2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
104 #include <linux/if_bridge.h>
105 #include <linux/if_macvlan.h>
107 #include <net/pkt_sched.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/kmod.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/wext.h>
118 #include <net/iw_handler.h>
119 #include <asm/current.h>
120 #include <linux/audit.h>
121 #include <linux/dmaengine.h>
122 #include <linux/err.h>
123 #include <linux/ctype.h>
124 #include <linux/if_arp.h>
125 #include <linux/if_vlan.h>
126 #include <linux/ip.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
133 #include <linux/pci.h>
135 #include "net-sysfs.h"
137 /* Instead of increasing this, you should create a hash table. */
138 #define MAX_GRO_SKBS 8
140 /* This should be increased if a protocol with a bigger head is added. */
141 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 * The list of packet types we will receive (as opposed to discard)
145 * and the routines to invoke.
147 * Why 16. Because with 16 the only overlap we get on a hash of the
148 * low nibble of the protocol value is RARP/SNAP/X.25.
150 * NOTE: That is no longer true with the addition of VLAN tags. Not
151 * sure which should go first, but I bet it won't make much
152 * difference if we are running VLANs. The good news is that
153 * this protocol won't be in the list unless compiled in, so
154 * the average user (w/out VLANs) will not be adversely affected.
171 #define PTYPE_HASH_SIZE (16)
172 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
174 static DEFINE_SPINLOCK(ptype_lock);
175 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
176 static struct list_head ptype_all __read_mostly; /* Taps */
179 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
182 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
184 * Writers must hold the rtnl semaphore while they loop through the
185 * dev_base_head list, and hold dev_base_lock for writing when they do the
186 * actual updates. This allows pure readers to access the list even
187 * while a writer is preparing to update it.
189 * To put it another way, dev_base_lock is held for writing only to
190 * protect against pure readers; the rtnl semaphore provides the
191 * protection against other writers.
193 * See, for example usages, register_netdevice() and
194 * unregister_netdevice(), which must be called with the rtnl
197 DEFINE_RWLOCK(dev_base_lock);
198 EXPORT_SYMBOL(dev_base_lock);
200 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
202 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
203 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
206 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
208 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
211 static inline void rps_lock(struct softnet_data *sd)
214 spin_lock(&sd->input_pkt_queue.lock);
218 static inline void rps_unlock(struct softnet_data *sd)
221 spin_unlock(&sd->input_pkt_queue.lock);
225 /* Device list insertion */
226 static int list_netdevice(struct net_device *dev)
228 struct net *net = dev_net(dev);
232 write_lock_bh(&dev_base_lock);
233 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
234 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
235 hlist_add_head_rcu(&dev->index_hlist,
236 dev_index_hash(net, dev->ifindex));
237 write_unlock_bh(&dev_base_lock);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device *dev)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock);
250 list_del_rcu(&dev->dev_list);
251 hlist_del_rcu(&dev->name_hlist);
252 hlist_del_rcu(&dev->index_hlist);
253 write_unlock_bh(&dev_base_lock);
260 static RAW_NOTIFIER_HEAD(netdev_chain);
263 * Device drivers call our routines to queue packets here. We empty the
264 * queue in the local softnet handler.
267 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
268 EXPORT_PER_CPU_SYMBOL(softnet_data);
270 #ifdef CONFIG_LOCKDEP
272 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
273 * according to dev->type
275 static const unsigned short netdev_lock_type[] =
276 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
277 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
278 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
279 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
280 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
281 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
282 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
283 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
284 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
285 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
286 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
287 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
288 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
289 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
290 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
291 ARPHRD_VOID, ARPHRD_NONE};
293 static const char *const netdev_lock_name[] =
294 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
295 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
296 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
297 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
298 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
299 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
300 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
301 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
302 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
303 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
304 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
305 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
306 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
307 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
308 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
309 "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
318 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
319 if (netdev_lock_type[i] == dev_type)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
326 unsigned short dev_type)
330 i = netdev_lock_pos(dev_type);
331 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
332 netdev_lock_name[i]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
339 i = netdev_lock_pos(dev->type);
340 lockdep_set_class_and_name(&dev->addr_list_lock,
341 &netdev_addr_lock_key[i],
342 netdev_lock_name[i]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
346 unsigned short dev_type)
349 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
377 * dev_add_pack - add packet handler
378 * @pt: packet type declaration
380 * Add a protocol handler to the networking stack. The passed &packet_type
381 * is linked into kernel lists and may not be freed until it has been
382 * removed from the kernel lists.
384 * This call does not sleep therefore it can not
385 * guarantee all CPU's that are in middle of receiving packets
386 * will see the new packet type (until the next received packet).
389 void dev_add_pack(struct packet_type *pt)
393 spin_lock_bh(&ptype_lock);
394 if (pt->type == htons(ETH_P_ALL))
395 list_add_rcu(&pt->list, &ptype_all);
397 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
398 list_add_rcu(&pt->list, &ptype_base[hash]);
400 spin_unlock_bh(&ptype_lock);
402 EXPORT_SYMBOL(dev_add_pack);
405 * __dev_remove_pack - remove packet handler
406 * @pt: packet type declaration
408 * Remove a protocol handler that was previously added to the kernel
409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
410 * from the kernel lists and can be freed or reused once this function
413 * The packet type might still be in use by receivers
414 * and must not be freed until after all the CPU's have gone
415 * through a quiescent state.
417 void __dev_remove_pack(struct packet_type *pt)
419 struct list_head *head;
420 struct packet_type *pt1;
422 spin_lock_bh(&ptype_lock);
424 if (pt->type == htons(ETH_P_ALL))
427 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
429 list_for_each_entry(pt1, head, list) {
431 list_del_rcu(&pt->list);
436 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
438 spin_unlock_bh(&ptype_lock);
440 EXPORT_SYMBOL(__dev_remove_pack);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type *pt)
456 __dev_remove_pack(pt);
460 EXPORT_SYMBOL(dev_remove_pack);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name, struct ifmap *map)
482 struct netdev_boot_setup *s;
486 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
487 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
488 memset(s[i].name, 0, sizeof(s[i].name));
489 strlcpy(s[i].name, name, IFNAMSIZ);
490 memcpy(&s[i].map, map, sizeof(s[i].map));
495 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device *dev)
509 struct netdev_boot_setup *s = dev_boot_setup;
512 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
513 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
514 !strcmp(dev->name, s[i].name)) {
515 dev->irq = s[i].map.irq;
516 dev->base_addr = s[i].map.base_addr;
517 dev->mem_start = s[i].map.mem_start;
518 dev->mem_end = s[i].map.mem_end;
524 EXPORT_SYMBOL(netdev_boot_setup_check);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix, int unit)
539 const struct netdev_boot_setup *s = dev_boot_setup;
543 sprintf(name, "%s%d", prefix, unit);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net, name))
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
553 if (!strcmp(name, s[i].name))
554 return s[i].map.base_addr;
559 * Saves at boot time configured settings for any netdevice.
561 int __init netdev_boot_setup(char *str)
566 str = get_options(str, ARRAY_SIZE(ints), ints);
571 memset(&map, 0, sizeof(map));
575 map.base_addr = ints[2];
577 map.mem_start = ints[3];
579 map.mem_end = ints[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str, &map);
585 __setup("netdev=", netdev_boot_setup);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device *__dev_get_by_name(struct net *net, const char *name)
607 struct hlist_node *p;
608 struct net_device *dev;
609 struct hlist_head *head = dev_name_hash(net, name);
611 hlist_for_each_entry(dev, p, head, name_hlist)
612 if (!strncmp(dev->name, name, IFNAMSIZ))
617 EXPORT_SYMBOL(__dev_get_by_name);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
633 struct hlist_node *p;
634 struct net_device *dev;
635 struct hlist_head *head = dev_name_hash(net, name);
637 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
638 if (!strncmp(dev->name, name, IFNAMSIZ))
643 EXPORT_SYMBOL(dev_get_by_name_rcu);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device *dev_get_by_name(struct net *net, const char *name)
659 struct net_device *dev;
662 dev = dev_get_by_name_rcu(net, name);
668 EXPORT_SYMBOL(dev_get_by_name);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
684 struct hlist_node *p;
685 struct net_device *dev;
686 struct hlist_head *head = dev_index_hash(net, ifindex);
688 hlist_for_each_entry(dev, p, head, index_hlist)
689 if (dev->ifindex == ifindex)
694 EXPORT_SYMBOL(__dev_get_by_index);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
709 struct hlist_node *p;
710 struct net_device *dev;
711 struct hlist_head *head = dev_index_hash(net, ifindex);
713 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
714 if (dev->ifindex == ifindex)
719 EXPORT_SYMBOL(dev_get_by_index_rcu);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device *dev_get_by_index(struct net *net, int ifindex)
735 struct net_device *dev;
738 dev = dev_get_by_index_rcu(net, ifindex);
744 EXPORT_SYMBOL(dev_get_by_index);
747 * dev_getbyhwaddr - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device. The caller must hold the
754 * rtnl semaphore. The returned device has not had its ref count increased
755 * and the caller must therefore be careful about locking
758 * If the API was consistent this would be __dev_get_by_hwaddr
761 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
763 struct net_device *dev;
767 for_each_netdev(net, dev)
768 if (dev->type == type &&
769 !memcmp(dev->dev_addr, ha, dev->addr_len))
774 EXPORT_SYMBOL(dev_getbyhwaddr);
776 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
778 struct net_device *dev;
781 for_each_netdev(net, dev)
782 if (dev->type == type)
787 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
789 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
791 struct net_device *dev, *ret = NULL;
794 for_each_netdev_rcu(net, dev)
795 if (dev->type == type) {
803 EXPORT_SYMBOL(dev_getfirstbyhwtype);
806 * dev_get_by_flags_rcu - find any device with given flags
807 * @net: the applicable net namespace
808 * @if_flags: IFF_* values
809 * @mask: bitmask of bits in if_flags to check
811 * Search for any interface with the given flags. Returns NULL if a device
812 * is not found or a pointer to the device. Must be called inside
813 * rcu_read_lock(), and result refcount is unchanged.
816 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
819 struct net_device *dev, *ret;
822 for_each_netdev_rcu(net, dev) {
823 if (((dev->flags ^ if_flags) & mask) == 0) {
830 EXPORT_SYMBOL(dev_get_by_flags_rcu);
833 * dev_valid_name - check if name is okay for network device
836 * Network device names need to be valid file names to
837 * to allow sysfs to work. We also disallow any kind of
840 int dev_valid_name(const char *name)
844 if (strlen(name) >= IFNAMSIZ)
846 if (!strcmp(name, ".") || !strcmp(name, ".."))
850 if (*name == '/' || isspace(*name))
856 EXPORT_SYMBOL(dev_valid_name);
859 * __dev_alloc_name - allocate a name for a device
860 * @net: network namespace to allocate the device name in
861 * @name: name format string
862 * @buf: scratch buffer and result name string
864 * Passed a format string - eg "lt%d" it will try and find a suitable
865 * id. It scans list of devices to build up a free map, then chooses
866 * the first empty slot. The caller must hold the dev_base or rtnl lock
867 * while allocating the name and adding the device in order to avoid
869 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
870 * Returns the number of the unit assigned or a negative errno code.
873 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
877 const int max_netdevices = 8*PAGE_SIZE;
878 unsigned long *inuse;
879 struct net_device *d;
881 p = strnchr(name, IFNAMSIZ-1, '%');
884 * Verify the string as this thing may have come from
885 * the user. There must be either one "%d" and no other "%"
888 if (p[1] != 'd' || strchr(p + 2, '%'))
891 /* Use one page as a bit array of possible slots */
892 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
896 for_each_netdev(net, d) {
897 if (!sscanf(d->name, name, &i))
899 if (i < 0 || i >= max_netdevices)
902 /* avoid cases where sscanf is not exact inverse of printf */
903 snprintf(buf, IFNAMSIZ, name, i);
904 if (!strncmp(buf, d->name, IFNAMSIZ))
908 i = find_first_zero_bit(inuse, max_netdevices);
909 free_page((unsigned long) inuse);
913 snprintf(buf, IFNAMSIZ, name, i);
914 if (!__dev_get_by_name(net, buf))
917 /* It is possible to run out of possible slots
918 * when the name is long and there isn't enough space left
919 * for the digits, or if all bits are used.
925 * dev_alloc_name - allocate a name for a device
927 * @name: name format string
929 * Passed a format string - eg "lt%d" it will try and find a suitable
930 * id. It scans list of devices to build up a free map, then chooses
931 * the first empty slot. The caller must hold the dev_base or rtnl lock
932 * while allocating the name and adding the device in order to avoid
934 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
935 * Returns the number of the unit assigned or a negative errno code.
938 int dev_alloc_name(struct net_device *dev, const char *name)
944 BUG_ON(!dev_net(dev));
946 ret = __dev_alloc_name(net, name, buf);
948 strlcpy(dev->name, buf, IFNAMSIZ);
951 EXPORT_SYMBOL(dev_alloc_name);
953 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
957 BUG_ON(!dev_net(dev));
960 if (!dev_valid_name(name))
963 if (fmt && strchr(name, '%'))
964 return dev_alloc_name(dev, name);
965 else if (__dev_get_by_name(net, name))
967 else if (dev->name != name)
968 strlcpy(dev->name, name, IFNAMSIZ);
974 * dev_change_name - change name of a device
976 * @newname: name (or format string) must be at least IFNAMSIZ
978 * Change name of a device, can pass format strings "eth%d".
981 int dev_change_name(struct net_device *dev, const char *newname)
983 char oldname[IFNAMSIZ];
989 BUG_ON(!dev_net(dev));
992 if (dev->flags & IFF_UP)
995 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
998 memcpy(oldname, dev->name, IFNAMSIZ);
1000 err = dev_get_valid_name(dev, newname, 1);
1005 ret = device_rename(&dev->dev, dev->name);
1007 memcpy(dev->name, oldname, IFNAMSIZ);
1011 write_lock_bh(&dev_base_lock);
1012 hlist_del(&dev->name_hlist);
1013 write_unlock_bh(&dev_base_lock);
1017 write_lock_bh(&dev_base_lock);
1018 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1019 write_unlock_bh(&dev_base_lock);
1021 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1022 ret = notifier_to_errno(ret);
1025 /* err >= 0 after dev_alloc_name() or stores the first errno */
1028 memcpy(dev->name, oldname, IFNAMSIZ);
1032 "%s: name change rollback failed: %d.\n",
1041 * dev_set_alias - change ifalias of a device
1043 * @alias: name up to IFALIASZ
1044 * @len: limit of bytes to copy from info
1046 * Set ifalias for a device,
1048 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1052 if (len >= IFALIASZ)
1057 kfree(dev->ifalias);
1058 dev->ifalias = NULL;
1063 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1067 strlcpy(dev->ifalias, alias, len+1);
1073 * netdev_features_change - device changes features
1074 * @dev: device to cause notification
1076 * Called to indicate a device has changed features.
1078 void netdev_features_change(struct net_device *dev)
1080 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1082 EXPORT_SYMBOL(netdev_features_change);
1085 * netdev_state_change - device changes state
1086 * @dev: device to cause notification
1088 * Called to indicate a device has changed state. This function calls
1089 * the notifier chains for netdev_chain and sends a NEWLINK message
1090 * to the routing socket.
1092 void netdev_state_change(struct net_device *dev)
1094 if (dev->flags & IFF_UP) {
1095 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1096 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1099 EXPORT_SYMBOL(netdev_state_change);
1101 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1103 return call_netdevice_notifiers(event, dev);
1105 EXPORT_SYMBOL(netdev_bonding_change);
1108 * dev_load - load a network module
1109 * @net: the applicable net namespace
1110 * @name: name of interface
1112 * If a network interface is not present and the process has suitable
1113 * privileges this function loads the module. If module loading is not
1114 * available in this kernel then it becomes a nop.
1117 void dev_load(struct net *net, const char *name)
1119 struct net_device *dev;
1122 dev = dev_get_by_name_rcu(net, name);
1125 if (!dev && capable(CAP_NET_ADMIN))
1126 request_module("%s", name);
1128 EXPORT_SYMBOL(dev_load);
1130 static int __dev_open(struct net_device *dev)
1132 const struct net_device_ops *ops = dev->netdev_ops;
1138 * Is it even present?
1140 if (!netif_device_present(dev))
1143 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1144 ret = notifier_to_errno(ret);
1149 * Call device private open method
1151 set_bit(__LINK_STATE_START, &dev->state);
1153 if (ops->ndo_validate_addr)
1154 ret = ops->ndo_validate_addr(dev);
1156 if (!ret && ops->ndo_open)
1157 ret = ops->ndo_open(dev);
1160 * If it went open OK then:
1164 clear_bit(__LINK_STATE_START, &dev->state);
1169 dev->flags |= IFF_UP;
1174 net_dmaengine_get();
1177 * Initialize multicasting status
1179 dev_set_rx_mode(dev);
1182 * Wakeup transmit queue engine
1191 * dev_open - prepare an interface for use.
1192 * @dev: device to open
1194 * Takes a device from down to up state. The device's private open
1195 * function is invoked and then the multicast lists are loaded. Finally
1196 * the device is moved into the up state and a %NETDEV_UP message is
1197 * sent to the netdev notifier chain.
1199 * Calling this function on an active interface is a nop. On a failure
1200 * a negative errno code is returned.
1202 int dev_open(struct net_device *dev)
1209 if (dev->flags & IFF_UP)
1215 ret = __dev_open(dev);
1220 * ... and announce new interface.
1222 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1223 call_netdevice_notifiers(NETDEV_UP, dev);
1227 EXPORT_SYMBOL(dev_open);
1229 static int __dev_close(struct net_device *dev)
1231 const struct net_device_ops *ops = dev->netdev_ops;
1237 * Tell people we are going down, so that they can
1238 * prepare to death, when device is still operating.
1240 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1242 clear_bit(__LINK_STATE_START, &dev->state);
1244 /* Synchronize to scheduled poll. We cannot touch poll list,
1245 * it can be even on different cpu. So just clear netif_running().
1247 * dev->stop() will invoke napi_disable() on all of it's
1248 * napi_struct instances on this device.
1250 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1252 dev_deactivate(dev);
1255 * Call the device specific close. This cannot fail.
1256 * Only if device is UP
1258 * We allow it to be called even after a DETACH hot-plug
1265 * Device is now down.
1268 dev->flags &= ~IFF_UP;
1273 net_dmaengine_put();
1279 * dev_close - shutdown an interface.
1280 * @dev: device to shutdown
1282 * This function moves an active device into down state. A
1283 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1284 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1287 int dev_close(struct net_device *dev)
1289 if (!(dev->flags & IFF_UP))
1295 * Tell people we are down
1297 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1298 call_netdevice_notifiers(NETDEV_DOWN, dev);
1302 EXPORT_SYMBOL(dev_close);
1306 * dev_disable_lro - disable Large Receive Offload on a device
1309 * Disable Large Receive Offload (LRO) on a net device. Must be
1310 * called under RTNL. This is needed if received packets may be
1311 * forwarded to another interface.
1313 void dev_disable_lro(struct net_device *dev)
1315 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1316 dev->ethtool_ops->set_flags) {
1317 u32 flags = dev->ethtool_ops->get_flags(dev);
1318 if (flags & ETH_FLAG_LRO) {
1319 flags &= ~ETH_FLAG_LRO;
1320 dev->ethtool_ops->set_flags(dev, flags);
1323 WARN_ON(dev->features & NETIF_F_LRO);
1325 EXPORT_SYMBOL(dev_disable_lro);
1328 static int dev_boot_phase = 1;
1331 * Device change register/unregister. These are not inline or static
1332 * as we export them to the world.
1336 * register_netdevice_notifier - register a network notifier block
1339 * Register a notifier to be called when network device events occur.
1340 * The notifier passed is linked into the kernel structures and must
1341 * not be reused until it has been unregistered. A negative errno code
1342 * is returned on a failure.
1344 * When registered all registration and up events are replayed
1345 * to the new notifier to allow device to have a race free
1346 * view of the network device list.
1349 int register_netdevice_notifier(struct notifier_block *nb)
1351 struct net_device *dev;
1352 struct net_device *last;
1357 err = raw_notifier_chain_register(&netdev_chain, nb);
1363 for_each_netdev(net, dev) {
1364 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1365 err = notifier_to_errno(err);
1369 if (!(dev->flags & IFF_UP))
1372 nb->notifier_call(nb, NETDEV_UP, dev);
1383 for_each_netdev(net, dev) {
1387 if (dev->flags & IFF_UP) {
1388 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1389 nb->notifier_call(nb, NETDEV_DOWN, dev);
1391 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1392 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1396 raw_notifier_chain_unregister(&netdev_chain, nb);
1399 EXPORT_SYMBOL(register_netdevice_notifier);
1402 * unregister_netdevice_notifier - unregister a network notifier block
1405 * Unregister a notifier previously registered by
1406 * register_netdevice_notifier(). The notifier is unlinked into the
1407 * kernel structures and may then be reused. A negative errno code
1408 * is returned on a failure.
1411 int unregister_netdevice_notifier(struct notifier_block *nb)
1416 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1420 EXPORT_SYMBOL(unregister_netdevice_notifier);
1423 * call_netdevice_notifiers - call all network notifier blocks
1424 * @val: value passed unmodified to notifier function
1425 * @dev: net_device pointer passed unmodified to notifier function
1427 * Call all network notifier blocks. Parameters and return value
1428 * are as for raw_notifier_call_chain().
1431 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1434 return raw_notifier_call_chain(&netdev_chain, val, dev);
1437 /* When > 0 there are consumers of rx skb time stamps */
1438 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1440 void net_enable_timestamp(void)
1442 atomic_inc(&netstamp_needed);
1444 EXPORT_SYMBOL(net_enable_timestamp);
1446 void net_disable_timestamp(void)
1448 atomic_dec(&netstamp_needed);
1450 EXPORT_SYMBOL(net_disable_timestamp);
1452 static inline void net_timestamp_set(struct sk_buff *skb)
1454 if (atomic_read(&netstamp_needed))
1455 __net_timestamp(skb);
1457 skb->tstamp.tv64 = 0;
1460 static inline void net_timestamp_check(struct sk_buff *skb)
1462 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1463 __net_timestamp(skb);
1467 * dev_forward_skb - loopback an skb to another netif
1469 * @dev: destination network device
1470 * @skb: buffer to forward
1473 * NET_RX_SUCCESS (no congestion)
1474 * NET_RX_DROP (packet was dropped, but freed)
1476 * dev_forward_skb can be used for injecting an skb from the
1477 * start_xmit function of one device into the receive queue
1478 * of another device.
1480 * The receiving device may be in another namespace, so
1481 * we have to clear all information in the skb that could
1482 * impact namespace isolation.
1484 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1489 if (!(dev->flags & IFF_UP) ||
1490 (skb->len > (dev->mtu + dev->hard_header_len))) {
1494 skb_set_dev(skb, dev);
1495 skb->tstamp.tv64 = 0;
1496 skb->pkt_type = PACKET_HOST;
1497 skb->protocol = eth_type_trans(skb, dev);
1498 return netif_rx(skb);
1500 EXPORT_SYMBOL_GPL(dev_forward_skb);
1503 * Support routine. Sends outgoing frames to any network
1504 * taps currently in use.
1507 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1509 struct packet_type *ptype;
1511 #ifdef CONFIG_NET_CLS_ACT
1512 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1513 net_timestamp_set(skb);
1515 net_timestamp_set(skb);
1519 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1520 /* Never send packets back to the socket
1521 * they originated from - MvS (miquels@drinkel.ow.org)
1523 if ((ptype->dev == dev || !ptype->dev) &&
1524 (ptype->af_packet_priv == NULL ||
1525 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1526 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1530 /* skb->nh should be correctly
1531 set by sender, so that the second statement is
1532 just protection against buggy protocols.
1534 skb_reset_mac_header(skb2);
1536 if (skb_network_header(skb2) < skb2->data ||
1537 skb2->network_header > skb2->tail) {
1538 if (net_ratelimit())
1539 printk(KERN_CRIT "protocol %04x is "
1541 ntohs(skb2->protocol),
1543 skb_reset_network_header(skb2);
1546 skb2->transport_header = skb2->network_header;
1547 skb2->pkt_type = PACKET_OUTGOING;
1548 ptype->func(skb2, skb->dev, ptype, skb->dev);
1555 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1556 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1558 void netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1560 unsigned int real_num = dev->real_num_tx_queues;
1562 if (unlikely(txq > dev->num_tx_queues))
1564 else if (txq > real_num)
1565 dev->real_num_tx_queues = txq;
1566 else if (txq < real_num) {
1567 dev->real_num_tx_queues = txq;
1568 qdisc_reset_all_tx_gt(dev, txq);
1571 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1573 static inline void __netif_reschedule(struct Qdisc *q)
1575 struct softnet_data *sd;
1576 unsigned long flags;
1578 local_irq_save(flags);
1579 sd = &__get_cpu_var(softnet_data);
1580 q->next_sched = NULL;
1581 *sd->output_queue_tailp = q;
1582 sd->output_queue_tailp = &q->next_sched;
1583 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1584 local_irq_restore(flags);
1587 void __netif_schedule(struct Qdisc *q)
1589 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1590 __netif_reschedule(q);
1592 EXPORT_SYMBOL(__netif_schedule);
1594 void dev_kfree_skb_irq(struct sk_buff *skb)
1596 if (!skb->destructor)
1598 else if (atomic_dec_and_test(&skb->users)) {
1599 struct softnet_data *sd;
1600 unsigned long flags;
1602 local_irq_save(flags);
1603 sd = &__get_cpu_var(softnet_data);
1604 skb->next = sd->completion_queue;
1605 sd->completion_queue = skb;
1606 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1607 local_irq_restore(flags);
1610 EXPORT_SYMBOL(dev_kfree_skb_irq);
1612 void dev_kfree_skb_any(struct sk_buff *skb)
1614 if (in_irq() || irqs_disabled())
1615 dev_kfree_skb_irq(skb);
1619 EXPORT_SYMBOL(dev_kfree_skb_any);
1623 * netif_device_detach - mark device as removed
1624 * @dev: network device
1626 * Mark device as removed from system and therefore no longer available.
1628 void netif_device_detach(struct net_device *dev)
1630 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1631 netif_running(dev)) {
1632 netif_tx_stop_all_queues(dev);
1635 EXPORT_SYMBOL(netif_device_detach);
1638 * netif_device_attach - mark device as attached
1639 * @dev: network device
1641 * Mark device as attached from system and restart if needed.
1643 void netif_device_attach(struct net_device *dev)
1645 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1646 netif_running(dev)) {
1647 netif_tx_wake_all_queues(dev);
1648 __netdev_watchdog_up(dev);
1651 EXPORT_SYMBOL(netif_device_attach);
1653 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1655 return ((features & NETIF_F_GEN_CSUM) ||
1656 ((features & NETIF_F_IP_CSUM) &&
1657 protocol == htons(ETH_P_IP)) ||
1658 ((features & NETIF_F_IPV6_CSUM) &&
1659 protocol == htons(ETH_P_IPV6)) ||
1660 ((features & NETIF_F_FCOE_CRC) &&
1661 protocol == htons(ETH_P_FCOE)));
1664 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1666 if (can_checksum_protocol(dev->features, skb->protocol))
1669 if (skb->protocol == htons(ETH_P_8021Q)) {
1670 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1671 if (can_checksum_protocol(dev->features & dev->vlan_features,
1672 veh->h_vlan_encapsulated_proto))
1680 * skb_dev_set -- assign a new device to a buffer
1681 * @skb: buffer for the new device
1682 * @dev: network device
1684 * If an skb is owned by a device already, we have to reset
1685 * all data private to the namespace a device belongs to
1686 * before assigning it a new device.
1688 #ifdef CONFIG_NET_NS
1689 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1692 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1695 skb_init_secmark(skb);
1699 skb->ipvs_property = 0;
1700 #ifdef CONFIG_NET_SCHED
1706 EXPORT_SYMBOL(skb_set_dev);
1707 #endif /* CONFIG_NET_NS */
1710 * Invalidate hardware checksum when packet is to be mangled, and
1711 * complete checksum manually on outgoing path.
1713 int skb_checksum_help(struct sk_buff *skb)
1716 int ret = 0, offset;
1718 if (skb->ip_summed == CHECKSUM_COMPLETE)
1719 goto out_set_summed;
1721 if (unlikely(skb_shinfo(skb)->gso_size)) {
1722 /* Let GSO fix up the checksum. */
1723 goto out_set_summed;
1726 offset = skb->csum_start - skb_headroom(skb);
1727 BUG_ON(offset >= skb_headlen(skb));
1728 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1730 offset += skb->csum_offset;
1731 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1733 if (skb_cloned(skb) &&
1734 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1735 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1740 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1742 skb->ip_summed = CHECKSUM_NONE;
1746 EXPORT_SYMBOL(skb_checksum_help);
1749 * skb_gso_segment - Perform segmentation on skb.
1750 * @skb: buffer to segment
1751 * @features: features for the output path (see dev->features)
1753 * This function segments the given skb and returns a list of segments.
1755 * It may return NULL if the skb requires no segmentation. This is
1756 * only possible when GSO is used for verifying header integrity.
1758 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1760 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1761 struct packet_type *ptype;
1762 __be16 type = skb->protocol;
1765 skb_reset_mac_header(skb);
1766 skb->mac_len = skb->network_header - skb->mac_header;
1767 __skb_pull(skb, skb->mac_len);
1769 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1770 struct net_device *dev = skb->dev;
1771 struct ethtool_drvinfo info = {};
1773 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1774 dev->ethtool_ops->get_drvinfo(dev, &info);
1776 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1778 info.driver, dev ? dev->features : 0L,
1779 skb->sk ? skb->sk->sk_route_caps : 0L,
1780 skb->len, skb->data_len, skb->ip_summed);
1782 if (skb_header_cloned(skb) &&
1783 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1784 return ERR_PTR(err);
1788 list_for_each_entry_rcu(ptype,
1789 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1790 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1791 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1792 err = ptype->gso_send_check(skb);
1793 segs = ERR_PTR(err);
1794 if (err || skb_gso_ok(skb, features))
1796 __skb_push(skb, (skb->data -
1797 skb_network_header(skb)));
1799 segs = ptype->gso_segment(skb, features);
1805 __skb_push(skb, skb->data - skb_mac_header(skb));
1809 EXPORT_SYMBOL(skb_gso_segment);
1811 /* Take action when hardware reception checksum errors are detected. */
1813 void netdev_rx_csum_fault(struct net_device *dev)
1815 if (net_ratelimit()) {
1816 printk(KERN_ERR "%s: hw csum failure.\n",
1817 dev ? dev->name : "<unknown>");
1821 EXPORT_SYMBOL(netdev_rx_csum_fault);
1824 /* Actually, we should eliminate this check as soon as we know, that:
1825 * 1. IOMMU is present and allows to map all the memory.
1826 * 2. No high memory really exists on this machine.
1829 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1831 #ifdef CONFIG_HIGHMEM
1833 if (!(dev->features & NETIF_F_HIGHDMA)) {
1834 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1835 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1839 if (PCI_DMA_BUS_IS_PHYS) {
1840 struct device *pdev = dev->dev.parent;
1844 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1845 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1846 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1855 void (*destructor)(struct sk_buff *skb);
1858 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1860 static void dev_gso_skb_destructor(struct sk_buff *skb)
1862 struct dev_gso_cb *cb;
1865 struct sk_buff *nskb = skb->next;
1867 skb->next = nskb->next;
1870 } while (skb->next);
1872 cb = DEV_GSO_CB(skb);
1874 cb->destructor(skb);
1878 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1879 * @skb: buffer to segment
1881 * This function segments the given skb and stores the list of segments
1884 static int dev_gso_segment(struct sk_buff *skb)
1886 struct net_device *dev = skb->dev;
1887 struct sk_buff *segs;
1888 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1891 segs = skb_gso_segment(skb, features);
1893 /* Verifying header integrity only. */
1898 return PTR_ERR(segs);
1901 DEV_GSO_CB(skb)->destructor = skb->destructor;
1902 skb->destructor = dev_gso_skb_destructor;
1908 * Try to orphan skb early, right before transmission by the device.
1909 * We cannot orphan skb if tx timestamp is requested, since
1910 * drivers need to call skb_tstamp_tx() to send the timestamp.
1912 static inline void skb_orphan_try(struct sk_buff *skb)
1914 struct sock *sk = skb->sk;
1916 if (sk && !skb_tx(skb)->flags) {
1917 /* skb_tx_hash() wont be able to get sk.
1918 * We copy sk_hash into skb->rxhash
1921 skb->rxhash = sk->sk_hash;
1927 * Returns true if either:
1928 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1929 * 2. skb is fragmented and the device does not support SG, or if
1930 * at least one of fragments is in highmem and device does not
1931 * support DMA from it.
1933 static inline int skb_needs_linearize(struct sk_buff *skb,
1934 struct net_device *dev)
1936 return skb_is_nonlinear(skb) &&
1937 ((skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
1938 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
1939 illegal_highdma(dev, skb))));
1942 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1943 struct netdev_queue *txq)
1945 const struct net_device_ops *ops = dev->netdev_ops;
1946 int rc = NETDEV_TX_OK;
1948 if (likely(!skb->next)) {
1949 if (!list_empty(&ptype_all))
1950 dev_queue_xmit_nit(skb, dev);
1953 * If device doesnt need skb->dst, release it right now while
1954 * its hot in this cpu cache
1956 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1959 skb_orphan_try(skb);
1961 if (netif_needs_gso(dev, skb)) {
1962 if (unlikely(dev_gso_segment(skb)))
1967 if (skb_needs_linearize(skb, dev) &&
1968 __skb_linearize(skb))
1971 /* If packet is not checksummed and device does not
1972 * support checksumming for this protocol, complete
1973 * checksumming here.
1975 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1976 skb_set_transport_header(skb, skb->csum_start -
1978 if (!dev_can_checksum(dev, skb) &&
1979 skb_checksum_help(skb))
1984 rc = ops->ndo_start_xmit(skb, dev);
1985 if (rc == NETDEV_TX_OK)
1986 txq_trans_update(txq);
1992 struct sk_buff *nskb = skb->next;
1994 skb->next = nskb->next;
1998 * If device doesnt need nskb->dst, release it right now while
1999 * its hot in this cpu cache
2001 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2004 rc = ops->ndo_start_xmit(nskb, dev);
2005 if (unlikely(rc != NETDEV_TX_OK)) {
2006 if (rc & ~NETDEV_TX_MASK)
2007 goto out_kfree_gso_skb;
2008 nskb->next = skb->next;
2012 txq_trans_update(txq);
2013 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2014 return NETDEV_TX_BUSY;
2015 } while (skb->next);
2018 if (likely(skb->next == NULL))
2019 skb->destructor = DEV_GSO_CB(skb)->destructor;
2025 static u32 hashrnd __read_mostly;
2027 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
2031 if (skb_rx_queue_recorded(skb)) {
2032 hash = skb_get_rx_queue(skb);
2033 while (unlikely(hash >= dev->real_num_tx_queues))
2034 hash -= dev->real_num_tx_queues;
2038 if (skb->sk && skb->sk->sk_hash)
2039 hash = skb->sk->sk_hash;
2041 hash = (__force u16) skb->protocol ^ skb->rxhash;
2042 hash = jhash_1word(hash, hashrnd);
2044 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2046 EXPORT_SYMBOL(skb_tx_hash);
2048 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2050 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2051 if (net_ratelimit()) {
2052 pr_warning("%s selects TX queue %d, but "
2053 "real number of TX queues is %d\n",
2054 dev->name, queue_index, dev->real_num_tx_queues);
2061 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2062 struct sk_buff *skb)
2065 struct sock *sk = skb->sk;
2067 queue_index = sk_tx_queue_get(sk);
2068 if (queue_index < 0) {
2069 const struct net_device_ops *ops = dev->netdev_ops;
2071 if (ops->ndo_select_queue) {
2072 queue_index = ops->ndo_select_queue(dev, skb);
2073 queue_index = dev_cap_txqueue(dev, queue_index);
2076 if (dev->real_num_tx_queues > 1)
2077 queue_index = skb_tx_hash(dev, skb);
2080 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2082 if (dst && skb_dst(skb) == dst)
2083 sk_tx_queue_set(sk, queue_index);
2088 skb_set_queue_mapping(skb, queue_index);
2089 return netdev_get_tx_queue(dev, queue_index);
2092 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2093 struct net_device *dev,
2094 struct netdev_queue *txq)
2096 spinlock_t *root_lock = qdisc_lock(q);
2097 bool contended = qdisc_is_running(q);
2101 * Heuristic to force contended enqueues to serialize on a
2102 * separate lock before trying to get qdisc main lock.
2103 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2104 * and dequeue packets faster.
2106 if (unlikely(contended))
2107 spin_lock(&q->busylock);
2109 spin_lock(root_lock);
2110 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2113 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2114 qdisc_run_begin(q)) {
2116 * This is a work-conserving queue; there are no old skbs
2117 * waiting to be sent out; and the qdisc is not running -
2118 * xmit the skb directly.
2120 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2122 __qdisc_update_bstats(q, skb->len);
2123 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2124 if (unlikely(contended)) {
2125 spin_unlock(&q->busylock);
2132 rc = NET_XMIT_SUCCESS;
2135 rc = qdisc_enqueue_root(skb, q);
2136 if (qdisc_run_begin(q)) {
2137 if (unlikely(contended)) {
2138 spin_unlock(&q->busylock);
2144 spin_unlock(root_lock);
2145 if (unlikely(contended))
2146 spin_unlock(&q->busylock);
2151 * dev_queue_xmit - transmit a buffer
2152 * @skb: buffer to transmit
2154 * Queue a buffer for transmission to a network device. The caller must
2155 * have set the device and priority and built the buffer before calling
2156 * this function. The function can be called from an interrupt.
2158 * A negative errno code is returned on a failure. A success does not
2159 * guarantee the frame will be transmitted as it may be dropped due
2160 * to congestion or traffic shaping.
2162 * -----------------------------------------------------------------------------------
2163 * I notice this method can also return errors from the queue disciplines,
2164 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2167 * Regardless of the return value, the skb is consumed, so it is currently
2168 * difficult to retry a send to this method. (You can bump the ref count
2169 * before sending to hold a reference for retry if you are careful.)
2171 * When calling this method, interrupts MUST be enabled. This is because
2172 * the BH enable code must have IRQs enabled so that it will not deadlock.
2175 int dev_queue_xmit(struct sk_buff *skb)
2177 struct net_device *dev = skb->dev;
2178 struct netdev_queue *txq;
2182 /* Disable soft irqs for various locks below. Also
2183 * stops preemption for RCU.
2187 txq = dev_pick_tx(dev, skb);
2188 q = rcu_dereference_bh(txq->qdisc);
2190 #ifdef CONFIG_NET_CLS_ACT
2191 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2194 rc = __dev_xmit_skb(skb, q, dev, txq);
2198 /* The device has no queue. Common case for software devices:
2199 loopback, all the sorts of tunnels...
2201 Really, it is unlikely that netif_tx_lock protection is necessary
2202 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2204 However, it is possible, that they rely on protection
2207 Check this and shot the lock. It is not prone from deadlocks.
2208 Either shot noqueue qdisc, it is even simpler 8)
2210 if (dev->flags & IFF_UP) {
2211 int cpu = smp_processor_id(); /* ok because BHs are off */
2213 if (txq->xmit_lock_owner != cpu) {
2215 HARD_TX_LOCK(dev, txq, cpu);
2217 if (!netif_tx_queue_stopped(txq)) {
2218 rc = dev_hard_start_xmit(skb, dev, txq);
2219 if (dev_xmit_complete(rc)) {
2220 HARD_TX_UNLOCK(dev, txq);
2224 HARD_TX_UNLOCK(dev, txq);
2225 if (net_ratelimit())
2226 printk(KERN_CRIT "Virtual device %s asks to "
2227 "queue packet!\n", dev->name);
2229 /* Recursion is detected! It is possible,
2231 if (net_ratelimit())
2232 printk(KERN_CRIT "Dead loop on virtual device "
2233 "%s, fix it urgently!\n", dev->name);
2238 rcu_read_unlock_bh();
2243 rcu_read_unlock_bh();
2246 EXPORT_SYMBOL(dev_queue_xmit);
2249 /*=======================================================================
2251 =======================================================================*/
2253 int netdev_max_backlog __read_mostly = 1000;
2254 int netdev_tstamp_prequeue __read_mostly = 1;
2255 int netdev_budget __read_mostly = 300;
2256 int weight_p __read_mostly = 64; /* old backlog weight */
2258 /* Called with irq disabled */
2259 static inline void ____napi_schedule(struct softnet_data *sd,
2260 struct napi_struct *napi)
2262 list_add_tail(&napi->poll_list, &sd->poll_list);
2263 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2268 /* One global table that all flow-based protocols share. */
2269 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2270 EXPORT_SYMBOL(rps_sock_flow_table);
2273 * get_rps_cpu is called from netif_receive_skb and returns the target
2274 * CPU from the RPS map of the receiving queue for a given skb.
2275 * rcu_read_lock must be held on entry.
2277 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2278 struct rps_dev_flow **rflowp)
2280 struct ipv6hdr *ip6;
2282 struct netdev_rx_queue *rxqueue;
2283 struct rps_map *map;
2284 struct rps_dev_flow_table *flow_table;
2285 struct rps_sock_flow_table *sock_flow_table;
2289 u32 addr1, addr2, ihl;
2295 if (skb_rx_queue_recorded(skb)) {
2296 u16 index = skb_get_rx_queue(skb);
2297 if (unlikely(index >= dev->num_rx_queues)) {
2298 WARN_ONCE(dev->num_rx_queues > 1, "%s received packet "
2299 "on queue %u, but number of RX queues is %u\n",
2300 dev->name, index, dev->num_rx_queues);
2303 rxqueue = dev->_rx + index;
2307 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2311 goto got_hash; /* Skip hash computation on packet header */
2313 switch (skb->protocol) {
2314 case __constant_htons(ETH_P_IP):
2315 if (!pskb_may_pull(skb, sizeof(*ip)))
2318 ip = (struct iphdr *) skb->data;
2319 ip_proto = ip->protocol;
2320 addr1 = (__force u32) ip->saddr;
2321 addr2 = (__force u32) ip->daddr;
2324 case __constant_htons(ETH_P_IPV6):
2325 if (!pskb_may_pull(skb, sizeof(*ip6)))
2328 ip6 = (struct ipv6hdr *) skb->data;
2329 ip_proto = ip6->nexthdr;
2330 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2331 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2344 case IPPROTO_UDPLITE:
2345 if (pskb_may_pull(skb, (ihl * 4) + 4)) {
2346 ports.v32 = * (__force u32 *) (skb->data + (ihl * 4));
2347 if (ports.v16[1] < ports.v16[0])
2348 swap(ports.v16[0], ports.v16[1]);
2356 /* get a consistent hash (same value on both flow directions) */
2359 skb->rxhash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2364 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2365 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2366 if (flow_table && sock_flow_table) {
2368 struct rps_dev_flow *rflow;
2370 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2373 next_cpu = sock_flow_table->ents[skb->rxhash &
2374 sock_flow_table->mask];
2377 * If the desired CPU (where last recvmsg was done) is
2378 * different from current CPU (one in the rx-queue flow
2379 * table entry), switch if one of the following holds:
2380 * - Current CPU is unset (equal to RPS_NO_CPU).
2381 * - Current CPU is offline.
2382 * - The current CPU's queue tail has advanced beyond the
2383 * last packet that was enqueued using this table entry.
2384 * This guarantees that all previous packets for the flow
2385 * have been dequeued, thus preserving in order delivery.
2387 if (unlikely(tcpu != next_cpu) &&
2388 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2389 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2390 rflow->last_qtail)) >= 0)) {
2391 tcpu = rflow->cpu = next_cpu;
2392 if (tcpu != RPS_NO_CPU)
2393 rflow->last_qtail = per_cpu(softnet_data,
2394 tcpu).input_queue_head;
2396 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2403 map = rcu_dereference(rxqueue->rps_map);
2405 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2407 if (cpu_online(tcpu)) {
2417 /* Called from hardirq (IPI) context */
2418 static void rps_trigger_softirq(void *data)
2420 struct softnet_data *sd = data;
2422 ____napi_schedule(sd, &sd->backlog);
2426 #endif /* CONFIG_RPS */
2429 * Check if this softnet_data structure is another cpu one
2430 * If yes, queue it to our IPI list and return 1
2433 static int rps_ipi_queued(struct softnet_data *sd)
2436 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2439 sd->rps_ipi_next = mysd->rps_ipi_list;
2440 mysd->rps_ipi_list = sd;
2442 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2445 #endif /* CONFIG_RPS */
2450 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2451 * queue (may be a remote CPU queue).
2453 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2454 unsigned int *qtail)
2456 struct softnet_data *sd;
2457 unsigned long flags;
2459 sd = &per_cpu(softnet_data, cpu);
2461 local_irq_save(flags);
2464 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2465 if (skb_queue_len(&sd->input_pkt_queue)) {
2467 __skb_queue_tail(&sd->input_pkt_queue, skb);
2468 input_queue_tail_incr_save(sd, qtail);
2470 local_irq_restore(flags);
2471 return NET_RX_SUCCESS;
2474 /* Schedule NAPI for backlog device
2475 * We can use non atomic operation since we own the queue lock
2477 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2478 if (!rps_ipi_queued(sd))
2479 ____napi_schedule(sd, &sd->backlog);
2487 local_irq_restore(flags);
2494 * netif_rx - post buffer to the network code
2495 * @skb: buffer to post
2497 * This function receives a packet from a device driver and queues it for
2498 * the upper (protocol) levels to process. It always succeeds. The buffer
2499 * may be dropped during processing for congestion control or by the
2503 * NET_RX_SUCCESS (no congestion)
2504 * NET_RX_DROP (packet was dropped)
2508 int netif_rx(struct sk_buff *skb)
2512 /* if netpoll wants it, pretend we never saw it */
2513 if (netpoll_rx(skb))
2516 if (netdev_tstamp_prequeue)
2517 net_timestamp_check(skb);
2521 struct rps_dev_flow voidflow, *rflow = &voidflow;
2526 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2528 cpu = smp_processor_id();
2530 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2537 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2543 EXPORT_SYMBOL(netif_rx);
2545 int netif_rx_ni(struct sk_buff *skb)
2550 err = netif_rx(skb);
2551 if (local_softirq_pending())
2557 EXPORT_SYMBOL(netif_rx_ni);
2559 static void net_tx_action(struct softirq_action *h)
2561 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2563 if (sd->completion_queue) {
2564 struct sk_buff *clist;
2566 local_irq_disable();
2567 clist = sd->completion_queue;
2568 sd->completion_queue = NULL;
2572 struct sk_buff *skb = clist;
2573 clist = clist->next;
2575 WARN_ON(atomic_read(&skb->users));
2580 if (sd->output_queue) {
2583 local_irq_disable();
2584 head = sd->output_queue;
2585 sd->output_queue = NULL;
2586 sd->output_queue_tailp = &sd->output_queue;
2590 struct Qdisc *q = head;
2591 spinlock_t *root_lock;
2593 head = head->next_sched;
2595 root_lock = qdisc_lock(q);
2596 if (spin_trylock(root_lock)) {
2597 smp_mb__before_clear_bit();
2598 clear_bit(__QDISC_STATE_SCHED,
2601 spin_unlock(root_lock);
2603 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2605 __netif_reschedule(q);
2607 smp_mb__before_clear_bit();
2608 clear_bit(__QDISC_STATE_SCHED,
2616 static inline int deliver_skb(struct sk_buff *skb,
2617 struct packet_type *pt_prev,
2618 struct net_device *orig_dev)
2620 atomic_inc(&skb->users);
2621 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2624 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2625 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2626 /* This hook is defined here for ATM LANE */
2627 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2628 unsigned char *addr) __read_mostly;
2629 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2632 #ifdef CONFIG_NET_CLS_ACT
2633 /* TODO: Maybe we should just force sch_ingress to be compiled in
2634 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2635 * a compare and 2 stores extra right now if we dont have it on
2636 * but have CONFIG_NET_CLS_ACT
2637 * NOTE: This doesnt stop any functionality; if you dont have
2638 * the ingress scheduler, you just cant add policies on ingress.
2641 static int ing_filter(struct sk_buff *skb)
2643 struct net_device *dev = skb->dev;
2644 u32 ttl = G_TC_RTTL(skb->tc_verd);
2645 struct netdev_queue *rxq;
2646 int result = TC_ACT_OK;
2649 if (unlikely(MAX_RED_LOOP < ttl++)) {
2650 if (net_ratelimit())
2651 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2652 skb->skb_iif, dev->ifindex);
2656 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2657 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2659 rxq = &dev->rx_queue;
2662 if (q != &noop_qdisc) {
2663 spin_lock(qdisc_lock(q));
2664 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2665 result = qdisc_enqueue_root(skb, q);
2666 spin_unlock(qdisc_lock(q));
2672 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2673 struct packet_type **pt_prev,
2674 int *ret, struct net_device *orig_dev)
2676 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2680 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2684 switch (ing_filter(skb)) {
2698 * netif_nit_deliver - deliver received packets to network taps
2701 * This function is used to deliver incoming packets to network
2702 * taps. It should be used when the normal netif_receive_skb path
2703 * is bypassed, for example because of VLAN acceleration.
2705 void netif_nit_deliver(struct sk_buff *skb)
2707 struct packet_type *ptype;
2709 if (list_empty(&ptype_all))
2712 skb_reset_network_header(skb);
2713 skb_reset_transport_header(skb);
2714 skb->mac_len = skb->network_header - skb->mac_header;
2717 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2718 if (!ptype->dev || ptype->dev == skb->dev)
2719 deliver_skb(skb, ptype, skb->dev);
2725 * netdev_rx_handler_register - register receive handler
2726 * @dev: device to register a handler for
2727 * @rx_handler: receive handler to register
2728 * @rx_handler_data: data pointer that is used by rx handler
2730 * Register a receive hander for a device. This handler will then be
2731 * called from __netif_receive_skb. A negative errno code is returned
2734 * The caller must hold the rtnl_mutex.
2736 int netdev_rx_handler_register(struct net_device *dev,
2737 rx_handler_func_t *rx_handler,
2738 void *rx_handler_data)
2742 if (dev->rx_handler)
2745 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2746 rcu_assign_pointer(dev->rx_handler, rx_handler);
2750 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2753 * netdev_rx_handler_unregister - unregister receive handler
2754 * @dev: device to unregister a handler from
2756 * Unregister a receive hander from a device.
2758 * The caller must hold the rtnl_mutex.
2760 void netdev_rx_handler_unregister(struct net_device *dev)
2764 rcu_assign_pointer(dev->rx_handler, NULL);
2765 rcu_assign_pointer(dev->rx_handler_data, NULL);
2767 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2769 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2770 struct net_device *master)
2772 if (skb->pkt_type == PACKET_HOST) {
2773 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2775 memcpy(dest, master->dev_addr, ETH_ALEN);
2779 /* On bonding slaves other than the currently active slave, suppress
2780 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2781 * ARP on active-backup slaves with arp_validate enabled.
2783 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2785 struct net_device *dev = skb->dev;
2787 if (master->priv_flags & IFF_MASTER_ARPMON)
2788 dev->last_rx = jiffies;
2790 if ((master->priv_flags & IFF_MASTER_ALB) &&
2791 (master->priv_flags & IFF_BRIDGE_PORT)) {
2792 /* Do address unmangle. The local destination address
2793 * will be always the one master has. Provides the right
2794 * functionality in a bridge.
2796 skb_bond_set_mac_by_master(skb, master);
2799 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2800 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2801 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2804 if (master->priv_flags & IFF_MASTER_ALB) {
2805 if (skb->pkt_type != PACKET_BROADCAST &&
2806 skb->pkt_type != PACKET_MULTICAST)
2809 if (master->priv_flags & IFF_MASTER_8023AD &&
2810 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2817 EXPORT_SYMBOL(__skb_bond_should_drop);
2819 static int __netif_receive_skb(struct sk_buff *skb)
2821 struct packet_type *ptype, *pt_prev;
2822 rx_handler_func_t *rx_handler;
2823 struct net_device *orig_dev;
2824 struct net_device *master;
2825 struct net_device *null_or_orig;
2826 struct net_device *orig_or_bond;
2827 int ret = NET_RX_DROP;
2830 if (!netdev_tstamp_prequeue)
2831 net_timestamp_check(skb);
2833 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2834 return NET_RX_SUCCESS;
2836 /* if we've gotten here through NAPI, check netpoll */
2837 if (netpoll_receive_skb(skb))
2841 skb->skb_iif = skb->dev->ifindex;
2844 * bonding note: skbs received on inactive slaves should only
2845 * be delivered to pkt handlers that are exact matches. Also
2846 * the deliver_no_wcard flag will be set. If packet handlers
2847 * are sensitive to duplicate packets these skbs will need to
2848 * be dropped at the handler. The vlan accel path may have
2849 * already set the deliver_no_wcard flag.
2851 null_or_orig = NULL;
2852 orig_dev = skb->dev;
2853 master = ACCESS_ONCE(orig_dev->master);
2854 if (skb->deliver_no_wcard)
2855 null_or_orig = orig_dev;
2857 if (skb_bond_should_drop(skb, master)) {
2858 skb->deliver_no_wcard = 1;
2859 null_or_orig = orig_dev; /* deliver only exact match */
2864 __this_cpu_inc(softnet_data.processed);
2865 skb_reset_network_header(skb);
2866 skb_reset_transport_header(skb);
2867 skb->mac_len = skb->network_header - skb->mac_header;
2873 #ifdef CONFIG_NET_CLS_ACT
2874 if (skb->tc_verd & TC_NCLS) {
2875 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2880 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2881 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2882 ptype->dev == orig_dev) {
2884 ret = deliver_skb(skb, pt_prev, orig_dev);
2889 #ifdef CONFIG_NET_CLS_ACT
2890 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2896 /* Handle special case of bridge or macvlan */
2897 rx_handler = rcu_dereference(skb->dev->rx_handler);
2900 ret = deliver_skb(skb, pt_prev, orig_dev);
2903 skb = rx_handler(skb);
2909 * Make sure frames received on VLAN interfaces stacked on
2910 * bonding interfaces still make their way to any base bonding
2911 * device that may have registered for a specific ptype. The
2912 * handler may have to adjust skb->dev and orig_dev.
2914 orig_or_bond = orig_dev;
2915 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2916 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2917 orig_or_bond = vlan_dev_real_dev(skb->dev);
2920 type = skb->protocol;
2921 list_for_each_entry_rcu(ptype,
2922 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2923 if (ptype->type == type && (ptype->dev == null_or_orig ||
2924 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2925 ptype->dev == orig_or_bond)) {
2927 ret = deliver_skb(skb, pt_prev, orig_dev);
2933 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2936 /* Jamal, now you will not able to escape explaining
2937 * me how you were going to use this. :-)
2948 * netif_receive_skb - process receive buffer from network
2949 * @skb: buffer to process
2951 * netif_receive_skb() is the main receive data processing function.
2952 * It always succeeds. The buffer may be dropped during processing
2953 * for congestion control or by the protocol layers.
2955 * This function may only be called from softirq context and interrupts
2956 * should be enabled.
2958 * Return values (usually ignored):
2959 * NET_RX_SUCCESS: no congestion
2960 * NET_RX_DROP: packet was dropped
2962 int netif_receive_skb(struct sk_buff *skb)
2964 if (netdev_tstamp_prequeue)
2965 net_timestamp_check(skb);
2967 if (skb_defer_rx_timestamp(skb))
2968 return NET_RX_SUCCESS;
2972 struct rps_dev_flow voidflow, *rflow = &voidflow;
2977 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2980 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2984 ret = __netif_receive_skb(skb);
2990 return __netif_receive_skb(skb);
2993 EXPORT_SYMBOL(netif_receive_skb);
2995 /* Network device is going away, flush any packets still pending
2996 * Called with irqs disabled.
2998 static void flush_backlog(void *arg)
3000 struct net_device *dev = arg;
3001 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3002 struct sk_buff *skb, *tmp;
3005 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3006 if (skb->dev == dev) {
3007 __skb_unlink(skb, &sd->input_pkt_queue);
3009 input_queue_head_incr(sd);
3014 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3015 if (skb->dev == dev) {
3016 __skb_unlink(skb, &sd->process_queue);
3018 input_queue_head_incr(sd);
3023 static int napi_gro_complete(struct sk_buff *skb)
3025 struct packet_type *ptype;
3026 __be16 type = skb->protocol;
3027 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3030 if (NAPI_GRO_CB(skb)->count == 1) {
3031 skb_shinfo(skb)->gso_size = 0;
3036 list_for_each_entry_rcu(ptype, head, list) {
3037 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3040 err = ptype->gro_complete(skb);
3046 WARN_ON(&ptype->list == head);
3048 return NET_RX_SUCCESS;
3052 return netif_receive_skb(skb);
3055 static void napi_gro_flush(struct napi_struct *napi)
3057 struct sk_buff *skb, *next;
3059 for (skb = napi->gro_list; skb; skb = next) {
3062 napi_gro_complete(skb);
3065 napi->gro_count = 0;
3066 napi->gro_list = NULL;
3069 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3071 struct sk_buff **pp = NULL;
3072 struct packet_type *ptype;
3073 __be16 type = skb->protocol;
3074 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3077 enum gro_result ret;
3079 if (!(skb->dev->features & NETIF_F_GRO))
3082 if (skb_is_gso(skb) || skb_has_frags(skb))
3086 list_for_each_entry_rcu(ptype, head, list) {
3087 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3090 skb_set_network_header(skb, skb_gro_offset(skb));
3091 mac_len = skb->network_header - skb->mac_header;
3092 skb->mac_len = mac_len;
3093 NAPI_GRO_CB(skb)->same_flow = 0;
3094 NAPI_GRO_CB(skb)->flush = 0;
3095 NAPI_GRO_CB(skb)->free = 0;
3097 pp = ptype->gro_receive(&napi->gro_list, skb);
3102 if (&ptype->list == head)
3105 same_flow = NAPI_GRO_CB(skb)->same_flow;
3106 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3109 struct sk_buff *nskb = *pp;
3113 napi_gro_complete(nskb);
3120 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3124 NAPI_GRO_CB(skb)->count = 1;
3125 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3126 skb->next = napi->gro_list;
3127 napi->gro_list = skb;
3131 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3132 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3134 BUG_ON(skb->end - skb->tail < grow);
3136 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3139 skb->data_len -= grow;
3141 skb_shinfo(skb)->frags[0].page_offset += grow;
3142 skb_shinfo(skb)->frags[0].size -= grow;
3144 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3145 put_page(skb_shinfo(skb)->frags[0].page);
3146 memmove(skb_shinfo(skb)->frags,
3147 skb_shinfo(skb)->frags + 1,
3148 --skb_shinfo(skb)->nr_frags);
3159 EXPORT_SYMBOL(dev_gro_receive);
3162 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3166 if (netpoll_rx_on(skb))
3169 for (p = napi->gro_list; p; p = p->next) {
3170 NAPI_GRO_CB(p)->same_flow =
3171 (p->dev == skb->dev) &&
3172 !compare_ether_header(skb_mac_header(p),
3173 skb_gro_mac_header(skb));
3174 NAPI_GRO_CB(p)->flush = 0;
3177 return dev_gro_receive(napi, skb);
3180 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3184 if (netif_receive_skb(skb))
3189 case GRO_MERGED_FREE:
3200 EXPORT_SYMBOL(napi_skb_finish);
3202 void skb_gro_reset_offset(struct sk_buff *skb)
3204 NAPI_GRO_CB(skb)->data_offset = 0;
3205 NAPI_GRO_CB(skb)->frag0 = NULL;
3206 NAPI_GRO_CB(skb)->frag0_len = 0;
3208 if (skb->mac_header == skb->tail &&
3209 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3210 NAPI_GRO_CB(skb)->frag0 =
3211 page_address(skb_shinfo(skb)->frags[0].page) +
3212 skb_shinfo(skb)->frags[0].page_offset;
3213 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3216 EXPORT_SYMBOL(skb_gro_reset_offset);
3218 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3220 skb_gro_reset_offset(skb);
3222 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3224 EXPORT_SYMBOL(napi_gro_receive);
3226 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3228 __skb_pull(skb, skb_headlen(skb));
3229 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3233 EXPORT_SYMBOL(napi_reuse_skb);
3235 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3237 struct sk_buff *skb = napi->skb;
3240 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3246 EXPORT_SYMBOL(napi_get_frags);
3248 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3254 skb->protocol = eth_type_trans(skb, skb->dev);
3256 if (ret == GRO_HELD)
3257 skb_gro_pull(skb, -ETH_HLEN);
3258 else if (netif_receive_skb(skb))
3263 case GRO_MERGED_FREE:
3264 napi_reuse_skb(napi, skb);
3273 EXPORT_SYMBOL(napi_frags_finish);
3275 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3277 struct sk_buff *skb = napi->skb;
3284 skb_reset_mac_header(skb);
3285 skb_gro_reset_offset(skb);
3287 off = skb_gro_offset(skb);
3288 hlen = off + sizeof(*eth);
3289 eth = skb_gro_header_fast(skb, off);
3290 if (skb_gro_header_hard(skb, hlen)) {
3291 eth = skb_gro_header_slow(skb, hlen, off);
3292 if (unlikely(!eth)) {
3293 napi_reuse_skb(napi, skb);
3299 skb_gro_pull(skb, sizeof(*eth));
3302 * This works because the only protocols we care about don't require
3303 * special handling. We'll fix it up properly at the end.
3305 skb->protocol = eth->h_proto;
3310 EXPORT_SYMBOL(napi_frags_skb);
3312 gro_result_t napi_gro_frags(struct napi_struct *napi)
3314 struct sk_buff *skb = napi_frags_skb(napi);
3319 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3321 EXPORT_SYMBOL(napi_gro_frags);
3324 * net_rps_action sends any pending IPI's for rps.
3325 * Note: called with local irq disabled, but exits with local irq enabled.
3327 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3330 struct softnet_data *remsd = sd->rps_ipi_list;
3333 sd->rps_ipi_list = NULL;
3337 /* Send pending IPI's to kick RPS processing on remote cpus. */
3339 struct softnet_data *next = remsd->rps_ipi_next;
3341 if (cpu_online(remsd->cpu))
3342 __smp_call_function_single(remsd->cpu,
3351 static int process_backlog(struct napi_struct *napi, int quota)
3354 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3357 /* Check if we have pending ipi, its better to send them now,
3358 * not waiting net_rx_action() end.
3360 if (sd->rps_ipi_list) {
3361 local_irq_disable();
3362 net_rps_action_and_irq_enable(sd);
3365 napi->weight = weight_p;
3366 local_irq_disable();
3367 while (work < quota) {
3368 struct sk_buff *skb;
3371 while ((skb = __skb_dequeue(&sd->process_queue))) {
3373 __netif_receive_skb(skb);
3374 local_irq_disable();
3375 input_queue_head_incr(sd);
3376 if (++work >= quota) {
3383 qlen = skb_queue_len(&sd->input_pkt_queue);
3385 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3386 &sd->process_queue);
3388 if (qlen < quota - work) {
3390 * Inline a custom version of __napi_complete().
3391 * only current cpu owns and manipulates this napi,
3392 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3393 * we can use a plain write instead of clear_bit(),
3394 * and we dont need an smp_mb() memory barrier.
3396 list_del(&napi->poll_list);
3399 quota = work + qlen;
3409 * __napi_schedule - schedule for receive
3410 * @n: entry to schedule
3412 * The entry's receive function will be scheduled to run
3414 void __napi_schedule(struct napi_struct *n)
3416 unsigned long flags;
3418 local_irq_save(flags);
3419 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3420 local_irq_restore(flags);
3422 EXPORT_SYMBOL(__napi_schedule);
3424 void __napi_complete(struct napi_struct *n)
3426 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3427 BUG_ON(n->gro_list);
3429 list_del(&n->poll_list);
3430 smp_mb__before_clear_bit();
3431 clear_bit(NAPI_STATE_SCHED, &n->state);
3433 EXPORT_SYMBOL(__napi_complete);
3435 void napi_complete(struct napi_struct *n)
3437 unsigned long flags;
3440 * don't let napi dequeue from the cpu poll list
3441 * just in case its running on a different cpu
3443 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3447 local_irq_save(flags);
3449 local_irq_restore(flags);
3451 EXPORT_SYMBOL(napi_complete);
3453 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3454 int (*poll)(struct napi_struct *, int), int weight)
3456 INIT_LIST_HEAD(&napi->poll_list);
3457 napi->gro_count = 0;
3458 napi->gro_list = NULL;
3461 napi->weight = weight;
3462 list_add(&napi->dev_list, &dev->napi_list);
3464 #ifdef CONFIG_NETPOLL
3465 spin_lock_init(&napi->poll_lock);
3466 napi->poll_owner = -1;
3468 set_bit(NAPI_STATE_SCHED, &napi->state);
3470 EXPORT_SYMBOL(netif_napi_add);
3472 void netif_napi_del(struct napi_struct *napi)
3474 struct sk_buff *skb, *next;
3476 list_del_init(&napi->dev_list);
3477 napi_free_frags(napi);
3479 for (skb = napi->gro_list; skb; skb = next) {
3485 napi->gro_list = NULL;
3486 napi->gro_count = 0;
3488 EXPORT_SYMBOL(netif_napi_del);
3490 static void net_rx_action(struct softirq_action *h)
3492 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3493 unsigned long time_limit = jiffies + 2;
3494 int budget = netdev_budget;
3497 local_irq_disable();
3499 while (!list_empty(&sd->poll_list)) {
3500 struct napi_struct *n;
3503 /* If softirq window is exhuasted then punt.
3504 * Allow this to run for 2 jiffies since which will allow
3505 * an average latency of 1.5/HZ.
3507 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3512 /* Even though interrupts have been re-enabled, this
3513 * access is safe because interrupts can only add new
3514 * entries to the tail of this list, and only ->poll()
3515 * calls can remove this head entry from the list.
3517 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3519 have = netpoll_poll_lock(n);
3523 /* This NAPI_STATE_SCHED test is for avoiding a race
3524 * with netpoll's poll_napi(). Only the entity which
3525 * obtains the lock and sees NAPI_STATE_SCHED set will
3526 * actually make the ->poll() call. Therefore we avoid
3527 * accidently calling ->poll() when NAPI is not scheduled.
3530 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3531 work = n->poll(n, weight);
3535 WARN_ON_ONCE(work > weight);
3539 local_irq_disable();
3541 /* Drivers must not modify the NAPI state if they
3542 * consume the entire weight. In such cases this code
3543 * still "owns" the NAPI instance and therefore can
3544 * move the instance around on the list at-will.
3546 if (unlikely(work == weight)) {
3547 if (unlikely(napi_disable_pending(n))) {
3550 local_irq_disable();
3552 list_move_tail(&n->poll_list, &sd->poll_list);
3555 netpoll_poll_unlock(have);
3558 net_rps_action_and_irq_enable(sd);
3560 #ifdef CONFIG_NET_DMA
3562 * There may not be any more sk_buffs coming right now, so push
3563 * any pending DMA copies to hardware
3565 dma_issue_pending_all();
3572 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3576 static gifconf_func_t *gifconf_list[NPROTO];
3579 * register_gifconf - register a SIOCGIF handler
3580 * @family: Address family
3581 * @gifconf: Function handler
3583 * Register protocol dependent address dumping routines. The handler
3584 * that is passed must not be freed or reused until it has been replaced
3585 * by another handler.
3587 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3589 if (family >= NPROTO)
3591 gifconf_list[family] = gifconf;
3594 EXPORT_SYMBOL(register_gifconf);
3598 * Map an interface index to its name (SIOCGIFNAME)
3602 * We need this ioctl for efficient implementation of the
3603 * if_indextoname() function required by the IPv6 API. Without
3604 * it, we would have to search all the interfaces to find a
3608 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3610 struct net_device *dev;
3614 * Fetch the caller's info block.
3617 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3621 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3627 strcpy(ifr.ifr_name, dev->name);
3630 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3636 * Perform a SIOCGIFCONF call. This structure will change
3637 * size eventually, and there is nothing I can do about it.
3638 * Thus we will need a 'compatibility mode'.
3641 static int dev_ifconf(struct net *net, char __user *arg)
3644 struct net_device *dev;
3651 * Fetch the caller's info block.
3654 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3661 * Loop over the interfaces, and write an info block for each.
3665 for_each_netdev(net, dev) {
3666 for (i = 0; i < NPROTO; i++) {
3667 if (gifconf_list[i]) {
3670 done = gifconf_list[i](dev, NULL, 0);
3672 done = gifconf_list[i](dev, pos + total,
3682 * All done. Write the updated control block back to the caller.
3684 ifc.ifc_len = total;
3687 * Both BSD and Solaris return 0 here, so we do too.
3689 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3692 #ifdef CONFIG_PROC_FS
3694 * This is invoked by the /proc filesystem handler to display a device
3697 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3700 struct net *net = seq_file_net(seq);
3702 struct net_device *dev;
3706 return SEQ_START_TOKEN;
3709 for_each_netdev_rcu(net, dev)
3716 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3718 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3719 first_net_device(seq_file_net(seq)) :
3720 next_net_device((struct net_device *)v);
3723 return rcu_dereference(dev);
3726 void dev_seq_stop(struct seq_file *seq, void *v)
3732 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3734 struct rtnl_link_stats64 temp;
3735 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3737 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3738 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3739 dev->name, stats->rx_bytes, stats->rx_packets,
3741 stats->rx_dropped + stats->rx_missed_errors,
3742 stats->rx_fifo_errors,
3743 stats->rx_length_errors + stats->rx_over_errors +
3744 stats->rx_crc_errors + stats->rx_frame_errors,
3745 stats->rx_compressed, stats->multicast,
3746 stats->tx_bytes, stats->tx_packets,
3747 stats->tx_errors, stats->tx_dropped,
3748 stats->tx_fifo_errors, stats->collisions,
3749 stats->tx_carrier_errors +
3750 stats->tx_aborted_errors +
3751 stats->tx_window_errors +
3752 stats->tx_heartbeat_errors,
3753 stats->tx_compressed);
3757 * Called from the PROCfs module. This now uses the new arbitrary sized
3758 * /proc/net interface to create /proc/net/dev
3760 static int dev_seq_show(struct seq_file *seq, void *v)
3762 if (v == SEQ_START_TOKEN)
3763 seq_puts(seq, "Inter-| Receive "
3765 " face |bytes packets errs drop fifo frame "
3766 "compressed multicast|bytes packets errs "
3767 "drop fifo colls carrier compressed\n");
3769 dev_seq_printf_stats(seq, v);
3773 static struct softnet_data *softnet_get_online(loff_t *pos)
3775 struct softnet_data *sd = NULL;
3777 while (*pos < nr_cpu_ids)
3778 if (cpu_online(*pos)) {
3779 sd = &per_cpu(softnet_data, *pos);
3786 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3788 return softnet_get_online(pos);
3791 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3794 return softnet_get_online(pos);
3797 static void softnet_seq_stop(struct seq_file *seq, void *v)
3801 static int softnet_seq_show(struct seq_file *seq, void *v)
3803 struct softnet_data *sd = v;
3805 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3806 sd->processed, sd->dropped, sd->time_squeeze, 0,
3807 0, 0, 0, 0, /* was fastroute */
3808 sd->cpu_collision, sd->received_rps);
3812 static const struct seq_operations dev_seq_ops = {
3813 .start = dev_seq_start,
3814 .next = dev_seq_next,
3815 .stop = dev_seq_stop,
3816 .show = dev_seq_show,
3819 static int dev_seq_open(struct inode *inode, struct file *file)
3821 return seq_open_net(inode, file, &dev_seq_ops,
3822 sizeof(struct seq_net_private));
3825 static const struct file_operations dev_seq_fops = {
3826 .owner = THIS_MODULE,
3827 .open = dev_seq_open,
3829 .llseek = seq_lseek,
3830 .release = seq_release_net,
3833 static const struct seq_operations softnet_seq_ops = {
3834 .start = softnet_seq_start,
3835 .next = softnet_seq_next,
3836 .stop = softnet_seq_stop,
3837 .show = softnet_seq_show,
3840 static int softnet_seq_open(struct inode *inode, struct file *file)
3842 return seq_open(file, &softnet_seq_ops);
3845 static const struct file_operations softnet_seq_fops = {
3846 .owner = THIS_MODULE,
3847 .open = softnet_seq_open,
3849 .llseek = seq_lseek,
3850 .release = seq_release,
3853 static void *ptype_get_idx(loff_t pos)
3855 struct packet_type *pt = NULL;
3859 list_for_each_entry_rcu(pt, &ptype_all, list) {
3865 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3866 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3875 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3879 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3882 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3884 struct packet_type *pt;
3885 struct list_head *nxt;
3889 if (v == SEQ_START_TOKEN)
3890 return ptype_get_idx(0);
3893 nxt = pt->list.next;
3894 if (pt->type == htons(ETH_P_ALL)) {
3895 if (nxt != &ptype_all)
3898 nxt = ptype_base[0].next;
3900 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3902 while (nxt == &ptype_base[hash]) {
3903 if (++hash >= PTYPE_HASH_SIZE)
3905 nxt = ptype_base[hash].next;
3908 return list_entry(nxt, struct packet_type, list);
3911 static void ptype_seq_stop(struct seq_file *seq, void *v)
3917 static int ptype_seq_show(struct seq_file *seq, void *v)
3919 struct packet_type *pt = v;
3921 if (v == SEQ_START_TOKEN)
3922 seq_puts(seq, "Type Device Function\n");
3923 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3924 if (pt->type == htons(ETH_P_ALL))
3925 seq_puts(seq, "ALL ");
3927 seq_printf(seq, "%04x", ntohs(pt->type));
3929 seq_printf(seq, " %-8s %pF\n",
3930 pt->dev ? pt->dev->name : "", pt->func);
3936 static const struct seq_operations ptype_seq_ops = {
3937 .start = ptype_seq_start,
3938 .next = ptype_seq_next,
3939 .stop = ptype_seq_stop,
3940 .show = ptype_seq_show,
3943 static int ptype_seq_open(struct inode *inode, struct file *file)
3945 return seq_open_net(inode, file, &ptype_seq_ops,
3946 sizeof(struct seq_net_private));
3949 static const struct file_operations ptype_seq_fops = {
3950 .owner = THIS_MODULE,
3951 .open = ptype_seq_open,
3953 .llseek = seq_lseek,
3954 .release = seq_release_net,
3958 static int __net_init dev_proc_net_init(struct net *net)
3962 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3964 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3966 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3969 if (wext_proc_init(net))
3975 proc_net_remove(net, "ptype");
3977 proc_net_remove(net, "softnet_stat");
3979 proc_net_remove(net, "dev");
3983 static void __net_exit dev_proc_net_exit(struct net *net)
3985 wext_proc_exit(net);
3987 proc_net_remove(net, "ptype");
3988 proc_net_remove(net, "softnet_stat");
3989 proc_net_remove(net, "dev");
3992 static struct pernet_operations __net_initdata dev_proc_ops = {
3993 .init = dev_proc_net_init,
3994 .exit = dev_proc_net_exit,
3997 static int __init dev_proc_init(void)
3999 return register_pernet_subsys(&dev_proc_ops);
4002 #define dev_proc_init() 0
4003 #endif /* CONFIG_PROC_FS */
4007 * netdev_set_master - set up master/slave pair
4008 * @slave: slave device
4009 * @master: new master device
4011 * Changes the master device of the slave. Pass %NULL to break the
4012 * bonding. The caller must hold the RTNL semaphore. On a failure
4013 * a negative errno code is returned. On success the reference counts
4014 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4015 * function returns zero.
4017 int netdev_set_master(struct net_device *slave, struct net_device *master)
4019 struct net_device *old = slave->master;
4029 slave->master = master;
4036 slave->flags |= IFF_SLAVE;
4038 slave->flags &= ~IFF_SLAVE;
4040 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4043 EXPORT_SYMBOL(netdev_set_master);
4045 static void dev_change_rx_flags(struct net_device *dev, int flags)
4047 const struct net_device_ops *ops = dev->netdev_ops;
4049 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4050 ops->ndo_change_rx_flags(dev, flags);
4053 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4055 unsigned short old_flags = dev->flags;
4061 dev->flags |= IFF_PROMISC;
4062 dev->promiscuity += inc;
4063 if (dev->promiscuity == 0) {
4066 * If inc causes overflow, untouch promisc and return error.
4069 dev->flags &= ~IFF_PROMISC;
4071 dev->promiscuity -= inc;
4072 printk(KERN_WARNING "%s: promiscuity touches roof, "
4073 "set promiscuity failed, promiscuity feature "
4074 "of device might be broken.\n", dev->name);
4078 if (dev->flags != old_flags) {
4079 printk(KERN_INFO "device %s %s promiscuous mode\n",
4080 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4082 if (audit_enabled) {
4083 current_uid_gid(&uid, &gid);
4084 audit_log(current->audit_context, GFP_ATOMIC,
4085 AUDIT_ANOM_PROMISCUOUS,
4086 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4087 dev->name, (dev->flags & IFF_PROMISC),
4088 (old_flags & IFF_PROMISC),
4089 audit_get_loginuid(current),
4091 audit_get_sessionid(current));
4094 dev_change_rx_flags(dev, IFF_PROMISC);
4100 * dev_set_promiscuity - update promiscuity count on a device
4104 * Add or remove promiscuity from a device. While the count in the device
4105 * remains above zero the interface remains promiscuous. Once it hits zero
4106 * the device reverts back to normal filtering operation. A negative inc
4107 * value is used to drop promiscuity on the device.
4108 * Return 0 if successful or a negative errno code on error.
4110 int dev_set_promiscuity(struct net_device *dev, int inc)
4112 unsigned short old_flags = dev->flags;
4115 err = __dev_set_promiscuity(dev, inc);
4118 if (dev->flags != old_flags)
4119 dev_set_rx_mode(dev);
4122 EXPORT_SYMBOL(dev_set_promiscuity);
4125 * dev_set_allmulti - update allmulti count on a device
4129 * Add or remove reception of all multicast frames to a device. While the
4130 * count in the device remains above zero the interface remains listening
4131 * to all interfaces. Once it hits zero the device reverts back to normal
4132 * filtering operation. A negative @inc value is used to drop the counter
4133 * when releasing a resource needing all multicasts.
4134 * Return 0 if successful or a negative errno code on error.
4137 int dev_set_allmulti(struct net_device *dev, int inc)
4139 unsigned short old_flags = dev->flags;
4143 dev->flags |= IFF_ALLMULTI;
4144 dev->allmulti += inc;
4145 if (dev->allmulti == 0) {
4148 * If inc causes overflow, untouch allmulti and return error.
4151 dev->flags &= ~IFF_ALLMULTI;
4153 dev->allmulti -= inc;
4154 printk(KERN_WARNING "%s: allmulti touches roof, "
4155 "set allmulti failed, allmulti feature of "
4156 "device might be broken.\n", dev->name);
4160 if (dev->flags ^ old_flags) {
4161 dev_change_rx_flags(dev, IFF_ALLMULTI);
4162 dev_set_rx_mode(dev);
4166 EXPORT_SYMBOL(dev_set_allmulti);
4169 * Upload unicast and multicast address lists to device and
4170 * configure RX filtering. When the device doesn't support unicast
4171 * filtering it is put in promiscuous mode while unicast addresses
4174 void __dev_set_rx_mode(struct net_device *dev)
4176 const struct net_device_ops *ops = dev->netdev_ops;
4178 /* dev_open will call this function so the list will stay sane. */
4179 if (!(dev->flags&IFF_UP))
4182 if (!netif_device_present(dev))
4185 if (ops->ndo_set_rx_mode)
4186 ops->ndo_set_rx_mode(dev);
4188 /* Unicast addresses changes may only happen under the rtnl,
4189 * therefore calling __dev_set_promiscuity here is safe.
4191 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4192 __dev_set_promiscuity(dev, 1);
4193 dev->uc_promisc = 1;
4194 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4195 __dev_set_promiscuity(dev, -1);
4196 dev->uc_promisc = 0;
4199 if (ops->ndo_set_multicast_list)
4200 ops->ndo_set_multicast_list(dev);
4204 void dev_set_rx_mode(struct net_device *dev)
4206 netif_addr_lock_bh(dev);
4207 __dev_set_rx_mode(dev);
4208 netif_addr_unlock_bh(dev);
4212 * dev_get_flags - get flags reported to userspace
4215 * Get the combination of flag bits exported through APIs to userspace.
4217 unsigned dev_get_flags(const struct net_device *dev)
4221 flags = (dev->flags & ~(IFF_PROMISC |
4226 (dev->gflags & (IFF_PROMISC |
4229 if (netif_running(dev)) {
4230 if (netif_oper_up(dev))
4231 flags |= IFF_RUNNING;
4232 if (netif_carrier_ok(dev))
4233 flags |= IFF_LOWER_UP;
4234 if (netif_dormant(dev))
4235 flags |= IFF_DORMANT;
4240 EXPORT_SYMBOL(dev_get_flags);
4242 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4244 int old_flags = dev->flags;
4250 * Set the flags on our device.
4253 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4254 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4256 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4260 * Load in the correct multicast list now the flags have changed.
4263 if ((old_flags ^ flags) & IFF_MULTICAST)
4264 dev_change_rx_flags(dev, IFF_MULTICAST);
4266 dev_set_rx_mode(dev);
4269 * Have we downed the interface. We handle IFF_UP ourselves
4270 * according to user attempts to set it, rather than blindly
4275 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4276 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4279 dev_set_rx_mode(dev);
4282 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4283 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4285 dev->gflags ^= IFF_PROMISC;
4286 dev_set_promiscuity(dev, inc);
4289 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4290 is important. Some (broken) drivers set IFF_PROMISC, when
4291 IFF_ALLMULTI is requested not asking us and not reporting.
4293 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4294 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4296 dev->gflags ^= IFF_ALLMULTI;
4297 dev_set_allmulti(dev, inc);
4303 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4305 unsigned int changes = dev->flags ^ old_flags;
4307 if (changes & IFF_UP) {
4308 if (dev->flags & IFF_UP)
4309 call_netdevice_notifiers(NETDEV_UP, dev);
4311 call_netdevice_notifiers(NETDEV_DOWN, dev);
4314 if (dev->flags & IFF_UP &&
4315 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4316 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4320 * dev_change_flags - change device settings
4322 * @flags: device state flags
4324 * Change settings on device based state flags. The flags are
4325 * in the userspace exported format.
4327 int dev_change_flags(struct net_device *dev, unsigned flags)
4330 int old_flags = dev->flags;
4332 ret = __dev_change_flags(dev, flags);
4336 changes = old_flags ^ dev->flags;
4338 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4340 __dev_notify_flags(dev, old_flags);
4343 EXPORT_SYMBOL(dev_change_flags);
4346 * dev_set_mtu - Change maximum transfer unit
4348 * @new_mtu: new transfer unit
4350 * Change the maximum transfer size of the network device.
4352 int dev_set_mtu(struct net_device *dev, int new_mtu)
4354 const struct net_device_ops *ops = dev->netdev_ops;
4357 if (new_mtu == dev->mtu)
4360 /* MTU must be positive. */
4364 if (!netif_device_present(dev))
4368 if (ops->ndo_change_mtu)
4369 err = ops->ndo_change_mtu(dev, new_mtu);
4373 if (!err && dev->flags & IFF_UP)
4374 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4377 EXPORT_SYMBOL(dev_set_mtu);
4380 * dev_set_mac_address - Change Media Access Control Address
4384 * Change the hardware (MAC) address of the device
4386 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4388 const struct net_device_ops *ops = dev->netdev_ops;
4391 if (!ops->ndo_set_mac_address)
4393 if (sa->sa_family != dev->type)
4395 if (!netif_device_present(dev))
4397 err = ops->ndo_set_mac_address(dev, sa);
4399 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4402 EXPORT_SYMBOL(dev_set_mac_address);
4405 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4407 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4410 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4416 case SIOCGIFFLAGS: /* Get interface flags */
4417 ifr->ifr_flags = (short) dev_get_flags(dev);
4420 case SIOCGIFMETRIC: /* Get the metric on the interface
4421 (currently unused) */
4422 ifr->ifr_metric = 0;
4425 case SIOCGIFMTU: /* Get the MTU of a device */
4426 ifr->ifr_mtu = dev->mtu;
4431 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4433 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4434 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4435 ifr->ifr_hwaddr.sa_family = dev->type;
4443 ifr->ifr_map.mem_start = dev->mem_start;
4444 ifr->ifr_map.mem_end = dev->mem_end;
4445 ifr->ifr_map.base_addr = dev->base_addr;
4446 ifr->ifr_map.irq = dev->irq;
4447 ifr->ifr_map.dma = dev->dma;
4448 ifr->ifr_map.port = dev->if_port;
4452 ifr->ifr_ifindex = dev->ifindex;
4456 ifr->ifr_qlen = dev->tx_queue_len;
4460 /* dev_ioctl() should ensure this case
4472 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4474 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4477 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4478 const struct net_device_ops *ops;
4483 ops = dev->netdev_ops;
4486 case SIOCSIFFLAGS: /* Set interface flags */
4487 return dev_change_flags(dev, ifr->ifr_flags);
4489 case SIOCSIFMETRIC: /* Set the metric on the interface
4490 (currently unused) */
4493 case SIOCSIFMTU: /* Set the MTU of a device */
4494 return dev_set_mtu(dev, ifr->ifr_mtu);
4497 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4499 case SIOCSIFHWBROADCAST:
4500 if (ifr->ifr_hwaddr.sa_family != dev->type)
4502 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4503 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4504 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4508 if (ops->ndo_set_config) {
4509 if (!netif_device_present(dev))
4511 return ops->ndo_set_config(dev, &ifr->ifr_map);
4516 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4517 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4519 if (!netif_device_present(dev))
4521 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4524 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4525 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4527 if (!netif_device_present(dev))
4529 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4532 if (ifr->ifr_qlen < 0)
4534 dev->tx_queue_len = ifr->ifr_qlen;
4538 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4539 return dev_change_name(dev, ifr->ifr_newname);
4542 * Unknown or private ioctl
4545 if ((cmd >= SIOCDEVPRIVATE &&
4546 cmd <= SIOCDEVPRIVATE + 15) ||
4547 cmd == SIOCBONDENSLAVE ||
4548 cmd == SIOCBONDRELEASE ||
4549 cmd == SIOCBONDSETHWADDR ||
4550 cmd == SIOCBONDSLAVEINFOQUERY ||
4551 cmd == SIOCBONDINFOQUERY ||
4552 cmd == SIOCBONDCHANGEACTIVE ||
4553 cmd == SIOCGMIIPHY ||
4554 cmd == SIOCGMIIREG ||
4555 cmd == SIOCSMIIREG ||
4556 cmd == SIOCBRADDIF ||
4557 cmd == SIOCBRDELIF ||
4558 cmd == SIOCSHWTSTAMP ||
4559 cmd == SIOCWANDEV) {
4561 if (ops->ndo_do_ioctl) {
4562 if (netif_device_present(dev))
4563 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4575 * This function handles all "interface"-type I/O control requests. The actual
4576 * 'doing' part of this is dev_ifsioc above.
4580 * dev_ioctl - network device ioctl
4581 * @net: the applicable net namespace
4582 * @cmd: command to issue
4583 * @arg: pointer to a struct ifreq in user space
4585 * Issue ioctl functions to devices. This is normally called by the
4586 * user space syscall interfaces but can sometimes be useful for
4587 * other purposes. The return value is the return from the syscall if
4588 * positive or a negative errno code on error.
4591 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4597 /* One special case: SIOCGIFCONF takes ifconf argument
4598 and requires shared lock, because it sleeps writing
4602 if (cmd == SIOCGIFCONF) {
4604 ret = dev_ifconf(net, (char __user *) arg);
4608 if (cmd == SIOCGIFNAME)
4609 return dev_ifname(net, (struct ifreq __user *)arg);
4611 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4614 ifr.ifr_name[IFNAMSIZ-1] = 0;
4616 colon = strchr(ifr.ifr_name, ':');
4621 * See which interface the caller is talking about.
4626 * These ioctl calls:
4627 * - can be done by all.
4628 * - atomic and do not require locking.
4639 dev_load(net, ifr.ifr_name);
4641 ret = dev_ifsioc_locked(net, &ifr, cmd);
4646 if (copy_to_user(arg, &ifr,
4647 sizeof(struct ifreq)))
4653 dev_load(net, ifr.ifr_name);
4655 ret = dev_ethtool(net, &ifr);
4660 if (copy_to_user(arg, &ifr,
4661 sizeof(struct ifreq)))
4667 * These ioctl calls:
4668 * - require superuser power.
4669 * - require strict serialization.
4675 if (!capable(CAP_NET_ADMIN))
4677 dev_load(net, ifr.ifr_name);
4679 ret = dev_ifsioc(net, &ifr, cmd);
4684 if (copy_to_user(arg, &ifr,
4685 sizeof(struct ifreq)))
4691 * These ioctl calls:
4692 * - require superuser power.
4693 * - require strict serialization.
4694 * - do not return a value
4704 case SIOCSIFHWBROADCAST:
4707 case SIOCBONDENSLAVE:
4708 case SIOCBONDRELEASE:
4709 case SIOCBONDSETHWADDR:
4710 case SIOCBONDCHANGEACTIVE:
4714 if (!capable(CAP_NET_ADMIN))
4717 case SIOCBONDSLAVEINFOQUERY:
4718 case SIOCBONDINFOQUERY:
4719 dev_load(net, ifr.ifr_name);
4721 ret = dev_ifsioc(net, &ifr, cmd);
4726 /* Get the per device memory space. We can add this but
4727 * currently do not support it */
4729 /* Set the per device memory buffer space.
4730 * Not applicable in our case */
4735 * Unknown or private ioctl.
4738 if (cmd == SIOCWANDEV ||
4739 (cmd >= SIOCDEVPRIVATE &&
4740 cmd <= SIOCDEVPRIVATE + 15)) {
4741 dev_load(net, ifr.ifr_name);
4743 ret = dev_ifsioc(net, &ifr, cmd);
4745 if (!ret && copy_to_user(arg, &ifr,
4746 sizeof(struct ifreq)))
4750 /* Take care of Wireless Extensions */
4751 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4752 return wext_handle_ioctl(net, &ifr, cmd, arg);
4759 * dev_new_index - allocate an ifindex
4760 * @net: the applicable net namespace
4762 * Returns a suitable unique value for a new device interface
4763 * number. The caller must hold the rtnl semaphore or the
4764 * dev_base_lock to be sure it remains unique.
4766 static int dev_new_index(struct net *net)
4772 if (!__dev_get_by_index(net, ifindex))
4777 /* Delayed registration/unregisteration */
4778 static LIST_HEAD(net_todo_list);
4780 static void net_set_todo(struct net_device *dev)
4782 list_add_tail(&dev->todo_list, &net_todo_list);
4785 static void rollback_registered_many(struct list_head *head)
4787 struct net_device *dev, *tmp;
4789 BUG_ON(dev_boot_phase);
4792 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4793 /* Some devices call without registering
4794 * for initialization unwind. Remove those
4795 * devices and proceed with the remaining.
4797 if (dev->reg_state == NETREG_UNINITIALIZED) {
4798 pr_debug("unregister_netdevice: device %s/%p never "
4799 "was registered\n", dev->name, dev);
4802 list_del(&dev->unreg_list);
4806 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4808 /* If device is running, close it first. */
4811 /* And unlink it from device chain. */
4812 unlist_netdevice(dev);
4814 dev->reg_state = NETREG_UNREGISTERING;
4819 list_for_each_entry(dev, head, unreg_list) {
4820 /* Shutdown queueing discipline. */
4824 /* Notify protocols, that we are about to destroy
4825 this device. They should clean all the things.
4827 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4829 if (!dev->rtnl_link_ops ||
4830 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4831 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4834 * Flush the unicast and multicast chains
4839 if (dev->netdev_ops->ndo_uninit)
4840 dev->netdev_ops->ndo_uninit(dev);
4842 /* Notifier chain MUST detach us from master device. */
4843 WARN_ON(dev->master);
4845 /* Remove entries from kobject tree */
4846 netdev_unregister_kobject(dev);
4849 /* Process any work delayed until the end of the batch */
4850 dev = list_first_entry(head, struct net_device, unreg_list);
4851 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4855 list_for_each_entry(dev, head, unreg_list)
4859 static void rollback_registered(struct net_device *dev)
4863 list_add(&dev->unreg_list, &single);
4864 rollback_registered_many(&single);
4867 static void __netdev_init_queue_locks_one(struct net_device *dev,
4868 struct netdev_queue *dev_queue,
4871 spin_lock_init(&dev_queue->_xmit_lock);
4872 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4873 dev_queue->xmit_lock_owner = -1;
4876 static void netdev_init_queue_locks(struct net_device *dev)
4878 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4879 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4882 unsigned long netdev_fix_features(unsigned long features, const char *name)
4884 /* Fix illegal SG+CSUM combinations. */
4885 if ((features & NETIF_F_SG) &&
4886 !(features & NETIF_F_ALL_CSUM)) {
4888 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4889 "checksum feature.\n", name);
4890 features &= ~NETIF_F_SG;
4893 /* TSO requires that SG is present as well. */
4894 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4896 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4897 "SG feature.\n", name);
4898 features &= ~NETIF_F_TSO;
4901 if (features & NETIF_F_UFO) {
4902 if (!(features & NETIF_F_GEN_CSUM)) {
4904 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4905 "since no NETIF_F_HW_CSUM feature.\n",
4907 features &= ~NETIF_F_UFO;
4910 if (!(features & NETIF_F_SG)) {
4912 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4913 "since no NETIF_F_SG feature.\n", name);
4914 features &= ~NETIF_F_UFO;
4920 EXPORT_SYMBOL(netdev_fix_features);
4923 * netif_stacked_transfer_operstate - transfer operstate
4924 * @rootdev: the root or lower level device to transfer state from
4925 * @dev: the device to transfer operstate to
4927 * Transfer operational state from root to device. This is normally
4928 * called when a stacking relationship exists between the root
4929 * device and the device(a leaf device).
4931 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4932 struct net_device *dev)
4934 if (rootdev->operstate == IF_OPER_DORMANT)
4935 netif_dormant_on(dev);
4937 netif_dormant_off(dev);
4939 if (netif_carrier_ok(rootdev)) {
4940 if (!netif_carrier_ok(dev))
4941 netif_carrier_on(dev);
4943 if (netif_carrier_ok(dev))
4944 netif_carrier_off(dev);
4947 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4950 * register_netdevice - register a network device
4951 * @dev: device to register
4953 * Take a completed network device structure and add it to the kernel
4954 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4955 * chain. 0 is returned on success. A negative errno code is returned
4956 * on a failure to set up the device, or if the name is a duplicate.
4958 * Callers must hold the rtnl semaphore. You may want
4959 * register_netdev() instead of this.
4962 * The locking appears insufficient to guarantee two parallel registers
4963 * will not get the same name.
4966 int register_netdevice(struct net_device *dev)
4969 struct net *net = dev_net(dev);
4971 BUG_ON(dev_boot_phase);
4976 /* When net_device's are persistent, this will be fatal. */
4977 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4980 spin_lock_init(&dev->addr_list_lock);
4981 netdev_set_addr_lockdep_class(dev);
4982 netdev_init_queue_locks(dev);
4987 if (!dev->num_rx_queues) {
4989 * Allocate a single RX queue if driver never called
4993 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
4999 dev->_rx->first = dev->_rx;
5000 atomic_set(&dev->_rx->count, 1);
5001 dev->num_rx_queues = 1;
5004 /* Init, if this function is available */
5005 if (dev->netdev_ops->ndo_init) {
5006 ret = dev->netdev_ops->ndo_init(dev);
5014 ret = dev_get_valid_name(dev, dev->name, 0);
5018 dev->ifindex = dev_new_index(net);
5019 if (dev->iflink == -1)
5020 dev->iflink = dev->ifindex;
5022 /* Fix illegal checksum combinations */
5023 if ((dev->features & NETIF_F_HW_CSUM) &&
5024 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5025 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5027 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5030 if ((dev->features & NETIF_F_NO_CSUM) &&
5031 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5032 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5034 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5037 dev->features = netdev_fix_features(dev->features, dev->name);
5039 /* Enable software GSO if SG is supported. */
5040 if (dev->features & NETIF_F_SG)
5041 dev->features |= NETIF_F_GSO;
5043 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5044 ret = notifier_to_errno(ret);
5048 ret = netdev_register_kobject(dev);
5051 dev->reg_state = NETREG_REGISTERED;
5054 * Default initial state at registry is that the
5055 * device is present.
5058 set_bit(__LINK_STATE_PRESENT, &dev->state);
5060 dev_init_scheduler(dev);
5062 list_netdevice(dev);
5064 /* Notify protocols, that a new device appeared. */
5065 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5066 ret = notifier_to_errno(ret);
5068 rollback_registered(dev);
5069 dev->reg_state = NETREG_UNREGISTERED;
5072 * Prevent userspace races by waiting until the network
5073 * device is fully setup before sending notifications.
5075 if (!dev->rtnl_link_ops ||
5076 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5077 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5083 if (dev->netdev_ops->ndo_uninit)
5084 dev->netdev_ops->ndo_uninit(dev);
5087 EXPORT_SYMBOL(register_netdevice);
5090 * init_dummy_netdev - init a dummy network device for NAPI
5091 * @dev: device to init
5093 * This takes a network device structure and initialize the minimum
5094 * amount of fields so it can be used to schedule NAPI polls without
5095 * registering a full blown interface. This is to be used by drivers
5096 * that need to tie several hardware interfaces to a single NAPI
5097 * poll scheduler due to HW limitations.
5099 int init_dummy_netdev(struct net_device *dev)
5101 /* Clear everything. Note we don't initialize spinlocks
5102 * are they aren't supposed to be taken by any of the
5103 * NAPI code and this dummy netdev is supposed to be
5104 * only ever used for NAPI polls
5106 memset(dev, 0, sizeof(struct net_device));
5108 /* make sure we BUG if trying to hit standard
5109 * register/unregister code path
5111 dev->reg_state = NETREG_DUMMY;
5113 /* initialize the ref count */
5114 atomic_set(&dev->refcnt, 1);
5116 /* NAPI wants this */
5117 INIT_LIST_HEAD(&dev->napi_list);
5119 /* a dummy interface is started by default */
5120 set_bit(__LINK_STATE_PRESENT, &dev->state);
5121 set_bit(__LINK_STATE_START, &dev->state);
5125 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5129 * register_netdev - register a network device
5130 * @dev: device to register
5132 * Take a completed network device structure and add it to the kernel
5133 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5134 * chain. 0 is returned on success. A negative errno code is returned
5135 * on a failure to set up the device, or if the name is a duplicate.
5137 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5138 * and expands the device name if you passed a format string to
5141 int register_netdev(struct net_device *dev)
5148 * If the name is a format string the caller wants us to do a
5151 if (strchr(dev->name, '%')) {
5152 err = dev_alloc_name(dev, dev->name);
5157 err = register_netdevice(dev);
5162 EXPORT_SYMBOL(register_netdev);
5165 * netdev_wait_allrefs - wait until all references are gone.
5167 * This is called when unregistering network devices.
5169 * Any protocol or device that holds a reference should register
5170 * for netdevice notification, and cleanup and put back the
5171 * reference if they receive an UNREGISTER event.
5172 * We can get stuck here if buggy protocols don't correctly
5175 static void netdev_wait_allrefs(struct net_device *dev)
5177 unsigned long rebroadcast_time, warning_time;
5179 linkwatch_forget_dev(dev);
5181 rebroadcast_time = warning_time = jiffies;
5182 while (atomic_read(&dev->refcnt) != 0) {
5183 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5186 /* Rebroadcast unregister notification */
5187 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5188 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5189 * should have already handle it the first time */
5191 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5193 /* We must not have linkwatch events
5194 * pending on unregister. If this
5195 * happens, we simply run the queue
5196 * unscheduled, resulting in a noop
5199 linkwatch_run_queue();
5204 rebroadcast_time = jiffies;
5209 if (time_after(jiffies, warning_time + 10 * HZ)) {
5210 printk(KERN_EMERG "unregister_netdevice: "
5211 "waiting for %s to become free. Usage "
5213 dev->name, atomic_read(&dev->refcnt));
5214 warning_time = jiffies;
5223 * register_netdevice(x1);
5224 * register_netdevice(x2);
5226 * unregister_netdevice(y1);
5227 * unregister_netdevice(y2);
5233 * We are invoked by rtnl_unlock().
5234 * This allows us to deal with problems:
5235 * 1) We can delete sysfs objects which invoke hotplug
5236 * without deadlocking with linkwatch via keventd.
5237 * 2) Since we run with the RTNL semaphore not held, we can sleep
5238 * safely in order to wait for the netdev refcnt to drop to zero.
5240 * We must not return until all unregister events added during
5241 * the interval the lock was held have been completed.
5243 void netdev_run_todo(void)
5245 struct list_head list;
5247 /* Snapshot list, allow later requests */
5248 list_replace_init(&net_todo_list, &list);
5252 while (!list_empty(&list)) {
5253 struct net_device *dev
5254 = list_first_entry(&list, struct net_device, todo_list);
5255 list_del(&dev->todo_list);
5257 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5258 printk(KERN_ERR "network todo '%s' but state %d\n",
5259 dev->name, dev->reg_state);
5264 dev->reg_state = NETREG_UNREGISTERED;
5266 on_each_cpu(flush_backlog, dev, 1);
5268 netdev_wait_allrefs(dev);
5271 BUG_ON(atomic_read(&dev->refcnt));
5272 WARN_ON(dev->ip_ptr);
5273 WARN_ON(dev->ip6_ptr);
5274 WARN_ON(dev->dn_ptr);
5276 if (dev->destructor)
5277 dev->destructor(dev);
5279 /* Free network device */
5280 kobject_put(&dev->dev.kobj);
5285 * dev_txq_stats_fold - fold tx_queues stats
5286 * @dev: device to get statistics from
5287 * @stats: struct rtnl_link_stats64 to hold results
5289 void dev_txq_stats_fold(const struct net_device *dev,
5290 struct rtnl_link_stats64 *stats)
5292 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5294 struct netdev_queue *txq;
5296 for (i = 0; i < dev->num_tx_queues; i++) {
5297 txq = netdev_get_tx_queue(dev, i);
5298 spin_lock_bh(&txq->_xmit_lock);
5299 tx_bytes += txq->tx_bytes;
5300 tx_packets += txq->tx_packets;
5301 tx_dropped += txq->tx_dropped;
5302 spin_unlock_bh(&txq->_xmit_lock);
5304 if (tx_bytes || tx_packets || tx_dropped) {
5305 stats->tx_bytes = tx_bytes;
5306 stats->tx_packets = tx_packets;
5307 stats->tx_dropped = tx_dropped;
5310 EXPORT_SYMBOL(dev_txq_stats_fold);
5312 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5313 * fields in the same order, with only the type differing.
5315 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5316 const struct net_device_stats *netdev_stats)
5318 #if BITS_PER_LONG == 64
5319 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5320 memcpy(stats64, netdev_stats, sizeof(*stats64));
5322 size_t i, n = sizeof(*stats64) / sizeof(u64);
5323 const unsigned long *src = (const unsigned long *)netdev_stats;
5324 u64 *dst = (u64 *)stats64;
5326 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5327 sizeof(*stats64) / sizeof(u64));
5328 for (i = 0; i < n; i++)
5334 * dev_get_stats - get network device statistics
5335 * @dev: device to get statistics from
5336 * @storage: place to store stats
5338 * Get network statistics from device. Return @storage.
5339 * The device driver may provide its own method by setting
5340 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5341 * otherwise the internal statistics structure is used.
5343 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5344 struct rtnl_link_stats64 *storage)
5346 const struct net_device_ops *ops = dev->netdev_ops;
5348 if (ops->ndo_get_stats64) {
5349 memset(storage, 0, sizeof(*storage));
5350 return ops->ndo_get_stats64(dev, storage);
5352 if (ops->ndo_get_stats) {
5353 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5356 netdev_stats_to_stats64(storage, &dev->stats);
5357 dev_txq_stats_fold(dev, storage);
5360 EXPORT_SYMBOL(dev_get_stats);
5362 static void netdev_init_one_queue(struct net_device *dev,
5363 struct netdev_queue *queue,
5369 static void netdev_init_queues(struct net_device *dev)
5371 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5372 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5373 spin_lock_init(&dev->tx_global_lock);
5377 * alloc_netdev_mq - allocate network device
5378 * @sizeof_priv: size of private data to allocate space for
5379 * @name: device name format string
5380 * @setup: callback to initialize device
5381 * @queue_count: the number of subqueues to allocate
5383 * Allocates a struct net_device with private data area for driver use
5384 * and performs basic initialization. Also allocates subquue structs
5385 * for each queue on the device at the end of the netdevice.
5387 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5388 void (*setup)(struct net_device *), unsigned int queue_count)
5390 struct netdev_queue *tx;
5391 struct net_device *dev;
5393 struct net_device *p;
5395 struct netdev_rx_queue *rx;
5399 BUG_ON(strlen(name) >= sizeof(dev->name));
5401 alloc_size = sizeof(struct net_device);
5403 /* ensure 32-byte alignment of private area */
5404 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5405 alloc_size += sizeof_priv;
5407 /* ensure 32-byte alignment of whole construct */
5408 alloc_size += NETDEV_ALIGN - 1;
5410 p = kzalloc(alloc_size, GFP_KERNEL);
5412 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5416 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5418 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5424 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5426 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5431 atomic_set(&rx->count, queue_count);
5434 * Set a pointer to first element in the array which holds the
5437 for (i = 0; i < queue_count; i++)
5441 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5442 dev->padded = (char *)dev - (char *)p;
5444 if (dev_addr_init(dev))
5450 dev_net_set(dev, &init_net);
5453 dev->num_tx_queues = queue_count;
5454 dev->real_num_tx_queues = queue_count;
5458 dev->num_rx_queues = queue_count;
5461 dev->gso_max_size = GSO_MAX_SIZE;
5463 netdev_init_queues(dev);
5465 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5466 dev->ethtool_ntuple_list.count = 0;
5467 INIT_LIST_HEAD(&dev->napi_list);
5468 INIT_LIST_HEAD(&dev->unreg_list);
5469 INIT_LIST_HEAD(&dev->link_watch_list);
5470 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5472 strcpy(dev->name, name);
5485 EXPORT_SYMBOL(alloc_netdev_mq);
5488 * free_netdev - free network device
5491 * This function does the last stage of destroying an allocated device
5492 * interface. The reference to the device object is released.
5493 * If this is the last reference then it will be freed.
5495 void free_netdev(struct net_device *dev)
5497 struct napi_struct *p, *n;
5499 release_net(dev_net(dev));
5503 /* Flush device addresses */
5504 dev_addr_flush(dev);
5506 /* Clear ethtool n-tuple list */
5507 ethtool_ntuple_flush(dev);
5509 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5512 /* Compatibility with error handling in drivers */
5513 if (dev->reg_state == NETREG_UNINITIALIZED) {
5514 kfree((char *)dev - dev->padded);
5518 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5519 dev->reg_state = NETREG_RELEASED;
5521 /* will free via device release */
5522 put_device(&dev->dev);
5524 EXPORT_SYMBOL(free_netdev);
5527 * synchronize_net - Synchronize with packet receive processing
5529 * Wait for packets currently being received to be done.
5530 * Does not block later packets from starting.
5532 void synchronize_net(void)
5537 EXPORT_SYMBOL(synchronize_net);
5540 * unregister_netdevice_queue - remove device from the kernel
5544 * This function shuts down a device interface and removes it
5545 * from the kernel tables.
5546 * If head not NULL, device is queued to be unregistered later.
5548 * Callers must hold the rtnl semaphore. You may want
5549 * unregister_netdev() instead of this.
5552 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5557 list_move_tail(&dev->unreg_list, head);
5559 rollback_registered(dev);
5560 /* Finish processing unregister after unlock */
5564 EXPORT_SYMBOL(unregister_netdevice_queue);
5567 * unregister_netdevice_many - unregister many devices
5568 * @head: list of devices
5570 void unregister_netdevice_many(struct list_head *head)
5572 struct net_device *dev;
5574 if (!list_empty(head)) {
5575 rollback_registered_many(head);
5576 list_for_each_entry(dev, head, unreg_list)
5580 EXPORT_SYMBOL(unregister_netdevice_many);
5583 * unregister_netdev - remove device from the kernel
5586 * This function shuts down a device interface and removes it
5587 * from the kernel tables.
5589 * This is just a wrapper for unregister_netdevice that takes
5590 * the rtnl semaphore. In general you want to use this and not
5591 * unregister_netdevice.
5593 void unregister_netdev(struct net_device *dev)
5596 unregister_netdevice(dev);
5599 EXPORT_SYMBOL(unregister_netdev);
5602 * dev_change_net_namespace - move device to different nethost namespace
5604 * @net: network namespace
5605 * @pat: If not NULL name pattern to try if the current device name
5606 * is already taken in the destination network namespace.
5608 * This function shuts down a device interface and moves it
5609 * to a new network namespace. On success 0 is returned, on
5610 * a failure a netagive errno code is returned.
5612 * Callers must hold the rtnl semaphore.
5615 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5621 /* Don't allow namespace local devices to be moved. */
5623 if (dev->features & NETIF_F_NETNS_LOCAL)
5626 /* Ensure the device has been registrered */
5628 if (dev->reg_state != NETREG_REGISTERED)
5631 /* Get out if there is nothing todo */
5633 if (net_eq(dev_net(dev), net))
5636 /* Pick the destination device name, and ensure
5637 * we can use it in the destination network namespace.
5640 if (__dev_get_by_name(net, dev->name)) {
5641 /* We get here if we can't use the current device name */
5644 if (dev_get_valid_name(dev, pat, 1))
5649 * And now a mini version of register_netdevice unregister_netdevice.
5652 /* If device is running close it first. */
5655 /* And unlink it from device chain */
5657 unlist_netdevice(dev);
5661 /* Shutdown queueing discipline. */
5664 /* Notify protocols, that we are about to destroy
5665 this device. They should clean all the things.
5667 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5668 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5671 * Flush the unicast and multicast chains
5676 /* Actually switch the network namespace */
5677 dev_net_set(dev, net);
5679 /* If there is an ifindex conflict assign a new one */
5680 if (__dev_get_by_index(net, dev->ifindex)) {
5681 int iflink = (dev->iflink == dev->ifindex);
5682 dev->ifindex = dev_new_index(net);
5684 dev->iflink = dev->ifindex;
5687 /* Fixup kobjects */
5688 err = device_rename(&dev->dev, dev->name);
5691 /* Add the device back in the hashes */
5692 list_netdevice(dev);
5694 /* Notify protocols, that a new device appeared. */
5695 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5698 * Prevent userspace races by waiting until the network
5699 * device is fully setup before sending notifications.
5701 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5708 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5710 static int dev_cpu_callback(struct notifier_block *nfb,
5711 unsigned long action,
5714 struct sk_buff **list_skb;
5715 struct sk_buff *skb;
5716 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5717 struct softnet_data *sd, *oldsd;
5719 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5722 local_irq_disable();
5723 cpu = smp_processor_id();
5724 sd = &per_cpu(softnet_data, cpu);
5725 oldsd = &per_cpu(softnet_data, oldcpu);
5727 /* Find end of our completion_queue. */
5728 list_skb = &sd->completion_queue;
5730 list_skb = &(*list_skb)->next;
5731 /* Append completion queue from offline CPU. */
5732 *list_skb = oldsd->completion_queue;
5733 oldsd->completion_queue = NULL;
5735 /* Append output queue from offline CPU. */
5736 if (oldsd->output_queue) {
5737 *sd->output_queue_tailp = oldsd->output_queue;
5738 sd->output_queue_tailp = oldsd->output_queue_tailp;
5739 oldsd->output_queue = NULL;
5740 oldsd->output_queue_tailp = &oldsd->output_queue;
5743 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5746 /* Process offline CPU's input_pkt_queue */
5747 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5749 input_queue_head_incr(oldsd);
5751 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5753 input_queue_head_incr(oldsd);
5761 * netdev_increment_features - increment feature set by one
5762 * @all: current feature set
5763 * @one: new feature set
5764 * @mask: mask feature set
5766 * Computes a new feature set after adding a device with feature set
5767 * @one to the master device with current feature set @all. Will not
5768 * enable anything that is off in @mask. Returns the new feature set.
5770 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5773 /* If device needs checksumming, downgrade to it. */
5774 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5775 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5776 else if (mask & NETIF_F_ALL_CSUM) {
5777 /* If one device supports v4/v6 checksumming, set for all. */
5778 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5779 !(all & NETIF_F_GEN_CSUM)) {
5780 all &= ~NETIF_F_ALL_CSUM;
5781 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5784 /* If one device supports hw checksumming, set for all. */
5785 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5786 all &= ~NETIF_F_ALL_CSUM;
5787 all |= NETIF_F_HW_CSUM;
5791 one |= NETIF_F_ALL_CSUM;
5793 one |= all & NETIF_F_ONE_FOR_ALL;
5794 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5795 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5799 EXPORT_SYMBOL(netdev_increment_features);
5801 static struct hlist_head *netdev_create_hash(void)
5804 struct hlist_head *hash;
5806 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5808 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5809 INIT_HLIST_HEAD(&hash[i]);
5814 /* Initialize per network namespace state */
5815 static int __net_init netdev_init(struct net *net)
5817 INIT_LIST_HEAD(&net->dev_base_head);
5819 net->dev_name_head = netdev_create_hash();
5820 if (net->dev_name_head == NULL)
5823 net->dev_index_head = netdev_create_hash();
5824 if (net->dev_index_head == NULL)
5830 kfree(net->dev_name_head);
5836 * netdev_drivername - network driver for the device
5837 * @dev: network device
5838 * @buffer: buffer for resulting name
5839 * @len: size of buffer
5841 * Determine network driver for device.
5843 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5845 const struct device_driver *driver;
5846 const struct device *parent;
5848 if (len <= 0 || !buffer)
5852 parent = dev->dev.parent;
5857 driver = parent->driver;
5858 if (driver && driver->name)
5859 strlcpy(buffer, driver->name, len);
5863 static int __netdev_printk(const char *level, const struct net_device *dev,
5864 struct va_format *vaf)
5868 if (dev && dev->dev.parent)
5869 r = dev_printk(level, dev->dev.parent, "%s: %pV",
5870 netdev_name(dev), vaf);
5872 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
5874 r = printk("%s(NULL net_device): %pV", level, vaf);
5879 int netdev_printk(const char *level, const struct net_device *dev,
5880 const char *format, ...)
5882 struct va_format vaf;
5886 va_start(args, format);
5891 r = __netdev_printk(level, dev, &vaf);
5896 EXPORT_SYMBOL(netdev_printk);
5898 #define define_netdev_printk_level(func, level) \
5899 int func(const struct net_device *dev, const char *fmt, ...) \
5902 struct va_format vaf; \
5905 va_start(args, fmt); \
5910 r = __netdev_printk(level, dev, &vaf); \
5915 EXPORT_SYMBOL(func);
5917 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
5918 define_netdev_printk_level(netdev_alert, KERN_ALERT);
5919 define_netdev_printk_level(netdev_crit, KERN_CRIT);
5920 define_netdev_printk_level(netdev_err, KERN_ERR);
5921 define_netdev_printk_level(netdev_warn, KERN_WARNING);
5922 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
5923 define_netdev_printk_level(netdev_info, KERN_INFO);
5925 static void __net_exit netdev_exit(struct net *net)
5927 kfree(net->dev_name_head);
5928 kfree(net->dev_index_head);
5931 static struct pernet_operations __net_initdata netdev_net_ops = {
5932 .init = netdev_init,
5933 .exit = netdev_exit,
5936 static void __net_exit default_device_exit(struct net *net)
5938 struct net_device *dev, *aux;
5940 * Push all migratable network devices back to the
5941 * initial network namespace
5944 for_each_netdev_safe(net, dev, aux) {
5946 char fb_name[IFNAMSIZ];
5948 /* Ignore unmoveable devices (i.e. loopback) */
5949 if (dev->features & NETIF_F_NETNS_LOCAL)
5952 /* Leave virtual devices for the generic cleanup */
5953 if (dev->rtnl_link_ops)
5956 /* Push remaing network devices to init_net */
5957 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5958 err = dev_change_net_namespace(dev, &init_net, fb_name);
5960 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5961 __func__, dev->name, err);
5968 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5970 /* At exit all network devices most be removed from a network
5971 * namespace. Do this in the reverse order of registeration.
5972 * Do this across as many network namespaces as possible to
5973 * improve batching efficiency.
5975 struct net_device *dev;
5977 LIST_HEAD(dev_kill_list);
5980 list_for_each_entry(net, net_list, exit_list) {
5981 for_each_netdev_reverse(net, dev) {
5982 if (dev->rtnl_link_ops)
5983 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5985 unregister_netdevice_queue(dev, &dev_kill_list);
5988 unregister_netdevice_many(&dev_kill_list);
5992 static struct pernet_operations __net_initdata default_device_ops = {
5993 .exit = default_device_exit,
5994 .exit_batch = default_device_exit_batch,
5998 * Initialize the DEV module. At boot time this walks the device list and
5999 * unhooks any devices that fail to initialise (normally hardware not
6000 * present) and leaves us with a valid list of present and active devices.
6005 * This is called single threaded during boot, so no need
6006 * to take the rtnl semaphore.
6008 static int __init net_dev_init(void)
6010 int i, rc = -ENOMEM;
6012 BUG_ON(!dev_boot_phase);
6014 if (dev_proc_init())
6017 if (netdev_kobject_init())
6020 INIT_LIST_HEAD(&ptype_all);
6021 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6022 INIT_LIST_HEAD(&ptype_base[i]);
6024 if (register_pernet_subsys(&netdev_net_ops))
6028 * Initialise the packet receive queues.
6031 for_each_possible_cpu(i) {
6032 struct softnet_data *sd = &per_cpu(softnet_data, i);
6034 memset(sd, 0, sizeof(*sd));
6035 skb_queue_head_init(&sd->input_pkt_queue);
6036 skb_queue_head_init(&sd->process_queue);
6037 sd->completion_queue = NULL;
6038 INIT_LIST_HEAD(&sd->poll_list);
6039 sd->output_queue = NULL;
6040 sd->output_queue_tailp = &sd->output_queue;
6042 sd->csd.func = rps_trigger_softirq;
6048 sd->backlog.poll = process_backlog;
6049 sd->backlog.weight = weight_p;
6050 sd->backlog.gro_list = NULL;
6051 sd->backlog.gro_count = 0;
6056 /* The loopback device is special if any other network devices
6057 * is present in a network namespace the loopback device must
6058 * be present. Since we now dynamically allocate and free the
6059 * loopback device ensure this invariant is maintained by
6060 * keeping the loopback device as the first device on the
6061 * list of network devices. Ensuring the loopback devices
6062 * is the first device that appears and the last network device
6065 if (register_pernet_device(&loopback_net_ops))
6068 if (register_pernet_device(&default_device_ops))
6071 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6072 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6074 hotcpu_notifier(dev_cpu_callback, 0);
6082 subsys_initcall(net_dev_init);
6084 static int __init initialize_hashrnd(void)
6086 get_random_bytes(&hashrnd, sizeof(hashrnd));
6090 late_initcall_sync(initialize_hashrnd);