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 - 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. The device returned has
813 * had a reference added and the pointer is safe until the user calls
814 * dev_put to indicate they have finished with it.
817 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
820 struct net_device *dev, *ret;
824 for_each_netdev_rcu(net, dev) {
825 if (((dev->flags ^ if_flags) & mask) == 0) {
834 EXPORT_SYMBOL(dev_get_by_flags);
837 * dev_valid_name - check if name is okay for network device
840 * Network device names need to be valid file names to
841 * to allow sysfs to work. We also disallow any kind of
844 int dev_valid_name(const char *name)
848 if (strlen(name) >= IFNAMSIZ)
850 if (!strcmp(name, ".") || !strcmp(name, ".."))
854 if (*name == '/' || isspace(*name))
860 EXPORT_SYMBOL(dev_valid_name);
863 * __dev_alloc_name - allocate a name for a device
864 * @net: network namespace to allocate the device name in
865 * @name: name format string
866 * @buf: scratch buffer and result name string
868 * Passed a format string - eg "lt%d" it will try and find a suitable
869 * id. It scans list of devices to build up a free map, then chooses
870 * the first empty slot. The caller must hold the dev_base or rtnl lock
871 * while allocating the name and adding the device in order to avoid
873 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
874 * Returns the number of the unit assigned or a negative errno code.
877 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
881 const int max_netdevices = 8*PAGE_SIZE;
882 unsigned long *inuse;
883 struct net_device *d;
885 p = strnchr(name, IFNAMSIZ-1, '%');
888 * Verify the string as this thing may have come from
889 * the user. There must be either one "%d" and no other "%"
892 if (p[1] != 'd' || strchr(p + 2, '%'))
895 /* Use one page as a bit array of possible slots */
896 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
900 for_each_netdev(net, d) {
901 if (!sscanf(d->name, name, &i))
903 if (i < 0 || i >= max_netdevices)
906 /* avoid cases where sscanf is not exact inverse of printf */
907 snprintf(buf, IFNAMSIZ, name, i);
908 if (!strncmp(buf, d->name, IFNAMSIZ))
912 i = find_first_zero_bit(inuse, max_netdevices);
913 free_page((unsigned long) inuse);
917 snprintf(buf, IFNAMSIZ, name, i);
918 if (!__dev_get_by_name(net, buf))
921 /* It is possible to run out of possible slots
922 * when the name is long and there isn't enough space left
923 * for the digits, or if all bits are used.
929 * dev_alloc_name - allocate a name for a device
931 * @name: name format string
933 * Passed a format string - eg "lt%d" it will try and find a suitable
934 * id. It scans list of devices to build up a free map, then chooses
935 * the first empty slot. The caller must hold the dev_base or rtnl lock
936 * while allocating the name and adding the device in order to avoid
938 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
939 * Returns the number of the unit assigned or a negative errno code.
942 int dev_alloc_name(struct net_device *dev, const char *name)
948 BUG_ON(!dev_net(dev));
950 ret = __dev_alloc_name(net, name, buf);
952 strlcpy(dev->name, buf, IFNAMSIZ);
955 EXPORT_SYMBOL(dev_alloc_name);
957 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
961 BUG_ON(!dev_net(dev));
964 if (!dev_valid_name(name))
967 if (fmt && strchr(name, '%'))
968 return dev_alloc_name(dev, name);
969 else if (__dev_get_by_name(net, name))
971 else if (dev->name != name)
972 strlcpy(dev->name, name, IFNAMSIZ);
978 * dev_change_name - change name of a device
980 * @newname: name (or format string) must be at least IFNAMSIZ
982 * Change name of a device, can pass format strings "eth%d".
985 int dev_change_name(struct net_device *dev, const char *newname)
987 char oldname[IFNAMSIZ];
993 BUG_ON(!dev_net(dev));
996 if (dev->flags & IFF_UP)
999 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1002 memcpy(oldname, dev->name, IFNAMSIZ);
1004 err = dev_get_valid_name(dev, newname, 1);
1009 ret = device_rename(&dev->dev, dev->name);
1011 memcpy(dev->name, oldname, IFNAMSIZ);
1015 write_lock_bh(&dev_base_lock);
1016 hlist_del(&dev->name_hlist);
1017 write_unlock_bh(&dev_base_lock);
1021 write_lock_bh(&dev_base_lock);
1022 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1023 write_unlock_bh(&dev_base_lock);
1025 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1026 ret = notifier_to_errno(ret);
1029 /* err >= 0 after dev_alloc_name() or stores the first errno */
1032 memcpy(dev->name, oldname, IFNAMSIZ);
1036 "%s: name change rollback failed: %d.\n",
1045 * dev_set_alias - change ifalias of a device
1047 * @alias: name up to IFALIASZ
1048 * @len: limit of bytes to copy from info
1050 * Set ifalias for a device,
1052 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1056 if (len >= IFALIASZ)
1061 kfree(dev->ifalias);
1062 dev->ifalias = NULL;
1067 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1071 strlcpy(dev->ifalias, alias, len+1);
1077 * netdev_features_change - device changes features
1078 * @dev: device to cause notification
1080 * Called to indicate a device has changed features.
1082 void netdev_features_change(struct net_device *dev)
1084 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1086 EXPORT_SYMBOL(netdev_features_change);
1089 * netdev_state_change - device changes state
1090 * @dev: device to cause notification
1092 * Called to indicate a device has changed state. This function calls
1093 * the notifier chains for netdev_chain and sends a NEWLINK message
1094 * to the routing socket.
1096 void netdev_state_change(struct net_device *dev)
1098 if (dev->flags & IFF_UP) {
1099 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1100 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1103 EXPORT_SYMBOL(netdev_state_change);
1105 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1107 return call_netdevice_notifiers(event, dev);
1109 EXPORT_SYMBOL(netdev_bonding_change);
1112 * dev_load - load a network module
1113 * @net: the applicable net namespace
1114 * @name: name of interface
1116 * If a network interface is not present and the process has suitable
1117 * privileges this function loads the module. If module loading is not
1118 * available in this kernel then it becomes a nop.
1121 void dev_load(struct net *net, const char *name)
1123 struct net_device *dev;
1126 dev = dev_get_by_name_rcu(net, name);
1129 if (!dev && capable(CAP_NET_ADMIN))
1130 request_module("%s", name);
1132 EXPORT_SYMBOL(dev_load);
1134 static int __dev_open(struct net_device *dev)
1136 const struct net_device_ops *ops = dev->netdev_ops;
1142 * Is it even present?
1144 if (!netif_device_present(dev))
1147 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1148 ret = notifier_to_errno(ret);
1153 * Call device private open method
1155 set_bit(__LINK_STATE_START, &dev->state);
1157 if (ops->ndo_validate_addr)
1158 ret = ops->ndo_validate_addr(dev);
1160 if (!ret && ops->ndo_open)
1161 ret = ops->ndo_open(dev);
1164 * If it went open OK then:
1168 clear_bit(__LINK_STATE_START, &dev->state);
1173 dev->flags |= IFF_UP;
1178 net_dmaengine_get();
1181 * Initialize multicasting status
1183 dev_set_rx_mode(dev);
1186 * Wakeup transmit queue engine
1195 * dev_open - prepare an interface for use.
1196 * @dev: device to open
1198 * Takes a device from down to up state. The device's private open
1199 * function is invoked and then the multicast lists are loaded. Finally
1200 * the device is moved into the up state and a %NETDEV_UP message is
1201 * sent to the netdev notifier chain.
1203 * Calling this function on an active interface is a nop. On a failure
1204 * a negative errno code is returned.
1206 int dev_open(struct net_device *dev)
1213 if (dev->flags & IFF_UP)
1219 ret = __dev_open(dev);
1224 * ... and announce new interface.
1226 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1227 call_netdevice_notifiers(NETDEV_UP, dev);
1231 EXPORT_SYMBOL(dev_open);
1233 static int __dev_close(struct net_device *dev)
1235 const struct net_device_ops *ops = dev->netdev_ops;
1241 * Tell people we are going down, so that they can
1242 * prepare to death, when device is still operating.
1244 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1246 clear_bit(__LINK_STATE_START, &dev->state);
1248 /* Synchronize to scheduled poll. We cannot touch poll list,
1249 * it can be even on different cpu. So just clear netif_running().
1251 * dev->stop() will invoke napi_disable() on all of it's
1252 * napi_struct instances on this device.
1254 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1256 dev_deactivate(dev);
1259 * Call the device specific close. This cannot fail.
1260 * Only if device is UP
1262 * We allow it to be called even after a DETACH hot-plug
1269 * Device is now down.
1272 dev->flags &= ~IFF_UP;
1277 net_dmaengine_put();
1283 * dev_close - shutdown an interface.
1284 * @dev: device to shutdown
1286 * This function moves an active device into down state. A
1287 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1288 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1291 int dev_close(struct net_device *dev)
1293 if (!(dev->flags & IFF_UP))
1299 * Tell people we are down
1301 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1302 call_netdevice_notifiers(NETDEV_DOWN, dev);
1306 EXPORT_SYMBOL(dev_close);
1310 * dev_disable_lro - disable Large Receive Offload on a device
1313 * Disable Large Receive Offload (LRO) on a net device. Must be
1314 * called under RTNL. This is needed if received packets may be
1315 * forwarded to another interface.
1317 void dev_disable_lro(struct net_device *dev)
1319 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1320 dev->ethtool_ops->set_flags) {
1321 u32 flags = dev->ethtool_ops->get_flags(dev);
1322 if (flags & ETH_FLAG_LRO) {
1323 flags &= ~ETH_FLAG_LRO;
1324 dev->ethtool_ops->set_flags(dev, flags);
1327 WARN_ON(dev->features & NETIF_F_LRO);
1329 EXPORT_SYMBOL(dev_disable_lro);
1332 static int dev_boot_phase = 1;
1335 * Device change register/unregister. These are not inline or static
1336 * as we export them to the world.
1340 * register_netdevice_notifier - register a network notifier block
1343 * Register a notifier to be called when network device events occur.
1344 * The notifier passed is linked into the kernel structures and must
1345 * not be reused until it has been unregistered. A negative errno code
1346 * is returned on a failure.
1348 * When registered all registration and up events are replayed
1349 * to the new notifier to allow device to have a race free
1350 * view of the network device list.
1353 int register_netdevice_notifier(struct notifier_block *nb)
1355 struct net_device *dev;
1356 struct net_device *last;
1361 err = raw_notifier_chain_register(&netdev_chain, nb);
1367 for_each_netdev(net, dev) {
1368 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1369 err = notifier_to_errno(err);
1373 if (!(dev->flags & IFF_UP))
1376 nb->notifier_call(nb, NETDEV_UP, dev);
1387 for_each_netdev(net, dev) {
1391 if (dev->flags & IFF_UP) {
1392 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1393 nb->notifier_call(nb, NETDEV_DOWN, dev);
1395 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1396 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1400 raw_notifier_chain_unregister(&netdev_chain, nb);
1403 EXPORT_SYMBOL(register_netdevice_notifier);
1406 * unregister_netdevice_notifier - unregister a network notifier block
1409 * Unregister a notifier previously registered by
1410 * register_netdevice_notifier(). The notifier is unlinked into the
1411 * kernel structures and may then be reused. A negative errno code
1412 * is returned on a failure.
1415 int unregister_netdevice_notifier(struct notifier_block *nb)
1420 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1424 EXPORT_SYMBOL(unregister_netdevice_notifier);
1427 * call_netdevice_notifiers - call all network notifier blocks
1428 * @val: value passed unmodified to notifier function
1429 * @dev: net_device pointer passed unmodified to notifier function
1431 * Call all network notifier blocks. Parameters and return value
1432 * are as for raw_notifier_call_chain().
1435 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1438 return raw_notifier_call_chain(&netdev_chain, val, dev);
1441 /* When > 0 there are consumers of rx skb time stamps */
1442 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1444 void net_enable_timestamp(void)
1446 atomic_inc(&netstamp_needed);
1448 EXPORT_SYMBOL(net_enable_timestamp);
1450 void net_disable_timestamp(void)
1452 atomic_dec(&netstamp_needed);
1454 EXPORT_SYMBOL(net_disable_timestamp);
1456 static inline void net_timestamp_set(struct sk_buff *skb)
1458 if (atomic_read(&netstamp_needed))
1459 __net_timestamp(skb);
1461 skb->tstamp.tv64 = 0;
1464 static inline void net_timestamp_check(struct sk_buff *skb)
1466 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1467 __net_timestamp(skb);
1471 * dev_forward_skb - loopback an skb to another netif
1473 * @dev: destination network device
1474 * @skb: buffer to forward
1477 * NET_RX_SUCCESS (no congestion)
1478 * NET_RX_DROP (packet was dropped, but freed)
1480 * dev_forward_skb can be used for injecting an skb from the
1481 * start_xmit function of one device into the receive queue
1482 * of another device.
1484 * The receiving device may be in another namespace, so
1485 * we have to clear all information in the skb that could
1486 * impact namespace isolation.
1488 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1492 if (!(dev->flags & IFF_UP) ||
1493 (skb->len > (dev->mtu + dev->hard_header_len))) {
1497 skb_set_dev(skb, dev);
1498 skb->tstamp.tv64 = 0;
1499 skb->pkt_type = PACKET_HOST;
1500 skb->protocol = eth_type_trans(skb, dev);
1501 return netif_rx(skb);
1503 EXPORT_SYMBOL_GPL(dev_forward_skb);
1506 * Support routine. Sends outgoing frames to any network
1507 * taps currently in use.
1510 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1512 struct packet_type *ptype;
1514 #ifdef CONFIG_NET_CLS_ACT
1515 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1516 net_timestamp_set(skb);
1518 net_timestamp_set(skb);
1522 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1523 /* Never send packets back to the socket
1524 * they originated from - MvS (miquels@drinkel.ow.org)
1526 if ((ptype->dev == dev || !ptype->dev) &&
1527 (ptype->af_packet_priv == NULL ||
1528 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1529 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1533 /* skb->nh should be correctly
1534 set by sender, so that the second statement is
1535 just protection against buggy protocols.
1537 skb_reset_mac_header(skb2);
1539 if (skb_network_header(skb2) < skb2->data ||
1540 skb2->network_header > skb2->tail) {
1541 if (net_ratelimit())
1542 printk(KERN_CRIT "protocol %04x is "
1544 skb2->protocol, dev->name);
1545 skb_reset_network_header(skb2);
1548 skb2->transport_header = skb2->network_header;
1549 skb2->pkt_type = PACKET_OUTGOING;
1550 ptype->func(skb2, skb->dev, ptype, skb->dev);
1557 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1558 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1560 void netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1562 unsigned int real_num = dev->real_num_tx_queues;
1564 if (unlikely(txq > dev->num_tx_queues))
1566 else if (txq > real_num)
1567 dev->real_num_tx_queues = txq;
1568 else if (txq < real_num) {
1569 dev->real_num_tx_queues = txq;
1570 qdisc_reset_all_tx_gt(dev, txq);
1573 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1575 static inline void __netif_reschedule(struct Qdisc *q)
1577 struct softnet_data *sd;
1578 unsigned long flags;
1580 local_irq_save(flags);
1581 sd = &__get_cpu_var(softnet_data);
1582 q->next_sched = NULL;
1583 *sd->output_queue_tailp = q;
1584 sd->output_queue_tailp = &q->next_sched;
1585 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1586 local_irq_restore(flags);
1589 void __netif_schedule(struct Qdisc *q)
1591 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1592 __netif_reschedule(q);
1594 EXPORT_SYMBOL(__netif_schedule);
1596 void dev_kfree_skb_irq(struct sk_buff *skb)
1598 if (atomic_dec_and_test(&skb->users)) {
1599 struct softnet_data *sd;
1600 unsigned long flags;
1602 local_irq_save(flags);
1603 sd = &__get_cpu_var(softnet_data);
1604 skb->next = sd->completion_queue;
1605 sd->completion_queue = skb;
1606 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1607 local_irq_restore(flags);
1610 EXPORT_SYMBOL(dev_kfree_skb_irq);
1612 void dev_kfree_skb_any(struct sk_buff *skb)
1614 if (in_irq() || irqs_disabled())
1615 dev_kfree_skb_irq(skb);
1619 EXPORT_SYMBOL(dev_kfree_skb_any);
1623 * netif_device_detach - mark device as removed
1624 * @dev: network device
1626 * Mark device as removed from system and therefore no longer available.
1628 void netif_device_detach(struct net_device *dev)
1630 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1631 netif_running(dev)) {
1632 netif_tx_stop_all_queues(dev);
1635 EXPORT_SYMBOL(netif_device_detach);
1638 * netif_device_attach - mark device as attached
1639 * @dev: network device
1641 * Mark device as attached from system and restart if needed.
1643 void netif_device_attach(struct net_device *dev)
1645 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1646 netif_running(dev)) {
1647 netif_tx_wake_all_queues(dev);
1648 __netdev_watchdog_up(dev);
1651 EXPORT_SYMBOL(netif_device_attach);
1653 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1655 return ((features & NETIF_F_GEN_CSUM) ||
1656 ((features & NETIF_F_IP_CSUM) &&
1657 protocol == htons(ETH_P_IP)) ||
1658 ((features & NETIF_F_IPV6_CSUM) &&
1659 protocol == htons(ETH_P_IPV6)) ||
1660 ((features & NETIF_F_FCOE_CRC) &&
1661 protocol == htons(ETH_P_FCOE)));
1664 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1666 if (can_checksum_protocol(dev->features, skb->protocol))
1669 if (skb->protocol == htons(ETH_P_8021Q)) {
1670 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1671 if (can_checksum_protocol(dev->features & dev->vlan_features,
1672 veh->h_vlan_encapsulated_proto))
1680 * skb_dev_set -- assign a new device to a buffer
1681 * @skb: buffer for the new device
1682 * @dev: network device
1684 * If an skb is owned by a device already, we have to reset
1685 * all data private to the namespace a device belongs to
1686 * before assigning it a new device.
1688 #ifdef CONFIG_NET_NS
1689 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1692 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1695 skb_init_secmark(skb);
1699 skb->ipvs_property = 0;
1700 #ifdef CONFIG_NET_SCHED
1706 EXPORT_SYMBOL(skb_set_dev);
1707 #endif /* CONFIG_NET_NS */
1710 * Invalidate hardware checksum when packet is to be mangled, and
1711 * complete checksum manually on outgoing path.
1713 int skb_checksum_help(struct sk_buff *skb)
1716 int ret = 0, offset;
1718 if (skb->ip_summed == CHECKSUM_COMPLETE)
1719 goto out_set_summed;
1721 if (unlikely(skb_shinfo(skb)->gso_size)) {
1722 /* Let GSO fix up the checksum. */
1723 goto out_set_summed;
1726 offset = skb->csum_start - skb_headroom(skb);
1727 BUG_ON(offset >= skb_headlen(skb));
1728 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1730 offset += skb->csum_offset;
1731 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1733 if (skb_cloned(skb) &&
1734 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1735 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1740 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1742 skb->ip_summed = CHECKSUM_NONE;
1746 EXPORT_SYMBOL(skb_checksum_help);
1749 * skb_gso_segment - Perform segmentation on skb.
1750 * @skb: buffer to segment
1751 * @features: features for the output path (see dev->features)
1753 * This function segments the given skb and returns a list of segments.
1755 * It may return NULL if the skb requires no segmentation. This is
1756 * only possible when GSO is used for verifying header integrity.
1758 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1760 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1761 struct packet_type *ptype;
1762 __be16 type = skb->protocol;
1765 skb_reset_mac_header(skb);
1766 skb->mac_len = skb->network_header - skb->mac_header;
1767 __skb_pull(skb, skb->mac_len);
1769 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1770 struct net_device *dev = skb->dev;
1771 struct ethtool_drvinfo info = {};
1773 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1774 dev->ethtool_ops->get_drvinfo(dev, &info);
1776 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1778 info.driver, dev ? dev->features : 0L,
1779 skb->sk ? skb->sk->sk_route_caps : 0L,
1780 skb->len, skb->data_len, skb->ip_summed);
1782 if (skb_header_cloned(skb) &&
1783 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1784 return ERR_PTR(err);
1788 list_for_each_entry_rcu(ptype,
1789 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1790 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1791 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1792 err = ptype->gso_send_check(skb);
1793 segs = ERR_PTR(err);
1794 if (err || skb_gso_ok(skb, features))
1796 __skb_push(skb, (skb->data -
1797 skb_network_header(skb)));
1799 segs = ptype->gso_segment(skb, features);
1805 __skb_push(skb, skb->data - skb_mac_header(skb));
1809 EXPORT_SYMBOL(skb_gso_segment);
1811 /* Take action when hardware reception checksum errors are detected. */
1813 void netdev_rx_csum_fault(struct net_device *dev)
1815 if (net_ratelimit()) {
1816 printk(KERN_ERR "%s: hw csum failure.\n",
1817 dev ? dev->name : "<unknown>");
1821 EXPORT_SYMBOL(netdev_rx_csum_fault);
1824 /* Actually, we should eliminate this check as soon as we know, that:
1825 * 1. IOMMU is present and allows to map all the memory.
1826 * 2. No high memory really exists on this machine.
1829 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1831 #ifdef CONFIG_HIGHMEM
1833 if (!(dev->features & NETIF_F_HIGHDMA)) {
1834 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1835 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1839 if (PCI_DMA_BUS_IS_PHYS) {
1840 struct device *pdev = dev->dev.parent;
1844 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1845 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1846 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1855 void (*destructor)(struct sk_buff *skb);
1858 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1860 static void dev_gso_skb_destructor(struct sk_buff *skb)
1862 struct dev_gso_cb *cb;
1865 struct sk_buff *nskb = skb->next;
1867 skb->next = nskb->next;
1870 } while (skb->next);
1872 cb = DEV_GSO_CB(skb);
1874 cb->destructor(skb);
1878 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1879 * @skb: buffer to segment
1881 * This function segments the given skb and stores the list of segments
1884 static int dev_gso_segment(struct sk_buff *skb)
1886 struct net_device *dev = skb->dev;
1887 struct sk_buff *segs;
1888 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1891 segs = skb_gso_segment(skb, features);
1893 /* Verifying header integrity only. */
1898 return PTR_ERR(segs);
1901 DEV_GSO_CB(skb)->destructor = skb->destructor;
1902 skb->destructor = dev_gso_skb_destructor;
1908 * Try to orphan skb early, right before transmission by the device.
1909 * We cannot orphan skb if tx timestamp is requested, since
1910 * drivers need to call skb_tstamp_tx() to send the timestamp.
1912 static inline void skb_orphan_try(struct sk_buff *skb)
1914 if (!skb_tx(skb)->flags)
1918 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1919 struct netdev_queue *txq)
1921 const struct net_device_ops *ops = dev->netdev_ops;
1922 int rc = NETDEV_TX_OK;
1924 if (likely(!skb->next)) {
1925 if (!list_empty(&ptype_all))
1926 dev_queue_xmit_nit(skb, dev);
1929 * If device doesnt need skb->dst, release it right now while
1930 * its hot in this cpu cache
1932 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1935 skb_orphan_try(skb);
1937 if (netif_needs_gso(dev, skb)) {
1938 if (unlikely(dev_gso_segment(skb)))
1944 rc = ops->ndo_start_xmit(skb, dev);
1945 if (rc == NETDEV_TX_OK)
1946 txq_trans_update(txq);
1952 struct sk_buff *nskb = skb->next;
1954 skb->next = nskb->next;
1958 * If device doesnt need nskb->dst, release it right now while
1959 * its hot in this cpu cache
1961 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1964 rc = ops->ndo_start_xmit(nskb, dev);
1965 if (unlikely(rc != NETDEV_TX_OK)) {
1966 if (rc & ~NETDEV_TX_MASK)
1967 goto out_kfree_gso_skb;
1968 nskb->next = skb->next;
1972 txq_trans_update(txq);
1973 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1974 return NETDEV_TX_BUSY;
1975 } while (skb->next);
1978 if (likely(skb->next == NULL))
1979 skb->destructor = DEV_GSO_CB(skb)->destructor;
1985 static u32 hashrnd __read_mostly;
1987 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1991 if (skb_rx_queue_recorded(skb)) {
1992 hash = skb_get_rx_queue(skb);
1993 while (unlikely(hash >= dev->real_num_tx_queues))
1994 hash -= dev->real_num_tx_queues;
1998 if (skb->sk && skb->sk->sk_hash)
1999 hash = skb->sk->sk_hash;
2001 hash = (__force u16) skb->protocol;
2003 hash = jhash_1word(hash, hashrnd);
2005 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2007 EXPORT_SYMBOL(skb_tx_hash);
2009 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2011 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2012 if (net_ratelimit()) {
2013 pr_warning("%s selects TX queue %d, but "
2014 "real number of TX queues is %d\n",
2015 dev->name, queue_index, dev->real_num_tx_queues);
2022 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2023 struct sk_buff *skb)
2026 struct sock *sk = skb->sk;
2028 if (sk_tx_queue_recorded(sk)) {
2029 queue_index = sk_tx_queue_get(sk);
2031 const struct net_device_ops *ops = dev->netdev_ops;
2033 if (ops->ndo_select_queue) {
2034 queue_index = ops->ndo_select_queue(dev, skb);
2035 queue_index = dev_cap_txqueue(dev, queue_index);
2038 if (dev->real_num_tx_queues > 1)
2039 queue_index = skb_tx_hash(dev, skb);
2042 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2044 if (dst && skb_dst(skb) == dst)
2045 sk_tx_queue_set(sk, queue_index);
2050 skb_set_queue_mapping(skb, queue_index);
2051 return netdev_get_tx_queue(dev, queue_index);
2054 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2055 struct net_device *dev,
2056 struct netdev_queue *txq)
2058 spinlock_t *root_lock = qdisc_lock(q);
2061 spin_lock(root_lock);
2062 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2065 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2066 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
2068 * This is a work-conserving queue; there are no old skbs
2069 * waiting to be sent out; and the qdisc is not running -
2070 * xmit the skb directly.
2072 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2074 __qdisc_update_bstats(q, skb->len);
2075 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
2078 clear_bit(__QDISC_STATE_RUNNING, &q->state);
2080 rc = NET_XMIT_SUCCESS;
2083 rc = qdisc_enqueue_root(skb, q);
2086 spin_unlock(root_lock);
2092 * Returns true if either:
2093 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2094 * 2. skb is fragmented and the device does not support SG, or if
2095 * at least one of fragments is in highmem and device does not
2096 * support DMA from it.
2098 static inline int skb_needs_linearize(struct sk_buff *skb,
2099 struct net_device *dev)
2101 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
2102 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
2103 illegal_highdma(dev, skb)));
2107 * dev_queue_xmit - transmit a buffer
2108 * @skb: buffer to transmit
2110 * Queue a buffer for transmission to a network device. The caller must
2111 * have set the device and priority and built the buffer before calling
2112 * this function. The function can be called from an interrupt.
2114 * A negative errno code is returned on a failure. A success does not
2115 * guarantee the frame will be transmitted as it may be dropped due
2116 * to congestion or traffic shaping.
2118 * -----------------------------------------------------------------------------------
2119 * I notice this method can also return errors from the queue disciplines,
2120 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2123 * Regardless of the return value, the skb is consumed, so it is currently
2124 * difficult to retry a send to this method. (You can bump the ref count
2125 * before sending to hold a reference for retry if you are careful.)
2127 * When calling this method, interrupts MUST be enabled. This is because
2128 * the BH enable code must have IRQs enabled so that it will not deadlock.
2131 int dev_queue_xmit(struct sk_buff *skb)
2133 struct net_device *dev = skb->dev;
2134 struct netdev_queue *txq;
2138 /* GSO will handle the following emulations directly. */
2139 if (netif_needs_gso(dev, skb))
2142 /* Convert a paged skb to linear, if required */
2143 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2146 /* If packet is not checksummed and device does not support
2147 * checksumming for this protocol, complete checksumming here.
2149 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2150 skb_set_transport_header(skb, skb->csum_start -
2152 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2157 /* Disable soft irqs for various locks below. Also
2158 * stops preemption for RCU.
2162 txq = dev_pick_tx(dev, skb);
2163 q = rcu_dereference_bh(txq->qdisc);
2165 #ifdef CONFIG_NET_CLS_ACT
2166 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2169 rc = __dev_xmit_skb(skb, q, dev, txq);
2173 /* The device has no queue. Common case for software devices:
2174 loopback, all the sorts of tunnels...
2176 Really, it is unlikely that netif_tx_lock protection is necessary
2177 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2179 However, it is possible, that they rely on protection
2182 Check this and shot the lock. It is not prone from deadlocks.
2183 Either shot noqueue qdisc, it is even simpler 8)
2185 if (dev->flags & IFF_UP) {
2186 int cpu = smp_processor_id(); /* ok because BHs are off */
2188 if (txq->xmit_lock_owner != cpu) {
2190 HARD_TX_LOCK(dev, txq, cpu);
2192 if (!netif_tx_queue_stopped(txq)) {
2193 rc = dev_hard_start_xmit(skb, dev, txq);
2194 if (dev_xmit_complete(rc)) {
2195 HARD_TX_UNLOCK(dev, txq);
2199 HARD_TX_UNLOCK(dev, txq);
2200 if (net_ratelimit())
2201 printk(KERN_CRIT "Virtual device %s asks to "
2202 "queue packet!\n", dev->name);
2204 /* Recursion is detected! It is possible,
2206 if (net_ratelimit())
2207 printk(KERN_CRIT "Dead loop on virtual device "
2208 "%s, fix it urgently!\n", dev->name);
2213 rcu_read_unlock_bh();
2219 rcu_read_unlock_bh();
2222 EXPORT_SYMBOL(dev_queue_xmit);
2225 /*=======================================================================
2227 =======================================================================*/
2229 int netdev_max_backlog __read_mostly = 1000;
2230 int netdev_tstamp_prequeue __read_mostly = 1;
2231 int netdev_budget __read_mostly = 300;
2232 int weight_p __read_mostly = 64; /* old backlog weight */
2234 /* Called with irq disabled */
2235 static inline void ____napi_schedule(struct softnet_data *sd,
2236 struct napi_struct *napi)
2238 list_add_tail(&napi->poll_list, &sd->poll_list);
2239 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2244 /* One global table that all flow-based protocols share. */
2245 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2246 EXPORT_SYMBOL(rps_sock_flow_table);
2249 * get_rps_cpu is called from netif_receive_skb and returns the target
2250 * CPU from the RPS map of the receiving queue for a given skb.
2251 * rcu_read_lock must be held on entry.
2253 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2254 struct rps_dev_flow **rflowp)
2256 struct ipv6hdr *ip6;
2258 struct netdev_rx_queue *rxqueue;
2259 struct rps_map *map;
2260 struct rps_dev_flow_table *flow_table;
2261 struct rps_sock_flow_table *sock_flow_table;
2265 u32 addr1, addr2, ihl;
2271 if (skb_rx_queue_recorded(skb)) {
2272 u16 index = skb_get_rx_queue(skb);
2273 if (unlikely(index >= dev->num_rx_queues)) {
2274 WARN_ONCE(dev->num_rx_queues > 1, "%s received packet "
2275 "on queue %u, but number of RX queues is %u\n",
2276 dev->name, index, dev->num_rx_queues);
2279 rxqueue = dev->_rx + index;
2283 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2287 goto got_hash; /* Skip hash computation on packet header */
2289 switch (skb->protocol) {
2290 case __constant_htons(ETH_P_IP):
2291 if (!pskb_may_pull(skb, sizeof(*ip)))
2294 ip = (struct iphdr *) skb->data;
2295 ip_proto = ip->protocol;
2296 addr1 = (__force u32) ip->saddr;
2297 addr2 = (__force u32) ip->daddr;
2300 case __constant_htons(ETH_P_IPV6):
2301 if (!pskb_may_pull(skb, sizeof(*ip6)))
2304 ip6 = (struct ipv6hdr *) skb->data;
2305 ip_proto = ip6->nexthdr;
2306 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2307 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2320 case IPPROTO_UDPLITE:
2321 if (pskb_may_pull(skb, (ihl * 4) + 4)) {
2322 ports.v32 = * (__force u32 *) (skb->data + (ihl * 4));
2323 if (ports.v16[1] < ports.v16[0])
2324 swap(ports.v16[0], ports.v16[1]);
2332 /* get a consistent hash (same value on both flow directions) */
2335 skb->rxhash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2340 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2341 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2342 if (flow_table && sock_flow_table) {
2344 struct rps_dev_flow *rflow;
2346 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2349 next_cpu = sock_flow_table->ents[skb->rxhash &
2350 sock_flow_table->mask];
2353 * If the desired CPU (where last recvmsg was done) is
2354 * different from current CPU (one in the rx-queue flow
2355 * table entry), switch if one of the following holds:
2356 * - Current CPU is unset (equal to RPS_NO_CPU).
2357 * - Current CPU is offline.
2358 * - The current CPU's queue tail has advanced beyond the
2359 * last packet that was enqueued using this table entry.
2360 * This guarantees that all previous packets for the flow
2361 * have been dequeued, thus preserving in order delivery.
2363 if (unlikely(tcpu != next_cpu) &&
2364 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2365 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2366 rflow->last_qtail)) >= 0)) {
2367 tcpu = rflow->cpu = next_cpu;
2368 if (tcpu != RPS_NO_CPU)
2369 rflow->last_qtail = per_cpu(softnet_data,
2370 tcpu).input_queue_head;
2372 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2379 map = rcu_dereference(rxqueue->rps_map);
2381 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2383 if (cpu_online(tcpu)) {
2393 /* Called from hardirq (IPI) context */
2394 static void rps_trigger_softirq(void *data)
2396 struct softnet_data *sd = data;
2398 ____napi_schedule(sd, &sd->backlog);
2402 #endif /* CONFIG_RPS */
2405 * Check if this softnet_data structure is another cpu one
2406 * If yes, queue it to our IPI list and return 1
2409 static int rps_ipi_queued(struct softnet_data *sd)
2412 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2415 sd->rps_ipi_next = mysd->rps_ipi_list;
2416 mysd->rps_ipi_list = sd;
2418 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2421 #endif /* CONFIG_RPS */
2426 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2427 * queue (may be a remote CPU queue).
2429 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2430 unsigned int *qtail)
2432 struct softnet_data *sd;
2433 unsigned long flags;
2435 sd = &per_cpu(softnet_data, cpu);
2437 local_irq_save(flags);
2440 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2441 if (skb_queue_len(&sd->input_pkt_queue)) {
2443 __skb_queue_tail(&sd->input_pkt_queue, skb);
2444 input_queue_tail_incr_save(sd, qtail);
2446 local_irq_restore(flags);
2447 return NET_RX_SUCCESS;
2450 /* Schedule NAPI for backlog device
2451 * We can use non atomic operation since we own the queue lock
2453 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2454 if (!rps_ipi_queued(sd))
2455 ____napi_schedule(sd, &sd->backlog);
2463 local_irq_restore(flags);
2470 * netif_rx - post buffer to the network code
2471 * @skb: buffer to post
2473 * This function receives a packet from a device driver and queues it for
2474 * the upper (protocol) levels to process. It always succeeds. The buffer
2475 * may be dropped during processing for congestion control or by the
2479 * NET_RX_SUCCESS (no congestion)
2480 * NET_RX_DROP (packet was dropped)
2484 int netif_rx(struct sk_buff *skb)
2488 /* if netpoll wants it, pretend we never saw it */
2489 if (netpoll_rx(skb))
2492 if (netdev_tstamp_prequeue)
2493 net_timestamp_check(skb);
2497 struct rps_dev_flow voidflow, *rflow = &voidflow;
2502 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2504 cpu = smp_processor_id();
2506 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2513 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2519 EXPORT_SYMBOL(netif_rx);
2521 int netif_rx_ni(struct sk_buff *skb)
2526 err = netif_rx(skb);
2527 if (local_softirq_pending())
2533 EXPORT_SYMBOL(netif_rx_ni);
2535 static void net_tx_action(struct softirq_action *h)
2537 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2539 if (sd->completion_queue) {
2540 struct sk_buff *clist;
2542 local_irq_disable();
2543 clist = sd->completion_queue;
2544 sd->completion_queue = NULL;
2548 struct sk_buff *skb = clist;
2549 clist = clist->next;
2551 WARN_ON(atomic_read(&skb->users));
2556 if (sd->output_queue) {
2559 local_irq_disable();
2560 head = sd->output_queue;
2561 sd->output_queue = NULL;
2562 sd->output_queue_tailp = &sd->output_queue;
2566 struct Qdisc *q = head;
2567 spinlock_t *root_lock;
2569 head = head->next_sched;
2571 root_lock = qdisc_lock(q);
2572 if (spin_trylock(root_lock)) {
2573 smp_mb__before_clear_bit();
2574 clear_bit(__QDISC_STATE_SCHED,
2577 spin_unlock(root_lock);
2579 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2581 __netif_reschedule(q);
2583 smp_mb__before_clear_bit();
2584 clear_bit(__QDISC_STATE_SCHED,
2592 static inline int deliver_skb(struct sk_buff *skb,
2593 struct packet_type *pt_prev,
2594 struct net_device *orig_dev)
2596 atomic_inc(&skb->users);
2597 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2600 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2602 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2603 /* This hook is defined here for ATM LANE */
2604 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2605 unsigned char *addr) __read_mostly;
2606 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2610 * If bridge module is loaded call bridging hook.
2611 * returns NULL if packet was consumed.
2613 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2614 struct sk_buff *skb) __read_mostly;
2615 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2617 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2618 struct packet_type **pt_prev, int *ret,
2619 struct net_device *orig_dev)
2621 struct net_bridge_port *port;
2623 if (skb->pkt_type == PACKET_LOOPBACK ||
2624 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2628 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2632 return br_handle_frame_hook(port, skb);
2635 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2638 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2639 struct sk_buff *(*macvlan_handle_frame_hook)(struct macvlan_port *p,
2640 struct sk_buff *skb) __read_mostly;
2641 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2643 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2644 struct packet_type **pt_prev,
2646 struct net_device *orig_dev)
2648 struct macvlan_port *port;
2650 port = rcu_dereference(skb->dev->macvlan_port);
2655 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2658 return macvlan_handle_frame_hook(port, skb);
2661 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2664 #ifdef CONFIG_NET_CLS_ACT
2665 /* TODO: Maybe we should just force sch_ingress to be compiled in
2666 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2667 * a compare and 2 stores extra right now if we dont have it on
2668 * but have CONFIG_NET_CLS_ACT
2669 * NOTE: This doesnt stop any functionality; if you dont have
2670 * the ingress scheduler, you just cant add policies on ingress.
2673 static int ing_filter(struct sk_buff *skb)
2675 struct net_device *dev = skb->dev;
2676 u32 ttl = G_TC_RTTL(skb->tc_verd);
2677 struct netdev_queue *rxq;
2678 int result = TC_ACT_OK;
2681 if (MAX_RED_LOOP < ttl++) {
2683 "Redir loop detected Dropping packet (%d->%d)\n",
2684 skb->skb_iif, dev->ifindex);
2688 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2689 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2691 rxq = &dev->rx_queue;
2694 if (q != &noop_qdisc) {
2695 spin_lock(qdisc_lock(q));
2696 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2697 result = qdisc_enqueue_root(skb, q);
2698 spin_unlock(qdisc_lock(q));
2704 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2705 struct packet_type **pt_prev,
2706 int *ret, struct net_device *orig_dev)
2708 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2712 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2715 /* Huh? Why does turning on AF_PACKET affect this? */
2716 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2719 switch (ing_filter(skb)) {
2733 * netif_nit_deliver - deliver received packets to network taps
2736 * This function is used to deliver incoming packets to network
2737 * taps. It should be used when the normal netif_receive_skb path
2738 * is bypassed, for example because of VLAN acceleration.
2740 void netif_nit_deliver(struct sk_buff *skb)
2742 struct packet_type *ptype;
2744 if (list_empty(&ptype_all))
2747 skb_reset_network_header(skb);
2748 skb_reset_transport_header(skb);
2749 skb->mac_len = skb->network_header - skb->mac_header;
2752 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2753 if (!ptype->dev || ptype->dev == skb->dev)
2754 deliver_skb(skb, ptype, skb->dev);
2759 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2760 struct net_device *master)
2762 if (skb->pkt_type == PACKET_HOST) {
2763 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2765 memcpy(dest, master->dev_addr, ETH_ALEN);
2769 /* On bonding slaves other than the currently active slave, suppress
2770 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2771 * ARP on active-backup slaves with arp_validate enabled.
2773 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2775 struct net_device *dev = skb->dev;
2777 if (master->priv_flags & IFF_MASTER_ARPMON)
2778 dev->last_rx = jiffies;
2780 if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) {
2781 /* Do address unmangle. The local destination address
2782 * will be always the one master has. Provides the right
2783 * functionality in a bridge.
2785 skb_bond_set_mac_by_master(skb, master);
2788 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2789 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2790 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2793 if (master->priv_flags & IFF_MASTER_ALB) {
2794 if (skb->pkt_type != PACKET_BROADCAST &&
2795 skb->pkt_type != PACKET_MULTICAST)
2798 if (master->priv_flags & IFF_MASTER_8023AD &&
2799 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2806 EXPORT_SYMBOL(__skb_bond_should_drop);
2808 static int __netif_receive_skb(struct sk_buff *skb)
2810 struct packet_type *ptype, *pt_prev;
2811 struct net_device *orig_dev;
2812 struct net_device *master;
2813 struct net_device *null_or_orig;
2814 struct net_device *orig_or_bond;
2815 int ret = NET_RX_DROP;
2818 if (!netdev_tstamp_prequeue)
2819 net_timestamp_check(skb);
2821 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2822 return NET_RX_SUCCESS;
2824 /* if we've gotten here through NAPI, check netpoll */
2825 if (netpoll_receive_skb(skb))
2829 skb->skb_iif = skb->dev->ifindex;
2832 * bonding note: skbs received on inactive slaves should only
2833 * be delivered to pkt handlers that are exact matches. Also
2834 * the deliver_no_wcard flag will be set. If packet handlers
2835 * are sensitive to duplicate packets these skbs will need to
2836 * be dropped at the handler. The vlan accel path may have
2837 * already set the deliver_no_wcard flag.
2839 null_or_orig = NULL;
2840 orig_dev = skb->dev;
2841 master = ACCESS_ONCE(orig_dev->master);
2842 if (skb->deliver_no_wcard)
2843 null_or_orig = orig_dev;
2845 if (skb_bond_should_drop(skb, master)) {
2846 skb->deliver_no_wcard = 1;
2847 null_or_orig = orig_dev; /* deliver only exact match */
2852 __get_cpu_var(softnet_data).processed++;
2854 skb_reset_network_header(skb);
2855 skb_reset_transport_header(skb);
2856 skb->mac_len = skb->network_header - skb->mac_header;
2862 #ifdef CONFIG_NET_CLS_ACT
2863 if (skb->tc_verd & TC_NCLS) {
2864 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2869 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2870 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2871 ptype->dev == orig_dev) {
2873 ret = deliver_skb(skb, pt_prev, orig_dev);
2878 #ifdef CONFIG_NET_CLS_ACT
2879 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2885 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2888 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2893 * Make sure frames received on VLAN interfaces stacked on
2894 * bonding interfaces still make their way to any base bonding
2895 * device that may have registered for a specific ptype. The
2896 * handler may have to adjust skb->dev and orig_dev.
2898 orig_or_bond = orig_dev;
2899 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2900 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2901 orig_or_bond = vlan_dev_real_dev(skb->dev);
2904 type = skb->protocol;
2905 list_for_each_entry_rcu(ptype,
2906 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2907 if (ptype->type == type && (ptype->dev == null_or_orig ||
2908 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2909 ptype->dev == orig_or_bond)) {
2911 ret = deliver_skb(skb, pt_prev, orig_dev);
2917 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2920 /* Jamal, now you will not able to escape explaining
2921 * me how you were going to use this. :-)
2932 * netif_receive_skb - process receive buffer from network
2933 * @skb: buffer to process
2935 * netif_receive_skb() is the main receive data processing function.
2936 * It always succeeds. The buffer may be dropped during processing
2937 * for congestion control or by the protocol layers.
2939 * This function may only be called from softirq context and interrupts
2940 * should be enabled.
2942 * Return values (usually ignored):
2943 * NET_RX_SUCCESS: no congestion
2944 * NET_RX_DROP: packet was dropped
2946 int netif_receive_skb(struct sk_buff *skb)
2948 if (netdev_tstamp_prequeue)
2949 net_timestamp_check(skb);
2953 struct rps_dev_flow voidflow, *rflow = &voidflow;
2958 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2961 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2965 ret = __netif_receive_skb(skb);
2971 return __netif_receive_skb(skb);
2974 EXPORT_SYMBOL(netif_receive_skb);
2976 /* Network device is going away, flush any packets still pending
2977 * Called with irqs disabled.
2979 static void flush_backlog(void *arg)
2981 struct net_device *dev = arg;
2982 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2983 struct sk_buff *skb, *tmp;
2986 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
2987 if (skb->dev == dev) {
2988 __skb_unlink(skb, &sd->input_pkt_queue);
2990 input_queue_head_incr(sd);
2995 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
2996 if (skb->dev == dev) {
2997 __skb_unlink(skb, &sd->process_queue);
2999 input_queue_head_incr(sd);
3004 static int napi_gro_complete(struct sk_buff *skb)
3006 struct packet_type *ptype;
3007 __be16 type = skb->protocol;
3008 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3011 if (NAPI_GRO_CB(skb)->count == 1) {
3012 skb_shinfo(skb)->gso_size = 0;
3017 list_for_each_entry_rcu(ptype, head, list) {
3018 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3021 err = ptype->gro_complete(skb);
3027 WARN_ON(&ptype->list == head);
3029 return NET_RX_SUCCESS;
3033 return netif_receive_skb(skb);
3036 static void napi_gro_flush(struct napi_struct *napi)
3038 struct sk_buff *skb, *next;
3040 for (skb = napi->gro_list; skb; skb = next) {
3043 napi_gro_complete(skb);
3046 napi->gro_count = 0;
3047 napi->gro_list = NULL;
3050 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3052 struct sk_buff **pp = NULL;
3053 struct packet_type *ptype;
3054 __be16 type = skb->protocol;
3055 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3058 enum gro_result ret;
3060 if (!(skb->dev->features & NETIF_F_GRO))
3063 if (skb_is_gso(skb) || skb_has_frags(skb))
3067 list_for_each_entry_rcu(ptype, head, list) {
3068 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3071 skb_set_network_header(skb, skb_gro_offset(skb));
3072 mac_len = skb->network_header - skb->mac_header;
3073 skb->mac_len = mac_len;
3074 NAPI_GRO_CB(skb)->same_flow = 0;
3075 NAPI_GRO_CB(skb)->flush = 0;
3076 NAPI_GRO_CB(skb)->free = 0;
3078 pp = ptype->gro_receive(&napi->gro_list, skb);
3083 if (&ptype->list == head)
3086 same_flow = NAPI_GRO_CB(skb)->same_flow;
3087 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3090 struct sk_buff *nskb = *pp;
3094 napi_gro_complete(nskb);
3101 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3105 NAPI_GRO_CB(skb)->count = 1;
3106 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3107 skb->next = napi->gro_list;
3108 napi->gro_list = skb;
3112 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3113 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3115 BUG_ON(skb->end - skb->tail < grow);
3117 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3120 skb->data_len -= grow;
3122 skb_shinfo(skb)->frags[0].page_offset += grow;
3123 skb_shinfo(skb)->frags[0].size -= grow;
3125 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3126 put_page(skb_shinfo(skb)->frags[0].page);
3127 memmove(skb_shinfo(skb)->frags,
3128 skb_shinfo(skb)->frags + 1,
3129 --skb_shinfo(skb)->nr_frags);
3140 EXPORT_SYMBOL(dev_gro_receive);
3143 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3147 if (netpoll_rx_on(skb))
3150 for (p = napi->gro_list; p; p = p->next) {
3151 NAPI_GRO_CB(p)->same_flow =
3152 (p->dev == skb->dev) &&
3153 !compare_ether_header(skb_mac_header(p),
3154 skb_gro_mac_header(skb));
3155 NAPI_GRO_CB(p)->flush = 0;
3158 return dev_gro_receive(napi, skb);
3161 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3165 if (netif_receive_skb(skb))
3170 case GRO_MERGED_FREE:
3181 EXPORT_SYMBOL(napi_skb_finish);
3183 void skb_gro_reset_offset(struct sk_buff *skb)
3185 NAPI_GRO_CB(skb)->data_offset = 0;
3186 NAPI_GRO_CB(skb)->frag0 = NULL;
3187 NAPI_GRO_CB(skb)->frag0_len = 0;
3189 if (skb->mac_header == skb->tail &&
3190 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3191 NAPI_GRO_CB(skb)->frag0 =
3192 page_address(skb_shinfo(skb)->frags[0].page) +
3193 skb_shinfo(skb)->frags[0].page_offset;
3194 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3197 EXPORT_SYMBOL(skb_gro_reset_offset);
3199 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3201 skb_gro_reset_offset(skb);
3203 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3205 EXPORT_SYMBOL(napi_gro_receive);
3207 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3209 __skb_pull(skb, skb_headlen(skb));
3210 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3214 EXPORT_SYMBOL(napi_reuse_skb);
3216 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3218 struct sk_buff *skb = napi->skb;
3221 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3227 EXPORT_SYMBOL(napi_get_frags);
3229 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3235 skb->protocol = eth_type_trans(skb, skb->dev);
3237 if (ret == GRO_HELD)
3238 skb_gro_pull(skb, -ETH_HLEN);
3239 else if (netif_receive_skb(skb))
3244 case GRO_MERGED_FREE:
3245 napi_reuse_skb(napi, skb);
3254 EXPORT_SYMBOL(napi_frags_finish);
3256 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3258 struct sk_buff *skb = napi->skb;
3265 skb_reset_mac_header(skb);
3266 skb_gro_reset_offset(skb);
3268 off = skb_gro_offset(skb);
3269 hlen = off + sizeof(*eth);
3270 eth = skb_gro_header_fast(skb, off);
3271 if (skb_gro_header_hard(skb, hlen)) {
3272 eth = skb_gro_header_slow(skb, hlen, off);
3273 if (unlikely(!eth)) {
3274 napi_reuse_skb(napi, skb);
3280 skb_gro_pull(skb, sizeof(*eth));
3283 * This works because the only protocols we care about don't require
3284 * special handling. We'll fix it up properly at the end.
3286 skb->protocol = eth->h_proto;
3291 EXPORT_SYMBOL(napi_frags_skb);
3293 gro_result_t napi_gro_frags(struct napi_struct *napi)
3295 struct sk_buff *skb = napi_frags_skb(napi);
3300 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3302 EXPORT_SYMBOL(napi_gro_frags);
3305 * net_rps_action sends any pending IPI's for rps.
3306 * Note: called with local irq disabled, but exits with local irq enabled.
3308 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3311 struct softnet_data *remsd = sd->rps_ipi_list;
3314 sd->rps_ipi_list = NULL;
3318 /* Send pending IPI's to kick RPS processing on remote cpus. */
3320 struct softnet_data *next = remsd->rps_ipi_next;
3322 if (cpu_online(remsd->cpu))
3323 __smp_call_function_single(remsd->cpu,
3332 static int process_backlog(struct napi_struct *napi, int quota)
3335 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3338 /* Check if we have pending ipi, its better to send them now,
3339 * not waiting net_rx_action() end.
3341 if (sd->rps_ipi_list) {
3342 local_irq_disable();
3343 net_rps_action_and_irq_enable(sd);
3346 napi->weight = weight_p;
3347 local_irq_disable();
3348 while (work < quota) {
3349 struct sk_buff *skb;
3352 while ((skb = __skb_dequeue(&sd->process_queue))) {
3354 __netif_receive_skb(skb);
3355 local_irq_disable();
3356 input_queue_head_incr(sd);
3357 if (++work >= quota) {
3364 qlen = skb_queue_len(&sd->input_pkt_queue);
3366 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3367 &sd->process_queue);
3369 if (qlen < quota - work) {
3371 * Inline a custom version of __napi_complete().
3372 * only current cpu owns and manipulates this napi,
3373 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3374 * we can use a plain write instead of clear_bit(),
3375 * and we dont need an smp_mb() memory barrier.
3377 list_del(&napi->poll_list);
3380 quota = work + qlen;
3390 * __napi_schedule - schedule for receive
3391 * @n: entry to schedule
3393 * The entry's receive function will be scheduled to run
3395 void __napi_schedule(struct napi_struct *n)
3397 unsigned long flags;
3399 local_irq_save(flags);
3400 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3401 local_irq_restore(flags);
3403 EXPORT_SYMBOL(__napi_schedule);
3405 void __napi_complete(struct napi_struct *n)
3407 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3408 BUG_ON(n->gro_list);
3410 list_del(&n->poll_list);
3411 smp_mb__before_clear_bit();
3412 clear_bit(NAPI_STATE_SCHED, &n->state);
3414 EXPORT_SYMBOL(__napi_complete);
3416 void napi_complete(struct napi_struct *n)
3418 unsigned long flags;
3421 * don't let napi dequeue from the cpu poll list
3422 * just in case its running on a different cpu
3424 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3428 local_irq_save(flags);
3430 local_irq_restore(flags);
3432 EXPORT_SYMBOL(napi_complete);
3434 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3435 int (*poll)(struct napi_struct *, int), int weight)
3437 INIT_LIST_HEAD(&napi->poll_list);
3438 napi->gro_count = 0;
3439 napi->gro_list = NULL;
3442 napi->weight = weight;
3443 list_add(&napi->dev_list, &dev->napi_list);
3445 #ifdef CONFIG_NETPOLL
3446 spin_lock_init(&napi->poll_lock);
3447 napi->poll_owner = -1;
3449 set_bit(NAPI_STATE_SCHED, &napi->state);
3451 EXPORT_SYMBOL(netif_napi_add);
3453 void netif_napi_del(struct napi_struct *napi)
3455 struct sk_buff *skb, *next;
3457 list_del_init(&napi->dev_list);
3458 napi_free_frags(napi);
3460 for (skb = napi->gro_list; skb; skb = next) {
3466 napi->gro_list = NULL;
3467 napi->gro_count = 0;
3469 EXPORT_SYMBOL(netif_napi_del);
3471 static void net_rx_action(struct softirq_action *h)
3473 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3474 unsigned long time_limit = jiffies + 2;
3475 int budget = netdev_budget;
3478 local_irq_disable();
3480 while (!list_empty(&sd->poll_list)) {
3481 struct napi_struct *n;
3484 /* If softirq window is exhuasted then punt.
3485 * Allow this to run for 2 jiffies since which will allow
3486 * an average latency of 1.5/HZ.
3488 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3493 /* Even though interrupts have been re-enabled, this
3494 * access is safe because interrupts can only add new
3495 * entries to the tail of this list, and only ->poll()
3496 * calls can remove this head entry from the list.
3498 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3500 have = netpoll_poll_lock(n);
3504 /* This NAPI_STATE_SCHED test is for avoiding a race
3505 * with netpoll's poll_napi(). Only the entity which
3506 * obtains the lock and sees NAPI_STATE_SCHED set will
3507 * actually make the ->poll() call. Therefore we avoid
3508 * accidently calling ->poll() when NAPI is not scheduled.
3511 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3512 work = n->poll(n, weight);
3516 WARN_ON_ONCE(work > weight);
3520 local_irq_disable();
3522 /* Drivers must not modify the NAPI state if they
3523 * consume the entire weight. In such cases this code
3524 * still "owns" the NAPI instance and therefore can
3525 * move the instance around on the list at-will.
3527 if (unlikely(work == weight)) {
3528 if (unlikely(napi_disable_pending(n))) {
3531 local_irq_disable();
3533 list_move_tail(&n->poll_list, &sd->poll_list);
3536 netpoll_poll_unlock(have);
3539 net_rps_action_and_irq_enable(sd);
3541 #ifdef CONFIG_NET_DMA
3543 * There may not be any more sk_buffs coming right now, so push
3544 * any pending DMA copies to hardware
3546 dma_issue_pending_all();
3553 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3557 static gifconf_func_t *gifconf_list[NPROTO];
3560 * register_gifconf - register a SIOCGIF handler
3561 * @family: Address family
3562 * @gifconf: Function handler
3564 * Register protocol dependent address dumping routines. The handler
3565 * that is passed must not be freed or reused until it has been replaced
3566 * by another handler.
3568 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3570 if (family >= NPROTO)
3572 gifconf_list[family] = gifconf;
3575 EXPORT_SYMBOL(register_gifconf);
3579 * Map an interface index to its name (SIOCGIFNAME)
3583 * We need this ioctl for efficient implementation of the
3584 * if_indextoname() function required by the IPv6 API. Without
3585 * it, we would have to search all the interfaces to find a
3589 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3591 struct net_device *dev;
3595 * Fetch the caller's info block.
3598 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3602 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3608 strcpy(ifr.ifr_name, dev->name);
3611 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3617 * Perform a SIOCGIFCONF call. This structure will change
3618 * size eventually, and there is nothing I can do about it.
3619 * Thus we will need a 'compatibility mode'.
3622 static int dev_ifconf(struct net *net, char __user *arg)
3625 struct net_device *dev;
3632 * Fetch the caller's info block.
3635 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3642 * Loop over the interfaces, and write an info block for each.
3646 for_each_netdev(net, dev) {
3647 for (i = 0; i < NPROTO; i++) {
3648 if (gifconf_list[i]) {
3651 done = gifconf_list[i](dev, NULL, 0);
3653 done = gifconf_list[i](dev, pos + total,
3663 * All done. Write the updated control block back to the caller.
3665 ifc.ifc_len = total;
3668 * Both BSD and Solaris return 0 here, so we do too.
3670 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3673 #ifdef CONFIG_PROC_FS
3675 * This is invoked by the /proc filesystem handler to display a device
3678 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3681 struct net *net = seq_file_net(seq);
3683 struct net_device *dev;
3687 return SEQ_START_TOKEN;
3690 for_each_netdev_rcu(net, dev)
3697 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3699 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3700 first_net_device(seq_file_net(seq)) :
3701 next_net_device((struct net_device *)v);
3704 return rcu_dereference(dev);
3707 void dev_seq_stop(struct seq_file *seq, void *v)
3713 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3715 const struct net_device_stats *stats = dev_get_stats(dev);
3717 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3718 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3719 dev->name, stats->rx_bytes, stats->rx_packets,
3721 stats->rx_dropped + stats->rx_missed_errors,
3722 stats->rx_fifo_errors,
3723 stats->rx_length_errors + stats->rx_over_errors +
3724 stats->rx_crc_errors + stats->rx_frame_errors,
3725 stats->rx_compressed, stats->multicast,
3726 stats->tx_bytes, stats->tx_packets,
3727 stats->tx_errors, stats->tx_dropped,
3728 stats->tx_fifo_errors, stats->collisions,
3729 stats->tx_carrier_errors +
3730 stats->tx_aborted_errors +
3731 stats->tx_window_errors +
3732 stats->tx_heartbeat_errors,
3733 stats->tx_compressed);
3737 * Called from the PROCfs module. This now uses the new arbitrary sized
3738 * /proc/net interface to create /proc/net/dev
3740 static int dev_seq_show(struct seq_file *seq, void *v)
3742 if (v == SEQ_START_TOKEN)
3743 seq_puts(seq, "Inter-| Receive "
3745 " face |bytes packets errs drop fifo frame "
3746 "compressed multicast|bytes packets errs "
3747 "drop fifo colls carrier compressed\n");
3749 dev_seq_printf_stats(seq, v);
3753 static struct softnet_data *softnet_get_online(loff_t *pos)
3755 struct softnet_data *sd = NULL;
3757 while (*pos < nr_cpu_ids)
3758 if (cpu_online(*pos)) {
3759 sd = &per_cpu(softnet_data, *pos);
3766 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3768 return softnet_get_online(pos);
3771 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3774 return softnet_get_online(pos);
3777 static void softnet_seq_stop(struct seq_file *seq, void *v)
3781 static int softnet_seq_show(struct seq_file *seq, void *v)
3783 struct softnet_data *sd = v;
3785 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3786 sd->processed, sd->dropped, sd->time_squeeze, 0,
3787 0, 0, 0, 0, /* was fastroute */
3788 sd->cpu_collision, sd->received_rps);
3792 static const struct seq_operations dev_seq_ops = {
3793 .start = dev_seq_start,
3794 .next = dev_seq_next,
3795 .stop = dev_seq_stop,
3796 .show = dev_seq_show,
3799 static int dev_seq_open(struct inode *inode, struct file *file)
3801 return seq_open_net(inode, file, &dev_seq_ops,
3802 sizeof(struct seq_net_private));
3805 static const struct file_operations dev_seq_fops = {
3806 .owner = THIS_MODULE,
3807 .open = dev_seq_open,
3809 .llseek = seq_lseek,
3810 .release = seq_release_net,
3813 static const struct seq_operations softnet_seq_ops = {
3814 .start = softnet_seq_start,
3815 .next = softnet_seq_next,
3816 .stop = softnet_seq_stop,
3817 .show = softnet_seq_show,
3820 static int softnet_seq_open(struct inode *inode, struct file *file)
3822 return seq_open(file, &softnet_seq_ops);
3825 static const struct file_operations softnet_seq_fops = {
3826 .owner = THIS_MODULE,
3827 .open = softnet_seq_open,
3829 .llseek = seq_lseek,
3830 .release = seq_release,
3833 static void *ptype_get_idx(loff_t pos)
3835 struct packet_type *pt = NULL;
3839 list_for_each_entry_rcu(pt, &ptype_all, list) {
3845 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3846 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3855 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3859 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3862 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3864 struct packet_type *pt;
3865 struct list_head *nxt;
3869 if (v == SEQ_START_TOKEN)
3870 return ptype_get_idx(0);
3873 nxt = pt->list.next;
3874 if (pt->type == htons(ETH_P_ALL)) {
3875 if (nxt != &ptype_all)
3878 nxt = ptype_base[0].next;
3880 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3882 while (nxt == &ptype_base[hash]) {
3883 if (++hash >= PTYPE_HASH_SIZE)
3885 nxt = ptype_base[hash].next;
3888 return list_entry(nxt, struct packet_type, list);
3891 static void ptype_seq_stop(struct seq_file *seq, void *v)
3897 static int ptype_seq_show(struct seq_file *seq, void *v)
3899 struct packet_type *pt = v;
3901 if (v == SEQ_START_TOKEN)
3902 seq_puts(seq, "Type Device Function\n");
3903 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3904 if (pt->type == htons(ETH_P_ALL))
3905 seq_puts(seq, "ALL ");
3907 seq_printf(seq, "%04x", ntohs(pt->type));
3909 seq_printf(seq, " %-8s %pF\n",
3910 pt->dev ? pt->dev->name : "", pt->func);
3916 static const struct seq_operations ptype_seq_ops = {
3917 .start = ptype_seq_start,
3918 .next = ptype_seq_next,
3919 .stop = ptype_seq_stop,
3920 .show = ptype_seq_show,
3923 static int ptype_seq_open(struct inode *inode, struct file *file)
3925 return seq_open_net(inode, file, &ptype_seq_ops,
3926 sizeof(struct seq_net_private));
3929 static const struct file_operations ptype_seq_fops = {
3930 .owner = THIS_MODULE,
3931 .open = ptype_seq_open,
3933 .llseek = seq_lseek,
3934 .release = seq_release_net,
3938 static int __net_init dev_proc_net_init(struct net *net)
3942 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3944 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3946 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3949 if (wext_proc_init(net))
3955 proc_net_remove(net, "ptype");
3957 proc_net_remove(net, "softnet_stat");
3959 proc_net_remove(net, "dev");
3963 static void __net_exit dev_proc_net_exit(struct net *net)
3965 wext_proc_exit(net);
3967 proc_net_remove(net, "ptype");
3968 proc_net_remove(net, "softnet_stat");
3969 proc_net_remove(net, "dev");
3972 static struct pernet_operations __net_initdata dev_proc_ops = {
3973 .init = dev_proc_net_init,
3974 .exit = dev_proc_net_exit,
3977 static int __init dev_proc_init(void)
3979 return register_pernet_subsys(&dev_proc_ops);
3982 #define dev_proc_init() 0
3983 #endif /* CONFIG_PROC_FS */
3987 * netdev_set_master - set up master/slave pair
3988 * @slave: slave device
3989 * @master: new master device
3991 * Changes the master device of the slave. Pass %NULL to break the
3992 * bonding. The caller must hold the RTNL semaphore. On a failure
3993 * a negative errno code is returned. On success the reference counts
3994 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3995 * function returns zero.
3997 int netdev_set_master(struct net_device *slave, struct net_device *master)
3999 struct net_device *old = slave->master;
4009 slave->master = master;
4016 slave->flags |= IFF_SLAVE;
4018 slave->flags &= ~IFF_SLAVE;
4020 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4023 EXPORT_SYMBOL(netdev_set_master);
4025 static void dev_change_rx_flags(struct net_device *dev, int flags)
4027 const struct net_device_ops *ops = dev->netdev_ops;
4029 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4030 ops->ndo_change_rx_flags(dev, flags);
4033 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4035 unsigned short old_flags = dev->flags;
4041 dev->flags |= IFF_PROMISC;
4042 dev->promiscuity += inc;
4043 if (dev->promiscuity == 0) {
4046 * If inc causes overflow, untouch promisc and return error.
4049 dev->flags &= ~IFF_PROMISC;
4051 dev->promiscuity -= inc;
4052 printk(KERN_WARNING "%s: promiscuity touches roof, "
4053 "set promiscuity failed, promiscuity feature "
4054 "of device might be broken.\n", dev->name);
4058 if (dev->flags != old_flags) {
4059 printk(KERN_INFO "device %s %s promiscuous mode\n",
4060 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4062 if (audit_enabled) {
4063 current_uid_gid(&uid, &gid);
4064 audit_log(current->audit_context, GFP_ATOMIC,
4065 AUDIT_ANOM_PROMISCUOUS,
4066 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4067 dev->name, (dev->flags & IFF_PROMISC),
4068 (old_flags & IFF_PROMISC),
4069 audit_get_loginuid(current),
4071 audit_get_sessionid(current));
4074 dev_change_rx_flags(dev, IFF_PROMISC);
4080 * dev_set_promiscuity - update promiscuity count on a device
4084 * Add or remove promiscuity from a device. While the count in the device
4085 * remains above zero the interface remains promiscuous. Once it hits zero
4086 * the device reverts back to normal filtering operation. A negative inc
4087 * value is used to drop promiscuity on the device.
4088 * Return 0 if successful or a negative errno code on error.
4090 int dev_set_promiscuity(struct net_device *dev, int inc)
4092 unsigned short old_flags = dev->flags;
4095 err = __dev_set_promiscuity(dev, inc);
4098 if (dev->flags != old_flags)
4099 dev_set_rx_mode(dev);
4102 EXPORT_SYMBOL(dev_set_promiscuity);
4105 * dev_set_allmulti - update allmulti count on a device
4109 * Add or remove reception of all multicast frames to a device. While the
4110 * count in the device remains above zero the interface remains listening
4111 * to all interfaces. Once it hits zero the device reverts back to normal
4112 * filtering operation. A negative @inc value is used to drop the counter
4113 * when releasing a resource needing all multicasts.
4114 * Return 0 if successful or a negative errno code on error.
4117 int dev_set_allmulti(struct net_device *dev, int inc)
4119 unsigned short old_flags = dev->flags;
4123 dev->flags |= IFF_ALLMULTI;
4124 dev->allmulti += inc;
4125 if (dev->allmulti == 0) {
4128 * If inc causes overflow, untouch allmulti and return error.
4131 dev->flags &= ~IFF_ALLMULTI;
4133 dev->allmulti -= inc;
4134 printk(KERN_WARNING "%s: allmulti touches roof, "
4135 "set allmulti failed, allmulti feature of "
4136 "device might be broken.\n", dev->name);
4140 if (dev->flags ^ old_flags) {
4141 dev_change_rx_flags(dev, IFF_ALLMULTI);
4142 dev_set_rx_mode(dev);
4146 EXPORT_SYMBOL(dev_set_allmulti);
4149 * Upload unicast and multicast address lists to device and
4150 * configure RX filtering. When the device doesn't support unicast
4151 * filtering it is put in promiscuous mode while unicast addresses
4154 void __dev_set_rx_mode(struct net_device *dev)
4156 const struct net_device_ops *ops = dev->netdev_ops;
4158 /* dev_open will call this function so the list will stay sane. */
4159 if (!(dev->flags&IFF_UP))
4162 if (!netif_device_present(dev))
4165 if (ops->ndo_set_rx_mode)
4166 ops->ndo_set_rx_mode(dev);
4168 /* Unicast addresses changes may only happen under the rtnl,
4169 * therefore calling __dev_set_promiscuity here is safe.
4171 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4172 __dev_set_promiscuity(dev, 1);
4173 dev->uc_promisc = 1;
4174 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4175 __dev_set_promiscuity(dev, -1);
4176 dev->uc_promisc = 0;
4179 if (ops->ndo_set_multicast_list)
4180 ops->ndo_set_multicast_list(dev);
4184 void dev_set_rx_mode(struct net_device *dev)
4186 netif_addr_lock_bh(dev);
4187 __dev_set_rx_mode(dev);
4188 netif_addr_unlock_bh(dev);
4192 * dev_get_flags - get flags reported to userspace
4195 * Get the combination of flag bits exported through APIs to userspace.
4197 unsigned dev_get_flags(const struct net_device *dev)
4201 flags = (dev->flags & ~(IFF_PROMISC |
4206 (dev->gflags & (IFF_PROMISC |
4209 if (netif_running(dev)) {
4210 if (netif_oper_up(dev))
4211 flags |= IFF_RUNNING;
4212 if (netif_carrier_ok(dev))
4213 flags |= IFF_LOWER_UP;
4214 if (netif_dormant(dev))
4215 flags |= IFF_DORMANT;
4220 EXPORT_SYMBOL(dev_get_flags);
4222 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4224 int old_flags = dev->flags;
4230 * Set the flags on our device.
4233 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4234 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4236 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4240 * Load in the correct multicast list now the flags have changed.
4243 if ((old_flags ^ flags) & IFF_MULTICAST)
4244 dev_change_rx_flags(dev, IFF_MULTICAST);
4246 dev_set_rx_mode(dev);
4249 * Have we downed the interface. We handle IFF_UP ourselves
4250 * according to user attempts to set it, rather than blindly
4255 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4256 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4259 dev_set_rx_mode(dev);
4262 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4263 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4265 dev->gflags ^= IFF_PROMISC;
4266 dev_set_promiscuity(dev, inc);
4269 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4270 is important. Some (broken) drivers set IFF_PROMISC, when
4271 IFF_ALLMULTI is requested not asking us and not reporting.
4273 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4274 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4276 dev->gflags ^= IFF_ALLMULTI;
4277 dev_set_allmulti(dev, inc);
4283 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4285 unsigned int changes = dev->flags ^ old_flags;
4287 if (changes & IFF_UP) {
4288 if (dev->flags & IFF_UP)
4289 call_netdevice_notifiers(NETDEV_UP, dev);
4291 call_netdevice_notifiers(NETDEV_DOWN, dev);
4294 if (dev->flags & IFF_UP &&
4295 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4296 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4300 * dev_change_flags - change device settings
4302 * @flags: device state flags
4304 * Change settings on device based state flags. The flags are
4305 * in the userspace exported format.
4307 int dev_change_flags(struct net_device *dev, unsigned flags)
4310 int old_flags = dev->flags;
4312 ret = __dev_change_flags(dev, flags);
4316 changes = old_flags ^ dev->flags;
4318 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4320 __dev_notify_flags(dev, old_flags);
4323 EXPORT_SYMBOL(dev_change_flags);
4326 * dev_set_mtu - Change maximum transfer unit
4328 * @new_mtu: new transfer unit
4330 * Change the maximum transfer size of the network device.
4332 int dev_set_mtu(struct net_device *dev, int new_mtu)
4334 const struct net_device_ops *ops = dev->netdev_ops;
4337 if (new_mtu == dev->mtu)
4340 /* MTU must be positive. */
4344 if (!netif_device_present(dev))
4348 if (ops->ndo_change_mtu)
4349 err = ops->ndo_change_mtu(dev, new_mtu);
4353 if (!err && dev->flags & IFF_UP)
4354 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4357 EXPORT_SYMBOL(dev_set_mtu);
4360 * dev_set_mac_address - Change Media Access Control Address
4364 * Change the hardware (MAC) address of the device
4366 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4368 const struct net_device_ops *ops = dev->netdev_ops;
4371 if (!ops->ndo_set_mac_address)
4373 if (sa->sa_family != dev->type)
4375 if (!netif_device_present(dev))
4377 err = ops->ndo_set_mac_address(dev, sa);
4379 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4382 EXPORT_SYMBOL(dev_set_mac_address);
4385 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4387 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4390 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4396 case SIOCGIFFLAGS: /* Get interface flags */
4397 ifr->ifr_flags = (short) dev_get_flags(dev);
4400 case SIOCGIFMETRIC: /* Get the metric on the interface
4401 (currently unused) */
4402 ifr->ifr_metric = 0;
4405 case SIOCGIFMTU: /* Get the MTU of a device */
4406 ifr->ifr_mtu = dev->mtu;
4411 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4413 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4414 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4415 ifr->ifr_hwaddr.sa_family = dev->type;
4423 ifr->ifr_map.mem_start = dev->mem_start;
4424 ifr->ifr_map.mem_end = dev->mem_end;
4425 ifr->ifr_map.base_addr = dev->base_addr;
4426 ifr->ifr_map.irq = dev->irq;
4427 ifr->ifr_map.dma = dev->dma;
4428 ifr->ifr_map.port = dev->if_port;
4432 ifr->ifr_ifindex = dev->ifindex;
4436 ifr->ifr_qlen = dev->tx_queue_len;
4440 /* dev_ioctl() should ensure this case
4452 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4454 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4457 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4458 const struct net_device_ops *ops;
4463 ops = dev->netdev_ops;
4466 case SIOCSIFFLAGS: /* Set interface flags */
4467 return dev_change_flags(dev, ifr->ifr_flags);
4469 case SIOCSIFMETRIC: /* Set the metric on the interface
4470 (currently unused) */
4473 case SIOCSIFMTU: /* Set the MTU of a device */
4474 return dev_set_mtu(dev, ifr->ifr_mtu);
4477 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4479 case SIOCSIFHWBROADCAST:
4480 if (ifr->ifr_hwaddr.sa_family != dev->type)
4482 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4483 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4484 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4488 if (ops->ndo_set_config) {
4489 if (!netif_device_present(dev))
4491 return ops->ndo_set_config(dev, &ifr->ifr_map);
4496 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4497 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4499 if (!netif_device_present(dev))
4501 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4504 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4505 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4507 if (!netif_device_present(dev))
4509 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4512 if (ifr->ifr_qlen < 0)
4514 dev->tx_queue_len = ifr->ifr_qlen;
4518 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4519 return dev_change_name(dev, ifr->ifr_newname);
4522 * Unknown or private ioctl
4525 if ((cmd >= SIOCDEVPRIVATE &&
4526 cmd <= SIOCDEVPRIVATE + 15) ||
4527 cmd == SIOCBONDENSLAVE ||
4528 cmd == SIOCBONDRELEASE ||
4529 cmd == SIOCBONDSETHWADDR ||
4530 cmd == SIOCBONDSLAVEINFOQUERY ||
4531 cmd == SIOCBONDINFOQUERY ||
4532 cmd == SIOCBONDCHANGEACTIVE ||
4533 cmd == SIOCGMIIPHY ||
4534 cmd == SIOCGMIIREG ||
4535 cmd == SIOCSMIIREG ||
4536 cmd == SIOCBRADDIF ||
4537 cmd == SIOCBRDELIF ||
4538 cmd == SIOCSHWTSTAMP ||
4539 cmd == SIOCWANDEV) {
4541 if (ops->ndo_do_ioctl) {
4542 if (netif_device_present(dev))
4543 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4555 * This function handles all "interface"-type I/O control requests. The actual
4556 * 'doing' part of this is dev_ifsioc above.
4560 * dev_ioctl - network device ioctl
4561 * @net: the applicable net namespace
4562 * @cmd: command to issue
4563 * @arg: pointer to a struct ifreq in user space
4565 * Issue ioctl functions to devices. This is normally called by the
4566 * user space syscall interfaces but can sometimes be useful for
4567 * other purposes. The return value is the return from the syscall if
4568 * positive or a negative errno code on error.
4571 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4577 /* One special case: SIOCGIFCONF takes ifconf argument
4578 and requires shared lock, because it sleeps writing
4582 if (cmd == SIOCGIFCONF) {
4584 ret = dev_ifconf(net, (char __user *) arg);
4588 if (cmd == SIOCGIFNAME)
4589 return dev_ifname(net, (struct ifreq __user *)arg);
4591 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4594 ifr.ifr_name[IFNAMSIZ-1] = 0;
4596 colon = strchr(ifr.ifr_name, ':');
4601 * See which interface the caller is talking about.
4606 * These ioctl calls:
4607 * - can be done by all.
4608 * - atomic and do not require locking.
4619 dev_load(net, ifr.ifr_name);
4621 ret = dev_ifsioc_locked(net, &ifr, cmd);
4626 if (copy_to_user(arg, &ifr,
4627 sizeof(struct ifreq)))
4633 dev_load(net, ifr.ifr_name);
4635 ret = dev_ethtool(net, &ifr);
4640 if (copy_to_user(arg, &ifr,
4641 sizeof(struct ifreq)))
4647 * These ioctl calls:
4648 * - require superuser power.
4649 * - require strict serialization.
4655 if (!capable(CAP_NET_ADMIN))
4657 dev_load(net, ifr.ifr_name);
4659 ret = dev_ifsioc(net, &ifr, cmd);
4664 if (copy_to_user(arg, &ifr,
4665 sizeof(struct ifreq)))
4671 * These ioctl calls:
4672 * - require superuser power.
4673 * - require strict serialization.
4674 * - do not return a value
4684 case SIOCSIFHWBROADCAST:
4687 case SIOCBONDENSLAVE:
4688 case SIOCBONDRELEASE:
4689 case SIOCBONDSETHWADDR:
4690 case SIOCBONDCHANGEACTIVE:
4694 if (!capable(CAP_NET_ADMIN))
4697 case SIOCBONDSLAVEINFOQUERY:
4698 case SIOCBONDINFOQUERY:
4699 dev_load(net, ifr.ifr_name);
4701 ret = dev_ifsioc(net, &ifr, cmd);
4706 /* Get the per device memory space. We can add this but
4707 * currently do not support it */
4709 /* Set the per device memory buffer space.
4710 * Not applicable in our case */
4715 * Unknown or private ioctl.
4718 if (cmd == SIOCWANDEV ||
4719 (cmd >= SIOCDEVPRIVATE &&
4720 cmd <= SIOCDEVPRIVATE + 15)) {
4721 dev_load(net, ifr.ifr_name);
4723 ret = dev_ifsioc(net, &ifr, cmd);
4725 if (!ret && copy_to_user(arg, &ifr,
4726 sizeof(struct ifreq)))
4730 /* Take care of Wireless Extensions */
4731 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4732 return wext_handle_ioctl(net, &ifr, cmd, arg);
4739 * dev_new_index - allocate an ifindex
4740 * @net: the applicable net namespace
4742 * Returns a suitable unique value for a new device interface
4743 * number. The caller must hold the rtnl semaphore or the
4744 * dev_base_lock to be sure it remains unique.
4746 static int dev_new_index(struct net *net)
4752 if (!__dev_get_by_index(net, ifindex))
4757 /* Delayed registration/unregisteration */
4758 static LIST_HEAD(net_todo_list);
4760 static void net_set_todo(struct net_device *dev)
4762 list_add_tail(&dev->todo_list, &net_todo_list);
4765 static void rollback_registered_many(struct list_head *head)
4767 struct net_device *dev, *tmp;
4769 BUG_ON(dev_boot_phase);
4772 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4773 /* Some devices call without registering
4774 * for initialization unwind. Remove those
4775 * devices and proceed with the remaining.
4777 if (dev->reg_state == NETREG_UNINITIALIZED) {
4778 pr_debug("unregister_netdevice: device %s/%p never "
4779 "was registered\n", dev->name, dev);
4782 list_del(&dev->unreg_list);
4786 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4788 /* If device is running, close it first. */
4791 /* And unlink it from device chain. */
4792 unlist_netdevice(dev);
4794 dev->reg_state = NETREG_UNREGISTERING;
4799 list_for_each_entry(dev, head, unreg_list) {
4800 /* Shutdown queueing discipline. */
4804 /* Notify protocols, that we are about to destroy
4805 this device. They should clean all the things.
4807 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4809 if (!dev->rtnl_link_ops ||
4810 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4811 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4814 * Flush the unicast and multicast chains
4819 if (dev->netdev_ops->ndo_uninit)
4820 dev->netdev_ops->ndo_uninit(dev);
4822 /* Notifier chain MUST detach us from master device. */
4823 WARN_ON(dev->master);
4825 /* Remove entries from kobject tree */
4826 netdev_unregister_kobject(dev);
4829 /* Process any work delayed until the end of the batch */
4830 dev = list_first_entry(head, struct net_device, unreg_list);
4831 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4835 list_for_each_entry(dev, head, unreg_list)
4839 static void rollback_registered(struct net_device *dev)
4843 list_add(&dev->unreg_list, &single);
4844 rollback_registered_many(&single);
4847 static void __netdev_init_queue_locks_one(struct net_device *dev,
4848 struct netdev_queue *dev_queue,
4851 spin_lock_init(&dev_queue->_xmit_lock);
4852 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4853 dev_queue->xmit_lock_owner = -1;
4856 static void netdev_init_queue_locks(struct net_device *dev)
4858 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4859 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4862 unsigned long netdev_fix_features(unsigned long features, const char *name)
4864 /* Fix illegal SG+CSUM combinations. */
4865 if ((features & NETIF_F_SG) &&
4866 !(features & NETIF_F_ALL_CSUM)) {
4868 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4869 "checksum feature.\n", name);
4870 features &= ~NETIF_F_SG;
4873 /* TSO requires that SG is present as well. */
4874 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4876 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4877 "SG feature.\n", name);
4878 features &= ~NETIF_F_TSO;
4881 if (features & NETIF_F_UFO) {
4882 if (!(features & NETIF_F_GEN_CSUM)) {
4884 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4885 "since no NETIF_F_HW_CSUM feature.\n",
4887 features &= ~NETIF_F_UFO;
4890 if (!(features & NETIF_F_SG)) {
4892 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4893 "since no NETIF_F_SG feature.\n", name);
4894 features &= ~NETIF_F_UFO;
4900 EXPORT_SYMBOL(netdev_fix_features);
4903 * netif_stacked_transfer_operstate - transfer operstate
4904 * @rootdev: the root or lower level device to transfer state from
4905 * @dev: the device to transfer operstate to
4907 * Transfer operational state from root to device. This is normally
4908 * called when a stacking relationship exists between the root
4909 * device and the device(a leaf device).
4911 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4912 struct net_device *dev)
4914 if (rootdev->operstate == IF_OPER_DORMANT)
4915 netif_dormant_on(dev);
4917 netif_dormant_off(dev);
4919 if (netif_carrier_ok(rootdev)) {
4920 if (!netif_carrier_ok(dev))
4921 netif_carrier_on(dev);
4923 if (netif_carrier_ok(dev))
4924 netif_carrier_off(dev);
4927 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4930 * register_netdevice - register a network device
4931 * @dev: device to register
4933 * Take a completed network device structure and add it to the kernel
4934 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4935 * chain. 0 is returned on success. A negative errno code is returned
4936 * on a failure to set up the device, or if the name is a duplicate.
4938 * Callers must hold the rtnl semaphore. You may want
4939 * register_netdev() instead of this.
4942 * The locking appears insufficient to guarantee two parallel registers
4943 * will not get the same name.
4946 int register_netdevice(struct net_device *dev)
4949 struct net *net = dev_net(dev);
4951 BUG_ON(dev_boot_phase);
4956 /* When net_device's are persistent, this will be fatal. */
4957 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4960 spin_lock_init(&dev->addr_list_lock);
4961 netdev_set_addr_lockdep_class(dev);
4962 netdev_init_queue_locks(dev);
4967 if (!dev->num_rx_queues) {
4969 * Allocate a single RX queue if driver never called
4973 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
4979 dev->_rx->first = dev->_rx;
4980 atomic_set(&dev->_rx->count, 1);
4981 dev->num_rx_queues = 1;
4984 /* Init, if this function is available */
4985 if (dev->netdev_ops->ndo_init) {
4986 ret = dev->netdev_ops->ndo_init(dev);
4994 ret = dev_get_valid_name(dev, dev->name, 0);
4998 dev->ifindex = dev_new_index(net);
4999 if (dev->iflink == -1)
5000 dev->iflink = dev->ifindex;
5002 /* Fix illegal checksum combinations */
5003 if ((dev->features & NETIF_F_HW_CSUM) &&
5004 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5005 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5007 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5010 if ((dev->features & NETIF_F_NO_CSUM) &&
5011 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5012 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5014 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5017 dev->features = netdev_fix_features(dev->features, dev->name);
5019 /* Enable software GSO if SG is supported. */
5020 if (dev->features & NETIF_F_SG)
5021 dev->features |= NETIF_F_GSO;
5023 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5024 ret = notifier_to_errno(ret);
5028 ret = netdev_register_kobject(dev);
5031 dev->reg_state = NETREG_REGISTERED;
5034 * Default initial state at registry is that the
5035 * device is present.
5038 set_bit(__LINK_STATE_PRESENT, &dev->state);
5040 dev_init_scheduler(dev);
5042 list_netdevice(dev);
5044 /* Notify protocols, that a new device appeared. */
5045 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5046 ret = notifier_to_errno(ret);
5048 rollback_registered(dev);
5049 dev->reg_state = NETREG_UNREGISTERED;
5052 * Prevent userspace races by waiting until the network
5053 * device is fully setup before sending notifications.
5055 if (!dev->rtnl_link_ops ||
5056 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5057 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5063 if (dev->netdev_ops->ndo_uninit)
5064 dev->netdev_ops->ndo_uninit(dev);
5067 EXPORT_SYMBOL(register_netdevice);
5070 * init_dummy_netdev - init a dummy network device for NAPI
5071 * @dev: device to init
5073 * This takes a network device structure and initialize the minimum
5074 * amount of fields so it can be used to schedule NAPI polls without
5075 * registering a full blown interface. This is to be used by drivers
5076 * that need to tie several hardware interfaces to a single NAPI
5077 * poll scheduler due to HW limitations.
5079 int init_dummy_netdev(struct net_device *dev)
5081 /* Clear everything. Note we don't initialize spinlocks
5082 * are they aren't supposed to be taken by any of the
5083 * NAPI code and this dummy netdev is supposed to be
5084 * only ever used for NAPI polls
5086 memset(dev, 0, sizeof(struct net_device));
5088 /* make sure we BUG if trying to hit standard
5089 * register/unregister code path
5091 dev->reg_state = NETREG_DUMMY;
5093 /* initialize the ref count */
5094 atomic_set(&dev->refcnt, 1);
5096 /* NAPI wants this */
5097 INIT_LIST_HEAD(&dev->napi_list);
5099 /* a dummy interface is started by default */
5100 set_bit(__LINK_STATE_PRESENT, &dev->state);
5101 set_bit(__LINK_STATE_START, &dev->state);
5105 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5109 * register_netdev - register a network device
5110 * @dev: device to register
5112 * Take a completed network device structure and add it to the kernel
5113 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5114 * chain. 0 is returned on success. A negative errno code is returned
5115 * on a failure to set up the device, or if the name is a duplicate.
5117 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5118 * and expands the device name if you passed a format string to
5121 int register_netdev(struct net_device *dev)
5128 * If the name is a format string the caller wants us to do a
5131 if (strchr(dev->name, '%')) {
5132 err = dev_alloc_name(dev, dev->name);
5137 err = register_netdevice(dev);
5142 EXPORT_SYMBOL(register_netdev);
5145 * netdev_wait_allrefs - wait until all references are gone.
5147 * This is called when unregistering network devices.
5149 * Any protocol or device that holds a reference should register
5150 * for netdevice notification, and cleanup and put back the
5151 * reference if they receive an UNREGISTER event.
5152 * We can get stuck here if buggy protocols don't correctly
5155 static void netdev_wait_allrefs(struct net_device *dev)
5157 unsigned long rebroadcast_time, warning_time;
5159 linkwatch_forget_dev(dev);
5161 rebroadcast_time = warning_time = jiffies;
5162 while (atomic_read(&dev->refcnt) != 0) {
5163 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5166 /* Rebroadcast unregister notification */
5167 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5168 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5169 * should have already handle it the first time */
5171 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5173 /* We must not have linkwatch events
5174 * pending on unregister. If this
5175 * happens, we simply run the queue
5176 * unscheduled, resulting in a noop
5179 linkwatch_run_queue();
5184 rebroadcast_time = jiffies;
5189 if (time_after(jiffies, warning_time + 10 * HZ)) {
5190 printk(KERN_EMERG "unregister_netdevice: "
5191 "waiting for %s to become free. Usage "
5193 dev->name, atomic_read(&dev->refcnt));
5194 warning_time = jiffies;
5203 * register_netdevice(x1);
5204 * register_netdevice(x2);
5206 * unregister_netdevice(y1);
5207 * unregister_netdevice(y2);
5213 * We are invoked by rtnl_unlock().
5214 * This allows us to deal with problems:
5215 * 1) We can delete sysfs objects which invoke hotplug
5216 * without deadlocking with linkwatch via keventd.
5217 * 2) Since we run with the RTNL semaphore not held, we can sleep
5218 * safely in order to wait for the netdev refcnt to drop to zero.
5220 * We must not return until all unregister events added during
5221 * the interval the lock was held have been completed.
5223 void netdev_run_todo(void)
5225 struct list_head list;
5227 /* Snapshot list, allow later requests */
5228 list_replace_init(&net_todo_list, &list);
5232 while (!list_empty(&list)) {
5233 struct net_device *dev
5234 = list_first_entry(&list, struct net_device, todo_list);
5235 list_del(&dev->todo_list);
5237 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5238 printk(KERN_ERR "network todo '%s' but state %d\n",
5239 dev->name, dev->reg_state);
5244 dev->reg_state = NETREG_UNREGISTERED;
5246 on_each_cpu(flush_backlog, dev, 1);
5248 netdev_wait_allrefs(dev);
5251 BUG_ON(atomic_read(&dev->refcnt));
5252 WARN_ON(dev->ip_ptr);
5253 WARN_ON(dev->ip6_ptr);
5254 WARN_ON(dev->dn_ptr);
5256 if (dev->destructor)
5257 dev->destructor(dev);
5259 /* Free network device */
5260 kobject_put(&dev->dev.kobj);
5265 * dev_txq_stats_fold - fold tx_queues stats
5266 * @dev: device to get statistics from
5267 * @stats: struct net_device_stats to hold results
5269 void dev_txq_stats_fold(const struct net_device *dev,
5270 struct net_device_stats *stats)
5272 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5274 struct netdev_queue *txq;
5276 for (i = 0; i < dev->num_tx_queues; i++) {
5277 txq = netdev_get_tx_queue(dev, i);
5278 tx_bytes += txq->tx_bytes;
5279 tx_packets += txq->tx_packets;
5280 tx_dropped += txq->tx_dropped;
5282 if (tx_bytes || tx_packets || tx_dropped) {
5283 stats->tx_bytes = tx_bytes;
5284 stats->tx_packets = tx_packets;
5285 stats->tx_dropped = tx_dropped;
5288 EXPORT_SYMBOL(dev_txq_stats_fold);
5291 * dev_get_stats - get network device statistics
5292 * @dev: device to get statistics from
5294 * Get network statistics from device. The device driver may provide
5295 * its own method by setting dev->netdev_ops->get_stats; otherwise
5296 * the internal statistics structure is used.
5298 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5300 const struct net_device_ops *ops = dev->netdev_ops;
5302 if (ops->ndo_get_stats)
5303 return ops->ndo_get_stats(dev);
5305 dev_txq_stats_fold(dev, &dev->stats);
5308 EXPORT_SYMBOL(dev_get_stats);
5310 static void netdev_init_one_queue(struct net_device *dev,
5311 struct netdev_queue *queue,
5317 static void netdev_init_queues(struct net_device *dev)
5319 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5320 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5321 spin_lock_init(&dev->tx_global_lock);
5325 * alloc_netdev_mq - allocate network device
5326 * @sizeof_priv: size of private data to allocate space for
5327 * @name: device name format string
5328 * @setup: callback to initialize device
5329 * @queue_count: the number of subqueues to allocate
5331 * Allocates a struct net_device with private data area for driver use
5332 * and performs basic initialization. Also allocates subquue structs
5333 * for each queue on the device at the end of the netdevice.
5335 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5336 void (*setup)(struct net_device *), unsigned int queue_count)
5338 struct netdev_queue *tx;
5339 struct net_device *dev;
5341 struct net_device *p;
5343 struct netdev_rx_queue *rx;
5347 BUG_ON(strlen(name) >= sizeof(dev->name));
5349 alloc_size = sizeof(struct net_device);
5351 /* ensure 32-byte alignment of private area */
5352 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5353 alloc_size += sizeof_priv;
5355 /* ensure 32-byte alignment of whole construct */
5356 alloc_size += NETDEV_ALIGN - 1;
5358 p = kzalloc(alloc_size, GFP_KERNEL);
5360 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5364 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5366 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5372 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5374 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5379 atomic_set(&rx->count, queue_count);
5382 * Set a pointer to first element in the array which holds the
5385 for (i = 0; i < queue_count; i++)
5389 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5390 dev->padded = (char *)dev - (char *)p;
5392 if (dev_addr_init(dev))
5398 dev_net_set(dev, &init_net);
5401 dev->num_tx_queues = queue_count;
5402 dev->real_num_tx_queues = queue_count;
5406 dev->num_rx_queues = queue_count;
5409 dev->gso_max_size = GSO_MAX_SIZE;
5411 netdev_init_queues(dev);
5413 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5414 dev->ethtool_ntuple_list.count = 0;
5415 INIT_LIST_HEAD(&dev->napi_list);
5416 INIT_LIST_HEAD(&dev->unreg_list);
5417 INIT_LIST_HEAD(&dev->link_watch_list);
5418 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5420 strcpy(dev->name, name);
5433 EXPORT_SYMBOL(alloc_netdev_mq);
5436 * free_netdev - free network device
5439 * This function does the last stage of destroying an allocated device
5440 * interface. The reference to the device object is released.
5441 * If this is the last reference then it will be freed.
5443 void free_netdev(struct net_device *dev)
5445 struct napi_struct *p, *n;
5447 release_net(dev_net(dev));
5451 /* Flush device addresses */
5452 dev_addr_flush(dev);
5454 /* Clear ethtool n-tuple list */
5455 ethtool_ntuple_flush(dev);
5457 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5460 /* Compatibility with error handling in drivers */
5461 if (dev->reg_state == NETREG_UNINITIALIZED) {
5462 kfree((char *)dev - dev->padded);
5466 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5467 dev->reg_state = NETREG_RELEASED;
5469 /* will free via device release */
5470 put_device(&dev->dev);
5472 EXPORT_SYMBOL(free_netdev);
5475 * synchronize_net - Synchronize with packet receive processing
5477 * Wait for packets currently being received to be done.
5478 * Does not block later packets from starting.
5480 void synchronize_net(void)
5485 EXPORT_SYMBOL(synchronize_net);
5488 * unregister_netdevice_queue - remove device from the kernel
5492 * This function shuts down a device interface and removes it
5493 * from the kernel tables.
5494 * If head not NULL, device is queued to be unregistered later.
5496 * Callers must hold the rtnl semaphore. You may want
5497 * unregister_netdev() instead of this.
5500 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5505 list_move_tail(&dev->unreg_list, head);
5507 rollback_registered(dev);
5508 /* Finish processing unregister after unlock */
5512 EXPORT_SYMBOL(unregister_netdevice_queue);
5515 * unregister_netdevice_many - unregister many devices
5516 * @head: list of devices
5518 void unregister_netdevice_many(struct list_head *head)
5520 struct net_device *dev;
5522 if (!list_empty(head)) {
5523 rollback_registered_many(head);
5524 list_for_each_entry(dev, head, unreg_list)
5528 EXPORT_SYMBOL(unregister_netdevice_many);
5531 * unregister_netdev - remove device from the kernel
5534 * This function shuts down a device interface and removes it
5535 * from the kernel tables.
5537 * This is just a wrapper for unregister_netdevice that takes
5538 * the rtnl semaphore. In general you want to use this and not
5539 * unregister_netdevice.
5541 void unregister_netdev(struct net_device *dev)
5544 unregister_netdevice(dev);
5547 EXPORT_SYMBOL(unregister_netdev);
5550 * dev_change_net_namespace - move device to different nethost namespace
5552 * @net: network namespace
5553 * @pat: If not NULL name pattern to try if the current device name
5554 * is already taken in the destination network namespace.
5556 * This function shuts down a device interface and moves it
5557 * to a new network namespace. On success 0 is returned, on
5558 * a failure a netagive errno code is returned.
5560 * Callers must hold the rtnl semaphore.
5563 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5569 /* Don't allow namespace local devices to be moved. */
5571 if (dev->features & NETIF_F_NETNS_LOCAL)
5574 /* Ensure the device has been registrered */
5576 if (dev->reg_state != NETREG_REGISTERED)
5579 /* Get out if there is nothing todo */
5581 if (net_eq(dev_net(dev), net))
5584 /* Pick the destination device name, and ensure
5585 * we can use it in the destination network namespace.
5588 if (__dev_get_by_name(net, dev->name)) {
5589 /* We get here if we can't use the current device name */
5592 if (dev_get_valid_name(dev, pat, 1))
5597 * And now a mini version of register_netdevice unregister_netdevice.
5600 /* If device is running close it first. */
5603 /* And unlink it from device chain */
5605 unlist_netdevice(dev);
5609 /* Shutdown queueing discipline. */
5612 /* Notify protocols, that we are about to destroy
5613 this device. They should clean all the things.
5615 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5616 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5619 * Flush the unicast and multicast chains
5624 /* Actually switch the network namespace */
5625 dev_net_set(dev, net);
5627 /* If there is an ifindex conflict assign a new one */
5628 if (__dev_get_by_index(net, dev->ifindex)) {
5629 int iflink = (dev->iflink == dev->ifindex);
5630 dev->ifindex = dev_new_index(net);
5632 dev->iflink = dev->ifindex;
5635 /* Fixup kobjects */
5636 err = device_rename(&dev->dev, dev->name);
5639 /* Add the device back in the hashes */
5640 list_netdevice(dev);
5642 /* Notify protocols, that a new device appeared. */
5643 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5646 * Prevent userspace races by waiting until the network
5647 * device is fully setup before sending notifications.
5649 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5656 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5658 static int dev_cpu_callback(struct notifier_block *nfb,
5659 unsigned long action,
5662 struct sk_buff **list_skb;
5663 struct sk_buff *skb;
5664 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5665 struct softnet_data *sd, *oldsd;
5667 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5670 local_irq_disable();
5671 cpu = smp_processor_id();
5672 sd = &per_cpu(softnet_data, cpu);
5673 oldsd = &per_cpu(softnet_data, oldcpu);
5675 /* Find end of our completion_queue. */
5676 list_skb = &sd->completion_queue;
5678 list_skb = &(*list_skb)->next;
5679 /* Append completion queue from offline CPU. */
5680 *list_skb = oldsd->completion_queue;
5681 oldsd->completion_queue = NULL;
5683 /* Append output queue from offline CPU. */
5684 if (oldsd->output_queue) {
5685 *sd->output_queue_tailp = oldsd->output_queue;
5686 sd->output_queue_tailp = oldsd->output_queue_tailp;
5687 oldsd->output_queue = NULL;
5688 oldsd->output_queue_tailp = &oldsd->output_queue;
5691 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5694 /* Process offline CPU's input_pkt_queue */
5695 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5697 input_queue_head_incr(oldsd);
5699 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5701 input_queue_head_incr(oldsd);
5709 * netdev_increment_features - increment feature set by one
5710 * @all: current feature set
5711 * @one: new feature set
5712 * @mask: mask feature set
5714 * Computes a new feature set after adding a device with feature set
5715 * @one to the master device with current feature set @all. Will not
5716 * enable anything that is off in @mask. Returns the new feature set.
5718 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5721 /* If device needs checksumming, downgrade to it. */
5722 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5723 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5724 else if (mask & NETIF_F_ALL_CSUM) {
5725 /* If one device supports v4/v6 checksumming, set for all. */
5726 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5727 !(all & NETIF_F_GEN_CSUM)) {
5728 all &= ~NETIF_F_ALL_CSUM;
5729 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5732 /* If one device supports hw checksumming, set for all. */
5733 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5734 all &= ~NETIF_F_ALL_CSUM;
5735 all |= NETIF_F_HW_CSUM;
5739 one |= NETIF_F_ALL_CSUM;
5741 one |= all & NETIF_F_ONE_FOR_ALL;
5742 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5743 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5747 EXPORT_SYMBOL(netdev_increment_features);
5749 static struct hlist_head *netdev_create_hash(void)
5752 struct hlist_head *hash;
5754 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5756 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5757 INIT_HLIST_HEAD(&hash[i]);
5762 /* Initialize per network namespace state */
5763 static int __net_init netdev_init(struct net *net)
5765 INIT_LIST_HEAD(&net->dev_base_head);
5767 net->dev_name_head = netdev_create_hash();
5768 if (net->dev_name_head == NULL)
5771 net->dev_index_head = netdev_create_hash();
5772 if (net->dev_index_head == NULL)
5778 kfree(net->dev_name_head);
5784 * netdev_drivername - network driver for the device
5785 * @dev: network device
5786 * @buffer: buffer for resulting name
5787 * @len: size of buffer
5789 * Determine network driver for device.
5791 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5793 const struct device_driver *driver;
5794 const struct device *parent;
5796 if (len <= 0 || !buffer)
5800 parent = dev->dev.parent;
5805 driver = parent->driver;
5806 if (driver && driver->name)
5807 strlcpy(buffer, driver->name, len);
5811 static void __net_exit netdev_exit(struct net *net)
5813 kfree(net->dev_name_head);
5814 kfree(net->dev_index_head);
5817 static struct pernet_operations __net_initdata netdev_net_ops = {
5818 .init = netdev_init,
5819 .exit = netdev_exit,
5822 static void __net_exit default_device_exit(struct net *net)
5824 struct net_device *dev, *aux;
5826 * Push all migratable network devices back to the
5827 * initial network namespace
5830 for_each_netdev_safe(net, dev, aux) {
5832 char fb_name[IFNAMSIZ];
5834 /* Ignore unmoveable devices (i.e. loopback) */
5835 if (dev->features & NETIF_F_NETNS_LOCAL)
5838 /* Leave virtual devices for the generic cleanup */
5839 if (dev->rtnl_link_ops)
5842 /* Push remaing network devices to init_net */
5843 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5844 err = dev_change_net_namespace(dev, &init_net, fb_name);
5846 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5847 __func__, dev->name, err);
5854 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5856 /* At exit all network devices most be removed from a network
5857 * namespace. Do this in the reverse order of registeration.
5858 * Do this across as many network namespaces as possible to
5859 * improve batching efficiency.
5861 struct net_device *dev;
5863 LIST_HEAD(dev_kill_list);
5866 list_for_each_entry(net, net_list, exit_list) {
5867 for_each_netdev_reverse(net, dev) {
5868 if (dev->rtnl_link_ops)
5869 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5871 unregister_netdevice_queue(dev, &dev_kill_list);
5874 unregister_netdevice_many(&dev_kill_list);
5878 static struct pernet_operations __net_initdata default_device_ops = {
5879 .exit = default_device_exit,
5880 .exit_batch = default_device_exit_batch,
5884 * Initialize the DEV module. At boot time this walks the device list and
5885 * unhooks any devices that fail to initialise (normally hardware not
5886 * present) and leaves us with a valid list of present and active devices.
5891 * This is called single threaded during boot, so no need
5892 * to take the rtnl semaphore.
5894 static int __init net_dev_init(void)
5896 int i, rc = -ENOMEM;
5898 BUG_ON(!dev_boot_phase);
5900 if (dev_proc_init())
5903 if (netdev_kobject_init())
5906 INIT_LIST_HEAD(&ptype_all);
5907 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5908 INIT_LIST_HEAD(&ptype_base[i]);
5910 if (register_pernet_subsys(&netdev_net_ops))
5914 * Initialise the packet receive queues.
5917 for_each_possible_cpu(i) {
5918 struct softnet_data *sd = &per_cpu(softnet_data, i);
5920 memset(sd, 0, sizeof(*sd));
5921 skb_queue_head_init(&sd->input_pkt_queue);
5922 skb_queue_head_init(&sd->process_queue);
5923 sd->completion_queue = NULL;
5924 INIT_LIST_HEAD(&sd->poll_list);
5925 sd->output_queue = NULL;
5926 sd->output_queue_tailp = &sd->output_queue;
5928 sd->csd.func = rps_trigger_softirq;
5934 sd->backlog.poll = process_backlog;
5935 sd->backlog.weight = weight_p;
5936 sd->backlog.gro_list = NULL;
5937 sd->backlog.gro_count = 0;
5942 /* The loopback device is special if any other network devices
5943 * is present in a network namespace the loopback device must
5944 * be present. Since we now dynamically allocate and free the
5945 * loopback device ensure this invariant is maintained by
5946 * keeping the loopback device as the first device on the
5947 * list of network devices. Ensuring the loopback devices
5948 * is the first device that appears and the last network device
5951 if (register_pernet_device(&loopback_net_ops))
5954 if (register_pernet_device(&default_device_ops))
5957 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5958 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5960 hotcpu_notifier(dev_cpu_callback, 0);
5968 subsys_initcall(net_dev_init);
5970 static int __init initialize_hashrnd(void)
5972 get_random_bytes(&hashrnd, sizeof(hashrnd));
5976 late_initcall_sync(initialize_hashrnd);