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)
1488 if (!(dev->flags & IFF_UP) ||
1489 (skb->len > (dev->mtu + dev->hard_header_len))) {
1493 skb_set_dev(skb, dev);
1494 skb->tstamp.tv64 = 0;
1495 skb->pkt_type = PACKET_HOST;
1496 skb->protocol = eth_type_trans(skb, dev);
1497 return netif_rx(skb);
1499 EXPORT_SYMBOL_GPL(dev_forward_skb);
1502 * Support routine. Sends outgoing frames to any network
1503 * taps currently in use.
1506 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1508 struct packet_type *ptype;
1510 #ifdef CONFIG_NET_CLS_ACT
1511 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1512 net_timestamp_set(skb);
1514 net_timestamp_set(skb);
1518 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1519 /* Never send packets back to the socket
1520 * they originated from - MvS (miquels@drinkel.ow.org)
1522 if ((ptype->dev == dev || !ptype->dev) &&
1523 (ptype->af_packet_priv == NULL ||
1524 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1525 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1529 /* skb->nh should be correctly
1530 set by sender, so that the second statement is
1531 just protection against buggy protocols.
1533 skb_reset_mac_header(skb2);
1535 if (skb_network_header(skb2) < skb2->data ||
1536 skb2->network_header > skb2->tail) {
1537 if (net_ratelimit())
1538 printk(KERN_CRIT "protocol %04x is "
1540 ntohs(skb2->protocol),
1542 skb_reset_network_header(skb2);
1545 skb2->transport_header = skb2->network_header;
1546 skb2->pkt_type = PACKET_OUTGOING;
1547 ptype->func(skb2, skb->dev, ptype, skb->dev);
1554 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1555 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1557 void netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1559 unsigned int real_num = dev->real_num_tx_queues;
1561 if (unlikely(txq > dev->num_tx_queues))
1563 else if (txq > real_num)
1564 dev->real_num_tx_queues = txq;
1565 else if (txq < real_num) {
1566 dev->real_num_tx_queues = txq;
1567 qdisc_reset_all_tx_gt(dev, txq);
1570 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1572 static inline void __netif_reschedule(struct Qdisc *q)
1574 struct softnet_data *sd;
1575 unsigned long flags;
1577 local_irq_save(flags);
1578 sd = &__get_cpu_var(softnet_data);
1579 q->next_sched = NULL;
1580 *sd->output_queue_tailp = q;
1581 sd->output_queue_tailp = &q->next_sched;
1582 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1583 local_irq_restore(flags);
1586 void __netif_schedule(struct Qdisc *q)
1588 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1589 __netif_reschedule(q);
1591 EXPORT_SYMBOL(__netif_schedule);
1593 void dev_kfree_skb_irq(struct sk_buff *skb)
1595 if (!skb->destructor)
1597 else if (atomic_dec_and_test(&skb->users)) {
1598 struct softnet_data *sd;
1599 unsigned long flags;
1601 local_irq_save(flags);
1602 sd = &__get_cpu_var(softnet_data);
1603 skb->next = sd->completion_queue;
1604 sd->completion_queue = skb;
1605 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1606 local_irq_restore(flags);
1609 EXPORT_SYMBOL(dev_kfree_skb_irq);
1611 void dev_kfree_skb_any(struct sk_buff *skb)
1613 if (in_irq() || irqs_disabled())
1614 dev_kfree_skb_irq(skb);
1618 EXPORT_SYMBOL(dev_kfree_skb_any);
1622 * netif_device_detach - mark device as removed
1623 * @dev: network device
1625 * Mark device as removed from system and therefore no longer available.
1627 void netif_device_detach(struct net_device *dev)
1629 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1630 netif_running(dev)) {
1631 netif_tx_stop_all_queues(dev);
1634 EXPORT_SYMBOL(netif_device_detach);
1637 * netif_device_attach - mark device as attached
1638 * @dev: network device
1640 * Mark device as attached from system and restart if needed.
1642 void netif_device_attach(struct net_device *dev)
1644 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1645 netif_running(dev)) {
1646 netif_tx_wake_all_queues(dev);
1647 __netdev_watchdog_up(dev);
1650 EXPORT_SYMBOL(netif_device_attach);
1652 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1654 return ((features & NETIF_F_GEN_CSUM) ||
1655 ((features & NETIF_F_IP_CSUM) &&
1656 protocol == htons(ETH_P_IP)) ||
1657 ((features & NETIF_F_IPV6_CSUM) &&
1658 protocol == htons(ETH_P_IPV6)) ||
1659 ((features & NETIF_F_FCOE_CRC) &&
1660 protocol == htons(ETH_P_FCOE)));
1663 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1665 if (can_checksum_protocol(dev->features, skb->protocol))
1668 if (skb->protocol == htons(ETH_P_8021Q)) {
1669 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1670 if (can_checksum_protocol(dev->features & dev->vlan_features,
1671 veh->h_vlan_encapsulated_proto))
1679 * skb_dev_set -- assign a new device to a buffer
1680 * @skb: buffer for the new device
1681 * @dev: network device
1683 * If an skb is owned by a device already, we have to reset
1684 * all data private to the namespace a device belongs to
1685 * before assigning it a new device.
1687 #ifdef CONFIG_NET_NS
1688 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1691 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1694 skb_init_secmark(skb);
1698 skb->ipvs_property = 0;
1699 #ifdef CONFIG_NET_SCHED
1705 EXPORT_SYMBOL(skb_set_dev);
1706 #endif /* CONFIG_NET_NS */
1709 * Invalidate hardware checksum when packet is to be mangled, and
1710 * complete checksum manually on outgoing path.
1712 int skb_checksum_help(struct sk_buff *skb)
1715 int ret = 0, offset;
1717 if (skb->ip_summed == CHECKSUM_COMPLETE)
1718 goto out_set_summed;
1720 if (unlikely(skb_shinfo(skb)->gso_size)) {
1721 /* Let GSO fix up the checksum. */
1722 goto out_set_summed;
1725 offset = skb->csum_start - skb_headroom(skb);
1726 BUG_ON(offset >= skb_headlen(skb));
1727 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1729 offset += skb->csum_offset;
1730 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1732 if (skb_cloned(skb) &&
1733 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1734 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1739 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1741 skb->ip_summed = CHECKSUM_NONE;
1745 EXPORT_SYMBOL(skb_checksum_help);
1748 * skb_gso_segment - Perform segmentation on skb.
1749 * @skb: buffer to segment
1750 * @features: features for the output path (see dev->features)
1752 * This function segments the given skb and returns a list of segments.
1754 * It may return NULL if the skb requires no segmentation. This is
1755 * only possible when GSO is used for verifying header integrity.
1757 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1759 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1760 struct packet_type *ptype;
1761 __be16 type = skb->protocol;
1764 skb_reset_mac_header(skb);
1765 skb->mac_len = skb->network_header - skb->mac_header;
1766 __skb_pull(skb, skb->mac_len);
1768 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1769 struct net_device *dev = skb->dev;
1770 struct ethtool_drvinfo info = {};
1772 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1773 dev->ethtool_ops->get_drvinfo(dev, &info);
1775 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1777 info.driver, dev ? dev->features : 0L,
1778 skb->sk ? skb->sk->sk_route_caps : 0L,
1779 skb->len, skb->data_len, skb->ip_summed);
1781 if (skb_header_cloned(skb) &&
1782 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1783 return ERR_PTR(err);
1787 list_for_each_entry_rcu(ptype,
1788 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1789 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1790 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1791 err = ptype->gso_send_check(skb);
1792 segs = ERR_PTR(err);
1793 if (err || skb_gso_ok(skb, features))
1795 __skb_push(skb, (skb->data -
1796 skb_network_header(skb)));
1798 segs = ptype->gso_segment(skb, features);
1804 __skb_push(skb, skb->data - skb_mac_header(skb));
1808 EXPORT_SYMBOL(skb_gso_segment);
1810 /* Take action when hardware reception checksum errors are detected. */
1812 void netdev_rx_csum_fault(struct net_device *dev)
1814 if (net_ratelimit()) {
1815 printk(KERN_ERR "%s: hw csum failure.\n",
1816 dev ? dev->name : "<unknown>");
1820 EXPORT_SYMBOL(netdev_rx_csum_fault);
1823 /* Actually, we should eliminate this check as soon as we know, that:
1824 * 1. IOMMU is present and allows to map all the memory.
1825 * 2. No high memory really exists on this machine.
1828 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1830 #ifdef CONFIG_HIGHMEM
1832 if (!(dev->features & NETIF_F_HIGHDMA)) {
1833 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1834 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1838 if (PCI_DMA_BUS_IS_PHYS) {
1839 struct device *pdev = dev->dev.parent;
1843 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1844 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1845 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1854 void (*destructor)(struct sk_buff *skb);
1857 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1859 static void dev_gso_skb_destructor(struct sk_buff *skb)
1861 struct dev_gso_cb *cb;
1864 struct sk_buff *nskb = skb->next;
1866 skb->next = nskb->next;
1869 } while (skb->next);
1871 cb = DEV_GSO_CB(skb);
1873 cb->destructor(skb);
1877 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1878 * @skb: buffer to segment
1880 * This function segments the given skb and stores the list of segments
1883 static int dev_gso_segment(struct sk_buff *skb)
1885 struct net_device *dev = skb->dev;
1886 struct sk_buff *segs;
1887 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1890 segs = skb_gso_segment(skb, features);
1892 /* Verifying header integrity only. */
1897 return PTR_ERR(segs);
1900 DEV_GSO_CB(skb)->destructor = skb->destructor;
1901 skb->destructor = dev_gso_skb_destructor;
1907 * Try to orphan skb early, right before transmission by the device.
1908 * We cannot orphan skb if tx timestamp is requested, since
1909 * drivers need to call skb_tstamp_tx() to send the timestamp.
1911 static inline void skb_orphan_try(struct sk_buff *skb)
1913 if (!skb_tx(skb)->flags)
1918 * Returns true if either:
1919 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1920 * 2. skb is fragmented and the device does not support SG, or if
1921 * at least one of fragments is in highmem and device does not
1922 * support DMA from it.
1924 static inline int skb_needs_linearize(struct sk_buff *skb,
1925 struct net_device *dev)
1927 return skb_is_nonlinear(skb) &&
1928 ((skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
1929 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
1930 illegal_highdma(dev, skb))));
1933 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1934 struct netdev_queue *txq)
1936 const struct net_device_ops *ops = dev->netdev_ops;
1937 int rc = NETDEV_TX_OK;
1939 if (likely(!skb->next)) {
1940 if (!list_empty(&ptype_all))
1941 dev_queue_xmit_nit(skb, dev);
1944 * If device doesnt need skb->dst, release it right now while
1945 * its hot in this cpu cache
1947 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1950 skb_orphan_try(skb);
1952 if (netif_needs_gso(dev, skb)) {
1953 if (unlikely(dev_gso_segment(skb)))
1958 if (skb_needs_linearize(skb, dev) &&
1959 __skb_linearize(skb))
1962 /* If packet is not checksummed and device does not
1963 * support checksumming for this protocol, complete
1964 * checksumming here.
1966 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1967 skb_set_transport_header(skb, skb->csum_start -
1969 if (!dev_can_checksum(dev, skb) &&
1970 skb_checksum_help(skb))
1975 rc = ops->ndo_start_xmit(skb, dev);
1976 if (rc == NETDEV_TX_OK)
1977 txq_trans_update(txq);
1983 struct sk_buff *nskb = skb->next;
1985 skb->next = nskb->next;
1989 * If device doesnt need nskb->dst, release it right now while
1990 * its hot in this cpu cache
1992 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1995 rc = ops->ndo_start_xmit(nskb, dev);
1996 if (unlikely(rc != NETDEV_TX_OK)) {
1997 if (rc & ~NETDEV_TX_MASK)
1998 goto out_kfree_gso_skb;
1999 nskb->next = skb->next;
2003 txq_trans_update(txq);
2004 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2005 return NETDEV_TX_BUSY;
2006 } while (skb->next);
2009 if (likely(skb->next == NULL))
2010 skb->destructor = DEV_GSO_CB(skb)->destructor;
2016 static u32 hashrnd __read_mostly;
2018 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
2022 if (skb_rx_queue_recorded(skb)) {
2023 hash = skb_get_rx_queue(skb);
2024 while (unlikely(hash >= dev->real_num_tx_queues))
2025 hash -= dev->real_num_tx_queues;
2029 if (skb->sk && skb->sk->sk_hash)
2030 hash = skb->sk->sk_hash;
2032 hash = (__force u16) skb->protocol;
2034 hash = jhash_1word(hash, hashrnd);
2036 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2038 EXPORT_SYMBOL(skb_tx_hash);
2040 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2042 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2043 if (net_ratelimit()) {
2044 pr_warning("%s selects TX queue %d, but "
2045 "real number of TX queues is %d\n",
2046 dev->name, queue_index, dev->real_num_tx_queues);
2053 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2054 struct sk_buff *skb)
2057 struct sock *sk = skb->sk;
2059 if (sk_tx_queue_recorded(sk)) {
2060 queue_index = sk_tx_queue_get(sk);
2062 const struct net_device_ops *ops = dev->netdev_ops;
2064 if (ops->ndo_select_queue) {
2065 queue_index = ops->ndo_select_queue(dev, skb);
2066 queue_index = dev_cap_txqueue(dev, queue_index);
2069 if (dev->real_num_tx_queues > 1)
2070 queue_index = skb_tx_hash(dev, skb);
2073 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2075 if (dst && skb_dst(skb) == dst)
2076 sk_tx_queue_set(sk, queue_index);
2081 skb_set_queue_mapping(skb, queue_index);
2082 return netdev_get_tx_queue(dev, queue_index);
2085 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2086 struct net_device *dev,
2087 struct netdev_queue *txq)
2089 spinlock_t *root_lock = qdisc_lock(q);
2090 bool contended = qdisc_is_running(q);
2094 * Heuristic to force contended enqueues to serialize on a
2095 * separate lock before trying to get qdisc main lock.
2096 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2097 * and dequeue packets faster.
2099 if (unlikely(contended))
2100 spin_lock(&q->busylock);
2102 spin_lock(root_lock);
2103 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2106 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2107 qdisc_run_begin(q)) {
2109 * This is a work-conserving queue; there are no old skbs
2110 * waiting to be sent out; and the qdisc is not running -
2111 * xmit the skb directly.
2113 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2115 __qdisc_update_bstats(q, skb->len);
2116 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2117 if (unlikely(contended)) {
2118 spin_unlock(&q->busylock);
2125 rc = NET_XMIT_SUCCESS;
2128 rc = qdisc_enqueue_root(skb, q);
2129 if (qdisc_run_begin(q)) {
2130 if (unlikely(contended)) {
2131 spin_unlock(&q->busylock);
2137 spin_unlock(root_lock);
2138 if (unlikely(contended))
2139 spin_unlock(&q->busylock);
2144 * dev_queue_xmit - transmit a buffer
2145 * @skb: buffer to transmit
2147 * Queue a buffer for transmission to a network device. The caller must
2148 * have set the device and priority and built the buffer before calling
2149 * this function. The function can be called from an interrupt.
2151 * A negative errno code is returned on a failure. A success does not
2152 * guarantee the frame will be transmitted as it may be dropped due
2153 * to congestion or traffic shaping.
2155 * -----------------------------------------------------------------------------------
2156 * I notice this method can also return errors from the queue disciplines,
2157 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2160 * Regardless of the return value, the skb is consumed, so it is currently
2161 * difficult to retry a send to this method. (You can bump the ref count
2162 * before sending to hold a reference for retry if you are careful.)
2164 * When calling this method, interrupts MUST be enabled. This is because
2165 * the BH enable code must have IRQs enabled so that it will not deadlock.
2168 int dev_queue_xmit(struct sk_buff *skb)
2170 struct net_device *dev = skb->dev;
2171 struct netdev_queue *txq;
2175 /* Disable soft irqs for various locks below. Also
2176 * stops preemption for RCU.
2180 txq = dev_pick_tx(dev, skb);
2181 q = rcu_dereference_bh(txq->qdisc);
2183 #ifdef CONFIG_NET_CLS_ACT
2184 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2187 rc = __dev_xmit_skb(skb, q, dev, txq);
2191 /* The device has no queue. Common case for software devices:
2192 loopback, all the sorts of tunnels...
2194 Really, it is unlikely that netif_tx_lock protection is necessary
2195 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2197 However, it is possible, that they rely on protection
2200 Check this and shot the lock. It is not prone from deadlocks.
2201 Either shot noqueue qdisc, it is even simpler 8)
2203 if (dev->flags & IFF_UP) {
2204 int cpu = smp_processor_id(); /* ok because BHs are off */
2206 if (txq->xmit_lock_owner != cpu) {
2208 HARD_TX_LOCK(dev, txq, cpu);
2210 if (!netif_tx_queue_stopped(txq)) {
2211 rc = dev_hard_start_xmit(skb, dev, txq);
2212 if (dev_xmit_complete(rc)) {
2213 HARD_TX_UNLOCK(dev, txq);
2217 HARD_TX_UNLOCK(dev, txq);
2218 if (net_ratelimit())
2219 printk(KERN_CRIT "Virtual device %s asks to "
2220 "queue packet!\n", dev->name);
2222 /* Recursion is detected! It is possible,
2224 if (net_ratelimit())
2225 printk(KERN_CRIT "Dead loop on virtual device "
2226 "%s, fix it urgently!\n", dev->name);
2231 rcu_read_unlock_bh();
2236 rcu_read_unlock_bh();
2239 EXPORT_SYMBOL(dev_queue_xmit);
2242 /*=======================================================================
2244 =======================================================================*/
2246 int netdev_max_backlog __read_mostly = 1000;
2247 int netdev_tstamp_prequeue __read_mostly = 1;
2248 int netdev_budget __read_mostly = 300;
2249 int weight_p __read_mostly = 64; /* old backlog weight */
2251 /* Called with irq disabled */
2252 static inline void ____napi_schedule(struct softnet_data *sd,
2253 struct napi_struct *napi)
2255 list_add_tail(&napi->poll_list, &sd->poll_list);
2256 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2261 /* One global table that all flow-based protocols share. */
2262 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2263 EXPORT_SYMBOL(rps_sock_flow_table);
2266 * get_rps_cpu is called from netif_receive_skb and returns the target
2267 * CPU from the RPS map of the receiving queue for a given skb.
2268 * rcu_read_lock must be held on entry.
2270 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2271 struct rps_dev_flow **rflowp)
2273 struct ipv6hdr *ip6;
2275 struct netdev_rx_queue *rxqueue;
2276 struct rps_map *map;
2277 struct rps_dev_flow_table *flow_table;
2278 struct rps_sock_flow_table *sock_flow_table;
2282 u32 addr1, addr2, ihl;
2288 if (skb_rx_queue_recorded(skb)) {
2289 u16 index = skb_get_rx_queue(skb);
2290 if (unlikely(index >= dev->num_rx_queues)) {
2291 WARN_ONCE(dev->num_rx_queues > 1, "%s received packet "
2292 "on queue %u, but number of RX queues is %u\n",
2293 dev->name, index, dev->num_rx_queues);
2296 rxqueue = dev->_rx + index;
2300 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2304 goto got_hash; /* Skip hash computation on packet header */
2306 switch (skb->protocol) {
2307 case __constant_htons(ETH_P_IP):
2308 if (!pskb_may_pull(skb, sizeof(*ip)))
2311 ip = (struct iphdr *) skb->data;
2312 ip_proto = ip->protocol;
2313 addr1 = (__force u32) ip->saddr;
2314 addr2 = (__force u32) ip->daddr;
2317 case __constant_htons(ETH_P_IPV6):
2318 if (!pskb_may_pull(skb, sizeof(*ip6)))
2321 ip6 = (struct ipv6hdr *) skb->data;
2322 ip_proto = ip6->nexthdr;
2323 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2324 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2337 case IPPROTO_UDPLITE:
2338 if (pskb_may_pull(skb, (ihl * 4) + 4)) {
2339 ports.v32 = * (__force u32 *) (skb->data + (ihl * 4));
2340 if (ports.v16[1] < ports.v16[0])
2341 swap(ports.v16[0], ports.v16[1]);
2349 /* get a consistent hash (same value on both flow directions) */
2352 skb->rxhash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2357 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2358 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2359 if (flow_table && sock_flow_table) {
2361 struct rps_dev_flow *rflow;
2363 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2366 next_cpu = sock_flow_table->ents[skb->rxhash &
2367 sock_flow_table->mask];
2370 * If the desired CPU (where last recvmsg was done) is
2371 * different from current CPU (one in the rx-queue flow
2372 * table entry), switch if one of the following holds:
2373 * - Current CPU is unset (equal to RPS_NO_CPU).
2374 * - Current CPU is offline.
2375 * - The current CPU's queue tail has advanced beyond the
2376 * last packet that was enqueued using this table entry.
2377 * This guarantees that all previous packets for the flow
2378 * have been dequeued, thus preserving in order delivery.
2380 if (unlikely(tcpu != next_cpu) &&
2381 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2382 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2383 rflow->last_qtail)) >= 0)) {
2384 tcpu = rflow->cpu = next_cpu;
2385 if (tcpu != RPS_NO_CPU)
2386 rflow->last_qtail = per_cpu(softnet_data,
2387 tcpu).input_queue_head;
2389 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2396 map = rcu_dereference(rxqueue->rps_map);
2398 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2400 if (cpu_online(tcpu)) {
2410 /* Called from hardirq (IPI) context */
2411 static void rps_trigger_softirq(void *data)
2413 struct softnet_data *sd = data;
2415 ____napi_schedule(sd, &sd->backlog);
2419 #endif /* CONFIG_RPS */
2422 * Check if this softnet_data structure is another cpu one
2423 * If yes, queue it to our IPI list and return 1
2426 static int rps_ipi_queued(struct softnet_data *sd)
2429 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2432 sd->rps_ipi_next = mysd->rps_ipi_list;
2433 mysd->rps_ipi_list = sd;
2435 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2438 #endif /* CONFIG_RPS */
2443 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2444 * queue (may be a remote CPU queue).
2446 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2447 unsigned int *qtail)
2449 struct softnet_data *sd;
2450 unsigned long flags;
2452 sd = &per_cpu(softnet_data, cpu);
2454 local_irq_save(flags);
2457 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2458 if (skb_queue_len(&sd->input_pkt_queue)) {
2460 __skb_queue_tail(&sd->input_pkt_queue, skb);
2461 input_queue_tail_incr_save(sd, qtail);
2463 local_irq_restore(flags);
2464 return NET_RX_SUCCESS;
2467 /* Schedule NAPI for backlog device
2468 * We can use non atomic operation since we own the queue lock
2470 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2471 if (!rps_ipi_queued(sd))
2472 ____napi_schedule(sd, &sd->backlog);
2480 local_irq_restore(flags);
2487 * netif_rx - post buffer to the network code
2488 * @skb: buffer to post
2490 * This function receives a packet from a device driver and queues it for
2491 * the upper (protocol) levels to process. It always succeeds. The buffer
2492 * may be dropped during processing for congestion control or by the
2496 * NET_RX_SUCCESS (no congestion)
2497 * NET_RX_DROP (packet was dropped)
2501 int netif_rx(struct sk_buff *skb)
2505 /* if netpoll wants it, pretend we never saw it */
2506 if (netpoll_rx(skb))
2509 if (netdev_tstamp_prequeue)
2510 net_timestamp_check(skb);
2514 struct rps_dev_flow voidflow, *rflow = &voidflow;
2519 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2521 cpu = smp_processor_id();
2523 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2530 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2536 EXPORT_SYMBOL(netif_rx);
2538 int netif_rx_ni(struct sk_buff *skb)
2543 err = netif_rx(skb);
2544 if (local_softirq_pending())
2550 EXPORT_SYMBOL(netif_rx_ni);
2552 static void net_tx_action(struct softirq_action *h)
2554 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2556 if (sd->completion_queue) {
2557 struct sk_buff *clist;
2559 local_irq_disable();
2560 clist = sd->completion_queue;
2561 sd->completion_queue = NULL;
2565 struct sk_buff *skb = clist;
2566 clist = clist->next;
2568 WARN_ON(atomic_read(&skb->users));
2573 if (sd->output_queue) {
2576 local_irq_disable();
2577 head = sd->output_queue;
2578 sd->output_queue = NULL;
2579 sd->output_queue_tailp = &sd->output_queue;
2583 struct Qdisc *q = head;
2584 spinlock_t *root_lock;
2586 head = head->next_sched;
2588 root_lock = qdisc_lock(q);
2589 if (spin_trylock(root_lock)) {
2590 smp_mb__before_clear_bit();
2591 clear_bit(__QDISC_STATE_SCHED,
2594 spin_unlock(root_lock);
2596 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2598 __netif_reschedule(q);
2600 smp_mb__before_clear_bit();
2601 clear_bit(__QDISC_STATE_SCHED,
2609 static inline int deliver_skb(struct sk_buff *skb,
2610 struct packet_type *pt_prev,
2611 struct net_device *orig_dev)
2613 atomic_inc(&skb->users);
2614 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2617 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2618 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2619 /* This hook is defined here for ATM LANE */
2620 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2621 unsigned char *addr) __read_mostly;
2622 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2625 #ifdef CONFIG_NET_CLS_ACT
2626 /* TODO: Maybe we should just force sch_ingress to be compiled in
2627 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2628 * a compare and 2 stores extra right now if we dont have it on
2629 * but have CONFIG_NET_CLS_ACT
2630 * NOTE: This doesnt stop any functionality; if you dont have
2631 * the ingress scheduler, you just cant add policies on ingress.
2634 static int ing_filter(struct sk_buff *skb)
2636 struct net_device *dev = skb->dev;
2637 u32 ttl = G_TC_RTTL(skb->tc_verd);
2638 struct netdev_queue *rxq;
2639 int result = TC_ACT_OK;
2642 if (MAX_RED_LOOP < ttl++) {
2644 "Redir loop detected Dropping packet (%d->%d)\n",
2645 skb->skb_iif, dev->ifindex);
2649 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2650 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2652 rxq = &dev->rx_queue;
2655 if (q != &noop_qdisc) {
2656 spin_lock(qdisc_lock(q));
2657 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2658 result = qdisc_enqueue_root(skb, q);
2659 spin_unlock(qdisc_lock(q));
2665 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2666 struct packet_type **pt_prev,
2667 int *ret, struct net_device *orig_dev)
2669 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2673 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2677 switch (ing_filter(skb)) {
2691 * netif_nit_deliver - deliver received packets to network taps
2694 * This function is used to deliver incoming packets to network
2695 * taps. It should be used when the normal netif_receive_skb path
2696 * is bypassed, for example because of VLAN acceleration.
2698 void netif_nit_deliver(struct sk_buff *skb)
2700 struct packet_type *ptype;
2702 if (list_empty(&ptype_all))
2705 skb_reset_network_header(skb);
2706 skb_reset_transport_header(skb);
2707 skb->mac_len = skb->network_header - skb->mac_header;
2710 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2711 if (!ptype->dev || ptype->dev == skb->dev)
2712 deliver_skb(skb, ptype, skb->dev);
2718 * netdev_rx_handler_register - register receive handler
2719 * @dev: device to register a handler for
2720 * @rx_handler: receive handler to register
2721 * @rx_handler_data: data pointer that is used by rx handler
2723 * Register a receive hander for a device. This handler will then be
2724 * called from __netif_receive_skb. A negative errno code is returned
2727 * The caller must hold the rtnl_mutex.
2729 int netdev_rx_handler_register(struct net_device *dev,
2730 rx_handler_func_t *rx_handler,
2731 void *rx_handler_data)
2735 if (dev->rx_handler)
2738 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2739 rcu_assign_pointer(dev->rx_handler, rx_handler);
2743 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2746 * netdev_rx_handler_unregister - unregister receive handler
2747 * @dev: device to unregister a handler from
2749 * Unregister a receive hander from a device.
2751 * The caller must hold the rtnl_mutex.
2753 void netdev_rx_handler_unregister(struct net_device *dev)
2757 rcu_assign_pointer(dev->rx_handler, NULL);
2758 rcu_assign_pointer(dev->rx_handler_data, NULL);
2760 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2762 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2763 struct net_device *master)
2765 if (skb->pkt_type == PACKET_HOST) {
2766 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2768 memcpy(dest, master->dev_addr, ETH_ALEN);
2772 /* On bonding slaves other than the currently active slave, suppress
2773 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2774 * ARP on active-backup slaves with arp_validate enabled.
2776 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2778 struct net_device *dev = skb->dev;
2780 if (master->priv_flags & IFF_MASTER_ARPMON)
2781 dev->last_rx = jiffies;
2783 if ((master->priv_flags & IFF_MASTER_ALB) &&
2784 (master->priv_flags & IFF_BRIDGE_PORT)) {
2785 /* Do address unmangle. The local destination address
2786 * will be always the one master has. Provides the right
2787 * functionality in a bridge.
2789 skb_bond_set_mac_by_master(skb, master);
2792 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2793 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2794 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2797 if (master->priv_flags & IFF_MASTER_ALB) {
2798 if (skb->pkt_type != PACKET_BROADCAST &&
2799 skb->pkt_type != PACKET_MULTICAST)
2802 if (master->priv_flags & IFF_MASTER_8023AD &&
2803 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2810 EXPORT_SYMBOL(__skb_bond_should_drop);
2812 static int __netif_receive_skb(struct sk_buff *skb)
2814 struct packet_type *ptype, *pt_prev;
2815 rx_handler_func_t *rx_handler;
2816 struct net_device *orig_dev;
2817 struct net_device *master;
2818 struct net_device *null_or_orig;
2819 struct net_device *orig_or_bond;
2820 int ret = NET_RX_DROP;
2823 if (!netdev_tstamp_prequeue)
2824 net_timestamp_check(skb);
2826 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2827 return NET_RX_SUCCESS;
2829 /* if we've gotten here through NAPI, check netpoll */
2830 if (netpoll_receive_skb(skb))
2834 skb->skb_iif = skb->dev->ifindex;
2837 * bonding note: skbs received on inactive slaves should only
2838 * be delivered to pkt handlers that are exact matches. Also
2839 * the deliver_no_wcard flag will be set. If packet handlers
2840 * are sensitive to duplicate packets these skbs will need to
2841 * be dropped at the handler. The vlan accel path may have
2842 * already set the deliver_no_wcard flag.
2844 null_or_orig = NULL;
2845 orig_dev = skb->dev;
2846 master = ACCESS_ONCE(orig_dev->master);
2847 if (skb->deliver_no_wcard)
2848 null_or_orig = orig_dev;
2850 if (skb_bond_should_drop(skb, master)) {
2851 skb->deliver_no_wcard = 1;
2852 null_or_orig = orig_dev; /* deliver only exact match */
2857 __this_cpu_inc(softnet_data.processed);
2858 skb_reset_network_header(skb);
2859 skb_reset_transport_header(skb);
2860 skb->mac_len = skb->network_header - skb->mac_header;
2866 #ifdef CONFIG_NET_CLS_ACT
2867 if (skb->tc_verd & TC_NCLS) {
2868 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2873 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2874 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2875 ptype->dev == orig_dev) {
2877 ret = deliver_skb(skb, pt_prev, orig_dev);
2882 #ifdef CONFIG_NET_CLS_ACT
2883 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2889 /* Handle special case of bridge or macvlan */
2890 rx_handler = rcu_dereference(skb->dev->rx_handler);
2893 ret = deliver_skb(skb, pt_prev, orig_dev);
2896 skb = rx_handler(skb);
2902 * Make sure frames received on VLAN interfaces stacked on
2903 * bonding interfaces still make their way to any base bonding
2904 * device that may have registered for a specific ptype. The
2905 * handler may have to adjust skb->dev and orig_dev.
2907 orig_or_bond = orig_dev;
2908 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2909 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2910 orig_or_bond = vlan_dev_real_dev(skb->dev);
2913 type = skb->protocol;
2914 list_for_each_entry_rcu(ptype,
2915 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2916 if (ptype->type == type && (ptype->dev == null_or_orig ||
2917 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2918 ptype->dev == orig_or_bond)) {
2920 ret = deliver_skb(skb, pt_prev, orig_dev);
2926 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2929 /* Jamal, now you will not able to escape explaining
2930 * me how you were going to use this. :-)
2941 * netif_receive_skb - process receive buffer from network
2942 * @skb: buffer to process
2944 * netif_receive_skb() is the main receive data processing function.
2945 * It always succeeds. The buffer may be dropped during processing
2946 * for congestion control or by the protocol layers.
2948 * This function may only be called from softirq context and interrupts
2949 * should be enabled.
2951 * Return values (usually ignored):
2952 * NET_RX_SUCCESS: no congestion
2953 * NET_RX_DROP: packet was dropped
2955 int netif_receive_skb(struct sk_buff *skb)
2957 if (netdev_tstamp_prequeue)
2958 net_timestamp_check(skb);
2960 if (skb_defer_rx_timestamp(skb))
2961 return NET_RX_SUCCESS;
2965 struct rps_dev_flow voidflow, *rflow = &voidflow;
2970 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2973 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2977 ret = __netif_receive_skb(skb);
2983 return __netif_receive_skb(skb);
2986 EXPORT_SYMBOL(netif_receive_skb);
2988 /* Network device is going away, flush any packets still pending
2989 * Called with irqs disabled.
2991 static void flush_backlog(void *arg)
2993 struct net_device *dev = arg;
2994 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2995 struct sk_buff *skb, *tmp;
2998 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
2999 if (skb->dev == dev) {
3000 __skb_unlink(skb, &sd->input_pkt_queue);
3002 input_queue_head_incr(sd);
3007 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3008 if (skb->dev == dev) {
3009 __skb_unlink(skb, &sd->process_queue);
3011 input_queue_head_incr(sd);
3016 static int napi_gro_complete(struct sk_buff *skb)
3018 struct packet_type *ptype;
3019 __be16 type = skb->protocol;
3020 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3023 if (NAPI_GRO_CB(skb)->count == 1) {
3024 skb_shinfo(skb)->gso_size = 0;
3029 list_for_each_entry_rcu(ptype, head, list) {
3030 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3033 err = ptype->gro_complete(skb);
3039 WARN_ON(&ptype->list == head);
3041 return NET_RX_SUCCESS;
3045 return netif_receive_skb(skb);
3048 static void napi_gro_flush(struct napi_struct *napi)
3050 struct sk_buff *skb, *next;
3052 for (skb = napi->gro_list; skb; skb = next) {
3055 napi_gro_complete(skb);
3058 napi->gro_count = 0;
3059 napi->gro_list = NULL;
3062 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3064 struct sk_buff **pp = NULL;
3065 struct packet_type *ptype;
3066 __be16 type = skb->protocol;
3067 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3070 enum gro_result ret;
3072 if (!(skb->dev->features & NETIF_F_GRO))
3075 if (skb_is_gso(skb) || skb_has_frags(skb))
3079 list_for_each_entry_rcu(ptype, head, list) {
3080 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3083 skb_set_network_header(skb, skb_gro_offset(skb));
3084 mac_len = skb->network_header - skb->mac_header;
3085 skb->mac_len = mac_len;
3086 NAPI_GRO_CB(skb)->same_flow = 0;
3087 NAPI_GRO_CB(skb)->flush = 0;
3088 NAPI_GRO_CB(skb)->free = 0;
3090 pp = ptype->gro_receive(&napi->gro_list, skb);
3095 if (&ptype->list == head)
3098 same_flow = NAPI_GRO_CB(skb)->same_flow;
3099 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3102 struct sk_buff *nskb = *pp;
3106 napi_gro_complete(nskb);
3113 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3117 NAPI_GRO_CB(skb)->count = 1;
3118 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3119 skb->next = napi->gro_list;
3120 napi->gro_list = skb;
3124 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3125 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3127 BUG_ON(skb->end - skb->tail < grow);
3129 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3132 skb->data_len -= grow;
3134 skb_shinfo(skb)->frags[0].page_offset += grow;
3135 skb_shinfo(skb)->frags[0].size -= grow;
3137 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3138 put_page(skb_shinfo(skb)->frags[0].page);
3139 memmove(skb_shinfo(skb)->frags,
3140 skb_shinfo(skb)->frags + 1,
3141 --skb_shinfo(skb)->nr_frags);
3152 EXPORT_SYMBOL(dev_gro_receive);
3155 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3159 if (netpoll_rx_on(skb))
3162 for (p = napi->gro_list; p; p = p->next) {
3163 NAPI_GRO_CB(p)->same_flow =
3164 (p->dev == skb->dev) &&
3165 !compare_ether_header(skb_mac_header(p),
3166 skb_gro_mac_header(skb));
3167 NAPI_GRO_CB(p)->flush = 0;
3170 return dev_gro_receive(napi, skb);
3173 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3177 if (netif_receive_skb(skb))
3182 case GRO_MERGED_FREE:
3193 EXPORT_SYMBOL(napi_skb_finish);
3195 void skb_gro_reset_offset(struct sk_buff *skb)
3197 NAPI_GRO_CB(skb)->data_offset = 0;
3198 NAPI_GRO_CB(skb)->frag0 = NULL;
3199 NAPI_GRO_CB(skb)->frag0_len = 0;
3201 if (skb->mac_header == skb->tail &&
3202 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3203 NAPI_GRO_CB(skb)->frag0 =
3204 page_address(skb_shinfo(skb)->frags[0].page) +
3205 skb_shinfo(skb)->frags[0].page_offset;
3206 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3209 EXPORT_SYMBOL(skb_gro_reset_offset);
3211 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3213 skb_gro_reset_offset(skb);
3215 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3217 EXPORT_SYMBOL(napi_gro_receive);
3219 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3221 __skb_pull(skb, skb_headlen(skb));
3222 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3226 EXPORT_SYMBOL(napi_reuse_skb);
3228 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3230 struct sk_buff *skb = napi->skb;
3233 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3239 EXPORT_SYMBOL(napi_get_frags);
3241 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3247 skb->protocol = eth_type_trans(skb, skb->dev);
3249 if (ret == GRO_HELD)
3250 skb_gro_pull(skb, -ETH_HLEN);
3251 else if (netif_receive_skb(skb))
3256 case GRO_MERGED_FREE:
3257 napi_reuse_skb(napi, skb);
3266 EXPORT_SYMBOL(napi_frags_finish);
3268 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3270 struct sk_buff *skb = napi->skb;
3277 skb_reset_mac_header(skb);
3278 skb_gro_reset_offset(skb);
3280 off = skb_gro_offset(skb);
3281 hlen = off + sizeof(*eth);
3282 eth = skb_gro_header_fast(skb, off);
3283 if (skb_gro_header_hard(skb, hlen)) {
3284 eth = skb_gro_header_slow(skb, hlen, off);
3285 if (unlikely(!eth)) {
3286 napi_reuse_skb(napi, skb);
3292 skb_gro_pull(skb, sizeof(*eth));
3295 * This works because the only protocols we care about don't require
3296 * special handling. We'll fix it up properly at the end.
3298 skb->protocol = eth->h_proto;
3303 EXPORT_SYMBOL(napi_frags_skb);
3305 gro_result_t napi_gro_frags(struct napi_struct *napi)
3307 struct sk_buff *skb = napi_frags_skb(napi);
3312 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3314 EXPORT_SYMBOL(napi_gro_frags);
3317 * net_rps_action sends any pending IPI's for rps.
3318 * Note: called with local irq disabled, but exits with local irq enabled.
3320 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3323 struct softnet_data *remsd = sd->rps_ipi_list;
3326 sd->rps_ipi_list = NULL;
3330 /* Send pending IPI's to kick RPS processing on remote cpus. */
3332 struct softnet_data *next = remsd->rps_ipi_next;
3334 if (cpu_online(remsd->cpu))
3335 __smp_call_function_single(remsd->cpu,
3344 static int process_backlog(struct napi_struct *napi, int quota)
3347 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3350 /* Check if we have pending ipi, its better to send them now,
3351 * not waiting net_rx_action() end.
3353 if (sd->rps_ipi_list) {
3354 local_irq_disable();
3355 net_rps_action_and_irq_enable(sd);
3358 napi->weight = weight_p;
3359 local_irq_disable();
3360 while (work < quota) {
3361 struct sk_buff *skb;
3364 while ((skb = __skb_dequeue(&sd->process_queue))) {
3366 __netif_receive_skb(skb);
3367 local_irq_disable();
3368 input_queue_head_incr(sd);
3369 if (++work >= quota) {
3376 qlen = skb_queue_len(&sd->input_pkt_queue);
3378 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3379 &sd->process_queue);
3381 if (qlen < quota - work) {
3383 * Inline a custom version of __napi_complete().
3384 * only current cpu owns and manipulates this napi,
3385 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3386 * we can use a plain write instead of clear_bit(),
3387 * and we dont need an smp_mb() memory barrier.
3389 list_del(&napi->poll_list);
3392 quota = work + qlen;
3402 * __napi_schedule - schedule for receive
3403 * @n: entry to schedule
3405 * The entry's receive function will be scheduled to run
3407 void __napi_schedule(struct napi_struct *n)
3409 unsigned long flags;
3411 local_irq_save(flags);
3412 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3413 local_irq_restore(flags);
3415 EXPORT_SYMBOL(__napi_schedule);
3417 void __napi_complete(struct napi_struct *n)
3419 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3420 BUG_ON(n->gro_list);
3422 list_del(&n->poll_list);
3423 smp_mb__before_clear_bit();
3424 clear_bit(NAPI_STATE_SCHED, &n->state);
3426 EXPORT_SYMBOL(__napi_complete);
3428 void napi_complete(struct napi_struct *n)
3430 unsigned long flags;
3433 * don't let napi dequeue from the cpu poll list
3434 * just in case its running on a different cpu
3436 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3440 local_irq_save(flags);
3442 local_irq_restore(flags);
3444 EXPORT_SYMBOL(napi_complete);
3446 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3447 int (*poll)(struct napi_struct *, int), int weight)
3449 INIT_LIST_HEAD(&napi->poll_list);
3450 napi->gro_count = 0;
3451 napi->gro_list = NULL;
3454 napi->weight = weight;
3455 list_add(&napi->dev_list, &dev->napi_list);
3457 #ifdef CONFIG_NETPOLL
3458 spin_lock_init(&napi->poll_lock);
3459 napi->poll_owner = -1;
3461 set_bit(NAPI_STATE_SCHED, &napi->state);
3463 EXPORT_SYMBOL(netif_napi_add);
3465 void netif_napi_del(struct napi_struct *napi)
3467 struct sk_buff *skb, *next;
3469 list_del_init(&napi->dev_list);
3470 napi_free_frags(napi);
3472 for (skb = napi->gro_list; skb; skb = next) {
3478 napi->gro_list = NULL;
3479 napi->gro_count = 0;
3481 EXPORT_SYMBOL(netif_napi_del);
3483 static void net_rx_action(struct softirq_action *h)
3485 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3486 unsigned long time_limit = jiffies + 2;
3487 int budget = netdev_budget;
3490 local_irq_disable();
3492 while (!list_empty(&sd->poll_list)) {
3493 struct napi_struct *n;
3496 /* If softirq window is exhuasted then punt.
3497 * Allow this to run for 2 jiffies since which will allow
3498 * an average latency of 1.5/HZ.
3500 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3505 /* Even though interrupts have been re-enabled, this
3506 * access is safe because interrupts can only add new
3507 * entries to the tail of this list, and only ->poll()
3508 * calls can remove this head entry from the list.
3510 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3512 have = netpoll_poll_lock(n);
3516 /* This NAPI_STATE_SCHED test is for avoiding a race
3517 * with netpoll's poll_napi(). Only the entity which
3518 * obtains the lock and sees NAPI_STATE_SCHED set will
3519 * actually make the ->poll() call. Therefore we avoid
3520 * accidently calling ->poll() when NAPI is not scheduled.
3523 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3524 work = n->poll(n, weight);
3528 WARN_ON_ONCE(work > weight);
3532 local_irq_disable();
3534 /* Drivers must not modify the NAPI state if they
3535 * consume the entire weight. In such cases this code
3536 * still "owns" the NAPI instance and therefore can
3537 * move the instance around on the list at-will.
3539 if (unlikely(work == weight)) {
3540 if (unlikely(napi_disable_pending(n))) {
3543 local_irq_disable();
3545 list_move_tail(&n->poll_list, &sd->poll_list);
3548 netpoll_poll_unlock(have);
3551 net_rps_action_and_irq_enable(sd);
3553 #ifdef CONFIG_NET_DMA
3555 * There may not be any more sk_buffs coming right now, so push
3556 * any pending DMA copies to hardware
3558 dma_issue_pending_all();
3565 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3569 static gifconf_func_t *gifconf_list[NPROTO];
3572 * register_gifconf - register a SIOCGIF handler
3573 * @family: Address family
3574 * @gifconf: Function handler
3576 * Register protocol dependent address dumping routines. The handler
3577 * that is passed must not be freed or reused until it has been replaced
3578 * by another handler.
3580 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3582 if (family >= NPROTO)
3584 gifconf_list[family] = gifconf;
3587 EXPORT_SYMBOL(register_gifconf);
3591 * Map an interface index to its name (SIOCGIFNAME)
3595 * We need this ioctl for efficient implementation of the
3596 * if_indextoname() function required by the IPv6 API. Without
3597 * it, we would have to search all the interfaces to find a
3601 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3603 struct net_device *dev;
3607 * Fetch the caller's info block.
3610 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3614 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3620 strcpy(ifr.ifr_name, dev->name);
3623 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3629 * Perform a SIOCGIFCONF call. This structure will change
3630 * size eventually, and there is nothing I can do about it.
3631 * Thus we will need a 'compatibility mode'.
3634 static int dev_ifconf(struct net *net, char __user *arg)
3637 struct net_device *dev;
3644 * Fetch the caller's info block.
3647 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3654 * Loop over the interfaces, and write an info block for each.
3658 for_each_netdev(net, dev) {
3659 for (i = 0; i < NPROTO; i++) {
3660 if (gifconf_list[i]) {
3663 done = gifconf_list[i](dev, NULL, 0);
3665 done = gifconf_list[i](dev, pos + total,
3675 * All done. Write the updated control block back to the caller.
3677 ifc.ifc_len = total;
3680 * Both BSD and Solaris return 0 here, so we do too.
3682 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3685 #ifdef CONFIG_PROC_FS
3687 * This is invoked by the /proc filesystem handler to display a device
3690 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3693 struct net *net = seq_file_net(seq);
3695 struct net_device *dev;
3699 return SEQ_START_TOKEN;
3702 for_each_netdev_rcu(net, dev)
3709 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3711 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3712 first_net_device(seq_file_net(seq)) :
3713 next_net_device((struct net_device *)v);
3716 return rcu_dereference(dev);
3719 void dev_seq_stop(struct seq_file *seq, void *v)
3725 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3727 struct rtnl_link_stats64 temp;
3728 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3730 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3731 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3732 dev->name, stats->rx_bytes, stats->rx_packets,
3734 stats->rx_dropped + stats->rx_missed_errors,
3735 stats->rx_fifo_errors,
3736 stats->rx_length_errors + stats->rx_over_errors +
3737 stats->rx_crc_errors + stats->rx_frame_errors,
3738 stats->rx_compressed, stats->multicast,
3739 stats->tx_bytes, stats->tx_packets,
3740 stats->tx_errors, stats->tx_dropped,
3741 stats->tx_fifo_errors, stats->collisions,
3742 stats->tx_carrier_errors +
3743 stats->tx_aborted_errors +
3744 stats->tx_window_errors +
3745 stats->tx_heartbeat_errors,
3746 stats->tx_compressed);
3750 * Called from the PROCfs module. This now uses the new arbitrary sized
3751 * /proc/net interface to create /proc/net/dev
3753 static int dev_seq_show(struct seq_file *seq, void *v)
3755 if (v == SEQ_START_TOKEN)
3756 seq_puts(seq, "Inter-| Receive "
3758 " face |bytes packets errs drop fifo frame "
3759 "compressed multicast|bytes packets errs "
3760 "drop fifo colls carrier compressed\n");
3762 dev_seq_printf_stats(seq, v);
3766 static struct softnet_data *softnet_get_online(loff_t *pos)
3768 struct softnet_data *sd = NULL;
3770 while (*pos < nr_cpu_ids)
3771 if (cpu_online(*pos)) {
3772 sd = &per_cpu(softnet_data, *pos);
3779 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3781 return softnet_get_online(pos);
3784 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3787 return softnet_get_online(pos);
3790 static void softnet_seq_stop(struct seq_file *seq, void *v)
3794 static int softnet_seq_show(struct seq_file *seq, void *v)
3796 struct softnet_data *sd = v;
3798 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3799 sd->processed, sd->dropped, sd->time_squeeze, 0,
3800 0, 0, 0, 0, /* was fastroute */
3801 sd->cpu_collision, sd->received_rps);
3805 static const struct seq_operations dev_seq_ops = {
3806 .start = dev_seq_start,
3807 .next = dev_seq_next,
3808 .stop = dev_seq_stop,
3809 .show = dev_seq_show,
3812 static int dev_seq_open(struct inode *inode, struct file *file)
3814 return seq_open_net(inode, file, &dev_seq_ops,
3815 sizeof(struct seq_net_private));
3818 static const struct file_operations dev_seq_fops = {
3819 .owner = THIS_MODULE,
3820 .open = dev_seq_open,
3822 .llseek = seq_lseek,
3823 .release = seq_release_net,
3826 static const struct seq_operations softnet_seq_ops = {
3827 .start = softnet_seq_start,
3828 .next = softnet_seq_next,
3829 .stop = softnet_seq_stop,
3830 .show = softnet_seq_show,
3833 static int softnet_seq_open(struct inode *inode, struct file *file)
3835 return seq_open(file, &softnet_seq_ops);
3838 static const struct file_operations softnet_seq_fops = {
3839 .owner = THIS_MODULE,
3840 .open = softnet_seq_open,
3842 .llseek = seq_lseek,
3843 .release = seq_release,
3846 static void *ptype_get_idx(loff_t pos)
3848 struct packet_type *pt = NULL;
3852 list_for_each_entry_rcu(pt, &ptype_all, list) {
3858 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3859 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3868 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3872 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3875 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3877 struct packet_type *pt;
3878 struct list_head *nxt;
3882 if (v == SEQ_START_TOKEN)
3883 return ptype_get_idx(0);
3886 nxt = pt->list.next;
3887 if (pt->type == htons(ETH_P_ALL)) {
3888 if (nxt != &ptype_all)
3891 nxt = ptype_base[0].next;
3893 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3895 while (nxt == &ptype_base[hash]) {
3896 if (++hash >= PTYPE_HASH_SIZE)
3898 nxt = ptype_base[hash].next;
3901 return list_entry(nxt, struct packet_type, list);
3904 static void ptype_seq_stop(struct seq_file *seq, void *v)
3910 static int ptype_seq_show(struct seq_file *seq, void *v)
3912 struct packet_type *pt = v;
3914 if (v == SEQ_START_TOKEN)
3915 seq_puts(seq, "Type Device Function\n");
3916 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3917 if (pt->type == htons(ETH_P_ALL))
3918 seq_puts(seq, "ALL ");
3920 seq_printf(seq, "%04x", ntohs(pt->type));
3922 seq_printf(seq, " %-8s %pF\n",
3923 pt->dev ? pt->dev->name : "", pt->func);
3929 static const struct seq_operations ptype_seq_ops = {
3930 .start = ptype_seq_start,
3931 .next = ptype_seq_next,
3932 .stop = ptype_seq_stop,
3933 .show = ptype_seq_show,
3936 static int ptype_seq_open(struct inode *inode, struct file *file)
3938 return seq_open_net(inode, file, &ptype_seq_ops,
3939 sizeof(struct seq_net_private));
3942 static const struct file_operations ptype_seq_fops = {
3943 .owner = THIS_MODULE,
3944 .open = ptype_seq_open,
3946 .llseek = seq_lseek,
3947 .release = seq_release_net,
3951 static int __net_init dev_proc_net_init(struct net *net)
3955 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3957 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3959 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3962 if (wext_proc_init(net))
3968 proc_net_remove(net, "ptype");
3970 proc_net_remove(net, "softnet_stat");
3972 proc_net_remove(net, "dev");
3976 static void __net_exit dev_proc_net_exit(struct net *net)
3978 wext_proc_exit(net);
3980 proc_net_remove(net, "ptype");
3981 proc_net_remove(net, "softnet_stat");
3982 proc_net_remove(net, "dev");
3985 static struct pernet_operations __net_initdata dev_proc_ops = {
3986 .init = dev_proc_net_init,
3987 .exit = dev_proc_net_exit,
3990 static int __init dev_proc_init(void)
3992 return register_pernet_subsys(&dev_proc_ops);
3995 #define dev_proc_init() 0
3996 #endif /* CONFIG_PROC_FS */
4000 * netdev_set_master - set up master/slave pair
4001 * @slave: slave device
4002 * @master: new master device
4004 * Changes the master device of the slave. Pass %NULL to break the
4005 * bonding. The caller must hold the RTNL semaphore. On a failure
4006 * a negative errno code is returned. On success the reference counts
4007 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4008 * function returns zero.
4010 int netdev_set_master(struct net_device *slave, struct net_device *master)
4012 struct net_device *old = slave->master;
4022 slave->master = master;
4029 slave->flags |= IFF_SLAVE;
4031 slave->flags &= ~IFF_SLAVE;
4033 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4036 EXPORT_SYMBOL(netdev_set_master);
4038 static void dev_change_rx_flags(struct net_device *dev, int flags)
4040 const struct net_device_ops *ops = dev->netdev_ops;
4042 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4043 ops->ndo_change_rx_flags(dev, flags);
4046 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4048 unsigned short old_flags = dev->flags;
4054 dev->flags |= IFF_PROMISC;
4055 dev->promiscuity += inc;
4056 if (dev->promiscuity == 0) {
4059 * If inc causes overflow, untouch promisc and return error.
4062 dev->flags &= ~IFF_PROMISC;
4064 dev->promiscuity -= inc;
4065 printk(KERN_WARNING "%s: promiscuity touches roof, "
4066 "set promiscuity failed, promiscuity feature "
4067 "of device might be broken.\n", dev->name);
4071 if (dev->flags != old_flags) {
4072 printk(KERN_INFO "device %s %s promiscuous mode\n",
4073 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4075 if (audit_enabled) {
4076 current_uid_gid(&uid, &gid);
4077 audit_log(current->audit_context, GFP_ATOMIC,
4078 AUDIT_ANOM_PROMISCUOUS,
4079 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4080 dev->name, (dev->flags & IFF_PROMISC),
4081 (old_flags & IFF_PROMISC),
4082 audit_get_loginuid(current),
4084 audit_get_sessionid(current));
4087 dev_change_rx_flags(dev, IFF_PROMISC);
4093 * dev_set_promiscuity - update promiscuity count on a device
4097 * Add or remove promiscuity from a device. While the count in the device
4098 * remains above zero the interface remains promiscuous. Once it hits zero
4099 * the device reverts back to normal filtering operation. A negative inc
4100 * value is used to drop promiscuity on the device.
4101 * Return 0 if successful or a negative errno code on error.
4103 int dev_set_promiscuity(struct net_device *dev, int inc)
4105 unsigned short old_flags = dev->flags;
4108 err = __dev_set_promiscuity(dev, inc);
4111 if (dev->flags != old_flags)
4112 dev_set_rx_mode(dev);
4115 EXPORT_SYMBOL(dev_set_promiscuity);
4118 * dev_set_allmulti - update allmulti count on a device
4122 * Add or remove reception of all multicast frames to a device. While the
4123 * count in the device remains above zero the interface remains listening
4124 * to all interfaces. Once it hits zero the device reverts back to normal
4125 * filtering operation. A negative @inc value is used to drop the counter
4126 * when releasing a resource needing all multicasts.
4127 * Return 0 if successful or a negative errno code on error.
4130 int dev_set_allmulti(struct net_device *dev, int inc)
4132 unsigned short old_flags = dev->flags;
4136 dev->flags |= IFF_ALLMULTI;
4137 dev->allmulti += inc;
4138 if (dev->allmulti == 0) {
4141 * If inc causes overflow, untouch allmulti and return error.
4144 dev->flags &= ~IFF_ALLMULTI;
4146 dev->allmulti -= inc;
4147 printk(KERN_WARNING "%s: allmulti touches roof, "
4148 "set allmulti failed, allmulti feature of "
4149 "device might be broken.\n", dev->name);
4153 if (dev->flags ^ old_flags) {
4154 dev_change_rx_flags(dev, IFF_ALLMULTI);
4155 dev_set_rx_mode(dev);
4159 EXPORT_SYMBOL(dev_set_allmulti);
4162 * Upload unicast and multicast address lists to device and
4163 * configure RX filtering. When the device doesn't support unicast
4164 * filtering it is put in promiscuous mode while unicast addresses
4167 void __dev_set_rx_mode(struct net_device *dev)
4169 const struct net_device_ops *ops = dev->netdev_ops;
4171 /* dev_open will call this function so the list will stay sane. */
4172 if (!(dev->flags&IFF_UP))
4175 if (!netif_device_present(dev))
4178 if (ops->ndo_set_rx_mode)
4179 ops->ndo_set_rx_mode(dev);
4181 /* Unicast addresses changes may only happen under the rtnl,
4182 * therefore calling __dev_set_promiscuity here is safe.
4184 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4185 __dev_set_promiscuity(dev, 1);
4186 dev->uc_promisc = 1;
4187 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4188 __dev_set_promiscuity(dev, -1);
4189 dev->uc_promisc = 0;
4192 if (ops->ndo_set_multicast_list)
4193 ops->ndo_set_multicast_list(dev);
4197 void dev_set_rx_mode(struct net_device *dev)
4199 netif_addr_lock_bh(dev);
4200 __dev_set_rx_mode(dev);
4201 netif_addr_unlock_bh(dev);
4205 * dev_get_flags - get flags reported to userspace
4208 * Get the combination of flag bits exported through APIs to userspace.
4210 unsigned dev_get_flags(const struct net_device *dev)
4214 flags = (dev->flags & ~(IFF_PROMISC |
4219 (dev->gflags & (IFF_PROMISC |
4222 if (netif_running(dev)) {
4223 if (netif_oper_up(dev))
4224 flags |= IFF_RUNNING;
4225 if (netif_carrier_ok(dev))
4226 flags |= IFF_LOWER_UP;
4227 if (netif_dormant(dev))
4228 flags |= IFF_DORMANT;
4233 EXPORT_SYMBOL(dev_get_flags);
4235 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4237 int old_flags = dev->flags;
4243 * Set the flags on our device.
4246 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4247 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4249 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4253 * Load in the correct multicast list now the flags have changed.
4256 if ((old_flags ^ flags) & IFF_MULTICAST)
4257 dev_change_rx_flags(dev, IFF_MULTICAST);
4259 dev_set_rx_mode(dev);
4262 * Have we downed the interface. We handle IFF_UP ourselves
4263 * according to user attempts to set it, rather than blindly
4268 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4269 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4272 dev_set_rx_mode(dev);
4275 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4276 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4278 dev->gflags ^= IFF_PROMISC;
4279 dev_set_promiscuity(dev, inc);
4282 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4283 is important. Some (broken) drivers set IFF_PROMISC, when
4284 IFF_ALLMULTI is requested not asking us and not reporting.
4286 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4287 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4289 dev->gflags ^= IFF_ALLMULTI;
4290 dev_set_allmulti(dev, inc);
4296 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4298 unsigned int changes = dev->flags ^ old_flags;
4300 if (changes & IFF_UP) {
4301 if (dev->flags & IFF_UP)
4302 call_netdevice_notifiers(NETDEV_UP, dev);
4304 call_netdevice_notifiers(NETDEV_DOWN, dev);
4307 if (dev->flags & IFF_UP &&
4308 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4309 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4313 * dev_change_flags - change device settings
4315 * @flags: device state flags
4317 * Change settings on device based state flags. The flags are
4318 * in the userspace exported format.
4320 int dev_change_flags(struct net_device *dev, unsigned flags)
4323 int old_flags = dev->flags;
4325 ret = __dev_change_flags(dev, flags);
4329 changes = old_flags ^ dev->flags;
4331 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4333 __dev_notify_flags(dev, old_flags);
4336 EXPORT_SYMBOL(dev_change_flags);
4339 * dev_set_mtu - Change maximum transfer unit
4341 * @new_mtu: new transfer unit
4343 * Change the maximum transfer size of the network device.
4345 int dev_set_mtu(struct net_device *dev, int new_mtu)
4347 const struct net_device_ops *ops = dev->netdev_ops;
4350 if (new_mtu == dev->mtu)
4353 /* MTU must be positive. */
4357 if (!netif_device_present(dev))
4361 if (ops->ndo_change_mtu)
4362 err = ops->ndo_change_mtu(dev, new_mtu);
4366 if (!err && dev->flags & IFF_UP)
4367 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4370 EXPORT_SYMBOL(dev_set_mtu);
4373 * dev_set_mac_address - Change Media Access Control Address
4377 * Change the hardware (MAC) address of the device
4379 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4381 const struct net_device_ops *ops = dev->netdev_ops;
4384 if (!ops->ndo_set_mac_address)
4386 if (sa->sa_family != dev->type)
4388 if (!netif_device_present(dev))
4390 err = ops->ndo_set_mac_address(dev, sa);
4392 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4395 EXPORT_SYMBOL(dev_set_mac_address);
4398 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4400 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4403 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4409 case SIOCGIFFLAGS: /* Get interface flags */
4410 ifr->ifr_flags = (short) dev_get_flags(dev);
4413 case SIOCGIFMETRIC: /* Get the metric on the interface
4414 (currently unused) */
4415 ifr->ifr_metric = 0;
4418 case SIOCGIFMTU: /* Get the MTU of a device */
4419 ifr->ifr_mtu = dev->mtu;
4424 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4426 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4427 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4428 ifr->ifr_hwaddr.sa_family = dev->type;
4436 ifr->ifr_map.mem_start = dev->mem_start;
4437 ifr->ifr_map.mem_end = dev->mem_end;
4438 ifr->ifr_map.base_addr = dev->base_addr;
4439 ifr->ifr_map.irq = dev->irq;
4440 ifr->ifr_map.dma = dev->dma;
4441 ifr->ifr_map.port = dev->if_port;
4445 ifr->ifr_ifindex = dev->ifindex;
4449 ifr->ifr_qlen = dev->tx_queue_len;
4453 /* dev_ioctl() should ensure this case
4465 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4467 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4470 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4471 const struct net_device_ops *ops;
4476 ops = dev->netdev_ops;
4479 case SIOCSIFFLAGS: /* Set interface flags */
4480 return dev_change_flags(dev, ifr->ifr_flags);
4482 case SIOCSIFMETRIC: /* Set the metric on the interface
4483 (currently unused) */
4486 case SIOCSIFMTU: /* Set the MTU of a device */
4487 return dev_set_mtu(dev, ifr->ifr_mtu);
4490 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4492 case SIOCSIFHWBROADCAST:
4493 if (ifr->ifr_hwaddr.sa_family != dev->type)
4495 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4496 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4497 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4501 if (ops->ndo_set_config) {
4502 if (!netif_device_present(dev))
4504 return ops->ndo_set_config(dev, &ifr->ifr_map);
4509 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4510 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4512 if (!netif_device_present(dev))
4514 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4517 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4518 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4520 if (!netif_device_present(dev))
4522 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4525 if (ifr->ifr_qlen < 0)
4527 dev->tx_queue_len = ifr->ifr_qlen;
4531 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4532 return dev_change_name(dev, ifr->ifr_newname);
4535 * Unknown or private ioctl
4538 if ((cmd >= SIOCDEVPRIVATE &&
4539 cmd <= SIOCDEVPRIVATE + 15) ||
4540 cmd == SIOCBONDENSLAVE ||
4541 cmd == SIOCBONDRELEASE ||
4542 cmd == SIOCBONDSETHWADDR ||
4543 cmd == SIOCBONDSLAVEINFOQUERY ||
4544 cmd == SIOCBONDINFOQUERY ||
4545 cmd == SIOCBONDCHANGEACTIVE ||
4546 cmd == SIOCGMIIPHY ||
4547 cmd == SIOCGMIIREG ||
4548 cmd == SIOCSMIIREG ||
4549 cmd == SIOCBRADDIF ||
4550 cmd == SIOCBRDELIF ||
4551 cmd == SIOCSHWTSTAMP ||
4552 cmd == SIOCWANDEV) {
4554 if (ops->ndo_do_ioctl) {
4555 if (netif_device_present(dev))
4556 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4568 * This function handles all "interface"-type I/O control requests. The actual
4569 * 'doing' part of this is dev_ifsioc above.
4573 * dev_ioctl - network device ioctl
4574 * @net: the applicable net namespace
4575 * @cmd: command to issue
4576 * @arg: pointer to a struct ifreq in user space
4578 * Issue ioctl functions to devices. This is normally called by the
4579 * user space syscall interfaces but can sometimes be useful for
4580 * other purposes. The return value is the return from the syscall if
4581 * positive or a negative errno code on error.
4584 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4590 /* One special case: SIOCGIFCONF takes ifconf argument
4591 and requires shared lock, because it sleeps writing
4595 if (cmd == SIOCGIFCONF) {
4597 ret = dev_ifconf(net, (char __user *) arg);
4601 if (cmd == SIOCGIFNAME)
4602 return dev_ifname(net, (struct ifreq __user *)arg);
4604 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4607 ifr.ifr_name[IFNAMSIZ-1] = 0;
4609 colon = strchr(ifr.ifr_name, ':');
4614 * See which interface the caller is talking about.
4619 * These ioctl calls:
4620 * - can be done by all.
4621 * - atomic and do not require locking.
4632 dev_load(net, ifr.ifr_name);
4634 ret = dev_ifsioc_locked(net, &ifr, cmd);
4639 if (copy_to_user(arg, &ifr,
4640 sizeof(struct ifreq)))
4646 dev_load(net, ifr.ifr_name);
4648 ret = dev_ethtool(net, &ifr);
4653 if (copy_to_user(arg, &ifr,
4654 sizeof(struct ifreq)))
4660 * These ioctl calls:
4661 * - require superuser power.
4662 * - require strict serialization.
4668 if (!capable(CAP_NET_ADMIN))
4670 dev_load(net, ifr.ifr_name);
4672 ret = dev_ifsioc(net, &ifr, cmd);
4677 if (copy_to_user(arg, &ifr,
4678 sizeof(struct ifreq)))
4684 * These ioctl calls:
4685 * - require superuser power.
4686 * - require strict serialization.
4687 * - do not return a value
4697 case SIOCSIFHWBROADCAST:
4700 case SIOCBONDENSLAVE:
4701 case SIOCBONDRELEASE:
4702 case SIOCBONDSETHWADDR:
4703 case SIOCBONDCHANGEACTIVE:
4707 if (!capable(CAP_NET_ADMIN))
4710 case SIOCBONDSLAVEINFOQUERY:
4711 case SIOCBONDINFOQUERY:
4712 dev_load(net, ifr.ifr_name);
4714 ret = dev_ifsioc(net, &ifr, cmd);
4719 /* Get the per device memory space. We can add this but
4720 * currently do not support it */
4722 /* Set the per device memory buffer space.
4723 * Not applicable in our case */
4728 * Unknown or private ioctl.
4731 if (cmd == SIOCWANDEV ||
4732 (cmd >= SIOCDEVPRIVATE &&
4733 cmd <= SIOCDEVPRIVATE + 15)) {
4734 dev_load(net, ifr.ifr_name);
4736 ret = dev_ifsioc(net, &ifr, cmd);
4738 if (!ret && copy_to_user(arg, &ifr,
4739 sizeof(struct ifreq)))
4743 /* Take care of Wireless Extensions */
4744 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4745 return wext_handle_ioctl(net, &ifr, cmd, arg);
4752 * dev_new_index - allocate an ifindex
4753 * @net: the applicable net namespace
4755 * Returns a suitable unique value for a new device interface
4756 * number. The caller must hold the rtnl semaphore or the
4757 * dev_base_lock to be sure it remains unique.
4759 static int dev_new_index(struct net *net)
4765 if (!__dev_get_by_index(net, ifindex))
4770 /* Delayed registration/unregisteration */
4771 static LIST_HEAD(net_todo_list);
4773 static void net_set_todo(struct net_device *dev)
4775 list_add_tail(&dev->todo_list, &net_todo_list);
4778 static void rollback_registered_many(struct list_head *head)
4780 struct net_device *dev, *tmp;
4782 BUG_ON(dev_boot_phase);
4785 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4786 /* Some devices call without registering
4787 * for initialization unwind. Remove those
4788 * devices and proceed with the remaining.
4790 if (dev->reg_state == NETREG_UNINITIALIZED) {
4791 pr_debug("unregister_netdevice: device %s/%p never "
4792 "was registered\n", dev->name, dev);
4795 list_del(&dev->unreg_list);
4799 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4801 /* If device is running, close it first. */
4804 /* And unlink it from device chain. */
4805 unlist_netdevice(dev);
4807 dev->reg_state = NETREG_UNREGISTERING;
4812 list_for_each_entry(dev, head, unreg_list) {
4813 /* Shutdown queueing discipline. */
4817 /* Notify protocols, that we are about to destroy
4818 this device. They should clean all the things.
4820 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4822 if (!dev->rtnl_link_ops ||
4823 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4824 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4827 * Flush the unicast and multicast chains
4832 if (dev->netdev_ops->ndo_uninit)
4833 dev->netdev_ops->ndo_uninit(dev);
4835 /* Notifier chain MUST detach us from master device. */
4836 WARN_ON(dev->master);
4838 /* Remove entries from kobject tree */
4839 netdev_unregister_kobject(dev);
4842 /* Process any work delayed until the end of the batch */
4843 dev = list_first_entry(head, struct net_device, unreg_list);
4844 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4848 list_for_each_entry(dev, head, unreg_list)
4852 static void rollback_registered(struct net_device *dev)
4856 list_add(&dev->unreg_list, &single);
4857 rollback_registered_many(&single);
4860 static void __netdev_init_queue_locks_one(struct net_device *dev,
4861 struct netdev_queue *dev_queue,
4864 spin_lock_init(&dev_queue->_xmit_lock);
4865 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4866 dev_queue->xmit_lock_owner = -1;
4869 static void netdev_init_queue_locks(struct net_device *dev)
4871 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4872 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4875 unsigned long netdev_fix_features(unsigned long features, const char *name)
4877 /* Fix illegal SG+CSUM combinations. */
4878 if ((features & NETIF_F_SG) &&
4879 !(features & NETIF_F_ALL_CSUM)) {
4881 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4882 "checksum feature.\n", name);
4883 features &= ~NETIF_F_SG;
4886 /* TSO requires that SG is present as well. */
4887 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4889 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4890 "SG feature.\n", name);
4891 features &= ~NETIF_F_TSO;
4894 if (features & NETIF_F_UFO) {
4895 if (!(features & NETIF_F_GEN_CSUM)) {
4897 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4898 "since no NETIF_F_HW_CSUM feature.\n",
4900 features &= ~NETIF_F_UFO;
4903 if (!(features & NETIF_F_SG)) {
4905 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4906 "since no NETIF_F_SG feature.\n", name);
4907 features &= ~NETIF_F_UFO;
4913 EXPORT_SYMBOL(netdev_fix_features);
4916 * netif_stacked_transfer_operstate - transfer operstate
4917 * @rootdev: the root or lower level device to transfer state from
4918 * @dev: the device to transfer operstate to
4920 * Transfer operational state from root to device. This is normally
4921 * called when a stacking relationship exists between the root
4922 * device and the device(a leaf device).
4924 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4925 struct net_device *dev)
4927 if (rootdev->operstate == IF_OPER_DORMANT)
4928 netif_dormant_on(dev);
4930 netif_dormant_off(dev);
4932 if (netif_carrier_ok(rootdev)) {
4933 if (!netif_carrier_ok(dev))
4934 netif_carrier_on(dev);
4936 if (netif_carrier_ok(dev))
4937 netif_carrier_off(dev);
4940 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4943 * register_netdevice - register a network device
4944 * @dev: device to register
4946 * Take a completed network device structure and add it to the kernel
4947 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4948 * chain. 0 is returned on success. A negative errno code is returned
4949 * on a failure to set up the device, or if the name is a duplicate.
4951 * Callers must hold the rtnl semaphore. You may want
4952 * register_netdev() instead of this.
4955 * The locking appears insufficient to guarantee two parallel registers
4956 * will not get the same name.
4959 int register_netdevice(struct net_device *dev)
4962 struct net *net = dev_net(dev);
4964 BUG_ON(dev_boot_phase);
4969 /* When net_device's are persistent, this will be fatal. */
4970 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4973 spin_lock_init(&dev->addr_list_lock);
4974 netdev_set_addr_lockdep_class(dev);
4975 netdev_init_queue_locks(dev);
4980 if (!dev->num_rx_queues) {
4982 * Allocate a single RX queue if driver never called
4986 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
4992 dev->_rx->first = dev->_rx;
4993 atomic_set(&dev->_rx->count, 1);
4994 dev->num_rx_queues = 1;
4997 /* Init, if this function is available */
4998 if (dev->netdev_ops->ndo_init) {
4999 ret = dev->netdev_ops->ndo_init(dev);
5007 ret = dev_get_valid_name(dev, dev->name, 0);
5011 dev->ifindex = dev_new_index(net);
5012 if (dev->iflink == -1)
5013 dev->iflink = dev->ifindex;
5015 /* Fix illegal checksum combinations */
5016 if ((dev->features & NETIF_F_HW_CSUM) &&
5017 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5018 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5020 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5023 if ((dev->features & NETIF_F_NO_CSUM) &&
5024 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5025 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5027 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5030 dev->features = netdev_fix_features(dev->features, dev->name);
5032 /* Enable software GSO if SG is supported. */
5033 if (dev->features & NETIF_F_SG)
5034 dev->features |= NETIF_F_GSO;
5036 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5037 ret = notifier_to_errno(ret);
5041 ret = netdev_register_kobject(dev);
5044 dev->reg_state = NETREG_REGISTERED;
5047 * Default initial state at registry is that the
5048 * device is present.
5051 set_bit(__LINK_STATE_PRESENT, &dev->state);
5053 dev_init_scheduler(dev);
5055 list_netdevice(dev);
5057 /* Notify protocols, that a new device appeared. */
5058 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5059 ret = notifier_to_errno(ret);
5061 rollback_registered(dev);
5062 dev->reg_state = NETREG_UNREGISTERED;
5065 * Prevent userspace races by waiting until the network
5066 * device is fully setup before sending notifications.
5068 if (!dev->rtnl_link_ops ||
5069 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5070 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5076 if (dev->netdev_ops->ndo_uninit)
5077 dev->netdev_ops->ndo_uninit(dev);
5080 EXPORT_SYMBOL(register_netdevice);
5083 * init_dummy_netdev - init a dummy network device for NAPI
5084 * @dev: device to init
5086 * This takes a network device structure and initialize the minimum
5087 * amount of fields so it can be used to schedule NAPI polls without
5088 * registering a full blown interface. This is to be used by drivers
5089 * that need to tie several hardware interfaces to a single NAPI
5090 * poll scheduler due to HW limitations.
5092 int init_dummy_netdev(struct net_device *dev)
5094 /* Clear everything. Note we don't initialize spinlocks
5095 * are they aren't supposed to be taken by any of the
5096 * NAPI code and this dummy netdev is supposed to be
5097 * only ever used for NAPI polls
5099 memset(dev, 0, sizeof(struct net_device));
5101 /* make sure we BUG if trying to hit standard
5102 * register/unregister code path
5104 dev->reg_state = NETREG_DUMMY;
5106 /* initialize the ref count */
5107 atomic_set(&dev->refcnt, 1);
5109 /* NAPI wants this */
5110 INIT_LIST_HEAD(&dev->napi_list);
5112 /* a dummy interface is started by default */
5113 set_bit(__LINK_STATE_PRESENT, &dev->state);
5114 set_bit(__LINK_STATE_START, &dev->state);
5118 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5122 * register_netdev - register a network device
5123 * @dev: device to register
5125 * Take a completed network device structure and add it to the kernel
5126 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5127 * chain. 0 is returned on success. A negative errno code is returned
5128 * on a failure to set up the device, or if the name is a duplicate.
5130 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5131 * and expands the device name if you passed a format string to
5134 int register_netdev(struct net_device *dev)
5141 * If the name is a format string the caller wants us to do a
5144 if (strchr(dev->name, '%')) {
5145 err = dev_alloc_name(dev, dev->name);
5150 err = register_netdevice(dev);
5155 EXPORT_SYMBOL(register_netdev);
5158 * netdev_wait_allrefs - wait until all references are gone.
5160 * This is called when unregistering network devices.
5162 * Any protocol or device that holds a reference should register
5163 * for netdevice notification, and cleanup and put back the
5164 * reference if they receive an UNREGISTER event.
5165 * We can get stuck here if buggy protocols don't correctly
5168 static void netdev_wait_allrefs(struct net_device *dev)
5170 unsigned long rebroadcast_time, warning_time;
5172 linkwatch_forget_dev(dev);
5174 rebroadcast_time = warning_time = jiffies;
5175 while (atomic_read(&dev->refcnt) != 0) {
5176 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5179 /* Rebroadcast unregister notification */
5180 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5181 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5182 * should have already handle it the first time */
5184 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5186 /* We must not have linkwatch events
5187 * pending on unregister. If this
5188 * happens, we simply run the queue
5189 * unscheduled, resulting in a noop
5192 linkwatch_run_queue();
5197 rebroadcast_time = jiffies;
5202 if (time_after(jiffies, warning_time + 10 * HZ)) {
5203 printk(KERN_EMERG "unregister_netdevice: "
5204 "waiting for %s to become free. Usage "
5206 dev->name, atomic_read(&dev->refcnt));
5207 warning_time = jiffies;
5216 * register_netdevice(x1);
5217 * register_netdevice(x2);
5219 * unregister_netdevice(y1);
5220 * unregister_netdevice(y2);
5226 * We are invoked by rtnl_unlock().
5227 * This allows us to deal with problems:
5228 * 1) We can delete sysfs objects which invoke hotplug
5229 * without deadlocking with linkwatch via keventd.
5230 * 2) Since we run with the RTNL semaphore not held, we can sleep
5231 * safely in order to wait for the netdev refcnt to drop to zero.
5233 * We must not return until all unregister events added during
5234 * the interval the lock was held have been completed.
5236 void netdev_run_todo(void)
5238 struct list_head list;
5240 /* Snapshot list, allow later requests */
5241 list_replace_init(&net_todo_list, &list);
5245 while (!list_empty(&list)) {
5246 struct net_device *dev
5247 = list_first_entry(&list, struct net_device, todo_list);
5248 list_del(&dev->todo_list);
5250 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5251 printk(KERN_ERR "network todo '%s' but state %d\n",
5252 dev->name, dev->reg_state);
5257 dev->reg_state = NETREG_UNREGISTERED;
5259 on_each_cpu(flush_backlog, dev, 1);
5261 netdev_wait_allrefs(dev);
5264 BUG_ON(atomic_read(&dev->refcnt));
5265 WARN_ON(dev->ip_ptr);
5266 WARN_ON(dev->ip6_ptr);
5267 WARN_ON(dev->dn_ptr);
5269 if (dev->destructor)
5270 dev->destructor(dev);
5272 /* Free network device */
5273 kobject_put(&dev->dev.kobj);
5278 * dev_txq_stats_fold - fold tx_queues stats
5279 * @dev: device to get statistics from
5280 * @stats: struct rtnl_link_stats64 to hold results
5282 void dev_txq_stats_fold(const struct net_device *dev,
5283 struct rtnl_link_stats64 *stats)
5285 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5287 struct netdev_queue *txq;
5289 for (i = 0; i < dev->num_tx_queues; i++) {
5290 txq = netdev_get_tx_queue(dev, i);
5291 tx_bytes += txq->tx_bytes;
5292 tx_packets += txq->tx_packets;
5293 tx_dropped += txq->tx_dropped;
5295 if (tx_bytes || tx_packets || tx_dropped) {
5296 stats->tx_bytes = tx_bytes;
5297 stats->tx_packets = tx_packets;
5298 stats->tx_dropped = tx_dropped;
5301 EXPORT_SYMBOL(dev_txq_stats_fold);
5303 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5304 * fields in the same order, with only the type differing.
5306 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5307 const struct net_device_stats *netdev_stats)
5309 #if BITS_PER_LONG == 64
5310 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5311 memcpy(stats64, netdev_stats, sizeof(*stats64));
5313 size_t i, n = sizeof(*stats64) / sizeof(u64);
5314 const unsigned long *src = (const unsigned long *)netdev_stats;
5315 u64 *dst = (u64 *)stats64;
5317 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5318 sizeof(*stats64) / sizeof(u64));
5319 for (i = 0; i < n; i++)
5325 * dev_get_stats - get network device statistics
5326 * @dev: device to get statistics from
5327 * @storage: place to store stats
5329 * Get network statistics from device. Return @storage.
5330 * The device driver may provide its own method by setting
5331 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5332 * otherwise the internal statistics structure is used.
5334 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5335 struct rtnl_link_stats64 *storage)
5337 const struct net_device_ops *ops = dev->netdev_ops;
5339 if (ops->ndo_get_stats64) {
5340 memset(storage, 0, sizeof(*storage));
5341 return ops->ndo_get_stats64(dev, storage);
5343 if (ops->ndo_get_stats) {
5344 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5347 netdev_stats_to_stats64(storage, &dev->stats);
5348 dev_txq_stats_fold(dev, storage);
5351 EXPORT_SYMBOL(dev_get_stats);
5353 static void netdev_init_one_queue(struct net_device *dev,
5354 struct netdev_queue *queue,
5360 static void netdev_init_queues(struct net_device *dev)
5362 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5363 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5364 spin_lock_init(&dev->tx_global_lock);
5368 * alloc_netdev_mq - allocate network device
5369 * @sizeof_priv: size of private data to allocate space for
5370 * @name: device name format string
5371 * @setup: callback to initialize device
5372 * @queue_count: the number of subqueues to allocate
5374 * Allocates a struct net_device with private data area for driver use
5375 * and performs basic initialization. Also allocates subquue structs
5376 * for each queue on the device at the end of the netdevice.
5378 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5379 void (*setup)(struct net_device *), unsigned int queue_count)
5381 struct netdev_queue *tx;
5382 struct net_device *dev;
5384 struct net_device *p;
5386 struct netdev_rx_queue *rx;
5390 BUG_ON(strlen(name) >= sizeof(dev->name));
5392 alloc_size = sizeof(struct net_device);
5394 /* ensure 32-byte alignment of private area */
5395 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5396 alloc_size += sizeof_priv;
5398 /* ensure 32-byte alignment of whole construct */
5399 alloc_size += NETDEV_ALIGN - 1;
5401 p = kzalloc(alloc_size, GFP_KERNEL);
5403 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5407 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5409 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5415 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5417 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5422 atomic_set(&rx->count, queue_count);
5425 * Set a pointer to first element in the array which holds the
5428 for (i = 0; i < queue_count; i++)
5432 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5433 dev->padded = (char *)dev - (char *)p;
5435 if (dev_addr_init(dev))
5441 dev_net_set(dev, &init_net);
5444 dev->num_tx_queues = queue_count;
5445 dev->real_num_tx_queues = queue_count;
5449 dev->num_rx_queues = queue_count;
5452 dev->gso_max_size = GSO_MAX_SIZE;
5454 netdev_init_queues(dev);
5456 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5457 dev->ethtool_ntuple_list.count = 0;
5458 INIT_LIST_HEAD(&dev->napi_list);
5459 INIT_LIST_HEAD(&dev->unreg_list);
5460 INIT_LIST_HEAD(&dev->link_watch_list);
5461 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5463 strcpy(dev->name, name);
5476 EXPORT_SYMBOL(alloc_netdev_mq);
5479 * free_netdev - free network device
5482 * This function does the last stage of destroying an allocated device
5483 * interface. The reference to the device object is released.
5484 * If this is the last reference then it will be freed.
5486 void free_netdev(struct net_device *dev)
5488 struct napi_struct *p, *n;
5490 release_net(dev_net(dev));
5494 /* Flush device addresses */
5495 dev_addr_flush(dev);
5497 /* Clear ethtool n-tuple list */
5498 ethtool_ntuple_flush(dev);
5500 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5503 /* Compatibility with error handling in drivers */
5504 if (dev->reg_state == NETREG_UNINITIALIZED) {
5505 kfree((char *)dev - dev->padded);
5509 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5510 dev->reg_state = NETREG_RELEASED;
5512 /* will free via device release */
5513 put_device(&dev->dev);
5515 EXPORT_SYMBOL(free_netdev);
5518 * synchronize_net - Synchronize with packet receive processing
5520 * Wait for packets currently being received to be done.
5521 * Does not block later packets from starting.
5523 void synchronize_net(void)
5528 EXPORT_SYMBOL(synchronize_net);
5531 * unregister_netdevice_queue - remove device from the kernel
5535 * This function shuts down a device interface and removes it
5536 * from the kernel tables.
5537 * If head not NULL, device is queued to be unregistered later.
5539 * Callers must hold the rtnl semaphore. You may want
5540 * unregister_netdev() instead of this.
5543 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5548 list_move_tail(&dev->unreg_list, head);
5550 rollback_registered(dev);
5551 /* Finish processing unregister after unlock */
5555 EXPORT_SYMBOL(unregister_netdevice_queue);
5558 * unregister_netdevice_many - unregister many devices
5559 * @head: list of devices
5561 void unregister_netdevice_many(struct list_head *head)
5563 struct net_device *dev;
5565 if (!list_empty(head)) {
5566 rollback_registered_many(head);
5567 list_for_each_entry(dev, head, unreg_list)
5571 EXPORT_SYMBOL(unregister_netdevice_many);
5574 * unregister_netdev - remove device from the kernel
5577 * This function shuts down a device interface and removes it
5578 * from the kernel tables.
5580 * This is just a wrapper for unregister_netdevice that takes
5581 * the rtnl semaphore. In general you want to use this and not
5582 * unregister_netdevice.
5584 void unregister_netdev(struct net_device *dev)
5587 unregister_netdevice(dev);
5590 EXPORT_SYMBOL(unregister_netdev);
5593 * dev_change_net_namespace - move device to different nethost namespace
5595 * @net: network namespace
5596 * @pat: If not NULL name pattern to try if the current device name
5597 * is already taken in the destination network namespace.
5599 * This function shuts down a device interface and moves it
5600 * to a new network namespace. On success 0 is returned, on
5601 * a failure a netagive errno code is returned.
5603 * Callers must hold the rtnl semaphore.
5606 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5612 /* Don't allow namespace local devices to be moved. */
5614 if (dev->features & NETIF_F_NETNS_LOCAL)
5617 /* Ensure the device has been registrered */
5619 if (dev->reg_state != NETREG_REGISTERED)
5622 /* Get out if there is nothing todo */
5624 if (net_eq(dev_net(dev), net))
5627 /* Pick the destination device name, and ensure
5628 * we can use it in the destination network namespace.
5631 if (__dev_get_by_name(net, dev->name)) {
5632 /* We get here if we can't use the current device name */
5635 if (dev_get_valid_name(dev, pat, 1))
5640 * And now a mini version of register_netdevice unregister_netdevice.
5643 /* If device is running close it first. */
5646 /* And unlink it from device chain */
5648 unlist_netdevice(dev);
5652 /* Shutdown queueing discipline. */
5655 /* Notify protocols, that we are about to destroy
5656 this device. They should clean all the things.
5658 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5659 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5662 * Flush the unicast and multicast chains
5667 /* Actually switch the network namespace */
5668 dev_net_set(dev, net);
5670 /* If there is an ifindex conflict assign a new one */
5671 if (__dev_get_by_index(net, dev->ifindex)) {
5672 int iflink = (dev->iflink == dev->ifindex);
5673 dev->ifindex = dev_new_index(net);
5675 dev->iflink = dev->ifindex;
5678 /* Fixup kobjects */
5679 err = device_rename(&dev->dev, dev->name);
5682 /* Add the device back in the hashes */
5683 list_netdevice(dev);
5685 /* Notify protocols, that a new device appeared. */
5686 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5689 * Prevent userspace races by waiting until the network
5690 * device is fully setup before sending notifications.
5692 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5699 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5701 static int dev_cpu_callback(struct notifier_block *nfb,
5702 unsigned long action,
5705 struct sk_buff **list_skb;
5706 struct sk_buff *skb;
5707 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5708 struct softnet_data *sd, *oldsd;
5710 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5713 local_irq_disable();
5714 cpu = smp_processor_id();
5715 sd = &per_cpu(softnet_data, cpu);
5716 oldsd = &per_cpu(softnet_data, oldcpu);
5718 /* Find end of our completion_queue. */
5719 list_skb = &sd->completion_queue;
5721 list_skb = &(*list_skb)->next;
5722 /* Append completion queue from offline CPU. */
5723 *list_skb = oldsd->completion_queue;
5724 oldsd->completion_queue = NULL;
5726 /* Append output queue from offline CPU. */
5727 if (oldsd->output_queue) {
5728 *sd->output_queue_tailp = oldsd->output_queue;
5729 sd->output_queue_tailp = oldsd->output_queue_tailp;
5730 oldsd->output_queue = NULL;
5731 oldsd->output_queue_tailp = &oldsd->output_queue;
5734 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5737 /* Process offline CPU's input_pkt_queue */
5738 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5740 input_queue_head_incr(oldsd);
5742 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5744 input_queue_head_incr(oldsd);
5752 * netdev_increment_features - increment feature set by one
5753 * @all: current feature set
5754 * @one: new feature set
5755 * @mask: mask feature set
5757 * Computes a new feature set after adding a device with feature set
5758 * @one to the master device with current feature set @all. Will not
5759 * enable anything that is off in @mask. Returns the new feature set.
5761 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5764 /* If device needs checksumming, downgrade to it. */
5765 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5766 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5767 else if (mask & NETIF_F_ALL_CSUM) {
5768 /* If one device supports v4/v6 checksumming, set for all. */
5769 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5770 !(all & NETIF_F_GEN_CSUM)) {
5771 all &= ~NETIF_F_ALL_CSUM;
5772 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5775 /* If one device supports hw checksumming, set for all. */
5776 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5777 all &= ~NETIF_F_ALL_CSUM;
5778 all |= NETIF_F_HW_CSUM;
5782 one |= NETIF_F_ALL_CSUM;
5784 one |= all & NETIF_F_ONE_FOR_ALL;
5785 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5786 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5790 EXPORT_SYMBOL(netdev_increment_features);
5792 static struct hlist_head *netdev_create_hash(void)
5795 struct hlist_head *hash;
5797 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5799 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5800 INIT_HLIST_HEAD(&hash[i]);
5805 /* Initialize per network namespace state */
5806 static int __net_init netdev_init(struct net *net)
5808 INIT_LIST_HEAD(&net->dev_base_head);
5810 net->dev_name_head = netdev_create_hash();
5811 if (net->dev_name_head == NULL)
5814 net->dev_index_head = netdev_create_hash();
5815 if (net->dev_index_head == NULL)
5821 kfree(net->dev_name_head);
5827 * netdev_drivername - network driver for the device
5828 * @dev: network device
5829 * @buffer: buffer for resulting name
5830 * @len: size of buffer
5832 * Determine network driver for device.
5834 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5836 const struct device_driver *driver;
5837 const struct device *parent;
5839 if (len <= 0 || !buffer)
5843 parent = dev->dev.parent;
5848 driver = parent->driver;
5849 if (driver && driver->name)
5850 strlcpy(buffer, driver->name, len);
5854 static int __netdev_printk(const char *level, const struct net_device *dev,
5855 struct va_format *vaf)
5859 if (dev && dev->dev.parent)
5860 r = dev_printk(level, dev->dev.parent, "%s: %pV",
5861 netdev_name(dev), vaf);
5863 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
5865 r = printk("%s(NULL net_device): %pV", level, vaf);
5870 int netdev_printk(const char *level, const struct net_device *dev,
5871 const char *format, ...)
5873 struct va_format vaf;
5877 va_start(args, format);
5882 r = __netdev_printk(level, dev, &vaf);
5887 EXPORT_SYMBOL(netdev_printk);
5889 #define define_netdev_printk_level(func, level) \
5890 int func(const struct net_device *dev, const char *fmt, ...) \
5893 struct va_format vaf; \
5896 va_start(args, fmt); \
5901 r = __netdev_printk(level, dev, &vaf); \
5906 EXPORT_SYMBOL(func);
5908 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
5909 define_netdev_printk_level(netdev_alert, KERN_ALERT);
5910 define_netdev_printk_level(netdev_crit, KERN_CRIT);
5911 define_netdev_printk_level(netdev_err, KERN_ERR);
5912 define_netdev_printk_level(netdev_warn, KERN_WARNING);
5913 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
5914 define_netdev_printk_level(netdev_info, KERN_INFO);
5916 static void __net_exit netdev_exit(struct net *net)
5918 kfree(net->dev_name_head);
5919 kfree(net->dev_index_head);
5922 static struct pernet_operations __net_initdata netdev_net_ops = {
5923 .init = netdev_init,
5924 .exit = netdev_exit,
5927 static void __net_exit default_device_exit(struct net *net)
5929 struct net_device *dev, *aux;
5931 * Push all migratable network devices back to the
5932 * initial network namespace
5935 for_each_netdev_safe(net, dev, aux) {
5937 char fb_name[IFNAMSIZ];
5939 /* Ignore unmoveable devices (i.e. loopback) */
5940 if (dev->features & NETIF_F_NETNS_LOCAL)
5943 /* Leave virtual devices for the generic cleanup */
5944 if (dev->rtnl_link_ops)
5947 /* Push remaing network devices to init_net */
5948 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5949 err = dev_change_net_namespace(dev, &init_net, fb_name);
5951 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5952 __func__, dev->name, err);
5959 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5961 /* At exit all network devices most be removed from a network
5962 * namespace. Do this in the reverse order of registeration.
5963 * Do this across as many network namespaces as possible to
5964 * improve batching efficiency.
5966 struct net_device *dev;
5968 LIST_HEAD(dev_kill_list);
5971 list_for_each_entry(net, net_list, exit_list) {
5972 for_each_netdev_reverse(net, dev) {
5973 if (dev->rtnl_link_ops)
5974 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5976 unregister_netdevice_queue(dev, &dev_kill_list);
5979 unregister_netdevice_many(&dev_kill_list);
5983 static struct pernet_operations __net_initdata default_device_ops = {
5984 .exit = default_device_exit,
5985 .exit_batch = default_device_exit_batch,
5989 * Initialize the DEV module. At boot time this walks the device list and
5990 * unhooks any devices that fail to initialise (normally hardware not
5991 * present) and leaves us with a valid list of present and active devices.
5996 * This is called single threaded during boot, so no need
5997 * to take the rtnl semaphore.
5999 static int __init net_dev_init(void)
6001 int i, rc = -ENOMEM;
6003 BUG_ON(!dev_boot_phase);
6005 if (dev_proc_init())
6008 if (netdev_kobject_init())
6011 INIT_LIST_HEAD(&ptype_all);
6012 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6013 INIT_LIST_HEAD(&ptype_base[i]);
6015 if (register_pernet_subsys(&netdev_net_ops))
6019 * Initialise the packet receive queues.
6022 for_each_possible_cpu(i) {
6023 struct softnet_data *sd = &per_cpu(softnet_data, i);
6025 memset(sd, 0, sizeof(*sd));
6026 skb_queue_head_init(&sd->input_pkt_queue);
6027 skb_queue_head_init(&sd->process_queue);
6028 sd->completion_queue = NULL;
6029 INIT_LIST_HEAD(&sd->poll_list);
6030 sd->output_queue = NULL;
6031 sd->output_queue_tailp = &sd->output_queue;
6033 sd->csd.func = rps_trigger_softirq;
6039 sd->backlog.poll = process_backlog;
6040 sd->backlog.weight = weight_p;
6041 sd->backlog.gro_list = NULL;
6042 sd->backlog.gro_count = 0;
6047 /* The loopback device is special if any other network devices
6048 * is present in a network namespace the loopback device must
6049 * be present. Since we now dynamically allocate and free the
6050 * loopback device ensure this invariant is maintained by
6051 * keeping the loopback device as the first device on the
6052 * list of network devices. Ensuring the loopback devices
6053 * is the first device that appears and the last network device
6056 if (register_pernet_device(&loopback_net_ops))
6059 if (register_pernet_device(&default_device_ops))
6062 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6063 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6065 hotcpu_notifier(dev_cpu_callback, 0);
6073 subsys_initcall(net_dev_init);
6075 static int __init initialize_hashrnd(void)
6077 get_random_bytes(&hashrnd, sizeof(hashrnd));
6081 late_initcall_sync(initialize_hashrnd);