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 *queue)
214 spin_lock(&queue->input_pkt_queue.lock);
218 static inline void rps_unlock(struct softnet_data *queue)
221 spin_unlock(&queue->input_pkt_queue.lock);
225 /* Device list insertion */
226 static int list_netdevice(struct net_device *dev)
228 struct net *net = dev_net(dev);
232 write_lock_bh(&dev_base_lock);
233 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
234 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
235 hlist_add_head_rcu(&dev->index_hlist,
236 dev_index_hash(net, dev->ifindex));
237 write_unlock_bh(&dev_base_lock);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device *dev)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock);
250 list_del_rcu(&dev->dev_list);
251 hlist_del_rcu(&dev->name_hlist);
252 hlist_del_rcu(&dev->index_hlist);
253 write_unlock_bh(&dev_base_lock);
260 static RAW_NOTIFIER_HEAD(netdev_chain);
263 * Device drivers call our routines to queue packets here. We empty the
264 * queue in the local softnet handler.
267 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
268 EXPORT_PER_CPU_SYMBOL(softnet_data);
270 #ifdef CONFIG_LOCKDEP
272 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
273 * according to dev->type
275 static const unsigned short netdev_lock_type[] =
276 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
277 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
278 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
279 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
280 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
281 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
282 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
283 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
284 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
285 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
286 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
287 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
288 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
289 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
290 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
291 ARPHRD_VOID, ARPHRD_NONE};
293 static const char *const netdev_lock_name[] =
294 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
295 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
296 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
297 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
298 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
299 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
300 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
301 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
302 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
303 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
304 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
305 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
306 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
307 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
308 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
309 "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
318 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
319 if (netdev_lock_type[i] == dev_type)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
326 unsigned short dev_type)
330 i = netdev_lock_pos(dev_type);
331 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
332 netdev_lock_name[i]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
339 i = netdev_lock_pos(dev->type);
340 lockdep_set_class_and_name(&dev->addr_list_lock,
341 &netdev_addr_lock_key[i],
342 netdev_lock_name[i]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
346 unsigned short dev_type)
349 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
377 * dev_add_pack - add packet handler
378 * @pt: packet type declaration
380 * Add a protocol handler to the networking stack. The passed &packet_type
381 * is linked into kernel lists and may not be freed until it has been
382 * removed from the kernel lists.
384 * This call does not sleep therefore it can not
385 * guarantee all CPU's that are in middle of receiving packets
386 * will see the new packet type (until the next received packet).
389 void dev_add_pack(struct packet_type *pt)
393 spin_lock_bh(&ptype_lock);
394 if (pt->type == htons(ETH_P_ALL))
395 list_add_rcu(&pt->list, &ptype_all);
397 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
398 list_add_rcu(&pt->list, &ptype_base[hash]);
400 spin_unlock_bh(&ptype_lock);
402 EXPORT_SYMBOL(dev_add_pack);
405 * __dev_remove_pack - remove packet handler
406 * @pt: packet type declaration
408 * Remove a protocol handler that was previously added to the kernel
409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
410 * from the kernel lists and can be freed or reused once this function
413 * The packet type might still be in use by receivers
414 * and must not be freed until after all the CPU's have gone
415 * through a quiescent state.
417 void __dev_remove_pack(struct packet_type *pt)
419 struct list_head *head;
420 struct packet_type *pt1;
422 spin_lock_bh(&ptype_lock);
424 if (pt->type == htons(ETH_P_ALL))
427 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
429 list_for_each_entry(pt1, head, list) {
431 list_del_rcu(&pt->list);
436 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
438 spin_unlock_bh(&ptype_lock);
440 EXPORT_SYMBOL(__dev_remove_pack);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type *pt)
456 __dev_remove_pack(pt);
460 EXPORT_SYMBOL(dev_remove_pack);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name, struct ifmap *map)
482 struct netdev_boot_setup *s;
486 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
487 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
488 memset(s[i].name, 0, sizeof(s[i].name));
489 strlcpy(s[i].name, name, IFNAMSIZ);
490 memcpy(&s[i].map, map, sizeof(s[i].map));
495 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device *dev)
509 struct netdev_boot_setup *s = dev_boot_setup;
512 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
513 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
514 !strcmp(dev->name, s[i].name)) {
515 dev->irq = s[i].map.irq;
516 dev->base_addr = s[i].map.base_addr;
517 dev->mem_start = s[i].map.mem_start;
518 dev->mem_end = s[i].map.mem_end;
524 EXPORT_SYMBOL(netdev_boot_setup_check);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix, int unit)
539 const struct netdev_boot_setup *s = dev_boot_setup;
543 sprintf(name, "%s%d", prefix, unit);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net, name))
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
553 if (!strcmp(name, s[i].name))
554 return s[i].map.base_addr;
559 * Saves at boot time configured settings for any netdevice.
561 int __init netdev_boot_setup(char *str)
566 str = get_options(str, ARRAY_SIZE(ints), ints);
571 memset(&map, 0, sizeof(map));
575 map.base_addr = ints[2];
577 map.mem_start = ints[3];
579 map.mem_end = ints[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str, &map);
585 __setup("netdev=", netdev_boot_setup);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device *__dev_get_by_name(struct net *net, const char *name)
607 struct hlist_node *p;
608 struct net_device *dev;
609 struct hlist_head *head = dev_name_hash(net, name);
611 hlist_for_each_entry(dev, p, head, name_hlist)
612 if (!strncmp(dev->name, name, IFNAMSIZ))
617 EXPORT_SYMBOL(__dev_get_by_name);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
633 struct hlist_node *p;
634 struct net_device *dev;
635 struct hlist_head *head = dev_name_hash(net, name);
637 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
638 if (!strncmp(dev->name, name, IFNAMSIZ))
643 EXPORT_SYMBOL(dev_get_by_name_rcu);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device *dev_get_by_name(struct net *net, const char *name)
659 struct net_device *dev;
662 dev = dev_get_by_name_rcu(net, name);
668 EXPORT_SYMBOL(dev_get_by_name);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
684 struct hlist_node *p;
685 struct net_device *dev;
686 struct hlist_head *head = dev_index_hash(net, ifindex);
688 hlist_for_each_entry(dev, p, head, index_hlist)
689 if (dev->ifindex == ifindex)
694 EXPORT_SYMBOL(__dev_get_by_index);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
709 struct hlist_node *p;
710 struct net_device *dev;
711 struct hlist_head *head = dev_index_hash(net, ifindex);
713 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
714 if (dev->ifindex == ifindex)
719 EXPORT_SYMBOL(dev_get_by_index_rcu);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device *dev_get_by_index(struct net *net, int ifindex)
735 struct net_device *dev;
738 dev = dev_get_by_index_rcu(net, ifindex);
744 EXPORT_SYMBOL(dev_get_by_index);
747 * dev_getbyhwaddr - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device. The caller must hold the
754 * rtnl semaphore. The returned device has not had its ref count increased
755 * and the caller must therefore be careful about locking
758 * If the API was consistent this would be __dev_get_by_hwaddr
761 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
763 struct net_device *dev;
767 for_each_netdev(net, dev)
768 if (dev->type == type &&
769 !memcmp(dev->dev_addr, ha, dev->addr_len))
774 EXPORT_SYMBOL(dev_getbyhwaddr);
776 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
778 struct net_device *dev;
781 for_each_netdev(net, dev)
782 if (dev->type == type)
787 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
789 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
791 struct net_device *dev, *ret = NULL;
794 for_each_netdev_rcu(net, dev)
795 if (dev->type == type) {
803 EXPORT_SYMBOL(dev_getfirstbyhwtype);
806 * dev_get_by_flags - find any device with given flags
807 * @net: the applicable net namespace
808 * @if_flags: IFF_* values
809 * @mask: bitmask of bits in if_flags to check
811 * Search for any interface with the given flags. Returns NULL if a device
812 * is not found or a pointer to the device. The device returned has
813 * had a reference added and the pointer is safe until the user calls
814 * dev_put to indicate they have finished with it.
817 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
820 struct net_device *dev, *ret;
824 for_each_netdev_rcu(net, dev) {
825 if (((dev->flags ^ if_flags) & mask) == 0) {
834 EXPORT_SYMBOL(dev_get_by_flags);
837 * dev_valid_name - check if name is okay for network device
840 * Network device names need to be valid file names to
841 * to allow sysfs to work. We also disallow any kind of
844 int dev_valid_name(const char *name)
848 if (strlen(name) >= IFNAMSIZ)
850 if (!strcmp(name, ".") || !strcmp(name, ".."))
854 if (*name == '/' || isspace(*name))
860 EXPORT_SYMBOL(dev_valid_name);
863 * __dev_alloc_name - allocate a name for a device
864 * @net: network namespace to allocate the device name in
865 * @name: name format string
866 * @buf: scratch buffer and result name string
868 * Passed a format string - eg "lt%d" it will try and find a suitable
869 * id. It scans list of devices to build up a free map, then chooses
870 * the first empty slot. The caller must hold the dev_base or rtnl lock
871 * while allocating the name and adding the device in order to avoid
873 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
874 * Returns the number of the unit assigned or a negative errno code.
877 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
881 const int max_netdevices = 8*PAGE_SIZE;
882 unsigned long *inuse;
883 struct net_device *d;
885 p = strnchr(name, IFNAMSIZ-1, '%');
888 * Verify the string as this thing may have come from
889 * the user. There must be either one "%d" and no other "%"
892 if (p[1] != 'd' || strchr(p + 2, '%'))
895 /* Use one page as a bit array of possible slots */
896 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
900 for_each_netdev(net, d) {
901 if (!sscanf(d->name, name, &i))
903 if (i < 0 || i >= max_netdevices)
906 /* avoid cases where sscanf is not exact inverse of printf */
907 snprintf(buf, IFNAMSIZ, name, i);
908 if (!strncmp(buf, d->name, IFNAMSIZ))
912 i = find_first_zero_bit(inuse, max_netdevices);
913 free_page((unsigned long) inuse);
917 snprintf(buf, IFNAMSIZ, name, i);
918 if (!__dev_get_by_name(net, buf))
921 /* It is possible to run out of possible slots
922 * when the name is long and there isn't enough space left
923 * for the digits, or if all bits are used.
929 * dev_alloc_name - allocate a name for a device
931 * @name: name format string
933 * Passed a format string - eg "lt%d" it will try and find a suitable
934 * id. It scans list of devices to build up a free map, then chooses
935 * the first empty slot. The caller must hold the dev_base or rtnl lock
936 * while allocating the name and adding the device in order to avoid
938 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
939 * Returns the number of the unit assigned or a negative errno code.
942 int dev_alloc_name(struct net_device *dev, const char *name)
948 BUG_ON(!dev_net(dev));
950 ret = __dev_alloc_name(net, name, buf);
952 strlcpy(dev->name, buf, IFNAMSIZ);
955 EXPORT_SYMBOL(dev_alloc_name);
957 static int dev_get_valid_name(struct net *net, const char *name, char *buf,
960 if (!dev_valid_name(name))
963 if (fmt && strchr(name, '%'))
964 return __dev_alloc_name(net, name, buf);
965 else if (__dev_get_by_name(net, name))
967 else if (buf != name)
968 strlcpy(buf, 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(net, newname, dev->name, 1);
1005 /* For now only devices in the initial network namespace
1008 if (net_eq(net, &init_net)) {
1009 ret = device_rename(&dev->dev, dev->name);
1011 memcpy(dev->name, oldname, IFNAMSIZ);
1016 write_lock_bh(&dev_base_lock);
1017 hlist_del(&dev->name_hlist);
1018 write_unlock_bh(&dev_base_lock);
1022 write_lock_bh(&dev_base_lock);
1023 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1024 write_unlock_bh(&dev_base_lock);
1026 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1027 ret = notifier_to_errno(ret);
1030 /* err >= 0 after dev_alloc_name() or stores the first errno */
1033 memcpy(dev->name, oldname, IFNAMSIZ);
1037 "%s: name change rollback failed: %d.\n",
1046 * dev_set_alias - change ifalias of a device
1048 * @alias: name up to IFALIASZ
1049 * @len: limit of bytes to copy from info
1051 * Set ifalias for a device,
1053 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1057 if (len >= IFALIASZ)
1062 kfree(dev->ifalias);
1063 dev->ifalias = NULL;
1068 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1072 strlcpy(dev->ifalias, alias, len+1);
1078 * netdev_features_change - device changes features
1079 * @dev: device to cause notification
1081 * Called to indicate a device has changed features.
1083 void netdev_features_change(struct net_device *dev)
1085 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1087 EXPORT_SYMBOL(netdev_features_change);
1090 * netdev_state_change - device changes state
1091 * @dev: device to cause notification
1093 * Called to indicate a device has changed state. This function calls
1094 * the notifier chains for netdev_chain and sends a NEWLINK message
1095 * to the routing socket.
1097 void netdev_state_change(struct net_device *dev)
1099 if (dev->flags & IFF_UP) {
1100 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1101 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1104 EXPORT_SYMBOL(netdev_state_change);
1106 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1108 return call_netdevice_notifiers(event, dev);
1110 EXPORT_SYMBOL(netdev_bonding_change);
1113 * dev_load - load a network module
1114 * @net: the applicable net namespace
1115 * @name: name of interface
1117 * If a network interface is not present and the process has suitable
1118 * privileges this function loads the module. If module loading is not
1119 * available in this kernel then it becomes a nop.
1122 void dev_load(struct net *net, const char *name)
1124 struct net_device *dev;
1127 dev = dev_get_by_name_rcu(net, name);
1130 if (!dev && capable(CAP_NET_ADMIN))
1131 request_module("%s", name);
1133 EXPORT_SYMBOL(dev_load);
1135 static int __dev_open(struct net_device *dev)
1137 const struct net_device_ops *ops = dev->netdev_ops;
1143 * Is it even present?
1145 if (!netif_device_present(dev))
1148 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1149 ret = notifier_to_errno(ret);
1154 * Call device private open method
1156 set_bit(__LINK_STATE_START, &dev->state);
1158 if (ops->ndo_validate_addr)
1159 ret = ops->ndo_validate_addr(dev);
1161 if (!ret && ops->ndo_open)
1162 ret = ops->ndo_open(dev);
1165 * If it went open OK then:
1169 clear_bit(__LINK_STATE_START, &dev->state);
1174 dev->flags |= IFF_UP;
1179 net_dmaengine_get();
1182 * Initialize multicasting status
1184 dev_set_rx_mode(dev);
1187 * Wakeup transmit queue engine
1196 * dev_open - prepare an interface for use.
1197 * @dev: device to open
1199 * Takes a device from down to up state. The device's private open
1200 * function is invoked and then the multicast lists are loaded. Finally
1201 * the device is moved into the up state and a %NETDEV_UP message is
1202 * sent to the netdev notifier chain.
1204 * Calling this function on an active interface is a nop. On a failure
1205 * a negative errno code is returned.
1207 int dev_open(struct net_device *dev)
1214 if (dev->flags & IFF_UP)
1220 ret = __dev_open(dev);
1225 * ... and announce new interface.
1227 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1228 call_netdevice_notifiers(NETDEV_UP, dev);
1232 EXPORT_SYMBOL(dev_open);
1234 static int __dev_close(struct net_device *dev)
1236 const struct net_device_ops *ops = dev->netdev_ops;
1242 * Tell people we are going down, so that they can
1243 * prepare to death, when device is still operating.
1245 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1247 clear_bit(__LINK_STATE_START, &dev->state);
1249 /* Synchronize to scheduled poll. We cannot touch poll list,
1250 * it can be even on different cpu. So just clear netif_running().
1252 * dev->stop() will invoke napi_disable() on all of it's
1253 * napi_struct instances on this device.
1255 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1257 dev_deactivate(dev);
1260 * Call the device specific close. This cannot fail.
1261 * Only if device is UP
1263 * We allow it to be called even after a DETACH hot-plug
1270 * Device is now down.
1273 dev->flags &= ~IFF_UP;
1278 net_dmaengine_put();
1284 * dev_close - shutdown an interface.
1285 * @dev: device to shutdown
1287 * This function moves an active device into down state. A
1288 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1289 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1292 int dev_close(struct net_device *dev)
1294 if (!(dev->flags & IFF_UP))
1300 * Tell people we are down
1302 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1303 call_netdevice_notifiers(NETDEV_DOWN, dev);
1307 EXPORT_SYMBOL(dev_close);
1311 * dev_disable_lro - disable Large Receive Offload on a device
1314 * Disable Large Receive Offload (LRO) on a net device. Must be
1315 * called under RTNL. This is needed if received packets may be
1316 * forwarded to another interface.
1318 void dev_disable_lro(struct net_device *dev)
1320 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1321 dev->ethtool_ops->set_flags) {
1322 u32 flags = dev->ethtool_ops->get_flags(dev);
1323 if (flags & ETH_FLAG_LRO) {
1324 flags &= ~ETH_FLAG_LRO;
1325 dev->ethtool_ops->set_flags(dev, flags);
1328 WARN_ON(dev->features & NETIF_F_LRO);
1330 EXPORT_SYMBOL(dev_disable_lro);
1333 static int dev_boot_phase = 1;
1336 * Device change register/unregister. These are not inline or static
1337 * as we export them to the world.
1341 * register_netdevice_notifier - register a network notifier block
1344 * Register a notifier to be called when network device events occur.
1345 * The notifier passed is linked into the kernel structures and must
1346 * not be reused until it has been unregistered. A negative errno code
1347 * is returned on a failure.
1349 * When registered all registration and up events are replayed
1350 * to the new notifier to allow device to have a race free
1351 * view of the network device list.
1354 int register_netdevice_notifier(struct notifier_block *nb)
1356 struct net_device *dev;
1357 struct net_device *last;
1362 err = raw_notifier_chain_register(&netdev_chain, nb);
1368 for_each_netdev(net, dev) {
1369 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1370 err = notifier_to_errno(err);
1374 if (!(dev->flags & IFF_UP))
1377 nb->notifier_call(nb, NETDEV_UP, dev);
1388 for_each_netdev(net, dev) {
1392 if (dev->flags & IFF_UP) {
1393 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1394 nb->notifier_call(nb, NETDEV_DOWN, dev);
1396 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1397 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1401 raw_notifier_chain_unregister(&netdev_chain, nb);
1404 EXPORT_SYMBOL(register_netdevice_notifier);
1407 * unregister_netdevice_notifier - unregister a network notifier block
1410 * Unregister a notifier previously registered by
1411 * register_netdevice_notifier(). The notifier is unlinked into the
1412 * kernel structures and may then be reused. A negative errno code
1413 * is returned on a failure.
1416 int unregister_netdevice_notifier(struct notifier_block *nb)
1421 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1425 EXPORT_SYMBOL(unregister_netdevice_notifier);
1428 * call_netdevice_notifiers - call all network notifier blocks
1429 * @val: value passed unmodified to notifier function
1430 * @dev: net_device pointer passed unmodified to notifier function
1432 * Call all network notifier blocks. Parameters and return value
1433 * are as for raw_notifier_call_chain().
1436 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1438 return raw_notifier_call_chain(&netdev_chain, val, dev);
1441 /* When > 0 there are consumers of rx skb time stamps */
1442 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1444 void net_enable_timestamp(void)
1446 atomic_inc(&netstamp_needed);
1448 EXPORT_SYMBOL(net_enable_timestamp);
1450 void net_disable_timestamp(void)
1452 atomic_dec(&netstamp_needed);
1454 EXPORT_SYMBOL(net_disable_timestamp);
1456 static inline void net_timestamp(struct sk_buff *skb)
1458 if (atomic_read(&netstamp_needed))
1459 __net_timestamp(skb);
1461 skb->tstamp.tv64 = 0;
1465 * dev_forward_skb - loopback an skb to another netif
1467 * @dev: destination network device
1468 * @skb: buffer to forward
1471 * NET_RX_SUCCESS (no congestion)
1472 * NET_RX_DROP (packet was dropped)
1474 * dev_forward_skb can be used for injecting an skb from the
1475 * start_xmit function of one device into the receive queue
1476 * of another device.
1478 * The receiving device may be in another namespace, so
1479 * we have to clear all information in the skb that could
1480 * impact namespace isolation.
1482 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1486 if (!(dev->flags & IFF_UP))
1489 if (skb->len > (dev->mtu + dev->hard_header_len))
1492 skb_set_dev(skb, dev);
1493 skb->tstamp.tv64 = 0;
1494 skb->pkt_type = PACKET_HOST;
1495 skb->protocol = eth_type_trans(skb, dev);
1496 return netif_rx(skb);
1498 EXPORT_SYMBOL_GPL(dev_forward_skb);
1501 * Support routine. Sends outgoing frames to any network
1502 * taps currently in use.
1505 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1507 struct packet_type *ptype;
1509 #ifdef CONFIG_NET_CLS_ACT
1510 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1517 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1518 /* Never send packets back to the socket
1519 * they originated from - MvS (miquels@drinkel.ow.org)
1521 if ((ptype->dev == dev || !ptype->dev) &&
1522 (ptype->af_packet_priv == NULL ||
1523 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1524 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1528 /* skb->nh should be correctly
1529 set by sender, so that the second statement is
1530 just protection against buggy protocols.
1532 skb_reset_mac_header(skb2);
1534 if (skb_network_header(skb2) < skb2->data ||
1535 skb2->network_header > skb2->tail) {
1536 if (net_ratelimit())
1537 printk(KERN_CRIT "protocol %04x is "
1539 skb2->protocol, dev->name);
1540 skb_reset_network_header(skb2);
1543 skb2->transport_header = skb2->network_header;
1544 skb2->pkt_type = PACKET_OUTGOING;
1545 ptype->func(skb2, skb->dev, ptype, skb->dev);
1552 static inline void __netif_reschedule(struct Qdisc *q)
1554 struct softnet_data *sd;
1555 unsigned long flags;
1557 local_irq_save(flags);
1558 sd = &__get_cpu_var(softnet_data);
1559 q->next_sched = sd->output_queue;
1560 sd->output_queue = q;
1561 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1562 local_irq_restore(flags);
1565 void __netif_schedule(struct Qdisc *q)
1567 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1568 __netif_reschedule(q);
1570 EXPORT_SYMBOL(__netif_schedule);
1572 void dev_kfree_skb_irq(struct sk_buff *skb)
1574 if (atomic_dec_and_test(&skb->users)) {
1575 struct softnet_data *sd;
1576 unsigned long flags;
1578 local_irq_save(flags);
1579 sd = &__get_cpu_var(softnet_data);
1580 skb->next = sd->completion_queue;
1581 sd->completion_queue = skb;
1582 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1583 local_irq_restore(flags);
1586 EXPORT_SYMBOL(dev_kfree_skb_irq);
1588 void dev_kfree_skb_any(struct sk_buff *skb)
1590 if (in_irq() || irqs_disabled())
1591 dev_kfree_skb_irq(skb);
1595 EXPORT_SYMBOL(dev_kfree_skb_any);
1599 * netif_device_detach - mark device as removed
1600 * @dev: network device
1602 * Mark device as removed from system and therefore no longer available.
1604 void netif_device_detach(struct net_device *dev)
1606 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1607 netif_running(dev)) {
1608 netif_tx_stop_all_queues(dev);
1611 EXPORT_SYMBOL(netif_device_detach);
1614 * netif_device_attach - mark device as attached
1615 * @dev: network device
1617 * Mark device as attached from system and restart if needed.
1619 void netif_device_attach(struct net_device *dev)
1621 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1622 netif_running(dev)) {
1623 netif_tx_wake_all_queues(dev);
1624 __netdev_watchdog_up(dev);
1627 EXPORT_SYMBOL(netif_device_attach);
1629 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1631 return ((features & NETIF_F_GEN_CSUM) ||
1632 ((features & NETIF_F_IP_CSUM) &&
1633 protocol == htons(ETH_P_IP)) ||
1634 ((features & NETIF_F_IPV6_CSUM) &&
1635 protocol == htons(ETH_P_IPV6)) ||
1636 ((features & NETIF_F_FCOE_CRC) &&
1637 protocol == htons(ETH_P_FCOE)));
1640 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1642 if (can_checksum_protocol(dev->features, skb->protocol))
1645 if (skb->protocol == htons(ETH_P_8021Q)) {
1646 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1647 if (can_checksum_protocol(dev->features & dev->vlan_features,
1648 veh->h_vlan_encapsulated_proto))
1656 * skb_dev_set -- assign a new device to a buffer
1657 * @skb: buffer for the new device
1658 * @dev: network device
1660 * If an skb is owned by a device already, we have to reset
1661 * all data private to the namespace a device belongs to
1662 * before assigning it a new device.
1664 #ifdef CONFIG_NET_NS
1665 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1668 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1671 skb_init_secmark(skb);
1675 skb->ipvs_property = 0;
1676 #ifdef CONFIG_NET_SCHED
1682 EXPORT_SYMBOL(skb_set_dev);
1683 #endif /* CONFIG_NET_NS */
1686 * Invalidate hardware checksum when packet is to be mangled, and
1687 * complete checksum manually on outgoing path.
1689 int skb_checksum_help(struct sk_buff *skb)
1692 int ret = 0, offset;
1694 if (skb->ip_summed == CHECKSUM_COMPLETE)
1695 goto out_set_summed;
1697 if (unlikely(skb_shinfo(skb)->gso_size)) {
1698 /* Let GSO fix up the checksum. */
1699 goto out_set_summed;
1702 offset = skb->csum_start - skb_headroom(skb);
1703 BUG_ON(offset >= skb_headlen(skb));
1704 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1706 offset += skb->csum_offset;
1707 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1709 if (skb_cloned(skb) &&
1710 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1711 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1716 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1718 skb->ip_summed = CHECKSUM_NONE;
1722 EXPORT_SYMBOL(skb_checksum_help);
1725 * skb_gso_segment - Perform segmentation on skb.
1726 * @skb: buffer to segment
1727 * @features: features for the output path (see dev->features)
1729 * This function segments the given skb and returns a list of segments.
1731 * It may return NULL if the skb requires no segmentation. This is
1732 * only possible when GSO is used for verifying header integrity.
1734 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1736 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1737 struct packet_type *ptype;
1738 __be16 type = skb->protocol;
1741 skb_reset_mac_header(skb);
1742 skb->mac_len = skb->network_header - skb->mac_header;
1743 __skb_pull(skb, skb->mac_len);
1745 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1746 struct net_device *dev = skb->dev;
1747 struct ethtool_drvinfo info = {};
1749 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1750 dev->ethtool_ops->get_drvinfo(dev, &info);
1752 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1754 info.driver, dev ? dev->features : 0L,
1755 skb->sk ? skb->sk->sk_route_caps : 0L,
1756 skb->len, skb->data_len, skb->ip_summed);
1758 if (skb_header_cloned(skb) &&
1759 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1760 return ERR_PTR(err);
1764 list_for_each_entry_rcu(ptype,
1765 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1766 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1767 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1768 err = ptype->gso_send_check(skb);
1769 segs = ERR_PTR(err);
1770 if (err || skb_gso_ok(skb, features))
1772 __skb_push(skb, (skb->data -
1773 skb_network_header(skb)));
1775 segs = ptype->gso_segment(skb, features);
1781 __skb_push(skb, skb->data - skb_mac_header(skb));
1785 EXPORT_SYMBOL(skb_gso_segment);
1787 /* Take action when hardware reception checksum errors are detected. */
1789 void netdev_rx_csum_fault(struct net_device *dev)
1791 if (net_ratelimit()) {
1792 printk(KERN_ERR "%s: hw csum failure.\n",
1793 dev ? dev->name : "<unknown>");
1797 EXPORT_SYMBOL(netdev_rx_csum_fault);
1800 /* Actually, we should eliminate this check as soon as we know, that:
1801 * 1. IOMMU is present and allows to map all the memory.
1802 * 2. No high memory really exists on this machine.
1805 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1807 #ifdef CONFIG_HIGHMEM
1809 if (!(dev->features & NETIF_F_HIGHDMA)) {
1810 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1811 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1815 if (PCI_DMA_BUS_IS_PHYS) {
1816 struct device *pdev = dev->dev.parent;
1820 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1821 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1822 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1831 void (*destructor)(struct sk_buff *skb);
1834 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1836 static void dev_gso_skb_destructor(struct sk_buff *skb)
1838 struct dev_gso_cb *cb;
1841 struct sk_buff *nskb = skb->next;
1843 skb->next = nskb->next;
1846 } while (skb->next);
1848 cb = DEV_GSO_CB(skb);
1850 cb->destructor(skb);
1854 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1855 * @skb: buffer to segment
1857 * This function segments the given skb and stores the list of segments
1860 static int dev_gso_segment(struct sk_buff *skb)
1862 struct net_device *dev = skb->dev;
1863 struct sk_buff *segs;
1864 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1867 segs = skb_gso_segment(skb, features);
1869 /* Verifying header integrity only. */
1874 return PTR_ERR(segs);
1877 DEV_GSO_CB(skb)->destructor = skb->destructor;
1878 skb->destructor = dev_gso_skb_destructor;
1884 * Try to orphan skb early, right before transmission by the device.
1885 * We cannot orphan skb if tx timestamp is requested, since
1886 * drivers need to call skb_tstamp_tx() to send the timestamp.
1888 static inline void skb_orphan_try(struct sk_buff *skb)
1890 if (!skb_tx(skb)->flags)
1894 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1895 struct netdev_queue *txq)
1897 const struct net_device_ops *ops = dev->netdev_ops;
1898 int rc = NETDEV_TX_OK;
1900 if (likely(!skb->next)) {
1901 if (!list_empty(&ptype_all))
1902 dev_queue_xmit_nit(skb, dev);
1904 if (netif_needs_gso(dev, skb)) {
1905 if (unlikely(dev_gso_segment(skb)))
1912 * If device doesnt need skb->dst, release it right now while
1913 * its hot in this cpu cache
1915 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1918 skb_orphan_try(skb);
1919 rc = ops->ndo_start_xmit(skb, dev);
1920 if (rc == NETDEV_TX_OK)
1921 txq_trans_update(txq);
1927 struct sk_buff *nskb = skb->next;
1929 skb->next = nskb->next;
1933 * If device doesnt need nskb->dst, release it right now while
1934 * its hot in this cpu cache
1936 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1939 skb_orphan_try(nskb);
1940 rc = ops->ndo_start_xmit(nskb, dev);
1941 if (unlikely(rc != NETDEV_TX_OK)) {
1942 if (rc & ~NETDEV_TX_MASK)
1943 goto out_kfree_gso_skb;
1944 nskb->next = skb->next;
1948 txq_trans_update(txq);
1949 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1950 return NETDEV_TX_BUSY;
1951 } while (skb->next);
1954 if (likely(skb->next == NULL))
1955 skb->destructor = DEV_GSO_CB(skb)->destructor;
1961 static u32 hashrnd __read_mostly;
1963 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1967 if (skb_rx_queue_recorded(skb)) {
1968 hash = skb_get_rx_queue(skb);
1969 while (unlikely(hash >= dev->real_num_tx_queues))
1970 hash -= dev->real_num_tx_queues;
1974 if (skb->sk && skb->sk->sk_hash)
1975 hash = skb->sk->sk_hash;
1977 hash = skb->protocol;
1979 hash = jhash_1word(hash, hashrnd);
1981 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1983 EXPORT_SYMBOL(skb_tx_hash);
1985 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1987 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1988 if (net_ratelimit()) {
1989 pr_warning("%s selects TX queue %d, but "
1990 "real number of TX queues is %d\n",
1991 dev->name, queue_index, dev->real_num_tx_queues);
1998 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1999 struct sk_buff *skb)
2002 struct sock *sk = skb->sk;
2004 if (sk_tx_queue_recorded(sk)) {
2005 queue_index = sk_tx_queue_get(sk);
2007 const struct net_device_ops *ops = dev->netdev_ops;
2009 if (ops->ndo_select_queue) {
2010 queue_index = ops->ndo_select_queue(dev, skb);
2011 queue_index = dev_cap_txqueue(dev, queue_index);
2014 if (dev->real_num_tx_queues > 1)
2015 queue_index = skb_tx_hash(dev, skb);
2017 if (sk && rcu_dereference_check(sk->sk_dst_cache, 1))
2018 sk_tx_queue_set(sk, queue_index);
2022 skb_set_queue_mapping(skb, queue_index);
2023 return netdev_get_tx_queue(dev, queue_index);
2026 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2027 struct net_device *dev,
2028 struct netdev_queue *txq)
2030 spinlock_t *root_lock = qdisc_lock(q);
2033 spin_lock(root_lock);
2034 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2037 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2038 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
2040 * This is a work-conserving queue; there are no old skbs
2041 * waiting to be sent out; and the qdisc is not running -
2042 * xmit the skb directly.
2044 __qdisc_update_bstats(q, skb->len);
2045 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
2048 clear_bit(__QDISC_STATE_RUNNING, &q->state);
2050 rc = NET_XMIT_SUCCESS;
2052 rc = qdisc_enqueue_root(skb, q);
2055 spin_unlock(root_lock);
2061 * Returns true if either:
2062 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2063 * 2. skb is fragmented and the device does not support SG, or if
2064 * at least one of fragments is in highmem and device does not
2065 * support DMA from it.
2067 static inline int skb_needs_linearize(struct sk_buff *skb,
2068 struct net_device *dev)
2070 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
2071 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
2072 illegal_highdma(dev, skb)));
2076 * dev_queue_xmit - transmit a buffer
2077 * @skb: buffer to transmit
2079 * Queue a buffer for transmission to a network device. The caller must
2080 * have set the device and priority and built the buffer before calling
2081 * this function. The function can be called from an interrupt.
2083 * A negative errno code is returned on a failure. A success does not
2084 * guarantee the frame will be transmitted as it may be dropped due
2085 * to congestion or traffic shaping.
2087 * -----------------------------------------------------------------------------------
2088 * I notice this method can also return errors from the queue disciplines,
2089 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2092 * Regardless of the return value, the skb is consumed, so it is currently
2093 * difficult to retry a send to this method. (You can bump the ref count
2094 * before sending to hold a reference for retry if you are careful.)
2096 * When calling this method, interrupts MUST be enabled. This is because
2097 * the BH enable code must have IRQs enabled so that it will not deadlock.
2100 int dev_queue_xmit(struct sk_buff *skb)
2102 struct net_device *dev = skb->dev;
2103 struct netdev_queue *txq;
2107 /* GSO will handle the following emulations directly. */
2108 if (netif_needs_gso(dev, skb))
2111 /* Convert a paged skb to linear, if required */
2112 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2115 /* If packet is not checksummed and device does not support
2116 * checksumming for this protocol, complete checksumming here.
2118 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2119 skb_set_transport_header(skb, skb->csum_start -
2121 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2126 /* Disable soft irqs for various locks below. Also
2127 * stops preemption for RCU.
2131 txq = dev_pick_tx(dev, skb);
2132 q = rcu_dereference_bh(txq->qdisc);
2134 #ifdef CONFIG_NET_CLS_ACT
2135 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2138 rc = __dev_xmit_skb(skb, q, dev, txq);
2142 /* The device has no queue. Common case for software devices:
2143 loopback, all the sorts of tunnels...
2145 Really, it is unlikely that netif_tx_lock protection is necessary
2146 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2148 However, it is possible, that they rely on protection
2151 Check this and shot the lock. It is not prone from deadlocks.
2152 Either shot noqueue qdisc, it is even simpler 8)
2154 if (dev->flags & IFF_UP) {
2155 int cpu = smp_processor_id(); /* ok because BHs are off */
2157 if (txq->xmit_lock_owner != cpu) {
2159 HARD_TX_LOCK(dev, txq, cpu);
2161 if (!netif_tx_queue_stopped(txq)) {
2162 rc = dev_hard_start_xmit(skb, dev, txq);
2163 if (dev_xmit_complete(rc)) {
2164 HARD_TX_UNLOCK(dev, txq);
2168 HARD_TX_UNLOCK(dev, txq);
2169 if (net_ratelimit())
2170 printk(KERN_CRIT "Virtual device %s asks to "
2171 "queue packet!\n", dev->name);
2173 /* Recursion is detected! It is possible,
2175 if (net_ratelimit())
2176 printk(KERN_CRIT "Dead loop on virtual device "
2177 "%s, fix it urgently!\n", dev->name);
2182 rcu_read_unlock_bh();
2188 rcu_read_unlock_bh();
2191 EXPORT_SYMBOL(dev_queue_xmit);
2194 /*=======================================================================
2196 =======================================================================*/
2198 int netdev_max_backlog __read_mostly = 1000;
2199 int netdev_budget __read_mostly = 300;
2200 int weight_p __read_mostly = 64; /* old backlog weight */
2202 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2206 /* One global table that all flow-based protocols share. */
2207 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2208 EXPORT_SYMBOL(rps_sock_flow_table);
2211 * get_rps_cpu is called from netif_receive_skb and returns the target
2212 * CPU from the RPS map of the receiving queue for a given skb.
2213 * rcu_read_lock must be held on entry.
2215 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2216 struct rps_dev_flow **rflowp)
2218 struct ipv6hdr *ip6;
2220 struct netdev_rx_queue *rxqueue;
2221 struct rps_map *map;
2222 struct rps_dev_flow_table *flow_table;
2223 struct rps_sock_flow_table *sock_flow_table;
2227 u32 addr1, addr2, ports, ihl;
2229 if (skb_rx_queue_recorded(skb)) {
2230 u16 index = skb_get_rx_queue(skb);
2231 if (unlikely(index >= dev->num_rx_queues)) {
2232 if (net_ratelimit()) {
2233 pr_warning("%s received packet on queue "
2234 "%u, but number of RX queues is %u\n",
2235 dev->name, index, dev->num_rx_queues);
2239 rxqueue = dev->_rx + index;
2243 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2247 goto got_hash; /* Skip hash computation on packet header */
2249 switch (skb->protocol) {
2250 case __constant_htons(ETH_P_IP):
2251 if (!pskb_may_pull(skb, sizeof(*ip)))
2254 ip = (struct iphdr *) skb->data;
2255 ip_proto = ip->protocol;
2260 case __constant_htons(ETH_P_IPV6):
2261 if (!pskb_may_pull(skb, sizeof(*ip6)))
2264 ip6 = (struct ipv6hdr *) skb->data;
2265 ip_proto = ip6->nexthdr;
2266 addr1 = ip6->saddr.s6_addr32[3];
2267 addr2 = ip6->daddr.s6_addr32[3];
2281 case IPPROTO_UDPLITE:
2282 if (pskb_may_pull(skb, (ihl * 4) + 4))
2283 ports = *((u32 *) (skb->data + (ihl * 4)));
2290 skb->rxhash = jhash_3words(addr1, addr2, ports, hashrnd);
2295 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2296 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2297 if (flow_table && sock_flow_table) {
2299 struct rps_dev_flow *rflow;
2301 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2304 next_cpu = sock_flow_table->ents[skb->rxhash &
2305 sock_flow_table->mask];
2308 * If the desired CPU (where last recvmsg was done) is
2309 * different from current CPU (one in the rx-queue flow
2310 * table entry), switch if one of the following holds:
2311 * - Current CPU is unset (equal to RPS_NO_CPU).
2312 * - Current CPU is offline.
2313 * - The current CPU's queue tail has advanced beyond the
2314 * last packet that was enqueued using this table entry.
2315 * This guarantees that all previous packets for the flow
2316 * have been dequeued, thus preserving in order delivery.
2318 if (unlikely(tcpu != next_cpu) &&
2319 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2320 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2321 rflow->last_qtail)) >= 0)) {
2322 tcpu = rflow->cpu = next_cpu;
2323 if (tcpu != RPS_NO_CPU)
2324 rflow->last_qtail = per_cpu(softnet_data,
2325 tcpu).input_queue_head;
2327 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2334 map = rcu_dereference(rxqueue->rps_map);
2336 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2338 if (cpu_online(tcpu)) {
2348 /* Called from hardirq (IPI) context */
2349 static void trigger_softirq(void *data)
2351 struct softnet_data *queue = data;
2352 __napi_schedule(&queue->backlog);
2353 __get_cpu_var(netdev_rx_stat).received_rps++;
2355 #endif /* CONFIG_RPS */
2358 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2359 * queue (may be a remote CPU queue).
2361 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2362 unsigned int *qtail)
2364 struct softnet_data *queue;
2365 unsigned long flags;
2367 queue = &per_cpu(softnet_data, cpu);
2369 local_irq_save(flags);
2370 __get_cpu_var(netdev_rx_stat).total++;
2373 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2374 if (queue->input_pkt_queue.qlen) {
2376 __skb_queue_tail(&queue->input_pkt_queue, skb);
2378 *qtail = queue->input_queue_head +
2379 queue->input_pkt_queue.qlen;
2382 local_irq_restore(flags);
2383 return NET_RX_SUCCESS;
2386 /* Schedule NAPI for backlog device */
2387 if (napi_schedule_prep(&queue->backlog)) {
2389 if (cpu != smp_processor_id()) {
2390 struct softnet_data *myqueue;
2392 myqueue = &__get_cpu_var(softnet_data);
2393 queue->rps_ipi_next = myqueue->rps_ipi_list;
2394 myqueue->rps_ipi_list = queue;
2396 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2400 __napi_schedule(&queue->backlog);
2407 __get_cpu_var(netdev_rx_stat).dropped++;
2408 local_irq_restore(flags);
2415 * netif_rx - post buffer to the network code
2416 * @skb: buffer to post
2418 * This function receives a packet from a device driver and queues it for
2419 * the upper (protocol) levels to process. It always succeeds. The buffer
2420 * may be dropped during processing for congestion control or by the
2424 * NET_RX_SUCCESS (no congestion)
2425 * NET_RX_DROP (packet was dropped)
2429 int netif_rx(struct sk_buff *skb)
2433 /* if netpoll wants it, pretend we never saw it */
2434 if (netpoll_rx(skb))
2437 if (!skb->tstamp.tv64)
2442 struct rps_dev_flow voidflow, *rflow = &voidflow;
2447 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2449 cpu = smp_processor_id();
2451 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2458 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2464 EXPORT_SYMBOL(netif_rx);
2466 int netif_rx_ni(struct sk_buff *skb)
2471 err = netif_rx(skb);
2472 if (local_softirq_pending())
2478 EXPORT_SYMBOL(netif_rx_ni);
2480 static void net_tx_action(struct softirq_action *h)
2482 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2484 if (sd->completion_queue) {
2485 struct sk_buff *clist;
2487 local_irq_disable();
2488 clist = sd->completion_queue;
2489 sd->completion_queue = NULL;
2493 struct sk_buff *skb = clist;
2494 clist = clist->next;
2496 WARN_ON(atomic_read(&skb->users));
2501 if (sd->output_queue) {
2504 local_irq_disable();
2505 head = sd->output_queue;
2506 sd->output_queue = NULL;
2510 struct Qdisc *q = head;
2511 spinlock_t *root_lock;
2513 head = head->next_sched;
2515 root_lock = qdisc_lock(q);
2516 if (spin_trylock(root_lock)) {
2517 smp_mb__before_clear_bit();
2518 clear_bit(__QDISC_STATE_SCHED,
2521 spin_unlock(root_lock);
2523 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2525 __netif_reschedule(q);
2527 smp_mb__before_clear_bit();
2528 clear_bit(__QDISC_STATE_SCHED,
2536 static inline int deliver_skb(struct sk_buff *skb,
2537 struct packet_type *pt_prev,
2538 struct net_device *orig_dev)
2540 atomic_inc(&skb->users);
2541 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2544 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2546 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2547 /* This hook is defined here for ATM LANE */
2548 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2549 unsigned char *addr) __read_mostly;
2550 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2554 * If bridge module is loaded call bridging hook.
2555 * returns NULL if packet was consumed.
2557 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2558 struct sk_buff *skb) __read_mostly;
2559 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2561 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2562 struct packet_type **pt_prev, int *ret,
2563 struct net_device *orig_dev)
2565 struct net_bridge_port *port;
2567 if (skb->pkt_type == PACKET_LOOPBACK ||
2568 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2572 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2576 return br_handle_frame_hook(port, skb);
2579 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2582 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2583 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2584 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2586 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2587 struct packet_type **pt_prev,
2589 struct net_device *orig_dev)
2591 if (skb->dev->macvlan_port == NULL)
2595 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2598 return macvlan_handle_frame_hook(skb);
2601 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2604 #ifdef CONFIG_NET_CLS_ACT
2605 /* TODO: Maybe we should just force sch_ingress to be compiled in
2606 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2607 * a compare and 2 stores extra right now if we dont have it on
2608 * but have CONFIG_NET_CLS_ACT
2609 * NOTE: This doesnt stop any functionality; if you dont have
2610 * the ingress scheduler, you just cant add policies on ingress.
2613 static int ing_filter(struct sk_buff *skb)
2615 struct net_device *dev = skb->dev;
2616 u32 ttl = G_TC_RTTL(skb->tc_verd);
2617 struct netdev_queue *rxq;
2618 int result = TC_ACT_OK;
2621 if (MAX_RED_LOOP < ttl++) {
2623 "Redir loop detected Dropping packet (%d->%d)\n",
2624 skb->skb_iif, dev->ifindex);
2628 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2629 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2631 rxq = &dev->rx_queue;
2634 if (q != &noop_qdisc) {
2635 spin_lock(qdisc_lock(q));
2636 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2637 result = qdisc_enqueue_root(skb, q);
2638 spin_unlock(qdisc_lock(q));
2644 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2645 struct packet_type **pt_prev,
2646 int *ret, struct net_device *orig_dev)
2648 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2652 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2655 /* Huh? Why does turning on AF_PACKET affect this? */
2656 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2659 switch (ing_filter(skb)) {
2673 * netif_nit_deliver - deliver received packets to network taps
2676 * This function is used to deliver incoming packets to network
2677 * taps. It should be used when the normal netif_receive_skb path
2678 * is bypassed, for example because of VLAN acceleration.
2680 void netif_nit_deliver(struct sk_buff *skb)
2682 struct packet_type *ptype;
2684 if (list_empty(&ptype_all))
2687 skb_reset_network_header(skb);
2688 skb_reset_transport_header(skb);
2689 skb->mac_len = skb->network_header - skb->mac_header;
2692 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2693 if (!ptype->dev || ptype->dev == skb->dev)
2694 deliver_skb(skb, ptype, skb->dev);
2699 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2700 struct net_device *master)
2702 if (skb->pkt_type == PACKET_HOST) {
2703 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2705 memcpy(dest, master->dev_addr, ETH_ALEN);
2709 /* On bonding slaves other than the currently active slave, suppress
2710 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2711 * ARP on active-backup slaves with arp_validate enabled.
2713 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2715 struct net_device *dev = skb->dev;
2717 if (master->priv_flags & IFF_MASTER_ARPMON)
2718 dev->last_rx = jiffies;
2720 if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) {
2721 /* Do address unmangle. The local destination address
2722 * will be always the one master has. Provides the right
2723 * functionality in a bridge.
2725 skb_bond_set_mac_by_master(skb, master);
2728 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2729 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2730 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2733 if (master->priv_flags & IFF_MASTER_ALB) {
2734 if (skb->pkt_type != PACKET_BROADCAST &&
2735 skb->pkt_type != PACKET_MULTICAST)
2738 if (master->priv_flags & IFF_MASTER_8023AD &&
2739 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2746 EXPORT_SYMBOL(__skb_bond_should_drop);
2748 static int __netif_receive_skb(struct sk_buff *skb)
2750 struct packet_type *ptype, *pt_prev;
2751 struct net_device *orig_dev;
2752 struct net_device *master;
2753 struct net_device *null_or_orig;
2754 struct net_device *null_or_bond;
2755 int ret = NET_RX_DROP;
2758 if (!skb->tstamp.tv64)
2761 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2762 return NET_RX_SUCCESS;
2764 /* if we've gotten here through NAPI, check netpoll */
2765 if (netpoll_receive_skb(skb))
2769 skb->skb_iif = skb->dev->ifindex;
2771 null_or_orig = NULL;
2772 orig_dev = skb->dev;
2773 master = ACCESS_ONCE(orig_dev->master);
2775 if (skb_bond_should_drop(skb, master))
2776 null_or_orig = orig_dev; /* deliver only exact match */
2781 __get_cpu_var(netdev_rx_stat).total++;
2783 skb_reset_network_header(skb);
2784 skb_reset_transport_header(skb);
2785 skb->mac_len = skb->network_header - skb->mac_header;
2791 #ifdef CONFIG_NET_CLS_ACT
2792 if (skb->tc_verd & TC_NCLS) {
2793 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2798 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2799 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2800 ptype->dev == orig_dev) {
2802 ret = deliver_skb(skb, pt_prev, orig_dev);
2807 #ifdef CONFIG_NET_CLS_ACT
2808 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2814 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2817 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2822 * Make sure frames received on VLAN interfaces stacked on
2823 * bonding interfaces still make their way to any base bonding
2824 * device that may have registered for a specific ptype. The
2825 * handler may have to adjust skb->dev and orig_dev.
2827 null_or_bond = NULL;
2828 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2829 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2830 null_or_bond = vlan_dev_real_dev(skb->dev);
2833 type = skb->protocol;
2834 list_for_each_entry_rcu(ptype,
2835 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2836 if (ptype->type == type && (ptype->dev == null_or_orig ||
2837 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2838 ptype->dev == null_or_bond)) {
2840 ret = deliver_skb(skb, pt_prev, orig_dev);
2846 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2849 /* Jamal, now you will not able to escape explaining
2850 * me how you were going to use this. :-)
2861 * netif_receive_skb - process receive buffer from network
2862 * @skb: buffer to process
2864 * netif_receive_skb() is the main receive data processing function.
2865 * It always succeeds. The buffer may be dropped during processing
2866 * for congestion control or by the protocol layers.
2868 * This function may only be called from softirq context and interrupts
2869 * should be enabled.
2871 * Return values (usually ignored):
2872 * NET_RX_SUCCESS: no congestion
2873 * NET_RX_DROP: packet was dropped
2875 int netif_receive_skb(struct sk_buff *skb)
2878 struct rps_dev_flow voidflow, *rflow = &voidflow;
2883 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2886 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2890 ret = __netif_receive_skb(skb);
2895 return __netif_receive_skb(skb);
2898 EXPORT_SYMBOL(netif_receive_skb);
2900 /* Network device is going away, flush any packets still pending
2901 * Called with irqs disabled.
2903 static void flush_backlog(void *arg)
2905 struct net_device *dev = arg;
2906 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2907 struct sk_buff *skb, *tmp;
2910 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2911 if (skb->dev == dev) {
2912 __skb_unlink(skb, &queue->input_pkt_queue);
2914 incr_input_queue_head(queue);
2919 static int napi_gro_complete(struct sk_buff *skb)
2921 struct packet_type *ptype;
2922 __be16 type = skb->protocol;
2923 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2926 if (NAPI_GRO_CB(skb)->count == 1) {
2927 skb_shinfo(skb)->gso_size = 0;
2932 list_for_each_entry_rcu(ptype, head, list) {
2933 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2936 err = ptype->gro_complete(skb);
2942 WARN_ON(&ptype->list == head);
2944 return NET_RX_SUCCESS;
2948 return netif_receive_skb(skb);
2951 static void napi_gro_flush(struct napi_struct *napi)
2953 struct sk_buff *skb, *next;
2955 for (skb = napi->gro_list; skb; skb = next) {
2958 napi_gro_complete(skb);
2961 napi->gro_count = 0;
2962 napi->gro_list = NULL;
2965 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2967 struct sk_buff **pp = NULL;
2968 struct packet_type *ptype;
2969 __be16 type = skb->protocol;
2970 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2973 enum gro_result ret;
2975 if (!(skb->dev->features & NETIF_F_GRO))
2978 if (skb_is_gso(skb) || skb_has_frags(skb))
2982 list_for_each_entry_rcu(ptype, head, list) {
2983 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2986 skb_set_network_header(skb, skb_gro_offset(skb));
2987 mac_len = skb->network_header - skb->mac_header;
2988 skb->mac_len = mac_len;
2989 NAPI_GRO_CB(skb)->same_flow = 0;
2990 NAPI_GRO_CB(skb)->flush = 0;
2991 NAPI_GRO_CB(skb)->free = 0;
2993 pp = ptype->gro_receive(&napi->gro_list, skb);
2998 if (&ptype->list == head)
3001 same_flow = NAPI_GRO_CB(skb)->same_flow;
3002 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3005 struct sk_buff *nskb = *pp;
3009 napi_gro_complete(nskb);
3016 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3020 NAPI_GRO_CB(skb)->count = 1;
3021 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3022 skb->next = napi->gro_list;
3023 napi->gro_list = skb;
3027 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3028 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3030 BUG_ON(skb->end - skb->tail < grow);
3032 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3035 skb->data_len -= grow;
3037 skb_shinfo(skb)->frags[0].page_offset += grow;
3038 skb_shinfo(skb)->frags[0].size -= grow;
3040 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3041 put_page(skb_shinfo(skb)->frags[0].page);
3042 memmove(skb_shinfo(skb)->frags,
3043 skb_shinfo(skb)->frags + 1,
3044 --skb_shinfo(skb)->nr_frags);
3055 EXPORT_SYMBOL(dev_gro_receive);
3058 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3062 if (netpoll_rx_on(skb))
3065 for (p = napi->gro_list; p; p = p->next) {
3066 NAPI_GRO_CB(p)->same_flow =
3067 (p->dev == skb->dev) &&
3068 !compare_ether_header(skb_mac_header(p),
3069 skb_gro_mac_header(skb));
3070 NAPI_GRO_CB(p)->flush = 0;
3073 return dev_gro_receive(napi, skb);
3076 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3080 if (netif_receive_skb(skb))
3085 case GRO_MERGED_FREE:
3096 EXPORT_SYMBOL(napi_skb_finish);
3098 void skb_gro_reset_offset(struct sk_buff *skb)
3100 NAPI_GRO_CB(skb)->data_offset = 0;
3101 NAPI_GRO_CB(skb)->frag0 = NULL;
3102 NAPI_GRO_CB(skb)->frag0_len = 0;
3104 if (skb->mac_header == skb->tail &&
3105 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3106 NAPI_GRO_CB(skb)->frag0 =
3107 page_address(skb_shinfo(skb)->frags[0].page) +
3108 skb_shinfo(skb)->frags[0].page_offset;
3109 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3112 EXPORT_SYMBOL(skb_gro_reset_offset);
3114 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3116 skb_gro_reset_offset(skb);
3118 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3120 EXPORT_SYMBOL(napi_gro_receive);
3122 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3124 __skb_pull(skb, skb_headlen(skb));
3125 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3129 EXPORT_SYMBOL(napi_reuse_skb);
3131 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3133 struct sk_buff *skb = napi->skb;
3136 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3142 EXPORT_SYMBOL(napi_get_frags);
3144 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3150 skb->protocol = eth_type_trans(skb, skb->dev);
3152 if (ret == GRO_HELD)
3153 skb_gro_pull(skb, -ETH_HLEN);
3154 else if (netif_receive_skb(skb))
3159 case GRO_MERGED_FREE:
3160 napi_reuse_skb(napi, skb);
3169 EXPORT_SYMBOL(napi_frags_finish);
3171 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3173 struct sk_buff *skb = napi->skb;
3180 skb_reset_mac_header(skb);
3181 skb_gro_reset_offset(skb);
3183 off = skb_gro_offset(skb);
3184 hlen = off + sizeof(*eth);
3185 eth = skb_gro_header_fast(skb, off);
3186 if (skb_gro_header_hard(skb, hlen)) {
3187 eth = skb_gro_header_slow(skb, hlen, off);
3188 if (unlikely(!eth)) {
3189 napi_reuse_skb(napi, skb);
3195 skb_gro_pull(skb, sizeof(*eth));
3198 * This works because the only protocols we care about don't require
3199 * special handling. We'll fix it up properly at the end.
3201 skb->protocol = eth->h_proto;
3206 EXPORT_SYMBOL(napi_frags_skb);
3208 gro_result_t napi_gro_frags(struct napi_struct *napi)
3210 struct sk_buff *skb = napi_frags_skb(napi);
3215 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3217 EXPORT_SYMBOL(napi_gro_frags);
3219 static int process_backlog(struct napi_struct *napi, int quota)
3222 struct softnet_data *queue = &__get_cpu_var(softnet_data);
3224 napi->weight = weight_p;
3226 struct sk_buff *skb;
3228 local_irq_disable();
3230 skb = __skb_dequeue(&queue->input_pkt_queue);
3232 __napi_complete(napi);
3237 incr_input_queue_head(queue);
3241 __netif_receive_skb(skb);
3242 } while (++work < quota);
3248 * __napi_schedule - schedule for receive
3249 * @n: entry to schedule
3251 * The entry's receive function will be scheduled to run
3253 void __napi_schedule(struct napi_struct *n)
3255 unsigned long flags;
3257 local_irq_save(flags);
3258 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
3259 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3260 local_irq_restore(flags);
3262 EXPORT_SYMBOL(__napi_schedule);
3264 void __napi_complete(struct napi_struct *n)
3266 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3267 BUG_ON(n->gro_list);
3269 list_del(&n->poll_list);
3270 smp_mb__before_clear_bit();
3271 clear_bit(NAPI_STATE_SCHED, &n->state);
3273 EXPORT_SYMBOL(__napi_complete);
3275 void napi_complete(struct napi_struct *n)
3277 unsigned long flags;
3280 * don't let napi dequeue from the cpu poll list
3281 * just in case its running on a different cpu
3283 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3287 local_irq_save(flags);
3289 local_irq_restore(flags);
3291 EXPORT_SYMBOL(napi_complete);
3293 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3294 int (*poll)(struct napi_struct *, int), int weight)
3296 INIT_LIST_HEAD(&napi->poll_list);
3297 napi->gro_count = 0;
3298 napi->gro_list = NULL;
3301 napi->weight = weight;
3302 list_add(&napi->dev_list, &dev->napi_list);
3304 #ifdef CONFIG_NETPOLL
3305 spin_lock_init(&napi->poll_lock);
3306 napi->poll_owner = -1;
3308 set_bit(NAPI_STATE_SCHED, &napi->state);
3310 EXPORT_SYMBOL(netif_napi_add);
3312 void netif_napi_del(struct napi_struct *napi)
3314 struct sk_buff *skb, *next;
3316 list_del_init(&napi->dev_list);
3317 napi_free_frags(napi);
3319 for (skb = napi->gro_list; skb; skb = next) {
3325 napi->gro_list = NULL;
3326 napi->gro_count = 0;
3328 EXPORT_SYMBOL(netif_napi_del);
3331 * net_rps_action sends any pending IPI's for rps.
3332 * Note: called with local irq disabled, but exits with local irq enabled.
3334 static void net_rps_action(void)
3337 struct softnet_data *locqueue = &__get_cpu_var(softnet_data);
3338 struct softnet_data *remqueue = locqueue->rps_ipi_list;
3341 locqueue->rps_ipi_list = NULL;
3345 /* Send pending IPI's to kick RPS processing on remote cpus. */
3347 struct softnet_data *next = remqueue->rps_ipi_next;
3348 if (cpu_online(remqueue->cpu))
3349 __smp_call_function_single(remqueue->cpu,
3358 static void net_rx_action(struct softirq_action *h)
3360 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
3361 unsigned long time_limit = jiffies + 2;
3362 int budget = netdev_budget;
3365 local_irq_disable();
3367 while (!list_empty(list)) {
3368 struct napi_struct *n;
3371 /* If softirq window is exhuasted then punt.
3372 * Allow this to run for 2 jiffies since which will allow
3373 * an average latency of 1.5/HZ.
3375 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3380 /* Even though interrupts have been re-enabled, this
3381 * access is safe because interrupts can only add new
3382 * entries to the tail of this list, and only ->poll()
3383 * calls can remove this head entry from the list.
3385 n = list_first_entry(list, struct napi_struct, poll_list);
3387 have = netpoll_poll_lock(n);
3391 /* This NAPI_STATE_SCHED test is for avoiding a race
3392 * with netpoll's poll_napi(). Only the entity which
3393 * obtains the lock and sees NAPI_STATE_SCHED set will
3394 * actually make the ->poll() call. Therefore we avoid
3395 * accidently calling ->poll() when NAPI is not scheduled.
3398 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3399 work = n->poll(n, weight);
3403 WARN_ON_ONCE(work > weight);
3407 local_irq_disable();
3409 /* Drivers must not modify the NAPI state if they
3410 * consume the entire weight. In such cases this code
3411 * still "owns" the NAPI instance and therefore can
3412 * move the instance around on the list at-will.
3414 if (unlikely(work == weight)) {
3415 if (unlikely(napi_disable_pending(n))) {
3418 local_irq_disable();
3420 list_move_tail(&n->poll_list, list);
3423 netpoll_poll_unlock(have);
3428 #ifdef CONFIG_NET_DMA
3430 * There may not be any more sk_buffs coming right now, so push
3431 * any pending DMA copies to hardware
3433 dma_issue_pending_all();
3439 __get_cpu_var(netdev_rx_stat).time_squeeze++;
3440 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3444 static gifconf_func_t *gifconf_list[NPROTO];
3447 * register_gifconf - register a SIOCGIF handler
3448 * @family: Address family
3449 * @gifconf: Function handler
3451 * Register protocol dependent address dumping routines. The handler
3452 * that is passed must not be freed or reused until it has been replaced
3453 * by another handler.
3455 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3457 if (family >= NPROTO)
3459 gifconf_list[family] = gifconf;
3462 EXPORT_SYMBOL(register_gifconf);
3466 * Map an interface index to its name (SIOCGIFNAME)
3470 * We need this ioctl for efficient implementation of the
3471 * if_indextoname() function required by the IPv6 API. Without
3472 * it, we would have to search all the interfaces to find a
3476 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3478 struct net_device *dev;
3482 * Fetch the caller's info block.
3485 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3489 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3495 strcpy(ifr.ifr_name, dev->name);
3498 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3504 * Perform a SIOCGIFCONF call. This structure will change
3505 * size eventually, and there is nothing I can do about it.
3506 * Thus we will need a 'compatibility mode'.
3509 static int dev_ifconf(struct net *net, char __user *arg)
3512 struct net_device *dev;
3519 * Fetch the caller's info block.
3522 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3529 * Loop over the interfaces, and write an info block for each.
3533 for_each_netdev(net, dev) {
3534 for (i = 0; i < NPROTO; i++) {
3535 if (gifconf_list[i]) {
3538 done = gifconf_list[i](dev, NULL, 0);
3540 done = gifconf_list[i](dev, pos + total,
3550 * All done. Write the updated control block back to the caller.
3552 ifc.ifc_len = total;
3555 * Both BSD and Solaris return 0 here, so we do too.
3557 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3560 #ifdef CONFIG_PROC_FS
3562 * This is invoked by the /proc filesystem handler to display a device
3565 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3568 struct net *net = seq_file_net(seq);
3570 struct net_device *dev;
3574 return SEQ_START_TOKEN;
3577 for_each_netdev_rcu(net, dev)
3584 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3586 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3587 first_net_device(seq_file_net(seq)) :
3588 next_net_device((struct net_device *)v);
3591 return rcu_dereference(dev);
3594 void dev_seq_stop(struct seq_file *seq, void *v)
3600 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3602 const struct net_device_stats *stats = dev_get_stats(dev);
3604 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3605 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3606 dev->name, stats->rx_bytes, stats->rx_packets,
3608 stats->rx_dropped + stats->rx_missed_errors,
3609 stats->rx_fifo_errors,
3610 stats->rx_length_errors + stats->rx_over_errors +
3611 stats->rx_crc_errors + stats->rx_frame_errors,
3612 stats->rx_compressed, stats->multicast,
3613 stats->tx_bytes, stats->tx_packets,
3614 stats->tx_errors, stats->tx_dropped,
3615 stats->tx_fifo_errors, stats->collisions,
3616 stats->tx_carrier_errors +
3617 stats->tx_aborted_errors +
3618 stats->tx_window_errors +
3619 stats->tx_heartbeat_errors,
3620 stats->tx_compressed);
3624 * Called from the PROCfs module. This now uses the new arbitrary sized
3625 * /proc/net interface to create /proc/net/dev
3627 static int dev_seq_show(struct seq_file *seq, void *v)
3629 if (v == SEQ_START_TOKEN)
3630 seq_puts(seq, "Inter-| Receive "
3632 " face |bytes packets errs drop fifo frame "
3633 "compressed multicast|bytes packets errs "
3634 "drop fifo colls carrier compressed\n");
3636 dev_seq_printf_stats(seq, v);
3640 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3642 struct netif_rx_stats *rc = NULL;
3644 while (*pos < nr_cpu_ids)
3645 if (cpu_online(*pos)) {
3646 rc = &per_cpu(netdev_rx_stat, *pos);
3653 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3655 return softnet_get_online(pos);
3658 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3661 return softnet_get_online(pos);
3664 static void softnet_seq_stop(struct seq_file *seq, void *v)
3668 static int softnet_seq_show(struct seq_file *seq, void *v)
3670 struct netif_rx_stats *s = v;
3672 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3673 s->total, s->dropped, s->time_squeeze, 0,
3674 0, 0, 0, 0, /* was fastroute */
3675 s->cpu_collision, s->received_rps);
3679 static const struct seq_operations dev_seq_ops = {
3680 .start = dev_seq_start,
3681 .next = dev_seq_next,
3682 .stop = dev_seq_stop,
3683 .show = dev_seq_show,
3686 static int dev_seq_open(struct inode *inode, struct file *file)
3688 return seq_open_net(inode, file, &dev_seq_ops,
3689 sizeof(struct seq_net_private));
3692 static const struct file_operations dev_seq_fops = {
3693 .owner = THIS_MODULE,
3694 .open = dev_seq_open,
3696 .llseek = seq_lseek,
3697 .release = seq_release_net,
3700 static const struct seq_operations softnet_seq_ops = {
3701 .start = softnet_seq_start,
3702 .next = softnet_seq_next,
3703 .stop = softnet_seq_stop,
3704 .show = softnet_seq_show,
3707 static int softnet_seq_open(struct inode *inode, struct file *file)
3709 return seq_open(file, &softnet_seq_ops);
3712 static const struct file_operations softnet_seq_fops = {
3713 .owner = THIS_MODULE,
3714 .open = softnet_seq_open,
3716 .llseek = seq_lseek,
3717 .release = seq_release,
3720 static void *ptype_get_idx(loff_t pos)
3722 struct packet_type *pt = NULL;
3726 list_for_each_entry_rcu(pt, &ptype_all, list) {
3732 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3733 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3742 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3746 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3749 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3751 struct packet_type *pt;
3752 struct list_head *nxt;
3756 if (v == SEQ_START_TOKEN)
3757 return ptype_get_idx(0);
3760 nxt = pt->list.next;
3761 if (pt->type == htons(ETH_P_ALL)) {
3762 if (nxt != &ptype_all)
3765 nxt = ptype_base[0].next;
3767 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3769 while (nxt == &ptype_base[hash]) {
3770 if (++hash >= PTYPE_HASH_SIZE)
3772 nxt = ptype_base[hash].next;
3775 return list_entry(nxt, struct packet_type, list);
3778 static void ptype_seq_stop(struct seq_file *seq, void *v)
3784 static int ptype_seq_show(struct seq_file *seq, void *v)
3786 struct packet_type *pt = v;
3788 if (v == SEQ_START_TOKEN)
3789 seq_puts(seq, "Type Device Function\n");
3790 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3791 if (pt->type == htons(ETH_P_ALL))
3792 seq_puts(seq, "ALL ");
3794 seq_printf(seq, "%04x", ntohs(pt->type));
3796 seq_printf(seq, " %-8s %pF\n",
3797 pt->dev ? pt->dev->name : "", pt->func);
3803 static const struct seq_operations ptype_seq_ops = {
3804 .start = ptype_seq_start,
3805 .next = ptype_seq_next,
3806 .stop = ptype_seq_stop,
3807 .show = ptype_seq_show,
3810 static int ptype_seq_open(struct inode *inode, struct file *file)
3812 return seq_open_net(inode, file, &ptype_seq_ops,
3813 sizeof(struct seq_net_private));
3816 static const struct file_operations ptype_seq_fops = {
3817 .owner = THIS_MODULE,
3818 .open = ptype_seq_open,
3820 .llseek = seq_lseek,
3821 .release = seq_release_net,
3825 static int __net_init dev_proc_net_init(struct net *net)
3829 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3831 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3833 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3836 if (wext_proc_init(net))
3842 proc_net_remove(net, "ptype");
3844 proc_net_remove(net, "softnet_stat");
3846 proc_net_remove(net, "dev");
3850 static void __net_exit dev_proc_net_exit(struct net *net)
3852 wext_proc_exit(net);
3854 proc_net_remove(net, "ptype");
3855 proc_net_remove(net, "softnet_stat");
3856 proc_net_remove(net, "dev");
3859 static struct pernet_operations __net_initdata dev_proc_ops = {
3860 .init = dev_proc_net_init,
3861 .exit = dev_proc_net_exit,
3864 static int __init dev_proc_init(void)
3866 return register_pernet_subsys(&dev_proc_ops);
3869 #define dev_proc_init() 0
3870 #endif /* CONFIG_PROC_FS */
3874 * netdev_set_master - set up master/slave pair
3875 * @slave: slave device
3876 * @master: new master device
3878 * Changes the master device of the slave. Pass %NULL to break the
3879 * bonding. The caller must hold the RTNL semaphore. On a failure
3880 * a negative errno code is returned. On success the reference counts
3881 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3882 * function returns zero.
3884 int netdev_set_master(struct net_device *slave, struct net_device *master)
3886 struct net_device *old = slave->master;
3896 slave->master = master;
3903 slave->flags |= IFF_SLAVE;
3905 slave->flags &= ~IFF_SLAVE;
3907 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3910 EXPORT_SYMBOL(netdev_set_master);
3912 static void dev_change_rx_flags(struct net_device *dev, int flags)
3914 const struct net_device_ops *ops = dev->netdev_ops;
3916 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3917 ops->ndo_change_rx_flags(dev, flags);
3920 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3922 unsigned short old_flags = dev->flags;
3928 dev->flags |= IFF_PROMISC;
3929 dev->promiscuity += inc;
3930 if (dev->promiscuity == 0) {
3933 * If inc causes overflow, untouch promisc and return error.
3936 dev->flags &= ~IFF_PROMISC;
3938 dev->promiscuity -= inc;
3939 printk(KERN_WARNING "%s: promiscuity touches roof, "
3940 "set promiscuity failed, promiscuity feature "
3941 "of device might be broken.\n", dev->name);
3945 if (dev->flags != old_flags) {
3946 printk(KERN_INFO "device %s %s promiscuous mode\n",
3947 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3949 if (audit_enabled) {
3950 current_uid_gid(&uid, &gid);
3951 audit_log(current->audit_context, GFP_ATOMIC,
3952 AUDIT_ANOM_PROMISCUOUS,
3953 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3954 dev->name, (dev->flags & IFF_PROMISC),
3955 (old_flags & IFF_PROMISC),
3956 audit_get_loginuid(current),
3958 audit_get_sessionid(current));
3961 dev_change_rx_flags(dev, IFF_PROMISC);
3967 * dev_set_promiscuity - update promiscuity count on a device
3971 * Add or remove promiscuity from a device. While the count in the device
3972 * remains above zero the interface remains promiscuous. Once it hits zero
3973 * the device reverts back to normal filtering operation. A negative inc
3974 * value is used to drop promiscuity on the device.
3975 * Return 0 if successful or a negative errno code on error.
3977 int dev_set_promiscuity(struct net_device *dev, int inc)
3979 unsigned short old_flags = dev->flags;
3982 err = __dev_set_promiscuity(dev, inc);
3985 if (dev->flags != old_flags)
3986 dev_set_rx_mode(dev);
3989 EXPORT_SYMBOL(dev_set_promiscuity);
3992 * dev_set_allmulti - update allmulti count on a device
3996 * Add or remove reception of all multicast frames to a device. While the
3997 * count in the device remains above zero the interface remains listening
3998 * to all interfaces. Once it hits zero the device reverts back to normal
3999 * filtering operation. A negative @inc value is used to drop the counter
4000 * when releasing a resource needing all multicasts.
4001 * Return 0 if successful or a negative errno code on error.
4004 int dev_set_allmulti(struct net_device *dev, int inc)
4006 unsigned short old_flags = dev->flags;
4010 dev->flags |= IFF_ALLMULTI;
4011 dev->allmulti += inc;
4012 if (dev->allmulti == 0) {
4015 * If inc causes overflow, untouch allmulti and return error.
4018 dev->flags &= ~IFF_ALLMULTI;
4020 dev->allmulti -= inc;
4021 printk(KERN_WARNING "%s: allmulti touches roof, "
4022 "set allmulti failed, allmulti feature of "
4023 "device might be broken.\n", dev->name);
4027 if (dev->flags ^ old_flags) {
4028 dev_change_rx_flags(dev, IFF_ALLMULTI);
4029 dev_set_rx_mode(dev);
4033 EXPORT_SYMBOL(dev_set_allmulti);
4036 * Upload unicast and multicast address lists to device and
4037 * configure RX filtering. When the device doesn't support unicast
4038 * filtering it is put in promiscuous mode while unicast addresses
4041 void __dev_set_rx_mode(struct net_device *dev)
4043 const struct net_device_ops *ops = dev->netdev_ops;
4045 /* dev_open will call this function so the list will stay sane. */
4046 if (!(dev->flags&IFF_UP))
4049 if (!netif_device_present(dev))
4052 if (ops->ndo_set_rx_mode)
4053 ops->ndo_set_rx_mode(dev);
4055 /* Unicast addresses changes may only happen under the rtnl,
4056 * therefore calling __dev_set_promiscuity here is safe.
4058 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4059 __dev_set_promiscuity(dev, 1);
4060 dev->uc_promisc = 1;
4061 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4062 __dev_set_promiscuity(dev, -1);
4063 dev->uc_promisc = 0;
4066 if (ops->ndo_set_multicast_list)
4067 ops->ndo_set_multicast_list(dev);
4071 void dev_set_rx_mode(struct net_device *dev)
4073 netif_addr_lock_bh(dev);
4074 __dev_set_rx_mode(dev);
4075 netif_addr_unlock_bh(dev);
4079 * dev_get_flags - get flags reported to userspace
4082 * Get the combination of flag bits exported through APIs to userspace.
4084 unsigned dev_get_flags(const struct net_device *dev)
4088 flags = (dev->flags & ~(IFF_PROMISC |
4093 (dev->gflags & (IFF_PROMISC |
4096 if (netif_running(dev)) {
4097 if (netif_oper_up(dev))
4098 flags |= IFF_RUNNING;
4099 if (netif_carrier_ok(dev))
4100 flags |= IFF_LOWER_UP;
4101 if (netif_dormant(dev))
4102 flags |= IFF_DORMANT;
4107 EXPORT_SYMBOL(dev_get_flags);
4109 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4111 int old_flags = dev->flags;
4117 * Set the flags on our device.
4120 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4121 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4123 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4127 * Load in the correct multicast list now the flags have changed.
4130 if ((old_flags ^ flags) & IFF_MULTICAST)
4131 dev_change_rx_flags(dev, IFF_MULTICAST);
4133 dev_set_rx_mode(dev);
4136 * Have we downed the interface. We handle IFF_UP ourselves
4137 * according to user attempts to set it, rather than blindly
4142 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4143 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4146 dev_set_rx_mode(dev);
4149 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4150 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4152 dev->gflags ^= IFF_PROMISC;
4153 dev_set_promiscuity(dev, inc);
4156 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4157 is important. Some (broken) drivers set IFF_PROMISC, when
4158 IFF_ALLMULTI is requested not asking us and not reporting.
4160 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4161 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4163 dev->gflags ^= IFF_ALLMULTI;
4164 dev_set_allmulti(dev, inc);
4170 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4172 unsigned int changes = dev->flags ^ old_flags;
4174 if (changes & IFF_UP) {
4175 if (dev->flags & IFF_UP)
4176 call_netdevice_notifiers(NETDEV_UP, dev);
4178 call_netdevice_notifiers(NETDEV_DOWN, dev);
4181 if (dev->flags & IFF_UP &&
4182 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4183 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4187 * dev_change_flags - change device settings
4189 * @flags: device state flags
4191 * Change settings on device based state flags. The flags are
4192 * in the userspace exported format.
4194 int dev_change_flags(struct net_device *dev, unsigned flags)
4197 int old_flags = dev->flags;
4199 ret = __dev_change_flags(dev, flags);
4203 changes = old_flags ^ dev->flags;
4205 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4207 __dev_notify_flags(dev, old_flags);
4210 EXPORT_SYMBOL(dev_change_flags);
4213 * dev_set_mtu - Change maximum transfer unit
4215 * @new_mtu: new transfer unit
4217 * Change the maximum transfer size of the network device.
4219 int dev_set_mtu(struct net_device *dev, int new_mtu)
4221 const struct net_device_ops *ops = dev->netdev_ops;
4224 if (new_mtu == dev->mtu)
4227 /* MTU must be positive. */
4231 if (!netif_device_present(dev))
4235 if (ops->ndo_change_mtu)
4236 err = ops->ndo_change_mtu(dev, new_mtu);
4240 if (!err && dev->flags & IFF_UP)
4241 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4244 EXPORT_SYMBOL(dev_set_mtu);
4247 * dev_set_mac_address - Change Media Access Control Address
4251 * Change the hardware (MAC) address of the device
4253 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4255 const struct net_device_ops *ops = dev->netdev_ops;
4258 if (!ops->ndo_set_mac_address)
4260 if (sa->sa_family != dev->type)
4262 if (!netif_device_present(dev))
4264 err = ops->ndo_set_mac_address(dev, sa);
4266 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4269 EXPORT_SYMBOL(dev_set_mac_address);
4272 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4274 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4277 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4283 case SIOCGIFFLAGS: /* Get interface flags */
4284 ifr->ifr_flags = (short) dev_get_flags(dev);
4287 case SIOCGIFMETRIC: /* Get the metric on the interface
4288 (currently unused) */
4289 ifr->ifr_metric = 0;
4292 case SIOCGIFMTU: /* Get the MTU of a device */
4293 ifr->ifr_mtu = dev->mtu;
4298 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4300 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4301 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4302 ifr->ifr_hwaddr.sa_family = dev->type;
4310 ifr->ifr_map.mem_start = dev->mem_start;
4311 ifr->ifr_map.mem_end = dev->mem_end;
4312 ifr->ifr_map.base_addr = dev->base_addr;
4313 ifr->ifr_map.irq = dev->irq;
4314 ifr->ifr_map.dma = dev->dma;
4315 ifr->ifr_map.port = dev->if_port;
4319 ifr->ifr_ifindex = dev->ifindex;
4323 ifr->ifr_qlen = dev->tx_queue_len;
4327 /* dev_ioctl() should ensure this case
4339 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4341 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4344 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4345 const struct net_device_ops *ops;
4350 ops = dev->netdev_ops;
4353 case SIOCSIFFLAGS: /* Set interface flags */
4354 return dev_change_flags(dev, ifr->ifr_flags);
4356 case SIOCSIFMETRIC: /* Set the metric on the interface
4357 (currently unused) */
4360 case SIOCSIFMTU: /* Set the MTU of a device */
4361 return dev_set_mtu(dev, ifr->ifr_mtu);
4364 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4366 case SIOCSIFHWBROADCAST:
4367 if (ifr->ifr_hwaddr.sa_family != dev->type)
4369 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4370 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4371 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4375 if (ops->ndo_set_config) {
4376 if (!netif_device_present(dev))
4378 return ops->ndo_set_config(dev, &ifr->ifr_map);
4383 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4384 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4386 if (!netif_device_present(dev))
4388 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4391 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4392 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4394 if (!netif_device_present(dev))
4396 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4399 if (ifr->ifr_qlen < 0)
4401 dev->tx_queue_len = ifr->ifr_qlen;
4405 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4406 return dev_change_name(dev, ifr->ifr_newname);
4409 * Unknown or private ioctl
4412 if ((cmd >= SIOCDEVPRIVATE &&
4413 cmd <= SIOCDEVPRIVATE + 15) ||
4414 cmd == SIOCBONDENSLAVE ||
4415 cmd == SIOCBONDRELEASE ||
4416 cmd == SIOCBONDSETHWADDR ||
4417 cmd == SIOCBONDSLAVEINFOQUERY ||
4418 cmd == SIOCBONDINFOQUERY ||
4419 cmd == SIOCBONDCHANGEACTIVE ||
4420 cmd == SIOCGMIIPHY ||
4421 cmd == SIOCGMIIREG ||
4422 cmd == SIOCSMIIREG ||
4423 cmd == SIOCBRADDIF ||
4424 cmd == SIOCBRDELIF ||
4425 cmd == SIOCSHWTSTAMP ||
4426 cmd == SIOCWANDEV) {
4428 if (ops->ndo_do_ioctl) {
4429 if (netif_device_present(dev))
4430 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4442 * This function handles all "interface"-type I/O control requests. The actual
4443 * 'doing' part of this is dev_ifsioc above.
4447 * dev_ioctl - network device ioctl
4448 * @net: the applicable net namespace
4449 * @cmd: command to issue
4450 * @arg: pointer to a struct ifreq in user space
4452 * Issue ioctl functions to devices. This is normally called by the
4453 * user space syscall interfaces but can sometimes be useful for
4454 * other purposes. The return value is the return from the syscall if
4455 * positive or a negative errno code on error.
4458 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4464 /* One special case: SIOCGIFCONF takes ifconf argument
4465 and requires shared lock, because it sleeps writing
4469 if (cmd == SIOCGIFCONF) {
4471 ret = dev_ifconf(net, (char __user *) arg);
4475 if (cmd == SIOCGIFNAME)
4476 return dev_ifname(net, (struct ifreq __user *)arg);
4478 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4481 ifr.ifr_name[IFNAMSIZ-1] = 0;
4483 colon = strchr(ifr.ifr_name, ':');
4488 * See which interface the caller is talking about.
4493 * These ioctl calls:
4494 * - can be done by all.
4495 * - atomic and do not require locking.
4506 dev_load(net, ifr.ifr_name);
4508 ret = dev_ifsioc_locked(net, &ifr, cmd);
4513 if (copy_to_user(arg, &ifr,
4514 sizeof(struct ifreq)))
4520 dev_load(net, ifr.ifr_name);
4522 ret = dev_ethtool(net, &ifr);
4527 if (copy_to_user(arg, &ifr,
4528 sizeof(struct ifreq)))
4534 * These ioctl calls:
4535 * - require superuser power.
4536 * - require strict serialization.
4542 if (!capable(CAP_NET_ADMIN))
4544 dev_load(net, ifr.ifr_name);
4546 ret = dev_ifsioc(net, &ifr, cmd);
4551 if (copy_to_user(arg, &ifr,
4552 sizeof(struct ifreq)))
4558 * These ioctl calls:
4559 * - require superuser power.
4560 * - require strict serialization.
4561 * - do not return a value
4571 case SIOCSIFHWBROADCAST:
4574 case SIOCBONDENSLAVE:
4575 case SIOCBONDRELEASE:
4576 case SIOCBONDSETHWADDR:
4577 case SIOCBONDCHANGEACTIVE:
4581 if (!capable(CAP_NET_ADMIN))
4584 case SIOCBONDSLAVEINFOQUERY:
4585 case SIOCBONDINFOQUERY:
4586 dev_load(net, ifr.ifr_name);
4588 ret = dev_ifsioc(net, &ifr, cmd);
4593 /* Get the per device memory space. We can add this but
4594 * currently do not support it */
4596 /* Set the per device memory buffer space.
4597 * Not applicable in our case */
4602 * Unknown or private ioctl.
4605 if (cmd == SIOCWANDEV ||
4606 (cmd >= SIOCDEVPRIVATE &&
4607 cmd <= SIOCDEVPRIVATE + 15)) {
4608 dev_load(net, ifr.ifr_name);
4610 ret = dev_ifsioc(net, &ifr, cmd);
4612 if (!ret && copy_to_user(arg, &ifr,
4613 sizeof(struct ifreq)))
4617 /* Take care of Wireless Extensions */
4618 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4619 return wext_handle_ioctl(net, &ifr, cmd, arg);
4626 * dev_new_index - allocate an ifindex
4627 * @net: the applicable net namespace
4629 * Returns a suitable unique value for a new device interface
4630 * number. The caller must hold the rtnl semaphore or the
4631 * dev_base_lock to be sure it remains unique.
4633 static int dev_new_index(struct net *net)
4639 if (!__dev_get_by_index(net, ifindex))
4644 /* Delayed registration/unregisteration */
4645 static LIST_HEAD(net_todo_list);
4647 static void net_set_todo(struct net_device *dev)
4649 list_add_tail(&dev->todo_list, &net_todo_list);
4652 static void rollback_registered_many(struct list_head *head)
4654 struct net_device *dev, *tmp;
4656 BUG_ON(dev_boot_phase);
4659 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4660 /* Some devices call without registering
4661 * for initialization unwind. Remove those
4662 * devices and proceed with the remaining.
4664 if (dev->reg_state == NETREG_UNINITIALIZED) {
4665 pr_debug("unregister_netdevice: device %s/%p never "
4666 "was registered\n", dev->name, dev);
4669 list_del(&dev->unreg_list);
4673 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4675 /* If device is running, close it first. */
4678 /* And unlink it from device chain. */
4679 unlist_netdevice(dev);
4681 dev->reg_state = NETREG_UNREGISTERING;
4686 list_for_each_entry(dev, head, unreg_list) {
4687 /* Shutdown queueing discipline. */
4691 /* Notify protocols, that we are about to destroy
4692 this device. They should clean all the things.
4694 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4696 if (!dev->rtnl_link_ops ||
4697 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4698 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4701 * Flush the unicast and multicast chains
4706 if (dev->netdev_ops->ndo_uninit)
4707 dev->netdev_ops->ndo_uninit(dev);
4709 /* Notifier chain MUST detach us from master device. */
4710 WARN_ON(dev->master);
4712 /* Remove entries from kobject tree */
4713 netdev_unregister_kobject(dev);
4716 /* Process any work delayed until the end of the batch */
4717 dev = list_first_entry(head, struct net_device, unreg_list);
4718 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4722 list_for_each_entry(dev, head, unreg_list)
4726 static void rollback_registered(struct net_device *dev)
4730 list_add(&dev->unreg_list, &single);
4731 rollback_registered_many(&single);
4734 static void __netdev_init_queue_locks_one(struct net_device *dev,
4735 struct netdev_queue *dev_queue,
4738 spin_lock_init(&dev_queue->_xmit_lock);
4739 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4740 dev_queue->xmit_lock_owner = -1;
4743 static void netdev_init_queue_locks(struct net_device *dev)
4745 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4746 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4749 unsigned long netdev_fix_features(unsigned long features, const char *name)
4751 /* Fix illegal SG+CSUM combinations. */
4752 if ((features & NETIF_F_SG) &&
4753 !(features & NETIF_F_ALL_CSUM)) {
4755 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4756 "checksum feature.\n", name);
4757 features &= ~NETIF_F_SG;
4760 /* TSO requires that SG is present as well. */
4761 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4763 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4764 "SG feature.\n", name);
4765 features &= ~NETIF_F_TSO;
4768 if (features & NETIF_F_UFO) {
4769 if (!(features & NETIF_F_GEN_CSUM)) {
4771 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4772 "since no NETIF_F_HW_CSUM feature.\n",
4774 features &= ~NETIF_F_UFO;
4777 if (!(features & NETIF_F_SG)) {
4779 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4780 "since no NETIF_F_SG feature.\n", name);
4781 features &= ~NETIF_F_UFO;
4787 EXPORT_SYMBOL(netdev_fix_features);
4790 * netif_stacked_transfer_operstate - transfer operstate
4791 * @rootdev: the root or lower level device to transfer state from
4792 * @dev: the device to transfer operstate to
4794 * Transfer operational state from root to device. This is normally
4795 * called when a stacking relationship exists between the root
4796 * device and the device(a leaf device).
4798 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4799 struct net_device *dev)
4801 if (rootdev->operstate == IF_OPER_DORMANT)
4802 netif_dormant_on(dev);
4804 netif_dormant_off(dev);
4806 if (netif_carrier_ok(rootdev)) {
4807 if (!netif_carrier_ok(dev))
4808 netif_carrier_on(dev);
4810 if (netif_carrier_ok(dev))
4811 netif_carrier_off(dev);
4814 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4817 * register_netdevice - register a network device
4818 * @dev: device to register
4820 * Take a completed network device structure and add it to the kernel
4821 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4822 * chain. 0 is returned on success. A negative errno code is returned
4823 * on a failure to set up the device, or if the name is a duplicate.
4825 * Callers must hold the rtnl semaphore. You may want
4826 * register_netdev() instead of this.
4829 * The locking appears insufficient to guarantee two parallel registers
4830 * will not get the same name.
4833 int register_netdevice(struct net_device *dev)
4836 struct net *net = dev_net(dev);
4838 BUG_ON(dev_boot_phase);
4843 /* When net_device's are persistent, this will be fatal. */
4844 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4847 spin_lock_init(&dev->addr_list_lock);
4848 netdev_set_addr_lockdep_class(dev);
4849 netdev_init_queue_locks(dev);
4854 if (!dev->num_rx_queues) {
4856 * Allocate a single RX queue if driver never called
4860 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
4866 dev->_rx->first = dev->_rx;
4867 atomic_set(&dev->_rx->count, 1);
4868 dev->num_rx_queues = 1;
4871 /* Init, if this function is available */
4872 if (dev->netdev_ops->ndo_init) {
4873 ret = dev->netdev_ops->ndo_init(dev);
4881 ret = dev_get_valid_name(net, dev->name, dev->name, 0);
4885 dev->ifindex = dev_new_index(net);
4886 if (dev->iflink == -1)
4887 dev->iflink = dev->ifindex;
4889 /* Fix illegal checksum combinations */
4890 if ((dev->features & NETIF_F_HW_CSUM) &&
4891 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4892 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4894 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4897 if ((dev->features & NETIF_F_NO_CSUM) &&
4898 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4899 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4901 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4904 dev->features = netdev_fix_features(dev->features, dev->name);
4906 /* Enable software GSO if SG is supported. */
4907 if (dev->features & NETIF_F_SG)
4908 dev->features |= NETIF_F_GSO;
4910 netdev_initialize_kobject(dev);
4912 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4913 ret = notifier_to_errno(ret);
4917 ret = netdev_register_kobject(dev);
4920 dev->reg_state = NETREG_REGISTERED;
4923 * Default initial state at registry is that the
4924 * device is present.
4927 set_bit(__LINK_STATE_PRESENT, &dev->state);
4929 dev_init_scheduler(dev);
4931 list_netdevice(dev);
4933 /* Notify protocols, that a new device appeared. */
4934 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4935 ret = notifier_to_errno(ret);
4937 rollback_registered(dev);
4938 dev->reg_state = NETREG_UNREGISTERED;
4941 * Prevent userspace races by waiting until the network
4942 * device is fully setup before sending notifications.
4944 if (!dev->rtnl_link_ops ||
4945 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4946 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
4952 if (dev->netdev_ops->ndo_uninit)
4953 dev->netdev_ops->ndo_uninit(dev);
4956 EXPORT_SYMBOL(register_netdevice);
4959 * init_dummy_netdev - init a dummy network device for NAPI
4960 * @dev: device to init
4962 * This takes a network device structure and initialize the minimum
4963 * amount of fields so it can be used to schedule NAPI polls without
4964 * registering a full blown interface. This is to be used by drivers
4965 * that need to tie several hardware interfaces to a single NAPI
4966 * poll scheduler due to HW limitations.
4968 int init_dummy_netdev(struct net_device *dev)
4970 /* Clear everything. Note we don't initialize spinlocks
4971 * are they aren't supposed to be taken by any of the
4972 * NAPI code and this dummy netdev is supposed to be
4973 * only ever used for NAPI polls
4975 memset(dev, 0, sizeof(struct net_device));
4977 /* make sure we BUG if trying to hit standard
4978 * register/unregister code path
4980 dev->reg_state = NETREG_DUMMY;
4982 /* initialize the ref count */
4983 atomic_set(&dev->refcnt, 1);
4985 /* NAPI wants this */
4986 INIT_LIST_HEAD(&dev->napi_list);
4988 /* a dummy interface is started by default */
4989 set_bit(__LINK_STATE_PRESENT, &dev->state);
4990 set_bit(__LINK_STATE_START, &dev->state);
4994 EXPORT_SYMBOL_GPL(init_dummy_netdev);
4998 * register_netdev - register a network device
4999 * @dev: device to register
5001 * Take a completed network device structure and add it to the kernel
5002 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5003 * chain. 0 is returned on success. A negative errno code is returned
5004 * on a failure to set up the device, or if the name is a duplicate.
5006 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5007 * and expands the device name if you passed a format string to
5010 int register_netdev(struct net_device *dev)
5017 * If the name is a format string the caller wants us to do a
5020 if (strchr(dev->name, '%')) {
5021 err = dev_alloc_name(dev, dev->name);
5026 err = register_netdevice(dev);
5031 EXPORT_SYMBOL(register_netdev);
5034 * netdev_wait_allrefs - wait until all references are gone.
5036 * This is called when unregistering network devices.
5038 * Any protocol or device that holds a reference should register
5039 * for netdevice notification, and cleanup and put back the
5040 * reference if they receive an UNREGISTER event.
5041 * We can get stuck here if buggy protocols don't correctly
5044 static void netdev_wait_allrefs(struct net_device *dev)
5046 unsigned long rebroadcast_time, warning_time;
5048 linkwatch_forget_dev(dev);
5050 rebroadcast_time = warning_time = jiffies;
5051 while (atomic_read(&dev->refcnt) != 0) {
5052 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5055 /* Rebroadcast unregister notification */
5056 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5057 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5058 * should have already handle it the first time */
5060 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5062 /* We must not have linkwatch events
5063 * pending on unregister. If this
5064 * happens, we simply run the queue
5065 * unscheduled, resulting in a noop
5068 linkwatch_run_queue();
5073 rebroadcast_time = jiffies;
5078 if (time_after(jiffies, warning_time + 10 * HZ)) {
5079 printk(KERN_EMERG "unregister_netdevice: "
5080 "waiting for %s to become free. Usage "
5082 dev->name, atomic_read(&dev->refcnt));
5083 warning_time = jiffies;
5092 * register_netdevice(x1);
5093 * register_netdevice(x2);
5095 * unregister_netdevice(y1);
5096 * unregister_netdevice(y2);
5102 * We are invoked by rtnl_unlock().
5103 * This allows us to deal with problems:
5104 * 1) We can delete sysfs objects which invoke hotplug
5105 * without deadlocking with linkwatch via keventd.
5106 * 2) Since we run with the RTNL semaphore not held, we can sleep
5107 * safely in order to wait for the netdev refcnt to drop to zero.
5109 * We must not return until all unregister events added during
5110 * the interval the lock was held have been completed.
5112 void netdev_run_todo(void)
5114 struct list_head list;
5116 /* Snapshot list, allow later requests */
5117 list_replace_init(&net_todo_list, &list);
5121 while (!list_empty(&list)) {
5122 struct net_device *dev
5123 = list_first_entry(&list, struct net_device, todo_list);
5124 list_del(&dev->todo_list);
5126 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5127 printk(KERN_ERR "network todo '%s' but state %d\n",
5128 dev->name, dev->reg_state);
5133 dev->reg_state = NETREG_UNREGISTERED;
5135 on_each_cpu(flush_backlog, dev, 1);
5137 netdev_wait_allrefs(dev);
5140 BUG_ON(atomic_read(&dev->refcnt));
5141 WARN_ON(dev->ip_ptr);
5142 WARN_ON(dev->ip6_ptr);
5143 WARN_ON(dev->dn_ptr);
5145 if (dev->destructor)
5146 dev->destructor(dev);
5148 /* Free network device */
5149 kobject_put(&dev->dev.kobj);
5154 * dev_txq_stats_fold - fold tx_queues stats
5155 * @dev: device to get statistics from
5156 * @stats: struct net_device_stats to hold results
5158 void dev_txq_stats_fold(const struct net_device *dev,
5159 struct net_device_stats *stats)
5161 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5163 struct netdev_queue *txq;
5165 for (i = 0; i < dev->num_tx_queues; i++) {
5166 txq = netdev_get_tx_queue(dev, i);
5167 tx_bytes += txq->tx_bytes;
5168 tx_packets += txq->tx_packets;
5169 tx_dropped += txq->tx_dropped;
5171 if (tx_bytes || tx_packets || tx_dropped) {
5172 stats->tx_bytes = tx_bytes;
5173 stats->tx_packets = tx_packets;
5174 stats->tx_dropped = tx_dropped;
5177 EXPORT_SYMBOL(dev_txq_stats_fold);
5180 * dev_get_stats - get network device statistics
5181 * @dev: device to get statistics from
5183 * Get network statistics from device. The device driver may provide
5184 * its own method by setting dev->netdev_ops->get_stats; otherwise
5185 * the internal statistics structure is used.
5187 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5189 const struct net_device_ops *ops = dev->netdev_ops;
5191 if (ops->ndo_get_stats)
5192 return ops->ndo_get_stats(dev);
5194 dev_txq_stats_fold(dev, &dev->stats);
5197 EXPORT_SYMBOL(dev_get_stats);
5199 static void netdev_init_one_queue(struct net_device *dev,
5200 struct netdev_queue *queue,
5206 static void netdev_init_queues(struct net_device *dev)
5208 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5209 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5210 spin_lock_init(&dev->tx_global_lock);
5214 * alloc_netdev_mq - allocate network device
5215 * @sizeof_priv: size of private data to allocate space for
5216 * @name: device name format string
5217 * @setup: callback to initialize device
5218 * @queue_count: the number of subqueues to allocate
5220 * Allocates a struct net_device with private data area for driver use
5221 * and performs basic initialization. Also allocates subquue structs
5222 * for each queue on the device at the end of the netdevice.
5224 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5225 void (*setup)(struct net_device *), unsigned int queue_count)
5227 struct netdev_queue *tx;
5228 struct net_device *dev;
5230 struct net_device *p;
5232 struct netdev_rx_queue *rx;
5236 BUG_ON(strlen(name) >= sizeof(dev->name));
5238 alloc_size = sizeof(struct net_device);
5240 /* ensure 32-byte alignment of private area */
5241 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5242 alloc_size += sizeof_priv;
5244 /* ensure 32-byte alignment of whole construct */
5245 alloc_size += NETDEV_ALIGN - 1;
5247 p = kzalloc(alloc_size, GFP_KERNEL);
5249 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5253 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5255 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5261 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5263 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5268 atomic_set(&rx->count, queue_count);
5271 * Set a pointer to first element in the array which holds the
5274 for (i = 0; i < queue_count; i++)
5278 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5279 dev->padded = (char *)dev - (char *)p;
5281 if (dev_addr_init(dev))
5287 dev_net_set(dev, &init_net);
5290 dev->num_tx_queues = queue_count;
5291 dev->real_num_tx_queues = queue_count;
5295 dev->num_rx_queues = queue_count;
5298 dev->gso_max_size = GSO_MAX_SIZE;
5300 netdev_init_queues(dev);
5302 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5303 dev->ethtool_ntuple_list.count = 0;
5304 INIT_LIST_HEAD(&dev->napi_list);
5305 INIT_LIST_HEAD(&dev->unreg_list);
5306 INIT_LIST_HEAD(&dev->link_watch_list);
5307 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5309 strcpy(dev->name, name);
5322 EXPORT_SYMBOL(alloc_netdev_mq);
5325 * free_netdev - free network device
5328 * This function does the last stage of destroying an allocated device
5329 * interface. The reference to the device object is released.
5330 * If this is the last reference then it will be freed.
5332 void free_netdev(struct net_device *dev)
5334 struct napi_struct *p, *n;
5336 release_net(dev_net(dev));
5340 /* Flush device addresses */
5341 dev_addr_flush(dev);
5343 /* Clear ethtool n-tuple list */
5344 ethtool_ntuple_flush(dev);
5346 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5349 /* Compatibility with error handling in drivers */
5350 if (dev->reg_state == NETREG_UNINITIALIZED) {
5351 kfree((char *)dev - dev->padded);
5355 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5356 dev->reg_state = NETREG_RELEASED;
5358 /* will free via device release */
5359 put_device(&dev->dev);
5361 EXPORT_SYMBOL(free_netdev);
5364 * synchronize_net - Synchronize with packet receive processing
5366 * Wait for packets currently being received to be done.
5367 * Does not block later packets from starting.
5369 void synchronize_net(void)
5374 EXPORT_SYMBOL(synchronize_net);
5377 * unregister_netdevice_queue - remove device from the kernel
5381 * This function shuts down a device interface and removes it
5382 * from the kernel tables.
5383 * If head not NULL, device is queued to be unregistered later.
5385 * Callers must hold the rtnl semaphore. You may want
5386 * unregister_netdev() instead of this.
5389 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5394 list_move_tail(&dev->unreg_list, head);
5396 rollback_registered(dev);
5397 /* Finish processing unregister after unlock */
5401 EXPORT_SYMBOL(unregister_netdevice_queue);
5404 * unregister_netdevice_many - unregister many devices
5405 * @head: list of devices
5407 void unregister_netdevice_many(struct list_head *head)
5409 struct net_device *dev;
5411 if (!list_empty(head)) {
5412 rollback_registered_many(head);
5413 list_for_each_entry(dev, head, unreg_list)
5417 EXPORT_SYMBOL(unregister_netdevice_many);
5420 * unregister_netdev - remove device from the kernel
5423 * This function shuts down a device interface and removes it
5424 * from the kernel tables.
5426 * This is just a wrapper for unregister_netdevice that takes
5427 * the rtnl semaphore. In general you want to use this and not
5428 * unregister_netdevice.
5430 void unregister_netdev(struct net_device *dev)
5433 unregister_netdevice(dev);
5436 EXPORT_SYMBOL(unregister_netdev);
5439 * dev_change_net_namespace - move device to different nethost namespace
5441 * @net: network namespace
5442 * @pat: If not NULL name pattern to try if the current device name
5443 * is already taken in the destination network namespace.
5445 * This function shuts down a device interface and moves it
5446 * to a new network namespace. On success 0 is returned, on
5447 * a failure a netagive errno code is returned.
5449 * Callers must hold the rtnl semaphore.
5452 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5458 /* Don't allow namespace local devices to be moved. */
5460 if (dev->features & NETIF_F_NETNS_LOCAL)
5464 /* Don't allow real devices to be moved when sysfs
5468 if (dev->dev.parent)
5472 /* Ensure the device has been registrered */
5474 if (dev->reg_state != NETREG_REGISTERED)
5477 /* Get out if there is nothing todo */
5479 if (net_eq(dev_net(dev), net))
5482 /* Pick the destination device name, and ensure
5483 * we can use it in the destination network namespace.
5486 if (__dev_get_by_name(net, dev->name)) {
5487 /* We get here if we can't use the current device name */
5490 if (dev_get_valid_name(net, pat, dev->name, 1))
5495 * And now a mini version of register_netdevice unregister_netdevice.
5498 /* If device is running close it first. */
5501 /* And unlink it from device chain */
5503 unlist_netdevice(dev);
5507 /* Shutdown queueing discipline. */
5510 /* Notify protocols, that we are about to destroy
5511 this device. They should clean all the things.
5513 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5514 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5517 * Flush the unicast and multicast chains
5522 netdev_unregister_kobject(dev);
5524 /* Actually switch the network namespace */
5525 dev_net_set(dev, net);
5527 /* If there is an ifindex conflict assign a new one */
5528 if (__dev_get_by_index(net, dev->ifindex)) {
5529 int iflink = (dev->iflink == dev->ifindex);
5530 dev->ifindex = dev_new_index(net);
5532 dev->iflink = dev->ifindex;
5535 /* Fixup kobjects */
5536 err = netdev_register_kobject(dev);
5539 /* Add the device back in the hashes */
5540 list_netdevice(dev);
5542 /* Notify protocols, that a new device appeared. */
5543 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5546 * Prevent userspace races by waiting until the network
5547 * device is fully setup before sending notifications.
5549 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5556 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5558 static int dev_cpu_callback(struct notifier_block *nfb,
5559 unsigned long action,
5562 struct sk_buff **list_skb;
5563 struct Qdisc **list_net;
5564 struct sk_buff *skb;
5565 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5566 struct softnet_data *sd, *oldsd;
5568 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5571 local_irq_disable();
5572 cpu = smp_processor_id();
5573 sd = &per_cpu(softnet_data, cpu);
5574 oldsd = &per_cpu(softnet_data, oldcpu);
5576 /* Find end of our completion_queue. */
5577 list_skb = &sd->completion_queue;
5579 list_skb = &(*list_skb)->next;
5580 /* Append completion queue from offline CPU. */
5581 *list_skb = oldsd->completion_queue;
5582 oldsd->completion_queue = NULL;
5584 /* Find end of our output_queue. */
5585 list_net = &sd->output_queue;
5587 list_net = &(*list_net)->next_sched;
5588 /* Append output queue from offline CPU. */
5589 *list_net = oldsd->output_queue;
5590 oldsd->output_queue = NULL;
5592 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5595 /* Process offline CPU's input_pkt_queue */
5596 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5598 incr_input_queue_head(oldsd);
5606 * netdev_increment_features - increment feature set by one
5607 * @all: current feature set
5608 * @one: new feature set
5609 * @mask: mask feature set
5611 * Computes a new feature set after adding a device with feature set
5612 * @one to the master device with current feature set @all. Will not
5613 * enable anything that is off in @mask. Returns the new feature set.
5615 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5618 /* If device needs checksumming, downgrade to it. */
5619 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5620 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5621 else if (mask & NETIF_F_ALL_CSUM) {
5622 /* If one device supports v4/v6 checksumming, set for all. */
5623 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5624 !(all & NETIF_F_GEN_CSUM)) {
5625 all &= ~NETIF_F_ALL_CSUM;
5626 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5629 /* If one device supports hw checksumming, set for all. */
5630 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5631 all &= ~NETIF_F_ALL_CSUM;
5632 all |= NETIF_F_HW_CSUM;
5636 one |= NETIF_F_ALL_CSUM;
5638 one |= all & NETIF_F_ONE_FOR_ALL;
5639 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5640 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5644 EXPORT_SYMBOL(netdev_increment_features);
5646 static struct hlist_head *netdev_create_hash(void)
5649 struct hlist_head *hash;
5651 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5653 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5654 INIT_HLIST_HEAD(&hash[i]);
5659 /* Initialize per network namespace state */
5660 static int __net_init netdev_init(struct net *net)
5662 INIT_LIST_HEAD(&net->dev_base_head);
5664 net->dev_name_head = netdev_create_hash();
5665 if (net->dev_name_head == NULL)
5668 net->dev_index_head = netdev_create_hash();
5669 if (net->dev_index_head == NULL)
5675 kfree(net->dev_name_head);
5681 * netdev_drivername - network driver for the device
5682 * @dev: network device
5683 * @buffer: buffer for resulting name
5684 * @len: size of buffer
5686 * Determine network driver for device.
5688 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5690 const struct device_driver *driver;
5691 const struct device *parent;
5693 if (len <= 0 || !buffer)
5697 parent = dev->dev.parent;
5702 driver = parent->driver;
5703 if (driver && driver->name)
5704 strlcpy(buffer, driver->name, len);
5708 static void __net_exit netdev_exit(struct net *net)
5710 kfree(net->dev_name_head);
5711 kfree(net->dev_index_head);
5714 static struct pernet_operations __net_initdata netdev_net_ops = {
5715 .init = netdev_init,
5716 .exit = netdev_exit,
5719 static void __net_exit default_device_exit(struct net *net)
5721 struct net_device *dev, *aux;
5723 * Push all migratable network devices back to the
5724 * initial network namespace
5727 for_each_netdev_safe(net, dev, aux) {
5729 char fb_name[IFNAMSIZ];
5731 /* Ignore unmoveable devices (i.e. loopback) */
5732 if (dev->features & NETIF_F_NETNS_LOCAL)
5735 /* Leave virtual devices for the generic cleanup */
5736 if (dev->rtnl_link_ops)
5739 /* Push remaing network devices to init_net */
5740 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5741 err = dev_change_net_namespace(dev, &init_net, fb_name);
5743 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5744 __func__, dev->name, err);
5751 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5753 /* At exit all network devices most be removed from a network
5754 * namespace. Do this in the reverse order of registeration.
5755 * Do this across as many network namespaces as possible to
5756 * improve batching efficiency.
5758 struct net_device *dev;
5760 LIST_HEAD(dev_kill_list);
5763 list_for_each_entry(net, net_list, exit_list) {
5764 for_each_netdev_reverse(net, dev) {
5765 if (dev->rtnl_link_ops)
5766 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5768 unregister_netdevice_queue(dev, &dev_kill_list);
5771 unregister_netdevice_many(&dev_kill_list);
5775 static struct pernet_operations __net_initdata default_device_ops = {
5776 .exit = default_device_exit,
5777 .exit_batch = default_device_exit_batch,
5781 * Initialize the DEV module. At boot time this walks the device list and
5782 * unhooks any devices that fail to initialise (normally hardware not
5783 * present) and leaves us with a valid list of present and active devices.
5788 * This is called single threaded during boot, so no need
5789 * to take the rtnl semaphore.
5791 static int __init net_dev_init(void)
5793 int i, rc = -ENOMEM;
5795 BUG_ON(!dev_boot_phase);
5797 if (dev_proc_init())
5800 if (netdev_kobject_init())
5803 INIT_LIST_HEAD(&ptype_all);
5804 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5805 INIT_LIST_HEAD(&ptype_base[i]);
5807 if (register_pernet_subsys(&netdev_net_ops))
5811 * Initialise the packet receive queues.
5814 for_each_possible_cpu(i) {
5815 struct softnet_data *queue;
5817 queue = &per_cpu(softnet_data, i);
5818 skb_queue_head_init(&queue->input_pkt_queue);
5819 queue->completion_queue = NULL;
5820 INIT_LIST_HEAD(&queue->poll_list);
5823 queue->csd.func = trigger_softirq;
5824 queue->csd.info = queue;
5825 queue->csd.flags = 0;
5829 queue->backlog.poll = process_backlog;
5830 queue->backlog.weight = weight_p;
5831 queue->backlog.gro_list = NULL;
5832 queue->backlog.gro_count = 0;
5837 /* The loopback device is special if any other network devices
5838 * is present in a network namespace the loopback device must
5839 * be present. Since we now dynamically allocate and free the
5840 * loopback device ensure this invariant is maintained by
5841 * keeping the loopback device as the first device on the
5842 * list of network devices. Ensuring the loopback devices
5843 * is the first device that appears and the last network device
5846 if (register_pernet_device(&loopback_net_ops))
5849 if (register_pernet_device(&default_device_ops))
5852 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5853 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5855 hotcpu_notifier(dev_cpu_callback, 0);
5863 subsys_initcall(net_dev_init);
5865 static int __init initialize_hashrnd(void)
5867 get_random_bytes(&hashrnd, sizeof(hashrnd));
5871 late_initcall_sync(initialize_hashrnd);