, *next;
4396 while (da != NULL) {
4398 if (!da->da_synced) {
4399 err = __dev_addr_add(to, to_count,
4400 da->da_addr, da->da_addrlen, 0);
4405 } else if (da->da_users == 1) {
4406 __dev_addr_delete(to, to_count,
4407 da->da_addr, da->da_addrlen, 0);
4408 __dev_addr_delete(from, from_count,
4409 da->da_addr, da->da_addrlen, 0);
4415 EXPORT_SYMBOL_GPL(__dev_addr_sync);
4417 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4418 struct dev_addr_list **from, int *from_count)
4420 struct dev_addr_list *da, *next;
4423 while (da != NULL) {
4425 if (da->da_synced) {
4426 __dev_addr_delete(to, to_count,
4427 da->da_addr, da->da_addrlen, 0);
4429 __dev_addr_delete(from, from_count,
4430 da->da_addr, da->da_addrlen, 0);
4435 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4438 * dev_unicast_sync - Synchronize device's unicast list to another device
4439 * @to: destination device
4440 * @from: source device
4442 * Add newly added addresses to the destination device and release
4443 * addresses that have no users left. The source device must be
4444 * locked by netif_tx_lock_bh.
4446 * This function is intended to be called from the dev->set_rx_mode
4447 * function of layered software devices.
4449 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4453 if (to->addr_len != from->addr_len)
4456 netif_addr_lock_bh(to);
4457 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4459 __dev_set_rx_mode(to);
4460 netif_addr_unlock_bh(to);
4463 EXPORT_SYMBOL(dev_unicast_sync);
4466 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4467 * @to: destination device
4468 * @from: source device
4470 * Remove all addresses that were added to the destination device by
4471 * dev_unicast_sync(). This function is intended to be called from the
4472 * dev->stop function of layered software devices.
4474 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4476 if (to->addr_len != from->addr_len)
4479 netif_addr_lock_bh(from);
4480 netif_addr_lock(to);
4481 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4482 __dev_set_rx_mode(to);
4483 netif_addr_unlock(to);
4484 netif_addr_unlock_bh(from);
4486 EXPORT_SYMBOL(dev_unicast_unsync);
4488 void dev_unicast_flush(struct net_device *dev)
4490 netif_addr_lock_bh(dev);
4491 __hw_addr_flush(&dev->uc);
4492 netif_addr_unlock_bh(dev);
4494 EXPORT_SYMBOL(dev_unicast_flush);
4496 static void dev_unicast_init(struct net_device *dev)
4498 __hw_addr_init(&dev->uc);
4502 static void __dev_addr_discard(struct dev_addr_list **list)
4504 struct dev_addr_list *tmp;
4506 while (*list != NULL) {
4509 if (tmp->da_users > tmp->da_gusers)
4510 printk("__dev_addr_discard: address leakage! "
4511 "da_users=%d\n", tmp->da_users);
4516 void dev_addr_discard(struct net_device *dev)
4518 netif_addr_lock_bh(dev);
4520 __dev_addr_discard(&dev->mc_list);
4521 netdev_mc_count(dev) = 0;
4523 netif_addr_unlock_bh(dev);
4525 EXPORT_SYMBOL(dev_addr_discard);
4528 * dev_get_flags - get flags reported to userspace
4531 * Get the combination of flag bits exported through APIs to userspace.
4533 unsigned dev_get_flags(const struct net_device *dev)
4537 flags = (dev->flags & ~(IFF_PROMISC |
4542 (dev->gflags & (IFF_PROMISC |
4545 if (netif_running(dev)) {
4546 if (netif_oper_up(dev))
4547 flags |= IFF_RUNNING;
4548 if (netif_carrier_ok(dev))
4549 flags |= IFF_LOWER_UP;
4550 if (netif_dormant(dev))
4551 flags |= IFF_DORMANT;
4556 EXPORT_SYMBOL(dev_get_flags);
4558 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4560 int old_flags = dev->flags;
4566 * Set the flags on our device.
4569 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4570 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4572 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4576 * Load in the correct multicast list now the flags have changed.
4579 if ((old_flags ^ flags) & IFF_MULTICAST)
4580 dev_change_rx_flags(dev, IFF_MULTICAST);
4582 dev_set_rx_mode(dev);
4585 * Have we downed the interface. We handle IFF_UP ourselves
4586 * according to user attempts to set it, rather than blindly
4591 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4592 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4595 dev_set_rx_mode(dev);
4598 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4599 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4601 dev->gflags ^= IFF_PROMISC;
4602 dev_set_promiscuity(dev, inc);
4605 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4606 is important. Some (broken) drivers set IFF_PROMISC, when
4607 IFF_ALLMULTI is requested not asking us and not reporting.
4609 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4610 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4612 dev->gflags ^= IFF_ALLMULTI;
4613 dev_set_allmulti(dev, inc);
4619 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4621 unsigned int changes = dev->flags ^ old_flags;
4623 if (changes & IFF_UP) {
4624 if (dev->flags & IFF_UP)
4625 call_netdevice_notifiers(NETDEV_UP, dev);
4627 call_netdevice_notifiers(NETDEV_DOWN, dev);
4630 if (dev->flags & IFF_UP &&
4631 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4632 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4636 * dev_change_flags - change device settings
4638 * @flags: device state flags
4640 * Change settings on device based state flags. The flags are
4641 * in the userspace exported format.
4643 int dev_change_flags(struct net_device *dev, unsigned flags)
4646 int old_flags = dev->flags;
4648 ret = __dev_change_flags(dev, flags);
4652 changes = old_flags ^ dev->flags;
4654 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4656 __dev_notify_flags(dev, old_flags);
4659 EXPORT_SYMBOL(dev_change_flags);
4662 * dev_set_mtu - Change maximum transfer unit
4664 * @new_mtu: new transfer unit
4666 * Change the maximum transfer size of the network device.
4668 int dev_set_mtu(struct net_device *dev, int new_mtu)
4670 const struct net_device_ops *ops = dev->netdev_ops;
4673 if (new_mtu == dev->mtu)
4676 /* MTU must be positive. */
4680 if (!netif_device_present(dev))
4684 if (ops->ndo_change_mtu)
4685 err = ops->ndo_change_mtu(dev, new_mtu);
4689 if (!err && dev->flags & IFF_UP)
4690 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4693 EXPORT_SYMBOL(dev_set_mtu);
4696 * dev_set_mac_address - Change Media Access Control Address
4700 * Change the hardware (MAC) address of the device
4702 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4704 const struct net_device_ops *ops = dev->netdev_ops;
4707 if (!ops->ndo_set_mac_address)
4709 if (sa->sa_family != dev->type)
4711 if (!netif_device_present(dev))
4713 err = ops->ndo_set_mac_address(dev, sa);
4715 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4718 EXPORT_SYMBOL(dev_set_mac_address);
4721 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4723 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4726 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4732 case SIOCGIFFLAGS: /* Get interface flags */
4733 ifr->ifr_flags = (short) dev_get_flags(dev);
4736 case SIOCGIFMETRIC: /* Get the metric on the interface
4737 (currently unused) */
4738 ifr->ifr_metric = 0;
4741 case SIOCGIFMTU: /* Get the MTU of a device */
4742 ifr->ifr_mtu = dev->mtu;
4747 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4749 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4750 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4751 ifr->ifr_hwaddr.sa_family = dev->type;
4759 ifr->ifr_map.mem_start = dev->mem_start;
4760 ifr->ifr_map.mem_end = dev->mem_end;
4761 ifr->ifr_map.base_addr = dev->base_addr;
4762 ifr->ifr_map.irq = dev->irq;
4763 ifr->ifr_map.dma = dev->dma;
4764 ifr->ifr_map.port = dev->if_port;
4768 ifr->ifr_ifindex = dev->ifindex;
4772 ifr->ifr_qlen = dev->tx_queue_len;
4776 /* dev_ioctl() should ensure this case
4788 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4790 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4793 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4794 const struct net_device_ops *ops;
4799 ops = dev->netdev_ops;
4802 case SIOCSIFFLAGS: /* Set interface flags */
4803 return dev_change_flags(dev, ifr->ifr_flags);
4805 case SIOCSIFMETRIC: /* Set the metric on the interface
4806 (currently unused) */
4809 case SIOCSIFMTU: /* Set the MTU of a device */
4810 return dev_set_mtu(dev, ifr->ifr_mtu);
4813 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4815 case SIOCSIFHWBROADCAST:
4816 if (ifr->ifr_hwaddr.sa_family != dev->type)
4818 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4819 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4820 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4824 if (ops->ndo_set_config) {
4825 if (!netif_device_present(dev))
4827 return ops->ndo_set_config(dev, &ifr->ifr_map);
4832 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4833 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4835 if (!netif_device_present(dev))
4837 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4841 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4842 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4844 if (!netif_device_present(dev))
4846 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4850 if (ifr->ifr_qlen < 0)
4852 dev->tx_queue_len = ifr->ifr_qlen;
4856 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4857 return dev_change_name(dev, ifr->ifr_newname);
4860 * Unknown or private ioctl
4863 if ((cmd >= SIOCDEVPRIVATE &&
4864 cmd <= SIOCDEVPRIVATE + 15) ||
4865 cmd == SIOCBONDENSLAVE ||
4866 cmd == SIOCBONDRELEASE ||
4867 cmd == SIOCBONDSETHWADDR ||
4868 cmd == SIOCBONDSLAVEINFOQUERY ||
4869 cmd == SIOCBONDINFOQUERY ||
4870 cmd == SIOCBONDCHANGEACTIVE ||
4871 cmd == SIOCGMIIPHY ||
4872 cmd == SIOCGMIIREG ||
4873 cmd == SIOCSMIIREG ||
4874 cmd == SIOCBRADDIF ||
4875 cmd == SIOCBRDELIF ||
4876 cmd == SIOCSHWTSTAMP ||
4877 cmd == SIOCWANDEV) {
4879 if (ops->ndo_do_ioctl) {
4880 if (netif_device_present(dev))
4881 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4893 * This function handles all "interface"-type I/O control requests. The actual
4894 * 'doing' part of this is dev_ifsioc above.
4898 * dev_ioctl - network device ioctl
4899 * @net: the applicable net namespace
4900 * @cmd: command to issue
4901 * @arg: pointer to a struct ifreq in user space
4903 * Issue ioctl functions to devices. This is normally called by the
4904 * user space syscall interfaces but can sometimes be useful for
4905 * other purposes. The return value is the return from the syscall if
4906 * positive or a negative errno code on error.
4909 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4915 /* One special case: SIOCGIFCONF takes ifconf argument
4916 and requires shared lock, because it sleeps writing
4920 if (cmd == SIOCGIFCONF) {
4922 ret = dev_ifconf(net, (char __user *) arg);
4926 if (cmd == SIOCGIFNAME)
4927 return dev_ifname(net, (struct ifreq __user *)arg);
4929 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4932 ifr.ifr_name[IFNAMSIZ-1] = 0;
4934 colon = strchr(ifr.ifr_name, ':');
4939 * See which interface the caller is talking about.
4944 * These ioctl calls:
4945 * - can be done by all.
4946 * - atomic and do not require locking.
4957 dev_load(net, ifr.ifr_name);
4959 ret = dev_ifsioc_locked(net, &ifr, cmd);
4964 if (copy_to_user(arg, &ifr,
4965 sizeof(struct ifreq)))
4971 dev_load(net, ifr.ifr_name);
4973 ret = dev_ethtool(net, &ifr);
4978 if (copy_to_user(arg, &ifr,
4979 sizeof(struct ifreq)))
4985 * These ioctl calls:
4986 * - require superuser power.
4987 * - require strict serialization.
4993 if (!capable(CAP_NET_ADMIN))
4995 dev_load(net, ifr.ifr_name);
4997 ret = dev_ifsioc(net, &ifr, cmd);
5002 if (copy_to_user(arg, &ifr,
5003 sizeof(struct ifreq)))
5009 * These ioctl calls:
5010 * - require superuser power.
5011 * - require strict serialization.
5012 * - do not return a value
5022 case SIOCSIFHWBROADCAST:
5025 case SIOCBONDENSLAVE:
5026 case SIOCBONDRELEASE:
5027 case SIOCBONDSETHWADDR:
5028 case SIOCBONDCHANGEACTIVE:
5032 if (!capable(CAP_NET_ADMIN))
5035 case SIOCBONDSLAVEINFOQUERY:
5036 case SIOCBONDINFOQUERY:
5037 dev_load(net, ifr.ifr_name);
5039 ret = dev_ifsioc(net, &ifr, cmd);
5044 /* Get the per device memory space. We can add this but
5045 * currently do not support it */
5047 /* Set the per device memory buffer space.
5048 * Not applicable in our case */
5053 * Unknown or private ioctl.
5056 if (cmd == SIOCWANDEV ||
5057 (cmd >= SIOCDEVPRIVATE &&
5058 cmd <= SIOCDEVPRIVATE + 15)) {
5059 dev_load(net, ifr.ifr_name);
5061 ret = dev_ifsioc(net, &ifr, cmd);
5063 if (!ret && copy_to_user(arg, &ifr,
5064 sizeof(struct ifreq)))
5068 /* Take care of Wireless Extensions */
5069 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5070 return wext_handle_ioctl(net, &ifr, cmd, arg);
5077 * dev_new_index - allocate an ifindex
5078 * @net: the applicable net namespace
5080 * Returns a suitable unique value for a new device interface
5081 * number. The caller must hold the rtnl semaphore or the
5082 * dev_base_lock to be sure it remains unique.
5084 static int dev_new_index(struct net *net)
5090 if (!__dev_get_by_index(net, ifindex))
5095 /* Delayed registration/unregisteration */
5096 static LIST_HEAD(net_todo_list);
5098 static void net_set_todo(struct net_device *dev)
5100 list_add_tail(&dev->todo_list, &net_todo_list);
5103 static void rollback_registered_many(struct list_head *head)
5105 struct net_device *dev, *tmp;
5107 BUG_ON(dev_boot_phase);
5110 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5111 /* Some devices call without registering
5112 * for initialization unwind. Remove those
5113 * devices and proceed with the remaining.
5115 if (dev->reg_state == NETREG_UNINITIALIZED) {
5116 pr_debug("unregister_netdevice: device %s/%p never "
5117 "was registered\n", dev->name, dev);
5120 list_del(&dev->unreg_list);
5124 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5126 /* If device is running, close it first. */
5129 /* And unlink it from device chain. */
5130 unlist_netdevice(dev);
5132 dev->reg_state = NETREG_UNREGISTERING;
5137 list_for_each_entry(dev, head, unreg_list) {
5138 /* Shutdown queueing discipline. */
5142 /* Notify protocols, that we are about to destroy
5143 this device. They should clean all the things.
5145 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5147 if (!dev->rtnl_link_ops ||
5148 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5149 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5152 * Flush the unicast and multicast chains
5154 dev_unicast_flush(dev);
5155 dev_addr_discard(dev);
5157 if (dev->netdev_ops->ndo_uninit)
5158 dev->netdev_ops->ndo_uninit(dev);
5160 /* Notifier chain MUST detach us from master device. */
5161 WARN_ON(dev->master);
5163 /* Remove entries from kobject tree */
5164 netdev_unregister_kobject(dev);
5167 /* Process any work delayed until the end of the batch */
5168 dev = list_first_entry(head, struct net_device, unreg_list);
5169 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5173 list_for_each_entry(dev, head, unreg_list)
5177 static void rollback_registered(struct net_device *dev)
5181 list_add(&dev->unreg_list, &single);
5182 rollback_registered_many(&single);
5185 static void __netdev_init_queue_locks_one(struct net_device *dev,
5186 struct netdev_queue *dev_queue,
5189 spin_lock_init(&dev_queue->_xmit_lock);
5190 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
5191 dev_queue->xmit_lock_owner = -1;
5194 static void netdev_init_queue_locks(struct net_device *dev)
5196 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
5197 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
5200 unsigned long netdev_fix_features(unsigned long features, const char *name)
5202 /* Fix illegal SG+CSUM combinations. */
5203 if ((features & NETIF_F_SG) &&
5204 !(features & NETIF_F_ALL_CSUM)) {
5206 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
5207 "checksum feature.\n", name);
5208 features &= ~NETIF_F_SG;
5211 /* TSO requires that SG is present as well. */
5212 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
5214 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
5215 "SG feature.\n", name);
5216 features &= ~NETIF_F_TSO;
5219 if (features & NETIF_F_UFO) {
5220 if (!(features & NETIF_F_GEN_CSUM)) {
5222 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5223 "since no NETIF_F_HW_CSUM feature.\n",
5225 features &= ~NETIF_F_UFO;
5228 if (!(features & NETIF_F_SG)) {
5230 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5231 "since no NETIF_F_SG feature.\n", name);
5232 features &= ~NETIF_F_UFO;
5238 EXPORT_SYMBOL(netdev_fix_features);
5241 * netif_stacked_transfer_operstate - transfer operstate
5242 * @rootdev: the root or lower level device to transfer state from
5243 * @dev: the device to transfer operstate to
5245 * Transfer operational state from root to device. This is normally
5246 * called when a stacking relationship exists between the root
5247 * device and the device(a leaf device).
5249 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5250 struct net_device *dev)
5252 if (rootdev->operstate == IF_OPER_DORMANT)
5253 netif_dormant_on(dev);
5255 netif_dormant_off(dev);
5257 if (netif_carrier_ok(rootdev)) {
5258 if (!netif_carrier_ok(dev))
5259 netif_carrier_on(dev);
5261 if (netif_carrier_ok(dev))
5262 netif_carrier_off(dev);
5265 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5268 * register_netdevice - register a network device
5269 * @dev: device to register
5271 * Take a completed network device structure and add it to the kernel
5272 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5273 * chain. 0 is returned on success. A negative errno code is returned
5274 * on a failure to set up the device, or if the name is a duplicate.
5276 * Callers must hold the rtnl semaphore. You may want
5277 * register_netdev() instead of this.
5280 * The locking appears insufficient to guarantee two parallel registers
5281 * will not get the same name.
5284 int register_netdevice(struct net_device *dev)
5287 struct net *net = dev_net(dev);
5289 BUG_ON(dev_boot_phase);
5294 /* When net_device's are persistent, this will be fatal. */
5295 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5298 spin_lock_init(&dev->addr_list_lock);
5299 netdev_set_addr_lockdep_class(dev);
5300 netdev_init_queue_locks(dev);
5305 if (!dev->num_rx_queues) {
5307 * Allocate a single RX queue if driver never called
5311 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
5317 dev->_rx->first = dev->_rx;
5318 atomic_set(&dev->_rx->count, 1);
5319 dev->num_rx_queues = 1;
5322 /* Init, if this function is available */
5323 if (dev->netdev_ops->ndo_init) {
5324 ret = dev->netdev_ops->ndo_init(dev);
5332 ret = dev_get_valid_name(net, dev->name, dev->name, 0);
5336 dev->ifindex = dev_new_index(net);
5337 if (dev->iflink == -1)
5338 dev->iflink = dev->ifindex;
5340 /* Fix illegal checksum combinations */
5341 if ((dev->features & NETIF_F_HW_CSUM) &&
5342 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5343 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5345 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5348 if ((dev->features & NETIF_F_NO_CSUM) &&
5349 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5350 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5352 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5355 dev->features = netdev_fix_features(dev->features, dev->name);
5357 /* Enable software GSO if SG is supported. */
5358 if (dev->features & NETIF_F_SG)
5359 dev->features |= NETIF_F_GSO;
5361 netdev_initialize_kobject(dev);
5363 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5364 ret = notifier_to_errno(ret);
5368 ret = netdev_register_kobject(dev);
5371 dev->reg_state = NETREG_REGISTERED;
5374 * Default initial state at registry is that the
5375 * device is present.
5378 set_bit(__LINK_STATE_PRESENT, &dev->state);
5380 dev_init_scheduler(dev);
5382 list_netdevice(dev);
5384 /* Notify protocols, that a new device appeared. */
5385 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5386 ret = notifier_to_errno(ret);
5388 rollback_registered(dev);
5389 dev->reg_state = NETREG_UNREGISTERED;
5392 * Prevent userspace races by waiting until the network
5393 * device is fully setup before sending notifications.
5395 if (!dev->rtnl_link_ops ||
5396 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5397 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5403 if (dev->netdev_ops->ndo_uninit)
5404 dev->netdev_ops->ndo_uninit(dev);
5407 EXPORT_SYMBOL(register_netdevice);
5410 * init_dummy_netdev - init a dummy network device for NAPI
5411 * @dev: device to init
5413 * This takes a network device structure and initialize the minimum
5414 * amount of fields so it can be used to schedule NAPI polls without
5415 * registering a full blown interface. This is to be used by drivers
5416 * that need to tie several hardware interfaces to a single NAPI
5417 * poll scheduler due to HW limitations.
5419 int init_dummy_netdev(struct net_device *dev)
5421 /* Clear everything. Note we don't initialize spinlocks
5422 * are they aren't supposed to be taken by any of the
5423 * NAPI code and this dummy netdev is supposed to be
5424 * only ever used for NAPI polls
5426 memset(dev, 0, sizeof(struct net_device));
5428 /* make sure we BUG if trying to hit standard
5429 * register/unregister code path
5431 dev->reg_state = NETREG_DUMMY;
5433 /* initialize the ref count */
5434 atomic_set(&dev->refcnt, 1);
5436 /* NAPI wants this */
5437 INIT_LIST_HEAD(&dev->napi_list);
5439 /* a dummy interface is started by default */
5440 set_bit(__LINK_STATE_PRESENT, &dev->state);
5441 set_bit(__LINK_STATE_START, &dev->state);
5445 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5449 * register_netdev - register a network device
5450 * @dev: device to register
5452 * Take a completed network device structure and add it to the kernel
5453 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5454 * chain. 0 is returned on success. A negative errno code is returned
5455 * on a failure to set up the device, or if the name is a duplicate.
5457 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5458 * and expands the device name if you passed a format string to
5461 int register_netdev(struct net_device *dev)
5468 * If the name is a format string the caller wants us to do a
5471 if (strchr(dev->name, '%')) {
5472 err = dev_alloc_name(dev, dev->name);
5477 err = register_netdevice(dev);
5482 EXPORT_SYMBOL(register_netdev);
5485 * netdev_wait_allrefs - wait until all references are gone.
5487 * This is called when unregistering network devices.
5489 * Any protocol or device that holds a reference should register
5490 * for netdevice notification, and cleanup and put back the
5491 * reference if they receive an UNREGISTER event.
5492 * We can get stuck here if buggy protocols don't correctly
5495 static void netdev_wait_allrefs(struct net_device *dev)
5497 unsigned long rebroadcast_time, warning_time;
5499 linkwatch_forget_dev(dev);
5501 rebroadcast_time = warning_time = jiffies;
5502 while (atomic_read(&dev->refcnt) != 0) {
5503 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5506 /* Rebroadcast unregister notification */
5507 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5508 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5509 * should have already handle it the first time */
5511 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5513 /* We must not have linkwatch events
5514 * pending on unregister. If this
5515 * happens, we simply run the queue
5516 * unscheduled, resulting in a noop
5519 linkwatch_run_queue();
5524 rebroadcast_time = jiffies;
5529 if (time_after(jiffies, warning_time + 10 * HZ)) {
5530 printk(KERN_EMERG "unregister_netdevice: "
5531 "waiting for %s to become free. Usage "
5533 dev->name, atomic_read(&dev->refcnt));
5534 warning_time = jiffies;
5543 * register_netdevice(x1);
5544 * register_netdevice(x2);
5546 * unregister_netdevice(y1);
5547 * unregister_netdevice(y2);
5553 * We are invoked by rtnl_unlock().
5554 * This allows us to deal with problems:
5555 * 1) We can delete sysfs objects which invoke hotplug
5556 * without deadlocking with linkwatch via keventd.
5557 * 2) Since we run with the RTNL semaphore not held, we can sleep
5558 * safely in order to wait for the netdev refcnt to drop to zero.
5560 * We must not return until all unregister events added during
5561 * the interval the lock was held have been completed.
5563 void netdev_run_todo(void)
5565 struct list_head list;
5567 /* Snapshot list, allow later requests */
5568 list_replace_init(&net_todo_list, &list);
5572 while (!list_empty(&list)) {
5573 struct net_device *dev
5574 = list_first_entry(&list, struct net_device, todo_list);
5575 list_del(&dev->todo_list);
5577 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5578 printk(KERN_ERR "network todo '%s' but state %d\n",
5579 dev->name, dev->reg_state);
5584 dev->reg_state = NETREG_UNREGISTERED;
5586 on_each_cpu(flush_backlog, dev, 1);
5588 netdev_wait_allrefs(dev);
5591 BUG_ON(atomic_read(&dev->refcnt));
5592 WARN_ON(dev->ip_ptr);
5593 WARN_ON(dev->ip6_ptr);
5594 WARN_ON(dev->dn_ptr);
5596 if (dev->destructor)
5597 dev->destructor(dev);
5599 /* Free network device */
5600 kobject_put(&dev->dev.kobj);
5605 * dev_txq_stats_fold - fold tx_queues stats
5606 * @dev: device to get statistics from
5607 * @stats: struct net_device_stats to hold results
5609 void dev_txq_stats_fold(const struct net_device *dev,
5610 struct net_device_stats *stats)
5612 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5614 struct netdev_queue *txq;
5616 for (i = 0; i < dev->num_tx_queues; i++) {
5617 txq = netdev_get_tx_queue(dev, i);
5618 tx_bytes += txq->tx_bytes;
5619 tx_packets += txq->tx_packets;
5620 tx_dropped += txq->tx_dropped;
5622 if (tx_bytes || tx_packets || tx_dropped) {
5623 stats->tx_bytes = tx_bytes;
5624 stats->tx_packets = tx_packets;
5625 stats->tx_dropped = tx_dropped;
5628 EXPORT_SYMBOL(dev_txq_stats_fold);
5631 * dev_get_stats - get network device statistics
5632 * @dev: device to get statistics from
5634 * Get network statistics from device. The device driver may provide
5635 * its own method by setting dev->netdev_ops->get_stats; otherwise
5636 * the internal statistics structure is used.
5638 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5640 const struct net_device_ops *ops = dev->netdev_ops;
5642 if (ops->ndo_get_stats)
5643 return ops->ndo_get_stats(dev);
5645 dev_txq_stats_fold(dev, &dev->stats);
5648 EXPORT_SYMBOL(dev_get_stats);
5650 static void netdev_init_one_queue(struct net_device *dev,
5651 struct netdev_queue *queue,
5657 static void netdev_init_queues(struct net_device *dev)
5659 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5660 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5661 spin_lock_init(&dev->tx_global_lock);
5665 * alloc_netdev_mq - allocate network device
5666 * @sizeof_priv: size of private data to allocate space for
5667 * @name: device name format string
5668 * @setup: callback to initialize device
5669 * @queue_count: the number of subqueues to allocate
5671 * Allocates a struct net_device with private data area for driver use
5672 * and performs basic initialization. Also allocates subquue structs
5673 * for each queue on the device at the end of the netdevice.
5675 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5676 void (*setup)(struct net_device *), unsigned int queue_count)
5678 struct netdev_queue *tx;
5679 struct net_device *dev;
5681 struct net_device *p;
5683 struct netdev_rx_queue *rx;
5687 BUG_ON(strlen(name) >= sizeof(dev->name));
5689 alloc_size = sizeof(struct net_device);
5691 /* ensure 32-byte alignment of private area */
5692 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5693 alloc_size += sizeof_priv;
5695 /* ensure 32-byte alignment of whole construct */
5696 alloc_size += NETDEV_ALIGN - 1;
5698 p = kzalloc(alloc_size, GFP_KERNEL);
5700 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5704 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5706 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5712 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5714 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5719 atomic_set(&rx->count, queue_count);
5722 * Set a pointer to first element in the array which holds the
5725 for (i = 0; i < queue_count; i++)
5729 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5730 dev->padded = (char *)dev - (char *)p;
5732 if (dev_addr_init(dev))
5735 dev_unicast_init(dev);
5737 dev_net_set(dev, &init_net);
5740 dev->num_tx_queues = queue_count;
5741 dev->real_num_tx_queues = queue_count;
5745 dev->num_rx_queues = queue_count;
5748 dev->gso_max_size = GSO_MAX_SIZE;
5750 netdev_init_queues(dev);
5752 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5753 dev->ethtool_ntuple_list.count = 0;
5754 INIT_LIST_HEAD(&dev->napi_list);
5755 INIT_LIST_HEAD(&dev->unreg_list);
5756 INIT_LIST_HEAD(&dev->link_watch_list);
5757 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5759 strcpy(dev->name, name);
5772 EXPORT_SYMBOL(alloc_netdev_mq);
5775 * free_netdev - free network device
5778 * This function does the last stage of destroying an allocated device
5779 * interface. The reference to the device object is released.
5780 * If this is the last reference then it will be freed.
5782 void free_netdev(struct net_device *dev)
5784 struct napi_struct *p, *n;
5786 release_net(dev_net(dev));
5790 /* Flush device addresses */
5791 dev_addr_flush(dev);
5793 /* Clear ethtool n-tuple list */
5794 ethtool_ntuple_flush(dev);
5796 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5799 /* Compatibility with error handling in drivers */
5800 if (dev->reg_state == NETREG_UNINITIALIZED) {
5801 kfree((char *)dev - dev->padded);
5805 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5806 dev->reg_state = NETREG_RELEASED;
5808 /* will free via device release */
5809 put_device(&dev->dev);
5811 EXPORT_SYMBOL(free_netdev);
5814 * synchronize_net - Synchronize with packet receive processing
5816 * Wait for packets currently being received to be done.
5817 * Does not block later packets from starting.
5819 void synchronize_net(void)
5824 EXPORT_SYMBOL(synchronize_net);
5827 * unregister_netdevice_queue - remove device from the kernel
5831 * This function shuts down a device interface and removes it
5832 * from the kernel tables.
5833 * If head not NULL, device is queued to be unregistered later.
5835 * Callers must hold the rtnl semaphore. You may want
5836 * unregister_netdev() instead of this.
5839 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5844 list_move_tail(&dev->unreg_list, head);
5846 rollback_registered(dev);
5847 /* Finish processing unregister after unlock */
5851 EXPORT_SYMBOL(unregister_netdevice_queue);
5854 * unregister_netdevice_many - unregister many devices
5855 * @head: list of devices
5857 void unregister_netdevice_many(struct list_head *head)
5859 struct net_device *dev;
5861 if (!list_empty(head)) {
5862 rollback_registered_many(head);
5863 list_for_each_entry(dev, head, unreg_list)
5867 EXPORT_SYMBOL(unregister_netdevice_many);
5870 * unregister_netdev - remove device from the kernel
5873 * This function shuts down a device interface and removes it
5874 * from the kernel tables.
5876 * This is just a wrapper for unregister_netdevice that takes
5877 * the rtnl semaphore. In general you want to use this and not
5878 * unregister_netdevice.
5880 void unregister_netdev(struct net_device *dev)
5883 unregister_netdevice(dev);
5886 EXPORT_SYMBOL(unregister_netdev);
5889 * dev_change_net_namespace - move device to different nethost namespace
5891 * @net: network namespace
5892 * @pat: If not NULL name pattern to try if the current device name
5893 * is already taken in the destination network namespace.
5895 * This function shuts down a device interface and moves it
5896 * to a new network namespace. On success 0 is returned, on
5897 * a failure a netagive errno code is returned.
5899 * Callers must hold the rtnl semaphore.
5902 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5908 /* Don't allow namespace local devices to be moved. */
5910 if (dev->features & NETIF_F_NETNS_LOCAL)
5914 /* Don't allow real devices to be moved when sysfs
5918 if (dev->dev.parent)
5922 /* Ensure the device has been registrered */
5924 if (dev->reg_state != NETREG_REGISTERED)
5927 /* Get out if there is nothing todo */
5929 if (net_eq(dev_net(dev), net))
5932 /* Pick the destination device name, and ensure
5933 * we can use it in the destination network namespace.
5936 if (__dev_get_by_name(net, dev->name)) {
5937 /* We get here if we can't use the current device name */
5940 if (dev_get_valid_name(net, pat, dev->name, 1))
5945 * And now a mini version of register_netdevice unregister_netdevice.
5948 /* If device is running close it first. */
5951 /* And unlink it from device chain */
5953 unlist_netdevice(dev);
5957 /* Shutdown queueing discipline. */
5960 /* Notify protocols, that we are about to destroy
5961 this device. They should clean all the things.
5963 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5964 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5967 * Flush the unicast and multicast chains
5969 dev_unicast_flush(dev);
5970 dev_addr_discard(dev);
5972 netdev_unregister_kobject(dev);
5974 /* Actually switch the network namespace */
5975 dev_net_set(dev, net);
5977 /* If there is an ifindex conflict assign a new one */
5978 if (__dev_get_by_index(net, dev->ifindex)) {
5979 int iflink = (dev->iflink == dev->ifindex);
5980 dev->ifindex = dev_new_index(net);
5982 dev->iflink = dev->ifindex;
5985 /* Fixup kobjects */
5986 err = netdev_register_kobject(dev);
5989 /* Add the device back in the hashes */
5990 list_netdevice(dev);
5992 /* Notify protocols, that a new device appeared. */
5993 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5996 * Prevent userspace races by waiting until the network
5997 * device is fully setup before sending notifications.
5999 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6006 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6008 static int dev_cpu_callback(struct notifier_block *nfb,
6009 unsigned long action,
6012 struct sk_buff **list_skb;
6013 struct Qdisc **list_net;
6014 struct sk_buff *skb;
6015 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6016 struct softnet_data *sd, *oldsd;
6018 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6021 local_irq_disable();
6022 cpu = smp_processor_id();
6023 sd = &per_cpu(softnet_data, cpu);
6024 oldsd = &per_cpu(softnet_data, oldcpu);
6026 /* Find end of our completion_queue. */
6027 list_skb = &sd->completion_queue;
6029 list_skb = &(*list_skb)->next;
6030 /* Append completion queue from offline CPU. */
6031 *list_skb = oldsd->completion_queue;
6032 oldsd->completion_queue = NULL;
6034 /* Find end of our output_queue. */
6035 list_net = &sd->output_queue;
6037 list_net = &(*list_net)->next_sched;
6038 /* Append output queue from offline CPU. */
6039 *list_net = oldsd->output_queue;
6040 oldsd->output_queue = NULL;
6042 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6045 /* Process offline CPU's input_pkt_queue */
6046 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
6054 * netdev_increment_features - increment feature set by one
6055 * @all: current feature set
6056 * @one: new feature set
6057 * @mask: mask feature set
6059 * Computes a new feature set after adding a device with feature set
6060 * @one to the master device with current feature set @all. Will not
6061 * enable anything that is off in @mask. Returns the new feature set.
6063 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
6066 /* If device needs checksumming, downgrade to it. */
6067 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
6068 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
6069 else if (mask & NETIF_F_ALL_CSUM) {
6070 /* If one device supports v4/v6 checksumming, set for all. */
6071 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
6072 !(all & NETIF_F_GEN_CSUM)) {
6073 all &= ~NETIF_F_ALL_CSUM;
6074 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
6077 /* If one device supports hw checksumming, set for all. */
6078 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
6079 all &= ~NETIF_F_ALL_CSUM;
6080 all |= NETIF_F_HW_CSUM;
6084 one |= NETIF_F_ALL_CSUM;
6086 one |= all & NETIF_F_ONE_FOR_ALL;
6087 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
6088 all |= one & mask & NETIF_F_ONE_FOR_ALL;
6092 EXPORT_SYMBOL(netdev_increment_features);
6094 static struct hlist_head *netdev_create_hash(void)
6097 struct hlist_head *hash;
6099 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6101 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6102 INIT_HLIST_HEAD(&hash[i]);
6107 /* Initialize per network namespace state */
6108 static int __net_init netdev_init(struct net *net)
6110 INIT_LIST_HEAD(&net->dev_base_head);
6112 net->dev_name_head = netdev_create_hash();
6113 if (net->dev_name_head == NULL)
6116 net->dev_index_head = netdev_create_hash();
6117 if (net->dev_index_head == NULL)
6123 kfree(net->dev_name_head);
6129 * netdev_drivername - network driver for the device
6130 * @dev: network device
6131 * @buffer: buffer for resulting name
6132 * @len: size of buffer
6134 * Determine network driver for device.
6136 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6138 const struct device_driver *driver;
6139 const struct device *parent;
6141 if (len <= 0 || !buffer)
6145 parent = dev->dev.parent;
6150 driver = parent->driver;
6151 if (driver && driver->name)
6152 strlcpy(buffer, driver->name, len);
6156 static void __net_exit netdev_exit(struct net *net)
6158 kfree(net->dev_name_head);
6159 kfree(net->dev_index_head);
6162 static struct pernet_operations __net_initdata netdev_net_ops = {
6163 .init = netdev_init,
6164 .exit = netdev_exit,
6167 static void __net_exit default_device_exit(struct net *net)
6169 struct net_device *dev, *aux;
6171 * Push all migratable network devices back to the
6172 * initial network namespace
6175 for_each_netdev_safe(net, dev, aux) {
6177 char fb_name[IFNAMSIZ];
6179 /* Ignore unmoveable devices (i.e. loopback) */
6180 if (dev->features & NETIF_F_NETNS_LOCAL)
6183 /* Leave virtual devices for the generic cleanup */
6184 if (dev->rtnl_link_ops)
6187 /* Push remaing network devices to init_net */
6188 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6189 err = dev_change_net_namespace(dev, &init_net, fb_name);
6191 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6192 __func__, dev->name, err);
6199 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6201 /* At exit all network devices most be removed from a network
6202 * namespace. Do this in the reverse order of registeration.
6203 * Do this across as many network namespaces as possible to
6204 * improve batching efficiency.
6206 struct net_device *dev;
6208 LIST_HEAD(dev_kill_list);
6211 list_for_each_entry(net, net_list, exit_list) {
6212 for_each_netdev_reverse(net, dev) {
6213 if (dev->rtnl_link_ops)
6214 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6216 unregister_netdevice_queue(dev, &dev_kill_list);
6219 unregister_netdevice_many(&dev_kill_list);
6223 static struct pernet_operations __net_initdata default_device_ops = {
6224 .exit = default_device_exit,
6225 .exit_batch = default_device_exit_batch,
6229 * Initialize the DEV module. At boot time this walks the device list and
6230 * unhooks any devices that fail to initialise (normally hardware not
6231 * present) and leaves us with a valid list of present and active devices.
6236 * This is called single threaded during boot, so no need
6237 * to take the rtnl semaphore.
6239 static int __init net_dev_init(void)
6241 int i, rc = -ENOMEM;
6243 BUG_ON(!dev_boot_phase);
6245 if (dev_proc_init())
6248 if (netdev_kobject_init())
6251 INIT_LIST_HEAD(&ptype_all);
6252 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6253 INIT_LIST_HEAD(&ptype_base[i]);
6255 if (register_pernet_subsys(&netdev_net_ops))
6259 * Initialise the packet receive queues.
6262 for_each_possible_cpu(i) {
6263 struct softnet_data *queue;
6265 queue = &per_cpu(softnet_data, i);
6266 skb_queue_head_init(&queue->input_pkt_queue);
6267 queue->completion_queue = NULL;
6268 INIT_LIST_HEAD(&queue->poll_list);
6271 queue->csd.func = trigger_softirq;
6272 queue->csd.info = queue;
6273 queue->csd.flags = 0;
6276 queue->backlog.poll = process_backlog;
6277 queue->backlog.weight = weight_p;
6278 queue->backlog.gro_list = NULL;
6279 queue->backlog.gro_count = 0;
6284 /* The loopback device is special if any other network devices
6285 * is present in a network namespace the loopback device must
6286 * be present. Since we now dynamically allocate and free the
6287 * loopback device ensure this invariant is maintained by
6288 * keeping the loopback device as the first device on the
6289 * list of network devices. Ensuring the loopback devices
6290 * is the first device that appears and the last network device
6293 if (register_pernet_device(&loopback_net_ops))
6296 if (register_pernet_device(&default_device_ops))
6299 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6300 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6302 hotcpu_notifier(dev_cpu_callback, 0);
6310 subsys_initcall(net_dev_init);
6312 static int __init initialize_hashrnd(void)
6314 get_random_bytes(&hashrnd, sizeof(hashrnd));
6318 late_initcall_sync(initialize_hashrnd);