2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
104 #include <linux/if_bridge.h>
105 #include <linux/if_macvlan.h>
107 #include <net/pkt_sched.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/kmod.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/wext.h>
118 #include <net/iw_handler.h>
119 #include <asm/current.h>
120 #include <linux/audit.h>
121 #include <linux/dmaengine.h>
122 #include <linux/err.h>
123 #include <linux/ctype.h>
124 #include <linux/if_arp.h>
125 #include <linux/if_vlan.h>
126 #include <linux/ip.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
133 #include <linux/pci.h>
135 #include "net-sysfs.h"
137 /* Instead of increasing this, you should create a hash table. */
138 #define MAX_GRO_SKBS 8
140 /* This should be increased if a protocol with a bigger head is added. */
141 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 * The list of packet types we will receive (as opposed to discard)
145 * and the routines to invoke.
147 * Why 16. Because with 16 the only overlap we get on a hash of the
148 * low nibble of the protocol value is RARP/SNAP/X.25.
150 * NOTE: That is no longer true with the addition of VLAN tags. Not
151 * sure which should go first, but I bet it won't make much
152 * difference if we are running VLANs. The good news is that
153 * this protocol won't be in the list unless compiled in, so
154 * the average user (w/out VLANs) will not be adversely affected.
171 #define PTYPE_HASH_SIZE (16)
172 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
174 static DEFINE_SPINLOCK(ptype_lock);
175 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
176 static struct list_head ptype_all __read_mostly; /* Taps */
179 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
182 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
184 * Writers must hold the rtnl semaphore while they loop through the
185 * dev_base_head list, and hold dev_base_lock for writing when they do the
186 * actual updates. This allows pure readers to access the list even
187 * while a writer is preparing to update it.
189 * To put it another way, dev_base_lock is held for writing only to
190 * protect against pure readers; the rtnl semaphore provides the
191 * protection against other writers.
193 * See, for example usages, register_netdevice() and
194 * unregister_netdevice(), which must be called with the rtnl
197 DEFINE_RWLOCK(dev_base_lock);
198 EXPORT_SYMBOL(dev_base_lock);
200 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
202 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
203 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
206 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
208 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
211 static inline void rps_lock(struct softnet_data *sd)
214 spin_lock(&sd->input_pkt_queue.lock);
218 static inline void rps_unlock(struct softnet_data *sd)
221 spin_unlock(&sd->input_pkt_queue.lock);
225 /* Device list insertion */
226 static int list_netdevice(struct net_device *dev)
228 struct net *net = dev_net(dev);
232 write_lock_bh(&dev_base_lock);
233 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
234 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
235 hlist_add_head_rcu(&dev->index_hlist,
236 dev_index_hash(net, dev->ifindex));
237 write_unlock_bh(&dev_base_lock);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device *dev)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock);
250 list_del_rcu(&dev->dev_list);
251 hlist_del_rcu(&dev->name_hlist);
252 hlist_del_rcu(&dev->index_hlist);
253 write_unlock_bh(&dev_base_lock);
260 static RAW_NOTIFIER_HEAD(netdev_chain);
263 * Device drivers call our routines to queue packets here. We empty the
264 * queue in the local softnet handler.
267 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
268 EXPORT_PER_CPU_SYMBOL(softnet_data);
270 #ifdef CONFIG_LOCKDEP
272 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
273 * according to dev->type
275 static const unsigned short netdev_lock_type[] =
276 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
277 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
278 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
279 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
280 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
281 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
282 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
283 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
284 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
285 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
286 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
287 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
288 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
289 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
290 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
291 ARPHRD_VOID, ARPHRD_NONE};
293 static const char *const netdev_lock_name[] =
294 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
295 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
296 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
297 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
298 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
299 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
300 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
301 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
302 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
303 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
304 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
305 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
306 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
307 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
308 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
309 "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
318 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
319 if (netdev_lock_type[i] == dev_type)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
326 unsigned short dev_type)
330 i = netdev_lock_pos(dev_type);
331 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
332 netdev_lock_name[i]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
339 i = netdev_lock_pos(dev->type);
340 lockdep_set_class_and_name(&dev->addr_list_lock,
341 &netdev_addr_lock_key[i],
342 netdev_lock_name[i]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
346 unsigned short dev_type)
349 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
377 * dev_add_pack - add packet handler
378 * @pt: packet type declaration
380 * Add a protocol handler to the networking stack. The passed &packet_type
381 * is linked into kernel lists and may not be freed until it has been
382 * removed from the kernel lists.
384 * This call does not sleep therefore it can not
385 * guarantee all CPU's that are in middle of receiving packets
386 * will see the new packet type (until the next received packet).
389 void dev_add_pack(struct packet_type *pt)
393 spin_lock_bh(&ptype_lock);
394 if (pt->type == htons(ETH_P_ALL))
395 list_add_rcu(&pt->list, &ptype_all);
397 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
398 list_add_rcu(&pt->list, &ptype_base[hash]);
400 spin_unlock_bh(&ptype_lock);
402 EXPORT_SYMBOL(dev_add_pack);
405 * __dev_remove_pack - remove packet handler
406 * @pt: packet type declaration
408 * Remove a protocol handler that was previously added to the kernel
409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
410 * from the kernel lists and can be freed or reused once this function
413 * The packet type might still be in use by receivers
414 * and must not be freed until after all the CPU's have gone
415 * through a quiescent state.
417 void __dev_remove_pack(struct packet_type *pt)
419 struct list_head *head;
420 struct packet_type *pt1;
422 spin_lock_bh(&ptype_lock);
424 if (pt->type == htons(ETH_P_ALL))
427 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
429 list_for_each_entry(pt1, head, list) {
431 list_del_rcu(&pt->list);
436 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
438 spin_unlock_bh(&ptype_lock);
440 EXPORT_SYMBOL(__dev_remove_pack);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type *pt)
456 __dev_remove_pack(pt);
460 EXPORT_SYMBOL(dev_remove_pack);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name, struct ifmap *map)
482 struct netdev_boot_setup *s;
486 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
487 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
488 memset(s[i].name, 0, sizeof(s[i].name));
489 strlcpy(s[i].name, name, IFNAMSIZ);
490 memcpy(&s[i].map, map, sizeof(s[i].map));
495 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device *dev)
509 struct netdev_boot_setup *s = dev_boot_setup;
512 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
513 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
514 !strcmp(dev->name, s[i].name)) {
515 dev->irq = s[i].map.irq;
516 dev->base_addr = s[i].map.base_addr;
517 dev->mem_start = s[i].map.mem_start;
518 dev->mem_end = s[i].map.mem_end;
524 EXPORT_SYMBOL(netdev_boot_setup_check);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix, int unit)
539 const struct netdev_boot_setup *s = dev_boot_setup;
543 sprintf(name, "%s%d", prefix, unit);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net, name))
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
553 if (!strcmp(name, s[i].name))
554 return s[i].map.base_addr;
559 * Saves at boot time configured settings for any netdevice.
561 int __init netdev_boot_setup(char *str)
566 str = get_options(str, ARRAY_SIZE(ints), ints);
571 memset(&map, 0, sizeof(map));
575 map.base_addr = ints[2];
577 map.mem_start = ints[3];
579 map.mem_end = ints[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str, &map);
585 __setup("netdev=", netdev_boot_setup);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device *__dev_get_by_name(struct net *net, const char *name)
607 struct hlist_node *p;
608 struct net_device *dev;
609 struct hlist_head *head = dev_name_hash(net, name);
611 hlist_for_each_entry(dev, p, head, name_hlist)
612 if (!strncmp(dev->name, name, IFNAMSIZ))
617 EXPORT_SYMBOL(__dev_get_by_name);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
633 struct hlist_node *p;
634 struct net_device *dev;
635 struct hlist_head *head = dev_name_hash(net, name);
637 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
638 if (!strncmp(dev->name, name, IFNAMSIZ))
643 EXPORT_SYMBOL(dev_get_by_name_rcu);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device *dev_get_by_name(struct net *net, const char *name)
659 struct net_device *dev;
662 dev = dev_get_by_name_rcu(net, name);
668 EXPORT_SYMBOL(dev_get_by_name);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
684 struct hlist_node *p;
685 struct net_device *dev;
686 struct hlist_head *head = dev_index_hash(net, ifindex);
688 hlist_for_each_entry(dev, p, head, index_hlist)
689 if (dev->ifindex == ifindex)
694 EXPORT_SYMBOL(__dev_get_by_index);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
709 struct hlist_node *p;
710 struct net_device *dev;
711 struct hlist_head *head = dev_index_hash(net, ifindex);
713 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
714 if (dev->ifindex == ifindex)
719 EXPORT_SYMBOL(dev_get_by_index_rcu);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device *dev_get_by_index(struct net *net, int ifindex)
735 struct net_device *dev;
738 dev = dev_get_by_index_rcu(net, ifindex);
744 EXPORT_SYMBOL(dev_get_by_index);
747 * dev_getbyhwaddr - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device. The caller must hold the
754 * rtnl semaphore. The returned device has not had its ref count increased
755 * and the caller must therefore be careful about locking
758 * If the API was consistent this would be __dev_get_by_hwaddr
761 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
763 struct net_device *dev;
767 for_each_netdev(net, dev)
768 if (dev->type == type &&
769 !memcmp(dev->dev_addr, ha, dev->addr_len))
774 EXPORT_SYMBOL(dev_getbyhwaddr);
776 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
778 struct net_device *dev;
781 for_each_netdev(net, dev)
782 if (dev->type == type)
787 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
789 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
791 struct net_device *dev, *ret = NULL;
794 for_each_netdev_rcu(net, dev)
795 if (dev->type == type) {
803 EXPORT_SYMBOL(dev_getfirstbyhwtype);
806 * dev_get_by_flags_rcu - find any device with given flags
807 * @net: the applicable net namespace
808 * @if_flags: IFF_* values
809 * @mask: bitmask of bits in if_flags to check
811 * Search for any interface with the given flags. Returns NULL if a device
812 * is not found or a pointer to the device. Must be called inside
813 * rcu_read_lock(), and result refcount is unchanged.
816 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
819 struct net_device *dev, *ret;
822 for_each_netdev_rcu(net, dev) {
823 if (((dev->flags ^ if_flags) & mask) == 0) {
830 EXPORT_SYMBOL(dev_get_by_flags_rcu);
833 * dev_valid_name - check if name is okay for network device
836 * Network device names need to be valid file names to
837 * to allow sysfs to work. We also disallow any kind of
840 int dev_valid_name(const char *name)
844 if (strlen(name) >= IFNAMSIZ)
846 if (!strcmp(name, ".") || !strcmp(name, ".."))
850 if (*name == '/' || isspace(*name))
856 EXPORT_SYMBOL(dev_valid_name);
859 * __dev_alloc_name - allocate a name for a device
860 * @net: network namespace to allocate the device name in
861 * @name: name format string
862 * @buf: scratch buffer and result name string
864 * Passed a format string - eg "lt%d" it will try and find a suitable
865 * id. It scans list of devices to build up a free map, then chooses
866 * the first empty slot. The caller must hold the dev_base or rtnl lock
867 * while allocating the name and adding the device in order to avoid
869 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
870 * Returns the number of the unit assigned or a negative errno code.
873 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
877 const int max_netdevices = 8*PAGE_SIZE;
878 unsigned long *inuse;
879 struct net_device *d;
881 p = strnchr(name, IFNAMSIZ-1, '%');
884 * Verify the string as this thing may have come from
885 * the user. There must be either one "%d" and no other "%"
888 if (p[1] != 'd' || strchr(p + 2, '%'))
891 /* Use one page as a bit array of possible slots */
892 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
896 for_each_netdev(net, d) {
897 if (!sscanf(d->name, name, &i))
899 if (i < 0 || i >= max_netdevices)
902 /* avoid cases where sscanf is not exact inverse of printf */
903 snprintf(buf, IFNAMSIZ, name, i);
904 if (!strncmp(buf, d->name, IFNAMSIZ))
908 i = find_first_zero_bit(inuse, max_netdevices);
909 free_page((unsigned long) inuse);
913 snprintf(buf, IFNAMSIZ, name, i);
914 if (!__dev_get_by_name(net, buf))
917 /* It is possible to run out of possible slots
918 * when the name is long and there isn't enough space left
919 * for the digits, or if all bits are used.
925 * dev_alloc_name - allocate a name for a device
927 * @name: name format string
929 * Passed a format string - eg "lt%d" it will try and find a suitable
930 * id. It scans list of devices to build up a free map, then chooses
931 * the first empty slot. The caller must hold the dev_base or rtnl lock
932 * while allocating the name and adding the device in order to avoid
934 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
935 * Returns the number of the unit assigned or a negative errno code.
938 int dev_alloc_name(struct net_device *dev, const char *name)
944 BUG_ON(!dev_net(dev));
946 ret = __dev_alloc_name(net, name, buf);
948 strlcpy(dev->name, buf, IFNAMSIZ);
951 EXPORT_SYMBOL(dev_alloc_name);
953 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
957 BUG_ON(!dev_net(dev));
960 if (!dev_valid_name(name))
963 if (fmt && strchr(name, '%'))
964 return dev_alloc_name(dev, name);
965 else if (__dev_get_by_name(net, name))
967 else if (dev->name != name)
968 strlcpy(dev->name, name, IFNAMSIZ);
974 * dev_change_name - change name of a device
976 * @newname: name (or format string) must be at least IFNAMSIZ
978 * Change name of a device, can pass format strings "eth%d".
981 int dev_change_name(struct net_device *dev, const char *newname)
983 char oldname[IFNAMSIZ];
989 BUG_ON(!dev_net(dev));
992 if (dev->flags & IFF_UP)
995 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
998 memcpy(oldname, dev->name, IFNAMSIZ);
1000 err = dev_get_valid_name(dev, newname, 1);
1005 ret = device_rename(&dev->dev, dev->name);
1007 memcpy(dev->name, oldname, IFNAMSIZ);
1011 write_lock_bh(&dev_base_lock);
1012 hlist_del(&dev->name_hlist);
1013 write_unlock_bh(&dev_base_lock);
1017 write_lock_bh(&dev_base_lock);
1018 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1019 write_unlock_bh(&dev_base_lock);
1021 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1022 ret = notifier_to_errno(ret);
1025 /* err >= 0 after dev_alloc_name() or stores the first errno */
1028 memcpy(dev->name, oldname, IFNAMSIZ);
1032 "%s: name change rollback failed: %d.\n",
1041 * dev_set_alias - change ifalias of a device
1043 * @alias: name up to IFALIASZ
1044 * @len: limit of bytes to copy from info
1046 * Set ifalias for a device,
1048 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1052 if (len >= IFALIASZ)
1057 kfree(dev->ifalias);
1058 dev->ifalias = NULL;
1063 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1067 strlcpy(dev->ifalias, alias, len+1);
1073 * netdev_features_change - device changes features
1074 * @dev: device to cause notification
1076 * Called to indicate a device has changed features.
1078 void netdev_features_change(struct net_device *dev)
1080 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1082 EXPORT_SYMBOL(netdev_features_change);
1085 * netdev_state_change - device changes state
1086 * @dev: device to cause notification
1088 * Called to indicate a device has changed state. This function calls
1089 * the notifier chains for netdev_chain and sends a NEWLINK message
1090 * to the routing socket.
1092 void netdev_state_change(struct net_device *dev)
1094 if (dev->flags & IFF_UP) {
1095 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1096 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1099 EXPORT_SYMBOL(netdev_state_change);
1101 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1103 return call_netdevice_notifiers(event, dev);
1105 EXPORT_SYMBOL(netdev_bonding_change);
1108 * dev_load - load a network module
1109 * @net: the applicable net namespace
1110 * @name: name of interface
1112 * If a network interface is not present and the process has suitable
1113 * privileges this function loads the module. If module loading is not
1114 * available in this kernel then it becomes a nop.
1117 void dev_load(struct net *net, const char *name)
1119 struct net_device *dev;
1122 dev = dev_get_by_name_rcu(net, name);
1125 if (!dev && capable(CAP_NET_ADMIN))
1126 request_module("%s", name);
1128 EXPORT_SYMBOL(dev_load);
1130 static int __dev_open(struct net_device *dev)
1132 const struct net_device_ops *ops = dev->netdev_ops;
1138 * Is it even present?
1140 if (!netif_device_present(dev))
1143 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1144 ret = notifier_to_errno(ret);
1149 * Call device private open method
1151 set_bit(__LINK_STATE_START, &dev->state);
1153 if (ops->ndo_validate_addr)
1154 ret = ops->ndo_validate_addr(dev);
1156 if (!ret && ops->ndo_open)
1157 ret = ops->ndo_open(dev);
1160 * If it went open OK then:
1164 clear_bit(__LINK_STATE_START, &dev->state);
1169 dev->flags |= IFF_UP;
1174 net_dmaengine_get();
1177 * Initialize multicasting status
1179 dev_set_rx_mode(dev);
1182 * Wakeup transmit queue engine
1191 * dev_open - prepare an interface for use.
1192 * @dev: device to open
1194 * Takes a device from down to up state. The device's private open
1195 * function is invoked and then the multicast lists are loaded. Finally
1196 * the device is moved into the up state and a %NETDEV_UP message is
1197 * sent to the netdev notifier chain.
1199 * Calling this function on an active interface is a nop. On a failure
1200 * a negative errno code is returned.
1202 int dev_open(struct net_device *dev)
1209 if (dev->flags & IFF_UP)
1215 ret = __dev_open(dev);
1220 * ... and announce new interface.
1222 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1223 call_netdevice_notifiers(NETDEV_UP, dev);
1227 EXPORT_SYMBOL(dev_open);
1229 static int __dev_close(struct net_device *dev)
1231 const struct net_device_ops *ops = dev->netdev_ops;
1237 * Tell people we are going down, so that they can
1238 * prepare to death, when device is still operating.
1240 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1242 clear_bit(__LINK_STATE_START, &dev->state);
1244 /* Synchronize to scheduled poll. We cannot touch poll list,
1245 * it can be even on different cpu. So just clear netif_running().
1247 * dev->stop() will invoke napi_disable() on all of it's
1248 * napi_struct instances on this device.
1250 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1252 dev_deactivate(dev);
1255 * Call the device specific close. This cannot fail.
1256 * Only if device is UP
1258 * We allow it to be called even after a DETACH hot-plug
1265 * Device is now down.
1268 dev->flags &= ~IFF_UP;
1273 net_dmaengine_put();
1279 * dev_close - shutdown an interface.
1280 * @dev: device to shutdown
1282 * This function moves an active device into down state. A
1283 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1284 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1287 int dev_close(struct net_device *dev)
1289 if (!(dev->flags & IFF_UP))
1295 * Tell people we are down
1297 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1298 call_netdevice_notifiers(NETDEV_DOWN, dev);
1302 EXPORT_SYMBOL(dev_close);
1306 * dev_disable_lro - disable Large Receive Offload on a device
1309 * Disable Large Receive Offload (LRO) on a net device. Must be
1310 * called under RTNL. This is needed if received packets may be
1311 * forwarded to another interface.
1313 void dev_disable_lro(struct net_device *dev)
1315 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1316 dev->ethtool_ops->set_flags) {
1317 u32 flags = dev->ethtool_ops->get_flags(dev);
1318 if (flags & ETH_FLAG_LRO) {
1319 flags &= ~ETH_FLAG_LRO;
1320 dev->ethtool_ops->set_flags(dev, flags);
1323 WARN_ON(dev->features & NETIF_F_LRO);
1325 EXPORT_SYMBOL(dev_disable_lro);
1328 static int dev_boot_phase = 1;
1331 * Device change register/unregister. These are not inline or static
1332 * as we export them to the world.
1336 * register_netdevice_notifier - register a network notifier block
1339 * Register a notifier to be called when network device events occur.
1340 * The notifier passed is linked into the kernel structures and must
1341 * not be reused until it has been unregistered. A negative errno code
1342 * is returned on a failure.
1344 * When registered all registration and up events are replayed
1345 * to the new notifier to allow device to have a race free
1346 * view of the network device list.
1349 int register_netdevice_notifier(struct notifier_block *nb)
1351 struct net_device *dev;
1352 struct net_device *last;
1357 err = raw_notifier_chain_register(&netdev_chain, nb);
1363 for_each_netdev(net, dev) {
1364 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1365 err = notifier_to_errno(err);
1369 if (!(dev->flags & IFF_UP))
1372 nb->notifier_call(nb, NETDEV_UP, dev);
1383 for_each_netdev(net, dev) {
1387 if (dev->flags & IFF_UP) {
1388 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1389 nb->notifier_call(nb, NETDEV_DOWN, dev);
1391 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1392 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1396 raw_notifier_chain_unregister(&netdev_chain, nb);
1399 EXPORT_SYMBOL(register_netdevice_notifier);
1402 * unregister_netdevice_notifier - unregister a network notifier block
1405 * Unregister a notifier previously registered by
1406 * register_netdevice_notifier(). The notifier is unlinked into the
1407 * kernel structures and may then be reused. A negative errno code
1408 * is returned on a failure.
1411 int unregister_netdevice_notifier(struct notifier_block *nb)
1416 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1420 EXPORT_SYMBOL(unregister_netdevice_notifier);
1423 * call_netdevice_notifiers - call all network notifier blocks
1424 * @val: value passed unmodified to notifier function
1425 * @dev: net_device pointer passed unmodified to notifier function
1427 * Call all network notifier blocks. Parameters and return value
1428 * are as for raw_notifier_call_chain().
1431 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1434 return raw_notifier_call_chain(&netdev_chain, val, dev);
1437 /* When > 0 there are consumers of rx skb time stamps */
1438 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1440 void net_enable_timestamp(void)
1442 atomic_inc(&netstamp_needed);
1444 EXPORT_SYMBOL(net_enable_timestamp);
1446 void net_disable_timestamp(void)
1448 atomic_dec(&netstamp_needed);
1450 EXPORT_SYMBOL(net_disable_timestamp);
1452 static inline void net_timestamp_set(struct sk_buff *skb)
1454 if (atomic_read(&netstamp_needed))
1455 __net_timestamp(skb);
1457 skb->tstamp.tv64 = 0;
1460 static inline void net_timestamp_check(struct sk_buff *skb)
1462 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1463 __net_timestamp(skb);
1467 * dev_forward_skb - loopback an skb to another netif
1469 * @dev: destination network device
1470 * @skb: buffer to forward
1473 * NET_RX_SUCCESS (no congestion)
1474 * NET_RX_DROP (packet was dropped, but freed)
1476 * dev_forward_skb can be used for injecting an skb from the
1477 * start_xmit function of one device into the receive queue
1478 * of another device.
1480 * The receiving device may be in another namespace, so
1481 * we have to clear all information in the skb that could
1482 * impact namespace isolation.
1484 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1488 if (!(dev->flags & IFF_UP) ||
1489 (skb->len > (dev->mtu + dev->hard_header_len))) {
1493 skb_set_dev(skb, dev);
1494 skb->tstamp.tv64 = 0;
1495 skb->pkt_type = PACKET_HOST;
1496 skb->protocol = eth_type_trans(skb, dev);
1497 return netif_rx(skb);
1499 EXPORT_SYMBOL_GPL(dev_forward_skb);
1502 * Support routine. Sends outgoing frames to any network
1503 * taps currently in use.
1506 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1508 struct packet_type *ptype;
1510 #ifdef CONFIG_NET_CLS_ACT
1511 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1512 net_timestamp_set(skb);
1514 net_timestamp_set(skb);
1518 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1519 /* Never send packets back to the socket
1520 * they originated from - MvS (miquels@drinkel.ow.org)
1522 if ((ptype->dev == dev || !ptype->dev) &&
1523 (ptype->af_packet_priv == NULL ||
1524 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1525 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1529 /* skb->nh should be correctly
1530 set by sender, so that the second statement is
1531 just protection against buggy protocols.
1533 skb_reset_mac_header(skb2);
1535 if (skb_network_header(skb2) < skb2->data ||
1536 skb2->network_header > skb2->tail) {
1537 if (net_ratelimit())
1538 printk(KERN_CRIT "protocol %04x is "
1540 ntohs(skb2->protocol),
1542 skb_reset_network_header(skb2);
1545 skb2->transport_header = skb2->network_header;
1546 skb2->pkt_type = PACKET_OUTGOING;
1547 ptype->func(skb2, skb->dev, ptype, skb->dev);
1554 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1555 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1557 void netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1559 unsigned int real_num = dev->real_num_tx_queues;
1561 if (unlikely(txq > dev->num_tx_queues))
1563 else if (txq > real_num)
1564 dev->real_num_tx_queues = txq;
1565 else if (txq < real_num) {
1566 dev->real_num_tx_queues = txq;
1567 qdisc_reset_all_tx_gt(dev, txq);
1570 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1572 static inline void __netif_reschedule(struct Qdisc *q)
1574 struct softnet_data *sd;
1575 unsigned long flags;
1577 local_irq_save(flags);
1578 sd = &__get_cpu_var(softnet_data);
1579 q->next_sched = NULL;
1580 *sd->output_queue_tailp = q;
1581 sd->output_queue_tailp = &q->next_sched;
1582 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1583 local_irq_restore(flags);
1586 void __netif_schedule(struct Qdisc *q)
1588 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1589 __netif_reschedule(q);
1591 EXPORT_SYMBOL(__netif_schedule);
1593 void dev_kfree_skb_irq(struct sk_buff *skb)
1595 if (!skb->destructor)
1597 else if (atomic_dec_and_test(&skb->users)) {
1598 struct softnet_data *sd;
1599 unsigned long flags;
1601 local_irq_save(flags);
1602 sd = &__get_cpu_var(softnet_data);
1603 skb->next = sd->completion_queue;
1604 sd->completion_queue = skb;
1605 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1606 local_irq_restore(flags);
1609 EXPORT_SYMBOL(dev_kfree_skb_irq);
1611 void dev_kfree_skb_any(struct sk_buff *skb)
1613 if (in_irq() || irqs_disabled())
1614 dev_kfree_skb_irq(skb);
1618 EXPORT_SYMBOL(dev_kfree_skb_any);
1622 * netif_device_detach - mark device as removed
1623 * @dev: network device
1625 * Mark device as removed from system and therefore no longer available.
1627 void netif_device_detach(struct net_device *dev)
1629 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1630 netif_running(dev)) {
1631 netif_tx_stop_all_queues(dev);
1634 EXPORT_SYMBOL(netif_device_detach);
1637 * netif_device_attach - mark device as attached
1638 * @dev: network device
1640 * Mark device as attached from system and restart if needed.
1642 void netif_device_attach(struct net_device *dev)
1644 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1645 netif_running(dev)) {
1646 netif_tx_wake_all_queues(dev);
1647 __netdev_watchdog_up(dev);
1650 EXPORT_SYMBOL(netif_device_attach);
1652 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1654 return ((features & NETIF_F_GEN_CSUM) ||
1655 ((features & NETIF_F_IP_CSUM) &&
1656 protocol == htons(ETH_P_IP)) ||
1657 ((features & NETIF_F_IPV6_CSUM) &&
1658 protocol == htons(ETH_P_IPV6)) ||
1659 ((features & NETIF_F_FCOE_CRC) &&
1660 protocol == htons(ETH_P_FCOE)));
1663 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1665 if (can_checksum_protocol(dev->features, skb->protocol))
1668 if (skb->protocol == htons(ETH_P_8021Q)) {
1669 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1670 if (can_checksum_protocol(dev->features & dev->vlan_features,
1671 veh->h_vlan_encapsulated_proto))
1679 * skb_dev_set -- assign a new device to a buffer
1680 * @skb: buffer for the new device
1681 * @dev: network device
1683 * If an skb is owned by a device already, we have to reset
1684 * all data private to the namespace a device belongs to
1685 * before assigning it a new device.
1687 #ifdef CONFIG_NET_NS
1688 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1691 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1694 skb_init_secmark(skb);
1698 skb->ipvs_property = 0;
1699 #ifdef CONFIG_NET_SCHED
1705 EXPORT_SYMBOL(skb_set_dev);
1706 #endif /* CONFIG_NET_NS */
1709 * Invalidate hardware checksum when packet is to be mangled, and
1710 * complete checksum manually on outgoing path.
1712 int skb_checksum_help(struct sk_buff *skb)
1715 int ret = 0, offset;
1717 if (skb->ip_summed == CHECKSUM_COMPLETE)
1718 goto out_set_summed;
1720 if (unlikely(skb_shinfo(skb)->gso_size)) {
1721 /* Let GSO fix up the checksum. */
1722 goto out_set_summed;
1725 offset = skb->csum_start - skb_headroom(skb);
1726 BUG_ON(offset >= skb_headlen(skb));
1727 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1729 offset += skb->csum_offset;
1730 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1732 if (skb_cloned(skb) &&
1733 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1734 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1739 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1741 skb->ip_summed = CHECKSUM_NONE;
1745 EXPORT_SYMBOL(skb_checksum_help);
1748 * skb_gso_segment - Perform segmentation on skb.
1749 * @skb: buffer to segment
1750 * @features: features for the output path (see dev->features)
1752 * This function segments the given skb and returns a list of segments.
1754 * It may return NULL if the skb requires no segmentation. This is
1755 * only possible when GSO is used for verifying header integrity.
1757 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1759 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1760 struct packet_type *ptype;
1761 __be16 type = skb->protocol;
1764 skb_reset_mac_header(skb);
1765 skb->mac_len = skb->network_header - skb->mac_header;
1766 __skb_pull(skb, skb->mac_len);
1768 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1769 struct net_device *dev = skb->dev;
1770 struct ethtool_drvinfo info = {};
1772 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1773 dev->ethtool_ops->get_drvinfo(dev, &info);
1775 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1777 info.driver, dev ? dev->features : 0L,
1778 skb->sk ? skb->sk->sk_route_caps : 0L,
1779 skb->len, skb->data_len, skb->ip_summed);
1781 if (skb_header_cloned(skb) &&
1782 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1783 return ERR_PTR(err);
1787 list_for_each_entry_rcu(ptype,
1788 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1789 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1790 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1791 err = ptype->gso_send_check(skb);
1792 segs = ERR_PTR(err);
1793 if (err || skb_gso_ok(skb, features))
1795 __skb_push(skb, (skb->data -
1796 skb_network_header(skb)));
1798 segs = ptype->gso_segment(skb, features);
1804 __skb_push(skb, skb->data - skb_mac_header(skb));
1808 EXPORT_SYMBOL(skb_gso_segment);
1810 /* Take action when hardware reception checksum errors are detected. */
1812 void netdev_rx_csum_fault(struct net_device *dev)
1814 if (net_ratelimit()) {
1815 printk(KERN_ERR "%s: hw csum failure.\n",
1816 dev ? dev->name : "<unknown>");
1820 EXPORT_SYMBOL(netdev_rx_csum_fault);
1823 /* Actually, we should eliminate this check as soon as we know, that:
1824 * 1. IOMMU is present and allows to map all the memory.
1825 * 2. No high memory really exists on this machine.
1828 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1830 #ifdef CONFIG_HIGHMEM
1832 if (!(dev->features & NETIF_F_HIGHDMA)) {
1833 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1834 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1838 if (PCI_DMA_BUS_IS_PHYS) {
1839 struct device *pdev = dev->dev.parent;
1843 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1844 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1845 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1854 void (*destructor)(struct sk_buff *skb);
1857 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1859 static void dev_gso_skb_destructor(struct sk_buff *skb)
1861 struct dev_gso_cb *cb;
1864 struct sk_buff *nskb = skb->next;
1866 skb->next = nskb->next;
1869 } while (skb->next);
1871 cb = DEV_GSO_CB(skb);
1873 cb->destructor(skb);
1877 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1878 * @skb: buffer to segment
1880 * This function segments the given skb and stores the list of segments
1883 static int dev_gso_segment(struct sk_buff *skb)
1885 struct net_device *dev = skb->dev;
1886 struct sk_buff *segs;
1887 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1890 segs = skb_gso_segment(skb, features);
1892 /* Verifying header integrity only. */
1897 return PTR_ERR(segs);
1900 DEV_GSO_CB(skb)->destructor = skb->destructor;
1901 skb->destructor = dev_gso_skb_destructor;
1907 * Try to orphan skb early, right before transmission by the device.
1908 * We cannot orphan skb if tx timestamp is requested, since
1909 * drivers need to call skb_tstamp_tx() to send the timestamp.
1911 static inline void skb_orphan_try(struct sk_buff *skb)
1913 struct sock *sk = skb->sk;
1915 if (sk && !skb_tx(skb)->flags) {
1916 /* skb_tx_hash() wont be able to get sk.
1917 * We copy sk_hash into skb->rxhash
1920 skb->rxhash = sk->sk_hash;
1926 * Returns true if either:
1927 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1928 * 2. skb is fragmented and the device does not support SG, or if
1929 * at least one of fragments is in highmem and device does not
1930 * support DMA from it.
1932 static inline int skb_needs_linearize(struct sk_buff *skb,
1933 struct net_device *dev)
1935 return skb_is_nonlinear(skb) &&
1936 ((skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
1937 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
1938 illegal_highdma(dev, skb))));
1941 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1942 struct netdev_queue *txq)
1944 const struct net_device_ops *ops = dev->netdev_ops;
1945 int rc = NETDEV_TX_OK;
1947 if (likely(!skb->next)) {
1948 if (!list_empty(&ptype_all))
1949 dev_queue_xmit_nit(skb, dev);
1952 * If device doesnt need skb->dst, release it right now while
1953 * its hot in this cpu cache
1955 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1958 skb_orphan_try(skb);
1960 if (netif_needs_gso(dev, skb)) {
1961 if (unlikely(dev_gso_segment(skb)))
1966 if (skb_needs_linearize(skb, dev) &&
1967 __skb_linearize(skb))
1970 /* If packet is not checksummed and device does not
1971 * support checksumming for this protocol, complete
1972 * checksumming here.
1974 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1975 skb_set_transport_header(skb, skb->csum_start -
1977 if (!dev_can_checksum(dev, skb) &&
1978 skb_checksum_help(skb))
1983 rc = ops->ndo_start_xmit(skb, dev);
1984 if (rc == NETDEV_TX_OK)
1985 txq_trans_update(txq);
1991 struct sk_buff *nskb = skb->next;
1993 skb->next = nskb->next;
1997 * If device doesnt need nskb->dst, release it right now while
1998 * its hot in this cpu cache
2000 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2003 rc = ops->ndo_start_xmit(nskb, dev);
2004 if (unlikely(rc != NETDEV_TX_OK)) {
2005 if (rc & ~NETDEV_TX_MASK)
2006 goto out_kfree_gso_skb;
2007 nskb->next = skb->next;
2011 txq_trans_update(txq);
2012 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2013 return NETDEV_TX_BUSY;
2014 } while (skb->next);
2017 if (likely(skb->next == NULL))
2018 skb->destructor = DEV_GSO_CB(skb)->destructor;
2024 static u32 hashrnd __read_mostly;
2026 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
2030 if (skb_rx_queue_recorded(skb)) {
2031 hash = skb_get_rx_queue(skb);
2032 while (unlikely(hash >= dev->real_num_tx_queues))
2033 hash -= dev->real_num_tx_queues;
2037 if (skb->sk && skb->sk->sk_hash)
2038 hash = skb->sk->sk_hash;
2040 hash = (__force u16) skb->protocol ^ skb->rxhash;
2041 hash = jhash_1word(hash, hashrnd);
2043 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2045 EXPORT_SYMBOL(skb_tx_hash);
2047 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2049 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2050 if (net_ratelimit()) {
2051 pr_warning("%s selects TX queue %d, but "
2052 "real number of TX queues is %d\n",
2053 dev->name, queue_index, dev->real_num_tx_queues);
2060 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2061 struct sk_buff *skb)
2064 struct sock *sk = skb->sk;
2066 queue_index = sk_tx_queue_get(sk);
2067 if (queue_index < 0) {
2068 const struct net_device_ops *ops = dev->netdev_ops;
2070 if (ops->ndo_select_queue) {
2071 queue_index = ops->ndo_select_queue(dev, skb);
2072 queue_index = dev_cap_txqueue(dev, queue_index);
2075 if (dev->real_num_tx_queues > 1)
2076 queue_index = skb_tx_hash(dev, skb);
2079 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2081 if (dst && skb_dst(skb) == dst)
2082 sk_tx_queue_set(sk, queue_index);
2087 skb_set_queue_mapping(skb, queue_index);
2088 return netdev_get_tx_queue(dev, queue_index);
2091 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2092 struct net_device *dev,
2093 struct netdev_queue *txq)
2095 spinlock_t *root_lock = qdisc_lock(q);
2096 bool contended = qdisc_is_running(q);
2100 * Heuristic to force contended enqueues to serialize on a
2101 * separate lock before trying to get qdisc main lock.
2102 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2103 * and dequeue packets faster.
2105 if (unlikely(contended))
2106 spin_lock(&q->busylock);
2108 spin_lock(root_lock);
2109 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2112 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2113 qdisc_run_begin(q)) {
2115 * This is a work-conserving queue; there are no old skbs
2116 * waiting to be sent out; and the qdisc is not running -
2117 * xmit the skb directly.
2119 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2121 __qdisc_update_bstats(q, skb->len);
2122 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2123 if (unlikely(contended)) {
2124 spin_unlock(&q->busylock);
2131 rc = NET_XMIT_SUCCESS;
2134 rc = qdisc_enqueue_root(skb, q);
2135 if (qdisc_run_begin(q)) {
2136 if (unlikely(contended)) {
2137 spin_unlock(&q->busylock);
2143 spin_unlock(root_lock);
2144 if (unlikely(contended))
2145 spin_unlock(&q->busylock);
2150 * dev_queue_xmit - transmit a buffer
2151 * @skb: buffer to transmit
2153 * Queue a buffer for transmission to a network device. The caller must
2154 * have set the device and priority and built the buffer before calling
2155 * this function. The function can be called from an interrupt.
2157 * A negative errno code is returned on a failure. A success does not
2158 * guarantee the frame will be transmitted as it may be dropped due
2159 * to congestion or traffic shaping.
2161 * -----------------------------------------------------------------------------------
2162 * I notice this method can also return errors from the queue disciplines,
2163 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2166 * Regardless of the return value, the skb is consumed, so it is currently
2167 * difficult to retry a send to this method. (You can bump the ref count
2168 * before sending to hold a reference for retry if you are careful.)
2170 * When calling this method, interrupts MUST be enabled. This is because
2171 * the BH enable code must have IRQs enabled so that it will not deadlock.
2174 int dev_queue_xmit(struct sk_buff *skb)
2176 struct net_device *dev = skb->dev;
2177 struct netdev_queue *txq;
2181 /* Disable soft irqs for various locks below. Also
2182 * stops preemption for RCU.
2186 txq = dev_pick_tx(dev, skb);
2187 q = rcu_dereference_bh(txq->qdisc);
2189 #ifdef CONFIG_NET_CLS_ACT
2190 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2193 rc = __dev_xmit_skb(skb, q, dev, txq);
2197 /* The device has no queue. Common case for software devices:
2198 loopback, all the sorts of tunnels...
2200 Really, it is unlikely that netif_tx_lock protection is necessary
2201 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2203 However, it is possible, that they rely on protection
2206 Check this and shot the lock. It is not prone from deadlocks.
2207 Either shot noqueue qdisc, it is even simpler 8)
2209 if (dev->flags & IFF_UP) {
2210 int cpu = smp_processor_id(); /* ok because BHs are off */
2212 if (txq->xmit_lock_owner != cpu) {
2214 HARD_TX_LOCK(dev, txq, cpu);
2216 if (!netif_tx_queue_stopped(txq)) {
2217 rc = dev_hard_start_xmit(skb, dev, txq);
2218 if (dev_xmit_complete(rc)) {
2219 HARD_TX_UNLOCK(dev, txq);
2223 HARD_TX_UNLOCK(dev, txq);
2224 if (net_ratelimit())
2225 printk(KERN_CRIT "Virtual device %s asks to "
2226 "queue packet!\n", dev->name);
2228 /* Recursion is detected! It is possible,
2230 if (net_ratelimit())
2231 printk(KERN_CRIT "Dead loop on virtual device "
2232 "%s, fix it urgently!\n", dev->name);
2237 rcu_read_unlock_bh();
2242 rcu_read_unlock_bh();
2245 EXPORT_SYMBOL(dev_queue_xmit);
2248 /*=======================================================================
2250 =======================================================================*/
2252 int netdev_max_backlog __read_mostly = 1000;
2253 int netdev_tstamp_prequeue __read_mostly = 1;
2254 int netdev_budget __read_mostly = 300;
2255 int weight_p __read_mostly = 64; /* old backlog weight */
2257 /* Called with irq disabled */
2258 static inline void ____napi_schedule(struct softnet_data *sd,
2259 struct napi_struct *napi)
2261 list_add_tail(&napi->poll_list, &sd->poll_list);
2262 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2267 /* One global table that all flow-based protocols share. */
2268 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2269 EXPORT_SYMBOL(rps_sock_flow_table);
2272 * get_rps_cpu is called from netif_receive_skb and returns the target
2273 * CPU from the RPS map of the receiving queue for a given skb.
2274 * rcu_read_lock must be held on entry.
2276 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2277 struct rps_dev_flow **rflowp)
2279 struct ipv6hdr *ip6;
2281 struct netdev_rx_queue *rxqueue;
2282 struct rps_map *map;
2283 struct rps_dev_flow_table *flow_table;
2284 struct rps_sock_flow_table *sock_flow_table;
2288 u32 addr1, addr2, ihl;
2294 if (skb_rx_queue_recorded(skb)) {
2295 u16 index = skb_get_rx_queue(skb);
2296 if (unlikely(index >= dev->num_rx_queues)) {
2297 WARN_ONCE(dev->num_rx_queues > 1, "%s received packet "
2298 "on queue %u, but number of RX queues is %u\n",
2299 dev->name, index, dev->num_rx_queues);
2302 rxqueue = dev->_rx + index;
2306 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2310 goto got_hash; /* Skip hash computation on packet header */
2312 switch (skb->protocol) {
2313 case __constant_htons(ETH_P_IP):
2314 if (!pskb_may_pull(skb, sizeof(*ip)))
2317 ip = (struct iphdr *) skb->data;
2318 ip_proto = ip->protocol;
2319 addr1 = (__force u32) ip->saddr;
2320 addr2 = (__force u32) ip->daddr;
2323 case __constant_htons(ETH_P_IPV6):
2324 if (!pskb_may_pull(skb, sizeof(*ip6)))
2327 ip6 = (struct ipv6hdr *) skb->data;
2328 ip_proto = ip6->nexthdr;
2329 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2330 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2343 case IPPROTO_UDPLITE:
2344 if (pskb_may_pull(skb, (ihl * 4) + 4)) {
2345 ports.v32 = * (__force u32 *) (skb->data + (ihl * 4));
2346 if (ports.v16[1] < ports.v16[0])
2347 swap(ports.v16[0], ports.v16[1]);
2355 /* get a consistent hash (same value on both flow directions) */
2358 skb->rxhash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2363 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2364 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2365 if (flow_table && sock_flow_table) {
2367 struct rps_dev_flow *rflow;
2369 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2372 next_cpu = sock_flow_table->ents[skb->rxhash &
2373 sock_flow_table->mask];
2376 * If the desired CPU (where last recvmsg was done) is
2377 * different from current CPU (one in the rx-queue flow
2378 * table entry), switch if one of the following holds:
2379 * - Current CPU is unset (equal to RPS_NO_CPU).
2380 * - Current CPU is offline.
2381 * - The current CPU's queue tail has advanced beyond the
2382 * last packet that was enqueued using this table entry.
2383 * This guarantees that all previous packets for the flow
2384 * have been dequeued, thus preserving in order delivery.
2386 if (unlikely(tcpu != next_cpu) &&
2387 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2388 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2389 rflow->last_qtail)) >= 0)) {
2390 tcpu = rflow->cpu = next_cpu;
2391 if (tcpu != RPS_NO_CPU)
2392 rflow->last_qtail = per_cpu(softnet_data,
2393 tcpu).input_queue_head;
2395 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2402 map = rcu_dereference(rxqueue->rps_map);
2404 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2406 if (cpu_online(tcpu)) {
2416 /* Called from hardirq (IPI) context */
2417 static void rps_trigger_softirq(void *data)
2419 struct softnet_data *sd = data;
2421 ____napi_schedule(sd, &sd->backlog);
2425 #endif /* CONFIG_RPS */
2428 * Check if this softnet_data structure is another cpu one
2429 * If yes, queue it to our IPI list and return 1
2432 static int rps_ipi_queued(struct softnet_data *sd)
2435 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2438 sd->rps_ipi_next = mysd->rps_ipi_list;
2439 mysd->rps_ipi_list = sd;
2441 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2444 #endif /* CONFIG_RPS */
2449 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2450 * queue (may be a remote CPU queue).
2452 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2453 unsigned int *qtail)
2455 struct softnet_data *sd;
2456 unsigned long flags;
2458 sd = &per_cpu(softnet_data, cpu);
2460 local_irq_save(flags);
2463 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2464 if (skb_queue_len(&sd->input_pkt_queue)) {
2466 __skb_queue_tail(&sd->input_pkt_queue, skb);
2467 input_queue_tail_incr_save(sd, qtail);
2469 local_irq_restore(flags);
2470 return NET_RX_SUCCESS;
2473 /* Schedule NAPI for backlog device
2474 * We can use non atomic operation since we own the queue lock
2476 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2477 if (!rps_ipi_queued(sd))
2478 ____napi_schedule(sd, &sd->backlog);
2486 local_irq_restore(flags);
2493 * netif_rx - post buffer to the network code
2494 * @skb: buffer to post
2496 * This function receives a packet from a device driver and queues it for
2497 * the upper (protocol) levels to process. It always succeeds. The buffer
2498 * may be dropped during processing for congestion control or by the
2502 * NET_RX_SUCCESS (no congestion)
2503 * NET_RX_DROP (packet was dropped)
2507 int netif_rx(struct sk_buff *skb)
2511 /* if netpoll wants it, pretend we never saw it */
2512 if (netpoll_rx(skb))
2515 if (netdev_tstamp_prequeue)
2516 net_timestamp_check(skb);
2520 struct rps_dev_flow voidflow, *rflow = &voidflow;
2525 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2527 cpu = smp_processor_id();
2529 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2536 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2542 EXPORT_SYMBOL(netif_rx);
2544 int netif_rx_ni(struct sk_buff *skb)
2549 err = netif_rx(skb);
2550 if (local_softirq_pending())
2556 EXPORT_SYMBOL(netif_rx_ni);
2558 static void net_tx_action(struct softirq_action *h)
2560 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2562 if (sd->completion_queue) {
2563 struct sk_buff *clist;
2565 local_irq_disable();
2566 clist = sd->completion_queue;
2567 sd->completion_queue = NULL;
2571 struct sk_buff *skb = clist;
2572 clist = clist->next;
2574 WARN_ON(atomic_read(&skb->users));
2579 if (sd->output_queue) {
2582 local_irq_disable();
2583 head = sd->output_queue;
2584 sd->output_queue = NULL;
2585 sd->output_queue_tailp = &sd->output_queue;
2589 struct Qdisc *q = head;
2590 spinlock_t *root_lock;
2592 head = head->next_sched;
2594 root_lock = qdisc_lock(q);
2595 if (spin_trylock(root_lock)) {
2596 smp_mb__before_clear_bit();
2597 clear_bit(__QDISC_STATE_SCHED,
2600 spin_unlock(root_lock);
2602 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2604 __netif_reschedule(q);
2606 smp_mb__before_clear_bit();
2607 clear_bit(__QDISC_STATE_SCHED,
2615 static inline int deliver_skb(struct sk_buff *skb,
2616 struct packet_type *pt_prev,
2617 struct net_device *orig_dev)
2619 atomic_inc(&skb->users);
2620 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2623 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2624 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2625 /* This hook is defined here for ATM LANE */
2626 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2627 unsigned char *addr) __read_mostly;
2628 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2631 #ifdef CONFIG_NET_CLS_ACT
2632 /* TODO: Maybe we should just force sch_ingress to be compiled in
2633 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2634 * a compare and 2 stores extra right now if we dont have it on
2635 * but have CONFIG_NET_CLS_ACT
2636 * NOTE: This doesnt stop any functionality; if you dont have
2637 * the ingress scheduler, you just cant add policies on ingress.
2640 static int ing_filter(struct sk_buff *skb)
2642 struct net_device *dev = skb->dev;
2643 u32 ttl = G_TC_RTTL(skb->tc_verd);
2644 struct netdev_queue *rxq;
2645 int result = TC_ACT_OK;
2648 if (MAX_RED_LOOP < ttl++) {
2650 "Redir loop detected Dropping packet (%d->%d)\n",
2651 skb->skb_iif, dev->ifindex);
2655 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2656 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2658 rxq = &dev->rx_queue;
2661 if (q != &noop_qdisc) {
2662 spin_lock(qdisc_lock(q));
2663 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2664 result = qdisc_enqueue_root(skb, q);
2665 spin_unlock(qdisc_lock(q));
2671 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2672 struct packet_type **pt_prev,
2673 int *ret, struct net_device *orig_dev)
2675 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2679 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2683 switch (ing_filter(skb)) {
2697 * netif_nit_deliver - deliver received packets to network taps
2700 * This function is used to deliver incoming packets to network
2701 * taps. It should be used when the normal netif_receive_skb path
2702 * is bypassed, for example because of VLAN acceleration.
2704 void netif_nit_deliver(struct sk_buff *skb)
2706 struct packet_type *ptype;
2708 if (list_empty(&ptype_all))
2711 skb_reset_network_header(skb);
2712 skb_reset_transport_header(skb);
2713 skb->mac_len = skb->network_header - skb->mac_header;
2716 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2717 if (!ptype->dev || ptype->dev == skb->dev)
2718 deliver_skb(skb, ptype, skb->dev);
2724 * netdev_rx_handler_register - register receive handler
2725 * @dev: device to register a handler for
2726 * @rx_handler: receive handler to register
2727 * @rx_handler_data: data pointer that is used by rx handler
2729 * Register a receive hander for a device. This handler will then be
2730 * called from __netif_receive_skb. A negative errno code is returned
2733 * The caller must hold the rtnl_mutex.
2735 int netdev_rx_handler_register(struct net_device *dev,
2736 rx_handler_func_t *rx_handler,
2737 void *rx_handler_data)
2741 if (dev->rx_handler)
2744 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2745 rcu_assign_pointer(dev->rx_handler, rx_handler);
2749 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2752 * netdev_rx_handler_unregister - unregister receive handler
2753 * @dev: device to unregister a handler from
2755 * Unregister a receive hander from a device.
2757 * The caller must hold the rtnl_mutex.
2759 void netdev_rx_handler_unregister(struct net_device *dev)
2763 rcu_assign_pointer(dev->rx_handler, NULL);
2764 rcu_assign_pointer(dev->rx_handler_data, NULL);
2766 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2768 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2769 struct net_device *master)
2771 if (skb->pkt_type == PACKET_HOST) {
2772 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2774 memcpy(dest, master->dev_addr, ETH_ALEN);
2778 /* On bonding slaves other than the currently active slave, suppress
2779 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2780 * ARP on active-backup slaves with arp_validate enabled.
2782 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2784 struct net_device *dev = skb->dev;
2786 if (master->priv_flags & IFF_MASTER_ARPMON)
2787 dev->last_rx = jiffies;
2789 if ((master->priv_flags & IFF_MASTER_ALB) &&
2790 (master->priv_flags & IFF_BRIDGE_PORT)) {
2791 /* Do address unmangle. The local destination address
2792 * will be always the one master has. Provides the right
2793 * functionality in a bridge.
2795 skb_bond_set_mac_by_master(skb, master);
2798 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2799 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2800 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2803 if (master->priv_flags & IFF_MASTER_ALB) {
2804 if (skb->pkt_type != PACKET_BROADCAST &&
2805 skb->pkt_type != PACKET_MULTICAST)
2808 if (master->priv_flags & IFF_MASTER_8023AD &&
2809 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2816 EXPORT_SYMBOL(__skb_bond_should_drop);
2818 static int __netif_receive_skb(struct sk_buff *skb)
2820 struct packet_type *ptype, *pt_prev;
2821 rx_handler_func_t *rx_handler;
2822 struct net_device *orig_dev;
2823 struct net_device *master;
2824 struct net_device *null_or_orig;
2825 struct net_device *orig_or_bond;
2826 int ret = NET_RX_DROP;
2829 if (!netdev_tstamp_prequeue)
2830 net_timestamp_check(skb);
2832 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2833 return NET_RX_SUCCESS;
2835 /* if we've gotten here through NAPI, check netpoll */
2836 if (netpoll_receive_skb(skb))
2840 skb->skb_iif = skb->dev->ifindex;
2843 * bonding note: skbs received on inactive slaves should only
2844 * be delivered to pkt handlers that are exact matches. Also
2845 * the deliver_no_wcard flag will be set. If packet handlers
2846 * are sensitive to duplicate packets these skbs will need to
2847 * be dropped at the handler. The vlan accel path may have
2848 * already set the deliver_no_wcard flag.
2850 null_or_orig = NULL;
2851 orig_dev = skb->dev;
2852 master = ACCESS_ONCE(orig_dev->master);
2853 if (skb->deliver_no_wcard)
2854 null_or_orig = orig_dev;
2856 if (skb_bond_should_drop(skb, master)) {
2857 skb->deliver_no_wcard = 1;
2858 null_or_orig = orig_dev; /* deliver only exact match */
2863 __this_cpu_inc(softnet_data.processed);
2864 skb_reset_network_header(skb);
2865 skb_reset_transport_header(skb);
2866 skb->mac_len = skb->network_header - skb->mac_header;
2872 #ifdef CONFIG_NET_CLS_ACT
2873 if (skb->tc_verd & TC_NCLS) {
2874 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2879 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2880 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2881 ptype->dev == orig_dev) {
2883 ret = deliver_skb(skb, pt_prev, orig_dev);
2888 #ifdef CONFIG_NET_CLS_ACT
2889 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2895 /* Handle special case of bridge or macvlan */
2896 rx_handler = rcu_dereference(skb->dev->rx_handler);
2899 ret = deliver_skb(skb, pt_prev, orig_dev);
2902 skb = rx_handler(skb);
2908 * Make sure frames received on VLAN interfaces stacked on
2909 * bonding interfaces still make their way to any base bonding
2910 * device that may have registered for a specific ptype. The
2911 * handler may have to adjust skb->dev and orig_dev.
2913 orig_or_bond = orig_dev;
2914 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2915 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2916 orig_or_bond = vlan_dev_real_dev(skb->dev);
2919 type = skb->protocol;
2920 list_for_each_entry_rcu(ptype,
2921 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2922 if (ptype->type == type && (ptype->dev == null_or_orig ||
2923 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2924 ptype->dev == orig_or_bond)) {
2926 ret = deliver_skb(skb, pt_prev, orig_dev);
2932 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2935 /* Jamal, now you will not able to escape explaining
2936 * me how you were going to use this. :-)
2947 * netif_receive_skb - process receive buffer from network
2948 * @skb: buffer to process
2950 * netif_receive_skb() is the main receive data processing function.
2951 * It always succeeds. The buffer may be dropped during processing
2952 * for congestion control or by the protocol layers.
2954 * This function may only be called from softirq context and interrupts
2955 * should be enabled.
2957 * Return values (usually ignored):
2958 * NET_RX_SUCCESS: no congestion
2959 * NET_RX_DROP: packet was dropped
2961 int netif_receive_skb(struct sk_buff *skb)
2963 if (netdev_tstamp_prequeue)
2964 net_timestamp_check(skb);
2966 if (skb_defer_rx_timestamp(skb))
2967 return NET_RX_SUCCESS;
2971 struct rps_dev_flow voidflow, *rflow = &voidflow;
2976 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2979 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2983 ret = __netif_receive_skb(skb);
2989 return __netif_receive_skb(skb);
2992 EXPORT_SYMBOL(netif_receive_skb);
2994 /* Network device is going away, flush any packets still pending
2995 * Called with irqs disabled.
2997 static void flush_backlog(void *arg)
2999 struct net_device *dev = arg;
3000 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3001 struct sk_buff *skb, *tmp;
3004 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3005 if (skb->dev == dev) {
3006 __skb_unlink(skb, &sd->input_pkt_queue);
3008 input_queue_head_incr(sd);
3013 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3014 if (skb->dev == dev) {
3015 __skb_unlink(skb, &sd->process_queue);
3017 input_queue_head_incr(sd);
3022 static int napi_gro_complete(struct sk_buff *skb)
3024 struct packet_type *ptype;
3025 __be16 type = skb->protocol;
3026 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3029 if (NAPI_GRO_CB(skb)->count == 1) {
3030 skb_shinfo(skb)->gso_size = 0;
3035 list_for_each_entry_rcu(ptype, head, list) {
3036 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3039 err = ptype->gro_complete(skb);
3045 WARN_ON(&ptype->list == head);
3047 return NET_RX_SUCCESS;
3051 return netif_receive_skb(skb);
3054 static void napi_gro_flush(struct napi_struct *napi)
3056 struct sk_buff *skb, *next;
3058 for (skb = napi->gro_list; skb; skb = next) {
3061 napi_gro_complete(skb);
3064 napi->gro_count = 0;
3065 napi->gro_list = NULL;
3068 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3070 struct sk_buff **pp = NULL;
3071 struct packet_type *ptype;
3072 __be16 type = skb->protocol;
3073 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3076 enum gro_result ret;
3078 if (!(skb->dev->features & NETIF_F_GRO))
3081 if (skb_is_gso(skb) || skb_has_frags(skb))
3085 list_for_each_entry_rcu(ptype, head, list) {
3086 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3089 skb_set_network_header(skb, skb_gro_offset(skb));
3090 mac_len = skb->network_header - skb->mac_header;
3091 skb->mac_len = mac_len;
3092 NAPI_GRO_CB(skb)->same_flow = 0;
3093 NAPI_GRO_CB(skb)->flush = 0;
3094 NAPI_GRO_CB(skb)->free = 0;
3096 pp = ptype->gro_receive(&napi->gro_list, skb);
3101 if (&ptype->list == head)
3104 same_flow = NAPI_GRO_CB(skb)->same_flow;
3105 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3108 struct sk_buff *nskb = *pp;
3112 napi_gro_complete(nskb);
3119 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3123 NAPI_GRO_CB(skb)->count = 1;
3124 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3125 skb->next = napi->gro_list;
3126 napi->gro_list = skb;
3130 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3131 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3133 BUG_ON(skb->end - skb->tail < grow);
3135 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3138 skb->data_len -= grow;
3140 skb_shinfo(skb)->frags[0].page_offset += grow;
3141 skb_shinfo(skb)->frags[0].size -= grow;
3143 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3144 put_page(skb_shinfo(skb)->frags[0].page);
3145 memmove(skb_shinfo(skb)->frags,
3146 skb_shinfo(skb)->frags + 1,
3147 --skb_shinfo(skb)->nr_frags);
3158 EXPORT_SYMBOL(dev_gro_receive);
3161 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3165 if (netpoll_rx_on(skb))
3168 for (p = napi->gro_list; p; p = p->next) {
3169 NAPI_GRO_CB(p)->same_flow =
3170 (p->dev == skb->dev) &&
3171 !compare_ether_header(skb_mac_header(p),
3172 skb_gro_mac_header(skb));
3173 NAPI_GRO_CB(p)->flush = 0;
3176 return dev_gro_receive(napi, skb);
3179 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3183 if (netif_receive_skb(skb))
3188 case GRO_MERGED_FREE:
3199 EXPORT_SYMBOL(napi_skb_finish);
3201 void skb_gro_reset_offset(struct sk_buff *skb)
3203 NAPI_GRO_CB(skb)->data_offset = 0;
3204 NAPI_GRO_CB(skb)->frag0 = NULL;
3205 NAPI_GRO_CB(skb)->frag0_len = 0;
3207 if (skb->mac_header == skb->tail &&
3208 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3209 NAPI_GRO_CB(skb)->frag0 =
3210 page_address(skb_shinfo(skb)->frags[0].page) +
3211 skb_shinfo(skb)->frags[0].page_offset;
3212 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3215 EXPORT_SYMBOL(skb_gro_reset_offset);
3217 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3219 skb_gro_reset_offset(skb);
3221 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3223 EXPORT_SYMBOL(napi_gro_receive);
3225 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3227 __skb_pull(skb, skb_headlen(skb));
3228 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3232 EXPORT_SYMBOL(napi_reuse_skb);
3234 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3236 struct sk_buff *skb = napi->skb;
3239 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3245 EXPORT_SYMBOL(napi_get_frags);
3247 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3253 skb->protocol = eth_type_trans(skb, skb->dev);
3255 if (ret == GRO_HELD)
3256 skb_gro_pull(skb, -ETH_HLEN);
3257 else if (netif_receive_skb(skb))
3262 case GRO_MERGED_FREE:
3263 napi_reuse_skb(napi, skb);
3272 EXPORT_SYMBOL(napi_frags_finish);
3274 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3276 struct sk_buff *skb = napi->skb;
3283 skb_reset_mac_header(skb);
3284 skb_gro_reset_offset(skb);
3286 off = skb_gro_offset(skb);
3287 hlen = off + sizeof(*eth);
3288 eth = skb_gro_header_fast(skb, off);
3289 if (skb_gro_header_hard(skb, hlen)) {
3290 eth = skb_gro_header_slow(skb, hlen, off);
3291 if (unlikely(!eth)) {
3292 napi_reuse_skb(napi, skb);
3298 skb_gro_pull(skb, sizeof(*eth));
3301 * This works because the only protocols we care about don't require
3302 * special handling. We'll fix it up properly at the end.
3304 skb->protocol = eth->h_proto;
3309 EXPORT_SYMBOL(napi_frags_skb);
3311 gro_result_t napi_gro_frags(struct napi_struct *napi)
3313 struct sk_buff *skb = napi_frags_skb(napi);
3318 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3320 EXPORT_SYMBOL(napi_gro_frags);
3323 * net_rps_action sends any pending IPI's for rps.
3324 * Note: called with local irq disabled, but exits with local irq enabled.
3326 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3329 struct softnet_data *remsd = sd->rps_ipi_list;
3332 sd->rps_ipi_list = NULL;
3336 /* Send pending IPI's to kick RPS processing on remote cpus. */
3338 struct softnet_data *next = remsd->rps_ipi_next;
3340 if (cpu_online(remsd->cpu))
3341 __smp_call_function_single(remsd->cpu,
3350 static int process_backlog(struct napi_struct *napi, int quota)
3353 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3356 /* Check if we have pending ipi, its better to send them now,
3357 * not waiting net_rx_action() end.
3359 if (sd->rps_ipi_list) {
3360 local_irq_disable();
3361 net_rps_action_and_irq_enable(sd);
3364 napi->weight = weight_p;
3365 local_irq_disable();
3366 while (work < quota) {
3367 struct sk_buff *skb;
3370 while ((skb = __skb_dequeue(&sd->process_queue))) {
3372 __netif_receive_skb(skb);
3373 local_irq_disable();
3374 input_queue_head_incr(sd);
3375 if (++work >= quota) {
3382 qlen = skb_queue_len(&sd->input_pkt_queue);
3384 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3385 &sd->process_queue);
3387 if (qlen < quota - work) {
3389 * Inline a custom version of __napi_complete().
3390 * only current cpu owns and manipulates this napi,
3391 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3392 * we can use a plain write instead of clear_bit(),
3393 * and we dont need an smp_mb() memory barrier.
3395 list_del(&napi->poll_list);
3398 quota = work + qlen;
3408 * __napi_schedule - schedule for receive
3409 * @n: entry to schedule
3411 * The entry's receive function will be scheduled to run
3413 void __napi_schedule(struct napi_struct *n)
3415 unsigned long flags;
3417 local_irq_save(flags);
3418 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3419 local_irq_restore(flags);
3421 EXPORT_SYMBOL(__napi_schedule);
3423 void __napi_complete(struct napi_struct *n)
3425 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3426 BUG_ON(n->gro_list);
3428 list_del(&n->poll_list);
3429 smp_mb__before_clear_bit();
3430 clear_bit(NAPI_STATE_SCHED, &n->state);
3432 EXPORT_SYMBOL(__napi_complete);
3434 void napi_complete(struct napi_struct *n)
3436 unsigned long flags;
3439 * don't let napi dequeue from the cpu poll list
3440 * just in case its running on a different cpu
3442 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3446 local_irq_save(flags);
3448 local_irq_restore(flags);
3450 EXPORT_SYMBOL(napi_complete);
3452 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3453 int (*poll)(struct napi_struct *, int), int weight)
3455 INIT_LIST_HEAD(&napi->poll_list);
3456 napi->gro_count = 0;
3457 napi->gro_list = NULL;
3460 napi->weight = weight;
3461 list_add(&napi->dev_list, &dev->napi_list);
3463 #ifdef CONFIG_NETPOLL
3464 spin_lock_init(&napi->poll_lock);
3465 napi->poll_owner = -1;
3467 set_bit(NAPI_STATE_SCHED, &napi->state);
3469 EXPORT_SYMBOL(netif_napi_add);
3471 void netif_napi_del(struct napi_struct *napi)
3473 struct sk_buff *skb, *next;
3475 list_del_init(&napi->dev_list);
3476 napi_free_frags(napi);
3478 for (skb = napi->gro_list; skb; skb = next) {
3484 napi->gro_list = NULL;
3485 napi->gro_count = 0;
3487 EXPORT_SYMBOL(netif_napi_del);
3489 static void net_rx_action(struct softirq_action *h)
3491 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3492 unsigned long time_limit = jiffies + 2;
3493 int budget = netdev_budget;
3496 local_irq_disable();
3498 while (!list_empty(&sd->poll_list)) {
3499 struct napi_struct *n;
3502 /* If softirq window is exhuasted then punt.
3503 * Allow this to run for 2 jiffies since which will allow
3504 * an average latency of 1.5/HZ.
3506 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3511 /* Even though interrupts have been re-enabled, this
3512 * access is safe because interrupts can only add new
3513 * entries to the tail of this list, and only ->poll()
3514 * calls can remove this head entry from the list.
3516 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3518 have = netpoll_poll_lock(n);
3522 /* This NAPI_STATE_SCHED test is for avoiding a race
3523 * with netpoll's poll_napi(). Only the entity which
3524 * obtains the lock and sees NAPI_STATE_SCHED set will
3525 * actually make the ->poll() call. Therefore we avoid
3526 * accidently calling ->poll() when NAPI is not scheduled.
3529 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3530 work = n->poll(n, weight);
3534 WARN_ON_ONCE(work > weight);
3538 local_irq_disable();
3540 /* Drivers must not modify the NAPI state if they
3541 * consume the entire weight. In such cases this code
3542 * still "owns" the NAPI instance and therefore can
3543 * move the instance around on the list at-will.
3545 if (unlikely(work == weight)) {
3546 if (unlikely(napi_disable_pending(n))) {
3549 local_irq_disable();
3551 list_move_tail(&n->poll_list, &sd->poll_list);
3554 netpoll_poll_unlock(have);
3557 net_rps_action_and_irq_enable(sd);
3559 #ifdef CONFIG_NET_DMA
3561 * There may not be any more sk_buffs coming right now, so push
3562 * any pending DMA copies to hardware
3564 dma_issue_pending_all();
3571 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3575 static gifconf_func_t *gifconf_list[NPROTO];
3578 * register_gifconf - register a SIOCGIF handler
3579 * @family: Address family
3580 * @gifconf: Function handler
3582 * Register protocol dependent address dumping routines. The handler
3583 * that is passed must not be freed or reused until it has been replaced
3584 * by another handler.
3586 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3588 if (family >= NPROTO)
3590 gifconf_list[family] = gifconf;
3593 EXPORT_SYMBOL(register_gifconf);
3597 * Map an interface index to its name (SIOCGIFNAME)
3601 * We need this ioctl for efficient implementation of the
3602 * if_indextoname() function required by the IPv6 API. Without
3603 * it, we would have to search all the interfaces to find a
3607 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3609 struct net_device *dev;
3613 * Fetch the caller's info block.
3616 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3620 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3626 strcpy(ifr.ifr_name, dev->name);
3629 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3635 * Perform a SIOCGIFCONF call. This structure will change
3636 * size eventually, and there is nothing I can do about it.
3637 * Thus we will need a 'compatibility mode'.
3640 static int dev_ifconf(struct net *net, char __user *arg)
3643 struct net_device *dev;
3650 * Fetch the caller's info block.
3653 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3660 * Loop over the interfaces, and write an info block for each.
3664 for_each_netdev(net, dev) {
3665 for (i = 0; i < NPROTO; i++) {
3666 if (gifconf_list[i]) {
3669 done = gifconf_list[i](dev, NULL, 0);
3671 done = gifconf_list[i](dev, pos + total,
3681 * All done. Write the updated control block back to the caller.
3683 ifc.ifc_len = total;
3686 * Both BSD and Solaris return 0 here, so we do too.
3688 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3691 #ifdef CONFIG_PROC_FS
3693 * This is invoked by the /proc filesystem handler to display a device
3696 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3699 struct net *net = seq_file_net(seq);
3701 struct net_device *dev;
3705 return SEQ_START_TOKEN;
3708 for_each_netdev_rcu(net, dev)
3715 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3717 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3718 first_net_device(seq_file_net(seq)) :
3719 next_net_device((struct net_device *)v);
3722 return rcu_dereference(dev);
3725 void dev_seq_stop(struct seq_file *seq, void *v)
3731 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3733 struct rtnl_link_stats64 temp;
3734 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3736 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3737 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3738 dev->name, stats->rx_bytes, stats->rx_packets,
3740 stats->rx_dropped + stats->rx_missed_errors,
3741 stats->rx_fifo_errors,
3742 stats->rx_length_errors + stats->rx_over_errors +
3743 stats->rx_crc_errors + stats->rx_frame_errors,
3744 stats->rx_compressed, stats->multicast,
3745 stats->tx_bytes, stats->tx_packets,
3746 stats->tx_errors, stats->tx_dropped,
3747 stats->tx_fifo_errors, stats->collisions,
3748 stats->tx_carrier_errors +
3749 stats->tx_aborted_errors +
3750 stats->tx_window_errors +
3751 stats->tx_heartbeat_errors,
3752 stats->tx_compressed);
3756 * Called from the PROCfs module. This now uses the new arbitrary sized
3757 * /proc/net interface to create /proc/net/dev
3759 static int dev_seq_show(struct seq_file *seq, void *v)
3761 if (v == SEQ_START_TOKEN)
3762 seq_puts(seq, "Inter-| Receive "
3764 " face |bytes packets errs drop fifo frame "
3765 "compressed multicast|bytes packets errs "
3766 "drop fifo colls carrier compressed\n");
3768 dev_seq_printf_stats(seq, v);
3772 static struct softnet_data *softnet_get_online(loff_t *pos)
3774 struct softnet_data *sd = NULL;
3776 while (*pos < nr_cpu_ids)
3777 if (cpu_online(*pos)) {
3778 sd = &per_cpu(softnet_data, *pos);
3785 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3787 return softnet_get_online(pos);
3790 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3793 return softnet_get_online(pos);
3796 static void softnet_seq_stop(struct seq_file *seq, void *v)
3800 static int softnet_seq_show(struct seq_file *seq, void *v)
3802 struct softnet_data *sd = v;
3804 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3805 sd->processed, sd->dropped, sd->time_squeeze, 0,
3806 0, 0, 0, 0, /* was fastroute */
3807 sd->cpu_collision, sd->received_rps);
3811 static const struct seq_operations dev_seq_ops = {
3812 .start = dev_seq_start,
3813 .next = dev_seq_next,
3814 .stop = dev_seq_stop,
3815 .show = dev_seq_show,
3818 static int dev_seq_open(struct inode *inode, struct file *file)
3820 return seq_open_net(inode, file, &dev_seq_ops,
3821 sizeof(struct seq_net_private));
3824 static const struct file_operations dev_seq_fops = {
3825 .owner = THIS_MODULE,
3826 .open = dev_seq_open,
3828 .llseek = seq_lseek,
3829 .release = seq_release_net,
3832 static const struct seq_operations softnet_seq_ops = {
3833 .start = softnet_seq_start,
3834 .next = softnet_seq_next,
3835 .stop = softnet_seq_stop,
3836 .show = softnet_seq_show,
3839 static int softnet_seq_open(struct inode *inode, struct file *file)
3841 return seq_open(file, &softnet_seq_ops);
3844 static const struct file_operations softnet_seq_fops = {
3845 .owner = THIS_MODULE,
3846 .open = softnet_seq_open,
3848 .llseek = seq_lseek,
3849 .release = seq_release,
3852 static void *ptype_get_idx(loff_t pos)
3854 struct packet_type *pt = NULL;
3858 list_for_each_entry_rcu(pt, &ptype_all, list) {
3864 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3865 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3874 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3878 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3881 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3883 struct packet_type *pt;
3884 struct list_head *nxt;
3888 if (v == SEQ_START_TOKEN)
3889 return ptype_get_idx(0);
3892 nxt = pt->list.next;
3893 if (pt->type == htons(ETH_P_ALL)) {
3894 if (nxt != &ptype_all)
3897 nxt = ptype_base[0].next;
3899 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3901 while (nxt == &ptype_base[hash]) {
3902 if (++hash >= PTYPE_HASH_SIZE)
3904 nxt = ptype_base[hash].next;
3907 return list_entry(nxt, struct packet_type, list);
3910 static void ptype_seq_stop(struct seq_file *seq, void *v)
3916 static int ptype_seq_show(struct seq_file *seq, void *v)
3918 struct packet_type *pt = v;
3920 if (v == SEQ_START_TOKEN)
3921 seq_puts(seq, "Type Device Function\n");
3922 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3923 if (pt->type == htons(ETH_P_ALL))
3924 seq_puts(seq, "ALL ");
3926 seq_printf(seq, "%04x", ntohs(pt->type));
3928 seq_printf(seq, " %-8s %pF\n",
3929 pt->dev ? pt->dev->name : "", pt->func);
3935 static const struct seq_operations ptype_seq_ops = {
3936 .start = ptype_seq_start,
3937 .next = ptype_seq_next,
3938 .stop = ptype_seq_stop,
3939 .show = ptype_seq_show,
3942 static int ptype_seq_open(struct inode *inode, struct file *file)
3944 return seq_open_net(inode, file, &ptype_seq_ops,
3945 sizeof(struct seq_net_private));
3948 static const struct file_operations ptype_seq_fops = {
3949 .owner = THIS_MODULE,
3950 .open = ptype_seq_open,
3952 .llseek = seq_lseek,
3953 .release = seq_release_net,
3957 static int __net_init dev_proc_net_init(struct net *net)
3961 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3963 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3965 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3968 if (wext_proc_init(net))
3974 proc_net_remove(net, "ptype");
3976 proc_net_remove(net, "softnet_stat");
3978 proc_net_remove(net, "dev");
3982 static void __net_exit dev_proc_net_exit(struct net *net)
3984 wext_proc_exit(net);
3986 proc_net_remove(net, "ptype");
3987 proc_net_remove(net, "softnet_stat");
3988 proc_net_remove(net, "dev");
3991 static struct pernet_operations __net_initdata dev_proc_ops = {
3992 .init = dev_proc_net_init,
3993 .exit = dev_proc_net_exit,
3996 static int __init dev_proc_init(void)
3998 return register_pernet_subsys(&dev_proc_ops);
4001 #define dev_proc_init() 0
4002 #endif /* CONFIG_PROC_FS */
4006 * netdev_set_master - set up master/slave pair
4007 * @slave: slave device
4008 * @master: new master device
4010 * Changes the master device of the slave. Pass %NULL to break the
4011 * bonding. The caller must hold the RTNL semaphore. On a failure
4012 * a negative errno code is returned. On success the reference counts
4013 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4014 * function returns zero.
4016 int netdev_set_master(struct net_device *slave, struct net_device *master)
4018 struct net_device *old = slave->master;
4028 slave->master = master;
4035 slave->flags |= IFF_SLAVE;
4037 slave->flags &= ~IFF_SLAVE;
4039 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4042 EXPORT_SYMBOL(netdev_set_master);
4044 static void dev_change_rx_flags(struct net_device *dev, int flags)
4046 const struct net_device_ops *ops = dev->netdev_ops;
4048 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4049 ops->ndo_change_rx_flags(dev, flags);
4052 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4054 unsigned short old_flags = dev->flags;
4060 dev->flags |= IFF_PROMISC;
4061 dev->promiscuity += inc;
4062 if (dev->promiscuity == 0) {
4065 * If inc causes overflow, untouch promisc and return error.
4068 dev->flags &= ~IFF_PROMISC;
4070 dev->promiscuity -= inc;
4071 printk(KERN_WARNING "%s: promiscuity touches roof, "
4072 "set promiscuity failed, promiscuity feature "
4073 "of device might be broken.\n", dev->name);
4077 if (dev->flags != old_flags) {
4078 printk(KERN_INFO "device %s %s promiscuous mode\n",
4079 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4081 if (audit_enabled) {
4082 current_uid_gid(&uid, &gid);
4083 audit_log(current->audit_context, GFP_ATOMIC,
4084 AUDIT_ANOM_PROMISCUOUS,
4085 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4086 dev->name, (dev->flags & IFF_PROMISC),
4087 (old_flags & IFF_PROMISC),
4088 audit_get_loginuid(current),
4090 audit_get_sessionid(current));
4093 dev_change_rx_flags(dev, IFF_PROMISC);
4099 * dev_set_promiscuity - update promiscuity count on a device
4103 * Add or remove promiscuity from a device. While the count in the device
4104 * remains above zero the interface remains promiscuous. Once it hits zero
4105 * the device reverts back to normal filtering operation. A negative inc
4106 * value is used to drop promiscuity on the device.
4107 * Return 0 if successful or a negative errno code on error.
4109 int dev_set_promiscuity(struct net_device *dev, int inc)
4111 unsigned short old_flags = dev->flags;
4114 err = __dev_set_promiscuity(dev, inc);
4117 if (dev->flags != old_flags)
4118 dev_set_rx_mode(dev);
4121 EXPORT_SYMBOL(dev_set_promiscuity);
4124 * dev_set_allmulti - update allmulti count on a device
4128 * Add or remove reception of all multicast frames to a device. While the
4129 * count in the device remains above zero the interface remains listening
4130 * to all interfaces. Once it hits zero the device reverts back to normal
4131 * filtering operation. A negative @inc value is used to drop the counter
4132 * when releasing a resource needing all multicasts.
4133 * Return 0 if successful or a negative errno code on error.
4136 int dev_set_allmulti(struct net_device *dev, int inc)
4138 unsigned short old_flags = dev->flags;
4142 dev->flags |= IFF_ALLMULTI;
4143 dev->allmulti += inc;
4144 if (dev->allmulti == 0) {
4147 * If inc causes overflow, untouch allmulti and return error.
4150 dev->flags &= ~IFF_ALLMULTI;
4152 dev->allmulti -= inc;
4153 printk(KERN_WARNING "%s: allmulti touches roof, "
4154 "set allmulti failed, allmulti feature of "
4155 "device might be broken.\n", dev->name);
4159 if (dev->flags ^ old_flags) {
4160 dev_change_rx_flags(dev, IFF_ALLMULTI);
4161 dev_set_rx_mode(dev);
4165 EXPORT_SYMBOL(dev_set_allmulti);
4168 * Upload unicast and multicast address lists to device and
4169 * configure RX filtering. When the device doesn't support unicast
4170 * filtering it is put in promiscuous mode while unicast addresses
4173 void __dev_set_rx_mode(struct net_device *dev)
4175 const struct net_device_ops *ops = dev->netdev_ops;
4177 /* dev_open will call this function so the list will stay sane. */
4178 if (!(dev->flags&IFF_UP))
4181 if (!netif_device_present(dev))
4184 if (ops->ndo_set_rx_mode)
4185 ops->ndo_set_rx_mode(dev);
4187 /* Unicast addresses changes may only happen under the rtnl,
4188 * therefore calling __dev_set_promiscuity here is safe.
4190 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4191 __dev_set_promiscuity(dev, 1);
4192 dev->uc_promisc = 1;
4193 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4194 __dev_set_promiscuity(dev, -1);
4195 dev->uc_promisc = 0;
4198 if (ops->ndo_set_multicast_list)
4199 ops->ndo_set_multicast_list(dev);
4203 void dev_set_rx_mode(struct net_device *dev)
4205 netif_addr_lock_bh(dev);
4206 __dev_set_rx_mode(dev);
4207 netif_addr_unlock_bh(dev);
4211 * dev_get_flags - get flags reported to userspace
4214 * Get the combination of flag bits exported through APIs to userspace.
4216 unsigned dev_get_flags(const struct net_device *dev)
4220 flags = (dev->flags & ~(IFF_PROMISC |
4225 (dev->gflags & (IFF_PROMISC |
4228 if (netif_running(dev)) {
4229 if (netif_oper_up(dev))
4230 flags |= IFF_RUNNING;
4231 if (netif_carrier_ok(dev))
4232 flags |= IFF_LOWER_UP;
4233 if (netif_dormant(dev))
4234 flags |= IFF_DORMANT;
4239 EXPORT_SYMBOL(dev_get_flags);
4241 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4243 int old_flags = dev->flags;
4249 * Set the flags on our device.
4252 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4253 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4255 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4259 * Load in the correct multicast list now the flags have changed.
4262 if ((old_flags ^ flags) & IFF_MULTICAST)
4263 dev_change_rx_flags(dev, IFF_MULTICAST);
4265 dev_set_rx_mode(dev);
4268 * Have we downed the interface. We handle IFF_UP ourselves
4269 * according to user attempts to set it, rather than blindly
4274 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4275 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4278 dev_set_rx_mode(dev);
4281 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4282 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4284 dev->gflags ^= IFF_PROMISC;
4285 dev_set_promiscuity(dev, inc);
4288 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4289 is important. Some (broken) drivers set IFF_PROMISC, when
4290 IFF_ALLMULTI is requested not asking us and not reporting.
4292 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4293 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4295 dev->gflags ^= IFF_ALLMULTI;
4296 dev_set_allmulti(dev, inc);
4302 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4304 unsigned int changes = dev->flags ^ old_flags;
4306 if (changes & IFF_UP) {
4307 if (dev->flags & IFF_UP)
4308 call_netdevice_notifiers(NETDEV_UP, dev);
4310 call_netdevice_notifiers(NETDEV_DOWN, dev);
4313 if (dev->flags & IFF_UP &&
4314 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4315 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4319 * dev_change_flags - change device settings
4321 * @flags: device state flags
4323 * Change settings on device based state flags. The flags are
4324 * in the userspace exported format.
4326 int dev_change_flags(struct net_device *dev, unsigned flags)
4329 int old_flags = dev->flags;
4331 ret = __dev_change_flags(dev, flags);
4335 changes = old_flags ^ dev->flags;
4337 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4339 __dev_notify_flags(dev, old_flags);
4342 EXPORT_SYMBOL(dev_change_flags);
4345 * dev_set_mtu - Change maximum transfer unit
4347 * @new_mtu: new transfer unit
4349 * Change the maximum transfer size of the network device.
4351 int dev_set_mtu(struct net_device *dev, int new_mtu)
4353 const struct net_device_ops *ops = dev->netdev_ops;
4356 if (new_mtu == dev->mtu)
4359 /* MTU must be positive. */
4363 if (!netif_device_present(dev))
4367 if (ops->ndo_change_mtu)
4368 err = ops->ndo_change_mtu(dev, new_mtu);
4372 if (!err && dev->flags & IFF_UP)
4373 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4376 EXPORT_SYMBOL(dev_set_mtu);
4379 * dev_set_mac_address - Change Media Access Control Address
4383 * Change the hardware (MAC) address of the device
4385 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4387 const struct net_device_ops *ops = dev->netdev_ops;
4390 if (!ops->ndo_set_mac_address)
4392 if (sa->sa_family != dev->type)
4394 if (!netif_device_present(dev))
4396 err = ops->ndo_set_mac_address(dev, sa);
4398 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4401 EXPORT_SYMBOL(dev_set_mac_address);
4404 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4406 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4409 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4415 case SIOCGIFFLAGS: /* Get interface flags */
4416 ifr->ifr_flags = (short) dev_get_flags(dev);
4419 case SIOCGIFMETRIC: /* Get the metric on the interface
4420 (currently unused) */
4421 ifr->ifr_metric = 0;
4424 case SIOCGIFMTU: /* Get the MTU of a device */
4425 ifr->ifr_mtu = dev->mtu;
4430 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4432 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4433 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4434 ifr->ifr_hwaddr.sa_family = dev->type;
4442 ifr->ifr_map.mem_start = dev->mem_start;
4443 ifr->ifr_map.mem_end = dev->mem_end;
4444 ifr->ifr_map.base_addr = dev->base_addr;
4445 ifr->ifr_map.irq = dev->irq;
4446 ifr->ifr_map.dma = dev->dma;
4447 ifr->ifr_map.port = dev->if_port;
4451 ifr->ifr_ifindex = dev->ifindex;
4455 ifr->ifr_qlen = dev->tx_queue_len;
4459 /* dev_ioctl() should ensure this case
4471 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4473 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4476 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4477 const struct net_device_ops *ops;
4482 ops = dev->netdev_ops;
4485 case SIOCSIFFLAGS: /* Set interface flags */
4486 return dev_change_flags(dev, ifr->ifr_flags);
4488 case SIOCSIFMETRIC: /* Set the metric on the interface
4489 (currently unused) */
4492 case SIOCSIFMTU: /* Set the MTU of a device */
4493 return dev_set_mtu(dev, ifr->ifr_mtu);
4496 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4498 case SIOCSIFHWBROADCAST:
4499 if (ifr->ifr_hwaddr.sa_family != dev->type)
4501 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4502 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4503 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4507 if (ops->ndo_set_config) {
4508 if (!netif_device_present(dev))
4510 return ops->ndo_set_config(dev, &ifr->ifr_map);
4515 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4516 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4518 if (!netif_device_present(dev))
4520 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4523 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4524 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4526 if (!netif_device_present(dev))
4528 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4531 if (ifr->ifr_qlen < 0)
4533 dev->tx_queue_len = ifr->ifr_qlen;
4537 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4538 return dev_change_name(dev, ifr->ifr_newname);
4541 * Unknown or private ioctl
4544 if ((cmd >= SIOCDEVPRIVATE &&
4545 cmd <= SIOCDEVPRIVATE + 15) ||
4546 cmd == SIOCBONDENSLAVE ||
4547 cmd == SIOCBONDRELEASE ||
4548 cmd == SIOCBONDSETHWADDR ||
4549 cmd == SIOCBONDSLAVEINFOQUERY ||
4550 cmd == SIOCBONDINFOQUERY ||
4551 cmd == SIOCBONDCHANGEACTIVE ||
4552 cmd == SIOCGMIIPHY ||
4553 cmd == SIOCGMIIREG ||
4554 cmd == SIOCSMIIREG ||
4555 cmd == SIOCBRADDIF ||
4556 cmd == SIOCBRDELIF ||
4557 cmd == SIOCSHWTSTAMP ||
4558 cmd == SIOCWANDEV) {
4560 if (ops->ndo_do_ioctl) {
4561 if (netif_device_present(dev))
4562 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4574 * This function handles all "interface"-type I/O control requests. The actual
4575 * 'doing' part of this is dev_ifsioc above.
4579 * dev_ioctl - network device ioctl
4580 * @net: the applicable net namespace
4581 * @cmd: command to issue
4582 * @arg: pointer to a struct ifreq in user space
4584 * Issue ioctl functions to devices. This is normally called by the
4585 * user space syscall interfaces but can sometimes be useful for
4586 * other purposes. The return value is the return from the syscall if
4587 * positive or a negative errno code on error.
4590 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4596 /* One special case: SIOCGIFCONF takes ifconf argument
4597 and requires shared lock, because it sleeps writing
4601 if (cmd == SIOCGIFCONF) {
4603 ret = dev_ifconf(net, (char __user *) arg);
4607 if (cmd == SIOCGIFNAME)
4608 return dev_ifname(net, (struct ifreq __user *)arg);
4610 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4613 ifr.ifr_name[IFNAMSIZ-1] = 0;
4615 colon = strchr(ifr.ifr_name, ':');
4620 * See which interface the caller is talking about.
4625 * These ioctl calls:
4626 * - can be done by all.
4627 * - atomic and do not require locking.
4638 dev_load(net, ifr.ifr_name);
4640 ret = dev_ifsioc_locked(net, &ifr, cmd);
4645 if (copy_to_user(arg, &ifr,
4646 sizeof(struct ifreq)))
4652 dev_load(net, ifr.ifr_name);
4654 ret = dev_ethtool(net, &ifr);
4659 if (copy_to_user(arg, &ifr,
4660 sizeof(struct ifreq)))
4666 * These ioctl calls:
4667 * - require superuser power.
4668 * - require strict serialization.
4674 if (!capable(CAP_NET_ADMIN))
4676 dev_load(net, ifr.ifr_name);
4678 ret = dev_ifsioc(net, &ifr, cmd);
4683 if (copy_to_user(arg, &ifr,
4684 sizeof(struct ifreq)))
4690 * These ioctl calls:
4691 * - require superuser power.
4692 * - require strict serialization.
4693 * - do not return a value
4703 case SIOCSIFHWBROADCAST:
4706 case SIOCBONDENSLAVE:
4707 case SIOCBONDRELEASE:
4708 case SIOCBONDSETHWADDR:
4709 case SIOCBONDCHANGEACTIVE:
4713 if (!capable(CAP_NET_ADMIN))
4716 case SIOCBONDSLAVEINFOQUERY:
4717 case SIOCBONDINFOQUERY:
4718 dev_load(net, ifr.ifr_name);
4720 ret = dev_ifsioc(net, &ifr, cmd);
4725 /* Get the per device memory space. We can add this but
4726 * currently do not support it */
4728 /* Set the per device memory buffer space.
4729 * Not applicable in our case */
4734 * Unknown or private ioctl.
4737 if (cmd == SIOCWANDEV ||
4738 (cmd >= SIOCDEVPRIVATE &&
4739 cmd <= SIOCDEVPRIVATE + 15)) {
4740 dev_load(net, ifr.ifr_name);
4742 ret = dev_ifsioc(net, &ifr, cmd);
4744 if (!ret && copy_to_user(arg, &ifr,
4745 sizeof(struct ifreq)))
4749 /* Take care of Wireless Extensions */
4750 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4751 return wext_handle_ioctl(net, &ifr, cmd, arg);
4758 * dev_new_index - allocate an ifindex
4759 * @net: the applicable net namespace
4761 * Returns a suitable unique value for a new device interface
4762 * number. The caller must hold the rtnl semaphore or the
4763 * dev_base_lock to be sure it remains unique.
4765 static int dev_new_index(struct net *net)
4771 if (!__dev_get_by_index(net, ifindex))
4776 /* Delayed registration/unregisteration */
4777 static LIST_HEAD(net_todo_list);
4779 static void net_set_todo(struct net_device *dev)
4781 list_add_tail(&dev->todo_list, &net_todo_list);
4784 static void rollback_registered_many(struct list_head *head)
4786 struct net_device *dev, *tmp;
4788 BUG_ON(dev_boot_phase);
4791 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4792 /* Some devices call without registering
4793 * for initialization unwind. Remove those
4794 * devices and proceed with the remaining.
4796 if (dev->reg_state == NETREG_UNINITIALIZED) {
4797 pr_debug("unregister_netdevice: device %s/%p never "
4798 "was registered\n", dev->name, dev);
4801 list_del(&dev->unreg_list);
4805 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4807 /* If device is running, close it first. */
4810 /* And unlink it from device chain. */
4811 unlist_netdevice(dev);
4813 dev->reg_state = NETREG_UNREGISTERING;
4818 list_for_each_entry(dev, head, unreg_list) {
4819 /* Shutdown queueing discipline. */
4823 /* Notify protocols, that we are about to destroy
4824 this device. They should clean all the things.
4826 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4828 if (!dev->rtnl_link_ops ||
4829 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4830 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4833 * Flush the unicast and multicast chains
4838 if (dev->netdev_ops->ndo_uninit)
4839 dev->netdev_ops->ndo_uninit(dev);
4841 /* Notifier chain MUST detach us from master device. */
4842 WARN_ON(dev->master);
4844 /* Remove entries from kobject tree */
4845 netdev_unregister_kobject(dev);
4848 /* Process any work delayed until the end of the batch */
4849 dev = list_first_entry(head, struct net_device, unreg_list);
4850 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4854 list_for_each_entry(dev, head, unreg_list)
4858 static void rollback_registered(struct net_device *dev)
4862 list_add(&dev->unreg_list, &single);
4863 rollback_registered_many(&single);
4866 static void __netdev_init_queue_locks_one(struct net_device *dev,
4867 struct netdev_queue *dev_queue,
4870 spin_lock_init(&dev_queue->_xmit_lock);
4871 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4872 dev_queue->xmit_lock_owner = -1;
4875 static void netdev_init_queue_locks(struct net_device *dev)
4877 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4878 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4881 unsigned long netdev_fix_features(unsigned long features, const char *name)
4883 /* Fix illegal SG+CSUM combinations. */
4884 if ((features & NETIF_F_SG) &&
4885 !(features & NETIF_F_ALL_CSUM)) {
4887 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4888 "checksum feature.\n", name);
4889 features &= ~NETIF_F_SG;
4892 /* TSO requires that SG is present as well. */
4893 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4895 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4896 "SG feature.\n", name);
4897 features &= ~NETIF_F_TSO;
4900 if (features & NETIF_F_UFO) {
4901 if (!(features & NETIF_F_GEN_CSUM)) {
4903 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4904 "since no NETIF_F_HW_CSUM feature.\n",
4906 features &= ~NETIF_F_UFO;
4909 if (!(features & NETIF_F_SG)) {
4911 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4912 "since no NETIF_F_SG feature.\n", name);
4913 features &= ~NETIF_F_UFO;
4919 EXPORT_SYMBOL(netdev_fix_features);
4922 * netif_stacked_transfer_operstate - transfer operstate
4923 * @rootdev: the root or lower level device to transfer state from
4924 * @dev: the device to transfer operstate to
4926 * Transfer operational state from root to device. This is normally
4927 * called when a stacking relationship exists between the root
4928 * device and the device(a leaf device).
4930 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4931 struct net_device *dev)
4933 if (rootdev->operstate == IF_OPER_DORMANT)
4934 netif_dormant_on(dev);
4936 netif_dormant_off(dev);
4938 if (netif_carrier_ok(rootdev)) {
4939 if (!netif_carrier_ok(dev))
4940 netif_carrier_on(dev);
4942 if (netif_carrier_ok(dev))
4943 netif_carrier_off(dev);
4946 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4949 * register_netdevice - register a network device
4950 * @dev: device to register
4952 * Take a completed network device structure and add it to the kernel
4953 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4954 * chain. 0 is returned on success. A negative errno code is returned
4955 * on a failure to set up the device, or if the name is a duplicate.
4957 * Callers must hold the rtnl semaphore. You may want
4958 * register_netdev() instead of this.
4961 * The locking appears insufficient to guarantee two parallel registers
4962 * will not get the same name.
4965 int register_netdevice(struct net_device *dev)
4968 struct net *net = dev_net(dev);
4970 BUG_ON(dev_boot_phase);
4975 /* When net_device's are persistent, this will be fatal. */
4976 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4979 spin_lock_init(&dev->addr_list_lock);
4980 netdev_set_addr_lockdep_class(dev);
4981 netdev_init_queue_locks(dev);
4986 if (!dev->num_rx_queues) {
4988 * Allocate a single RX queue if driver never called
4992 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
4998 dev->_rx->first = dev->_rx;
4999 atomic_set(&dev->_rx->count, 1);
5000 dev->num_rx_queues = 1;
5003 /* Init, if this function is available */
5004 if (dev->netdev_ops->ndo_init) {
5005 ret = dev->netdev_ops->ndo_init(dev);
5013 ret = dev_get_valid_name(dev, dev->name, 0);
5017 dev->ifindex = dev_new_index(net);
5018 if (dev->iflink == -1)
5019 dev->iflink = dev->ifindex;
5021 /* Fix illegal checksum combinations */
5022 if ((dev->features & NETIF_F_HW_CSUM) &&
5023 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5024 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5026 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5029 if ((dev->features & NETIF_F_NO_CSUM) &&
5030 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5031 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5033 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5036 dev->features = netdev_fix_features(dev->features, dev->name);
5038 /* Enable software GSO if SG is supported. */
5039 if (dev->features & NETIF_F_SG)
5040 dev->features |= NETIF_F_GSO;
5042 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5043 ret = notifier_to_errno(ret);
5047 ret = netdev_register_kobject(dev);
5050 dev->reg_state = NETREG_REGISTERED;
5053 * Default initial state at registry is that the
5054 * device is present.
5057 set_bit(__LINK_STATE_PRESENT, &dev->state);
5059 dev_init_scheduler(dev);
5061 list_netdevice(dev);
5063 /* Notify protocols, that a new device appeared. */
5064 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5065 ret = notifier_to_errno(ret);
5067 rollback_registered(dev);
5068 dev->reg_state = NETREG_UNREGISTERED;
5071 * Prevent userspace races by waiting until the network
5072 * device is fully setup before sending notifications.
5074 if (!dev->rtnl_link_ops ||
5075 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5076 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5082 if (dev->netdev_ops->ndo_uninit)
5083 dev->netdev_ops->ndo_uninit(dev);
5086 EXPORT_SYMBOL(register_netdevice);
5089 * init_dummy_netdev - init a dummy network device for NAPI
5090 * @dev: device to init
5092 * This takes a network device structure and initialize the minimum
5093 * amount of fields so it can be used to schedule NAPI polls without
5094 * registering a full blown interface. This is to be used by drivers
5095 * that need to tie several hardware interfaces to a single NAPI
5096 * poll scheduler due to HW limitations.
5098 int init_dummy_netdev(struct net_device *dev)
5100 /* Clear everything. Note we don't initialize spinlocks
5101 * are they aren't supposed to be taken by any of the
5102 * NAPI code and this dummy netdev is supposed to be
5103 * only ever used for NAPI polls
5105 memset(dev, 0, sizeof(struct net_device));
5107 /* make sure we BUG if trying to hit standard
5108 * register/unregister code path
5110 dev->reg_state = NETREG_DUMMY;
5112 /* initialize the ref count */
5113 atomic_set(&dev->refcnt, 1);
5115 /* NAPI wants this */
5116 INIT_LIST_HEAD(&dev->napi_list);
5118 /* a dummy interface is started by default */
5119 set_bit(__LINK_STATE_PRESENT, &dev->state);
5120 set_bit(__LINK_STATE_START, &dev->state);
5124 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5128 * register_netdev - register a network device
5129 * @dev: device to register
5131 * Take a completed network device structure and add it to the kernel
5132 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5133 * chain. 0 is returned on success. A negative errno code is returned
5134 * on a failure to set up the device, or if the name is a duplicate.
5136 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5137 * and expands the device name if you passed a format string to
5140 int register_netdev(struct net_device *dev)
5147 * If the name is a format string the caller wants us to do a
5150 if (strchr(dev->name, '%')) {
5151 err = dev_alloc_name(dev, dev->name);
5156 err = register_netdevice(dev);
5161 EXPORT_SYMBOL(register_netdev);
5164 * netdev_wait_allrefs - wait until all references are gone.
5166 * This is called when unregistering network devices.
5168 * Any protocol or device that holds a reference should register
5169 * for netdevice notification, and cleanup and put back the
5170 * reference if they receive an UNREGISTER event.
5171 * We can get stuck here if buggy protocols don't correctly
5174 static void netdev_wait_allrefs(struct net_device *dev)
5176 unsigned long rebroadcast_time, warning_time;
5178 linkwatch_forget_dev(dev);
5180 rebroadcast_time = warning_time = jiffies;
5181 while (atomic_read(&dev->refcnt) != 0) {
5182 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5185 /* Rebroadcast unregister notification */
5186 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5187 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5188 * should have already handle it the first time */
5190 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5192 /* We must not have linkwatch events
5193 * pending on unregister. If this
5194 * happens, we simply run the queue
5195 * unscheduled, resulting in a noop
5198 linkwatch_run_queue();
5203 rebroadcast_time = jiffies;
5208 if (time_after(jiffies, warning_time + 10 * HZ)) {
5209 printk(KERN_EMERG "unregister_netdevice: "
5210 "waiting for %s to become free. Usage "
5212 dev->name, atomic_read(&dev->refcnt));
5213 warning_time = jiffies;
5222 * register_netdevice(x1);
5223 * register_netdevice(x2);
5225 * unregister_netdevice(y1);
5226 * unregister_netdevice(y2);
5232 * We are invoked by rtnl_unlock().
5233 * This allows us to deal with problems:
5234 * 1) We can delete sysfs objects which invoke hotplug
5235 * without deadlocking with linkwatch via keventd.
5236 * 2) Since we run with the RTNL semaphore not held, we can sleep
5237 * safely in order to wait for the netdev refcnt to drop to zero.
5239 * We must not return until all unregister events added during
5240 * the interval the lock was held have been completed.
5242 void netdev_run_todo(void)
5244 struct list_head list;
5246 /* Snapshot list, allow later requests */
5247 list_replace_init(&net_todo_list, &list);
5251 while (!list_empty(&list)) {
5252 struct net_device *dev
5253 = list_first_entry(&list, struct net_device, todo_list);
5254 list_del(&dev->todo_list);
5256 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5257 printk(KERN_ERR "network todo '%s' but state %d\n",
5258 dev->name, dev->reg_state);
5263 dev->reg_state = NETREG_UNREGISTERED;
5265 on_each_cpu(flush_backlog, dev, 1);
5267 netdev_wait_allrefs(dev);
5270 BUG_ON(atomic_read(&dev->refcnt));
5271 WARN_ON(dev->ip_ptr);
5272 WARN_ON(dev->ip6_ptr);
5273 WARN_ON(dev->dn_ptr);
5275 if (dev->destructor)
5276 dev->destructor(dev);
5278 /* Free network device */
5279 kobject_put(&dev->dev.kobj);
5284 * dev_txq_stats_fold - fold tx_queues stats
5285 * @dev: device to get statistics from
5286 * @stats: struct rtnl_link_stats64 to hold results
5288 void dev_txq_stats_fold(const struct net_device *dev,
5289 struct rtnl_link_stats64 *stats)
5291 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5293 struct netdev_queue *txq;
5295 for (i = 0; i < dev->num_tx_queues; i++) {
5296 txq = netdev_get_tx_queue(dev, i);
5297 spin_lock_bh(&txq->_xmit_lock);
5298 tx_bytes += txq->tx_bytes;
5299 tx_packets += txq->tx_packets;
5300 tx_dropped += txq->tx_dropped;
5301 spin_unlock_bh(&txq->_xmit_lock);
5303 if (tx_bytes || tx_packets || tx_dropped) {
5304 stats->tx_bytes = tx_bytes;
5305 stats->tx_packets = tx_packets;
5306 stats->tx_dropped = tx_dropped;
5309 EXPORT_SYMBOL(dev_txq_stats_fold);
5311 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5312 * fields in the same order, with only the type differing.
5314 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5315 const struct net_device_stats *netdev_stats)
5317 #if BITS_PER_LONG == 64
5318 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5319 memcpy(stats64, netdev_stats, sizeof(*stats64));
5321 size_t i, n = sizeof(*stats64) / sizeof(u64);
5322 const unsigned long *src = (const unsigned long *)netdev_stats;
5323 u64 *dst = (u64 *)stats64;
5325 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5326 sizeof(*stats64) / sizeof(u64));
5327 for (i = 0; i < n; i++)
5333 * dev_get_stats - get network device statistics
5334 * @dev: device to get statistics from
5335 * @storage: place to store stats
5337 * Get network statistics from device. Return @storage.
5338 * The device driver may provide its own method by setting
5339 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5340 * otherwise the internal statistics structure is used.
5342 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5343 struct rtnl_link_stats64 *storage)
5345 const struct net_device_ops *ops = dev->netdev_ops;
5347 if (ops->ndo_get_stats64) {
5348 memset(storage, 0, sizeof(*storage));
5349 return ops->ndo_get_stats64(dev, storage);
5351 if (ops->ndo_get_stats) {
5352 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5355 netdev_stats_to_stats64(storage, &dev->stats);
5356 dev_txq_stats_fold(dev, storage);
5359 EXPORT_SYMBOL(dev_get_stats);
5361 static void netdev_init_one_queue(struct net_device *dev,
5362 struct netdev_queue *queue,
5368 static void netdev_init_queues(struct net_device *dev)
5370 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5371 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5372 spin_lock_init(&dev->tx_global_lock);
5376 * alloc_netdev_mq - allocate network device
5377 * @sizeof_priv: size of private data to allocate space for
5378 * @name: device name format string
5379 * @setup: callback to initialize device
5380 * @queue_count: the number of subqueues to allocate
5382 * Allocates a struct net_device with private data area for driver use
5383 * and performs basic initialization. Also allocates subquue structs
5384 * for each queue on the device at the end of the netdevice.
5386 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5387 void (*setup)(struct net_device *), unsigned int queue_count)
5389 struct netdev_queue *tx;
5390 struct net_device *dev;
5392 struct net_device *p;
5394 struct netdev_rx_queue *rx;
5398 BUG_ON(strlen(name) >= sizeof(dev->name));
5400 alloc_size = sizeof(struct net_device);
5402 /* ensure 32-byte alignment of private area */
5403 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5404 alloc_size += sizeof_priv;
5406 /* ensure 32-byte alignment of whole construct */
5407 alloc_size += NETDEV_ALIGN - 1;
5409 p = kzalloc(alloc_size, GFP_KERNEL);
5411 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5415 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5417 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5423 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5425 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5430 atomic_set(&rx->count, queue_count);
5433 * Set a pointer to first element in the array which holds the
5436 for (i = 0; i < queue_count; i++)
5440 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5441 dev->padded = (char *)dev - (char *)p;
5443 if (dev_addr_init(dev))
5449 dev_net_set(dev, &init_net);
5452 dev->num_tx_queues = queue_count;
5453 dev->real_num_tx_queues = queue_count;
5457 dev->num_rx_queues = queue_count;
5460 dev->gso_max_size = GSO_MAX_SIZE;
5462 netdev_init_queues(dev);
5464 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5465 dev->ethtool_ntuple_list.count = 0;
5466 INIT_LIST_HEAD(&dev->napi_list);
5467 INIT_LIST_HEAD(&dev->unreg_list);
5468 INIT_LIST_HEAD(&dev->link_watch_list);
5469 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5471 strcpy(dev->name, name);
5484 EXPORT_SYMBOL(alloc_netdev_mq);
5487 * free_netdev - free network device
5490 * This function does the last stage of destroying an allocated device
5491 * interface. The reference to the device object is released.
5492 * If this is the last reference then it will be freed.
5494 void free_netdev(struct net_device *dev)
5496 struct napi_struct *p, *n;
5498 release_net(dev_net(dev));
5502 /* Flush device addresses */
5503 dev_addr_flush(dev);
5505 /* Clear ethtool n-tuple list */
5506 ethtool_ntuple_flush(dev);
5508 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5511 /* Compatibility with error handling in drivers */
5512 if (dev->reg_state == NETREG_UNINITIALIZED) {
5513 kfree((char *)dev - dev->padded);
5517 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5518 dev->reg_state = NETREG_RELEASED;
5520 /* will free via device release */
5521 put_device(&dev->dev);
5523 EXPORT_SYMBOL(free_netdev);
5526 * synchronize_net - Synchronize with packet receive processing
5528 * Wait for packets currently being received to be done.
5529 * Does not block later packets from starting.
5531 void synchronize_net(void)
5536 EXPORT_SYMBOL(synchronize_net);
5539 * unregister_netdevice_queue - remove device from the kernel
5543 * This function shuts down a device interface and removes it
5544 * from the kernel tables.
5545 * If head not NULL, device is queued to be unregistered later.
5547 * Callers must hold the rtnl semaphore. You may want
5548 * unregister_netdev() instead of this.
5551 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5556 list_move_tail(&dev->unreg_list, head);
5558 rollback_registered(dev);
5559 /* Finish processing unregister after unlock */
5563 EXPORT_SYMBOL(unregister_netdevice_queue);
5566 * unregister_netdevice_many - unregister many devices
5567 * @head: list of devices
5569 void unregister_netdevice_many(struct list_head *head)
5571 struct net_device *dev;
5573 if (!list_empty(head)) {
5574 rollback_registered_many(head);
5575 list_for_each_entry(dev, head, unreg_list)
5579 EXPORT_SYMBOL(unregister_netdevice_many);
5582 * unregister_netdev - remove device from the kernel
5585 * This function shuts down a device interface and removes it
5586 * from the kernel tables.
5588 * This is just a wrapper for unregister_netdevice that takes
5589 * the rtnl semaphore. In general you want to use this and not
5590 * unregister_netdevice.
5592 void unregister_netdev(struct net_device *dev)
5595 unregister_netdevice(dev);
5598 EXPORT_SYMBOL(unregister_netdev);
5601 * dev_change_net_namespace - move device to different nethost namespace
5603 * @net: network namespace
5604 * @pat: If not NULL name pattern to try if the current device name
5605 * is already taken in the destination network namespace.
5607 * This function shuts down a device interface and moves it
5608 * to a new network namespace. On success 0 is returned, on
5609 * a failure a netagive errno code is returned.
5611 * Callers must hold the rtnl semaphore.
5614 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5620 /* Don't allow namespace local devices to be moved. */
5622 if (dev->features & NETIF_F_NETNS_LOCAL)
5625 /* Ensure the device has been registrered */
5627 if (dev->reg_state != NETREG_REGISTERED)
5630 /* Get out if there is nothing todo */
5632 if (net_eq(dev_net(dev), net))
5635 /* Pick the destination device name, and ensure
5636 * we can use it in the destination network namespace.
5639 if (__dev_get_by_name(net, dev->name)) {
5640 /* We get here if we can't use the current device name */
5643 if (dev_get_valid_name(dev, pat, 1))
5648 * And now a mini version of register_netdevice unregister_netdevice.
5651 /* If device is running close it first. */
5654 /* And unlink it from device chain */
5656 unlist_netdevice(dev);
5660 /* Shutdown queueing discipline. */
5663 /* Notify protocols, that we are about to destroy
5664 this device. They should clean all the things.
5666 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5667 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5670 * Flush the unicast and multicast chains
5675 /* Actually switch the network namespace */
5676 dev_net_set(dev, net);
5678 /* If there is an ifindex conflict assign a new one */
5679 if (__dev_get_by_index(net, dev->ifindex)) {
5680 int iflink = (dev->iflink == dev->ifindex);
5681 dev->ifindex = dev_new_index(net);
5683 dev->iflink = dev->ifindex;
5686 /* Fixup kobjects */
5687 err = device_rename(&dev->dev, dev->name);
5690 /* Add the device back in the hashes */
5691 list_netdevice(dev);
5693 /* Notify protocols, that a new device appeared. */
5694 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5697 * Prevent userspace races by waiting until the network
5698 * device is fully setup before sending notifications.
5700 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5707 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5709 static int dev_cpu_callback(struct notifier_block *nfb,
5710 unsigned long action,
5713 struct sk_buff **list_skb;
5714 struct sk_buff *skb;
5715 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5716 struct softnet_data *sd, *oldsd;
5718 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5721 local_irq_disable();
5722 cpu = smp_processor_id();
5723 sd = &per_cpu(softnet_data, cpu);
5724 oldsd = &per_cpu(softnet_data, oldcpu);
5726 /* Find end of our completion_queue. */
5727 list_skb = &sd->completion_queue;
5729 list_skb = &(*list_skb)->next;
5730 /* Append completion queue from offline CPU. */
5731 *list_skb = oldsd->completion_queue;
5732 oldsd->completion_queue = NULL;
5734 /* Append output queue from offline CPU. */
5735 if (oldsd->output_queue) {
5736 *sd->output_queue_tailp = oldsd->output_queue;
5737 sd->output_queue_tailp = oldsd->output_queue_tailp;
5738 oldsd->output_queue = NULL;
5739 oldsd->output_queue_tailp = &oldsd->output_queue;
5742 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5745 /* Process offline CPU's input_pkt_queue */
5746 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5748 input_queue_head_incr(oldsd);
5750 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5752 input_queue_head_incr(oldsd);
5760 * netdev_increment_features - increment feature set by one
5761 * @all: current feature set
5762 * @one: new feature set
5763 * @mask: mask feature set
5765 * Computes a new feature set after adding a device with feature set
5766 * @one to the master device with current feature set @all. Will not
5767 * enable anything that is off in @mask. Returns the new feature set.
5769 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5772 /* If device needs checksumming, downgrade to it. */
5773 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5774 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5775 else if (mask & NETIF_F_ALL_CSUM) {
5776 /* If one device supports v4/v6 checksumming, set for all. */
5777 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5778 !(all & NETIF_F_GEN_CSUM)) {
5779 all &= ~NETIF_F_ALL_CSUM;
5780 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5783 /* If one device supports hw checksumming, set for all. */
5784 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5785 all &= ~NETIF_F_ALL_CSUM;
5786 all |= NETIF_F_HW_CSUM;
5790 one |= NETIF_F_ALL_CSUM;
5792 one |= all & NETIF_F_ONE_FOR_ALL;
5793 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5794 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5798 EXPORT_SYMBOL(netdev_increment_features);
5800 static struct hlist_head *netdev_create_hash(void)
5803 struct hlist_head *hash;
5805 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5807 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5808 INIT_HLIST_HEAD(&hash[i]);
5813 /* Initialize per network namespace state */
5814 static int __net_init netdev_init(struct net *net)
5816 INIT_LIST_HEAD(&net->dev_base_head);
5818 net->dev_name_head = netdev_create_hash();
5819 if (net->dev_name_head == NULL)
5822 net->dev_index_head = netdev_create_hash();
5823 if (net->dev_index_head == NULL)
5829 kfree(net->dev_name_head);
5835 * netdev_drivername - network driver for the device
5836 * @dev: network device
5837 * @buffer: buffer for resulting name
5838 * @len: size of buffer
5840 * Determine network driver for device.
5842 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5844 const struct device_driver *driver;
5845 const struct device *parent;
5847 if (len <= 0 || !buffer)
5851 parent = dev->dev.parent;
5856 driver = parent->driver;
5857 if (driver && driver->name)
5858 strlcpy(buffer, driver->name, len);
5862 static int __netdev_printk(const char *level, const struct net_device *dev,
5863 struct va_format *vaf)
5867 if (dev && dev->dev.parent)
5868 r = dev_printk(level, dev->dev.parent, "%s: %pV",
5869 netdev_name(dev), vaf);
5871 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
5873 r = printk("%s(NULL net_device): %pV", level, vaf);
5878 int netdev_printk(const char *level, const struct net_device *dev,
5879 const char *format, ...)
5881 struct va_format vaf;
5885 va_start(args, format);
5890 r = __netdev_printk(level, dev, &vaf);
5895 EXPORT_SYMBOL(netdev_printk);
5897 #define define_netdev_printk_level(func, level) \
5898 int func(const struct net_device *dev, const char *fmt, ...) \
5901 struct va_format vaf; \
5904 va_start(args, fmt); \
5909 r = __netdev_printk(level, dev, &vaf); \
5914 EXPORT_SYMBOL(func);
5916 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
5917 define_netdev_printk_level(netdev_alert, KERN_ALERT);
5918 define_netdev_printk_level(netdev_crit, KERN_CRIT);
5919 define_netdev_printk_level(netdev_err, KERN_ERR);
5920 define_netdev_printk_level(netdev_warn, KERN_WARNING);
5921 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
5922 define_netdev_printk_level(netdev_info, KERN_INFO);
5924 static void __net_exit netdev_exit(struct net *net)
5926 kfree(net->dev_name_head);
5927 kfree(net->dev_index_head);
5930 static struct pernet_operations __net_initdata netdev_net_ops = {
5931 .init = netdev_init,
5932 .exit = netdev_exit,
5935 static void __net_exit default_device_exit(struct net *net)
5937 struct net_device *dev, *aux;
5939 * Push all migratable network devices back to the
5940 * initial network namespace
5943 for_each_netdev_safe(net, dev, aux) {
5945 char fb_name[IFNAMSIZ];
5947 /* Ignore unmoveable devices (i.e. loopback) */
5948 if (dev->features & NETIF_F_NETNS_LOCAL)
5951 /* Leave virtual devices for the generic cleanup */
5952 if (dev->rtnl_link_ops)
5955 /* Push remaing network devices to init_net */
5956 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5957 err = dev_change_net_namespace(dev, &init_net, fb_name);
5959 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5960 __func__, dev->name, err);
5967 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5969 /* At exit all network devices most be removed from a network
5970 * namespace. Do this in the reverse order of registeration.
5971 * Do this across as many network namespaces as possible to
5972 * improve batching efficiency.
5974 struct net_device *dev;
5976 LIST_HEAD(dev_kill_list);
5979 list_for_each_entry(net, net_list, exit_list) {
5980 for_each_netdev_reverse(net, dev) {
5981 if (dev->rtnl_link_ops)
5982 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5984 unregister_netdevice_queue(dev, &dev_kill_list);
5987 unregister_netdevice_many(&dev_kill_list);
5991 static struct pernet_operations __net_initdata default_device_ops = {
5992 .exit = default_device_exit,
5993 .exit_batch = default_device_exit_batch,
5997 * Initialize the DEV module. At boot time this walks the device list and
5998 * unhooks any devices that fail to initialise (normally hardware not
5999 * present) and leaves us with a valid list of present and active devices.
6004 * This is called single threaded during boot, so no need
6005 * to take the rtnl semaphore.
6007 static int __init net_dev_init(void)
6009 int i, rc = -ENOMEM;
6011 BUG_ON(!dev_boot_phase);
6013 if (dev_proc_init())
6016 if (netdev_kobject_init())
6019 INIT_LIST_HEAD(&ptype_all);
6020 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6021 INIT_LIST_HEAD(&ptype_base[i]);
6023 if (register_pernet_subsys(&netdev_net_ops))
6027 * Initialise the packet receive queues.
6030 for_each_possible_cpu(i) {
6031 struct softnet_data *sd = &per_cpu(softnet_data, i);
6033 memset(sd, 0, sizeof(*sd));
6034 skb_queue_head_init(&sd->input_pkt_queue);
6035 skb_queue_head_init(&sd->process_queue);
6036 sd->completion_queue = NULL;
6037 INIT_LIST_HEAD(&sd->poll_list);
6038 sd->output_queue = NULL;
6039 sd->output_queue_tailp = &sd->output_queue;
6041 sd->csd.func = rps_trigger_softirq;
6047 sd->backlog.poll = process_backlog;
6048 sd->backlog.weight = weight_p;
6049 sd->backlog.gro_list = NULL;
6050 sd->backlog.gro_count = 0;
6055 /* The loopback device is special if any other network devices
6056 * is present in a network namespace the loopback device must
6057 * be present. Since we now dynamically allocate and free the
6058 * loopback device ensure this invariant is maintained by
6059 * keeping the loopback device as the first device on the
6060 * list of network devices. Ensuring the loopback devices
6061 * is the first device that appears and the last network device
6064 if (register_pernet_device(&loopback_net_ops))
6067 if (register_pernet_device(&default_device_ops))
6070 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6071 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6073 hotcpu_notifier(dev_cpu_callback, 0);
6081 subsys_initcall(net_dev_init);
6083 static int __init initialize_hashrnd(void)
6085 get_random_bytes(&hashrnd, sizeof(hashrnd));
6089 late_initcall_sync(initialize_hashrnd);