2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
104 #include <linux/if_bridge.h>
105 #include <linux/if_macvlan.h>
107 #include <net/pkt_sched.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/kmod.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/wext.h>
118 #include <net/iw_handler.h>
119 #include <asm/current.h>
120 #include <linux/audit.h>
121 #include <linux/dmaengine.h>
122 #include <linux/err.h>
123 #include <linux/ctype.h>
124 #include <linux/if_arp.h>
125 #include <linux/if_vlan.h>
126 #include <linux/ip.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
133 #include <linux/pci.h>
135 #include "net-sysfs.h"
137 /* Instead of increasing this, you should create a hash table. */
138 #define MAX_GRO_SKBS 8
140 /* This should be increased if a protocol with a bigger head is added. */
141 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 * The list of packet types we will receive (as opposed to discard)
145 * and the routines to invoke.
147 * Why 16. Because with 16 the only overlap we get on a hash of the
148 * low nibble of the protocol value is RARP/SNAP/X.25.
150 * NOTE: That is no longer true with the addition of VLAN tags. Not
151 * sure which should go first, but I bet it won't make much
152 * difference if we are running VLANs. The good news is that
153 * this protocol won't be in the list unless compiled in, so
154 * the average user (w/out VLANs) will not be adversely affected.
171 #define PTYPE_HASH_SIZE (16)
172 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
174 static DEFINE_SPINLOCK(ptype_lock);
175 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
176 static struct list_head ptype_all __read_mostly; /* Taps */
179 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
182 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
184 * Writers must hold the rtnl semaphore while they loop through the
185 * dev_base_head list, and hold dev_base_lock for writing when they do the
186 * actual updates. This allows pure readers to access the list even
187 * while a writer is preparing to update it.
189 * To put it another way, dev_base_lock is held for writing only to
190 * protect against pure readers; the rtnl semaphore provides the
191 * protection against other writers.
193 * See, for example usages, register_netdevice() and
194 * unregister_netdevice(), which must be called with the rtnl
197 DEFINE_RWLOCK(dev_base_lock);
198 EXPORT_SYMBOL(dev_base_lock);
200 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
202 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
203 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
206 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
208 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
211 static inline void rps_lock(struct softnet_data *sd)
214 spin_lock(&sd->input_pkt_queue.lock);
218 static inline void rps_unlock(struct softnet_data *sd)
221 spin_unlock(&sd->input_pkt_queue.lock);
225 /* Device list insertion */
226 static int list_netdevice(struct net_device *dev)
228 struct net *net = dev_net(dev);
232 write_lock_bh(&dev_base_lock);
233 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
234 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
235 hlist_add_head_rcu(&dev->index_hlist,
236 dev_index_hash(net, dev->ifindex));
237 write_unlock_bh(&dev_base_lock);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device *dev)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock);
250 list_del_rcu(&dev->dev_list);
251 hlist_del_rcu(&dev->name_hlist);
252 hlist_del_rcu(&dev->index_hlist);
253 write_unlock_bh(&dev_base_lock);
260 static RAW_NOTIFIER_HEAD(netdev_chain);
263 * Device drivers call our routines to queue packets here. We empty the
264 * queue in the local softnet handler.
267 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
268 EXPORT_PER_CPU_SYMBOL(softnet_data);
270 #ifdef CONFIG_LOCKDEP
272 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
273 * according to dev->type
275 static const unsigned short netdev_lock_type[] =
276 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
277 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
278 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
279 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
280 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
281 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
282 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
283 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
284 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
285 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
286 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
287 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
288 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
289 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
290 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
291 ARPHRD_VOID, ARPHRD_NONE};
293 static const char *const netdev_lock_name[] =
294 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
295 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
296 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
297 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
298 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
299 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
300 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
301 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
302 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
303 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
304 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
305 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
306 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
307 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
308 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
309 "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
318 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
319 if (netdev_lock_type[i] == dev_type)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
326 unsigned short dev_type)
330 i = netdev_lock_pos(dev_type);
331 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
332 netdev_lock_name[i]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
339 i = netdev_lock_pos(dev->type);
340 lockdep_set_class_and_name(&dev->addr_list_lock,
341 &netdev_addr_lock_key[i],
342 netdev_lock_name[i]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
346 unsigned short dev_type)
349 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
377 * dev_add_pack - add packet handler
378 * @pt: packet type declaration
380 * Add a protocol handler to the networking stack. The passed &packet_type
381 * is linked into kernel lists and may not be freed until it has been
382 * removed from the kernel lists.
384 * This call does not sleep therefore it can not
385 * guarantee all CPU's that are in middle of receiving packets
386 * will see the new packet type (until the next received packet).
389 void dev_add_pack(struct packet_type *pt)
393 spin_lock_bh(&ptype_lock);
394 if (pt->type == htons(ETH_P_ALL))
395 list_add_rcu(&pt->list, &ptype_all);
397 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
398 list_add_rcu(&pt->list, &ptype_base[hash]);
400 spin_unlock_bh(&ptype_lock);
402 EXPORT_SYMBOL(dev_add_pack);
405 * __dev_remove_pack - remove packet handler
406 * @pt: packet type declaration
408 * Remove a protocol handler that was previously added to the kernel
409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
410 * from the kernel lists and can be freed or reused once this function
413 * The packet type might still be in use by receivers
414 * and must not be freed until after all the CPU's have gone
415 * through a quiescent state.
417 void __dev_remove_pack(struct packet_type *pt)
419 struct list_head *head;
420 struct packet_type *pt1;
422 spin_lock_bh(&ptype_lock);
424 if (pt->type == htons(ETH_P_ALL))
427 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
429 list_for_each_entry(pt1, head, list) {
431 list_del_rcu(&pt->list);
436 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
438 spin_unlock_bh(&ptype_lock);
440 EXPORT_SYMBOL(__dev_remove_pack);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type *pt)
456 __dev_remove_pack(pt);
460 EXPORT_SYMBOL(dev_remove_pack);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name, struct ifmap *map)
482 struct netdev_boot_setup *s;
486 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
487 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
488 memset(s[i].name, 0, sizeof(s[i].name));
489 strlcpy(s[i].name, name, IFNAMSIZ);
490 memcpy(&s[i].map, map, sizeof(s[i].map));
495 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device *dev)
509 struct netdev_boot_setup *s = dev_boot_setup;
512 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
513 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
514 !strcmp(dev->name, s[i].name)) {
515 dev->irq = s[i].map.irq;
516 dev->base_addr = s[i].map.base_addr;
517 dev->mem_start = s[i].map.mem_start;
518 dev->mem_end = s[i].map.mem_end;
524 EXPORT_SYMBOL(netdev_boot_setup_check);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix, int unit)
539 const struct netdev_boot_setup *s = dev_boot_setup;
543 sprintf(name, "%s%d", prefix, unit);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net, name))
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
553 if (!strcmp(name, s[i].name))
554 return s[i].map.base_addr;
559 * Saves at boot time configured settings for any netdevice.
561 int __init netdev_boot_setup(char *str)
566 str = get_options(str, ARRAY_SIZE(ints), ints);
571 memset(&map, 0, sizeof(map));
575 map.base_addr = ints[2];
577 map.mem_start = ints[3];
579 map.mem_end = ints[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str, &map);
585 __setup("netdev=", netdev_boot_setup);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device *__dev_get_by_name(struct net *net, const char *name)
607 struct hlist_node *p;
608 struct net_device *dev;
609 struct hlist_head *head = dev_name_hash(net, name);
611 hlist_for_each_entry(dev, p, head, name_hlist)
612 if (!strncmp(dev->name, name, IFNAMSIZ))
617 EXPORT_SYMBOL(__dev_get_by_name);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
633 struct hlist_node *p;
634 struct net_device *dev;
635 struct hlist_head *head = dev_name_hash(net, name);
637 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
638 if (!strncmp(dev->name, name, IFNAMSIZ))
643 EXPORT_SYMBOL(dev_get_by_name_rcu);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device *dev_get_by_name(struct net *net, const char *name)
659 struct net_device *dev;
662 dev = dev_get_by_name_rcu(net, name);
668 EXPORT_SYMBOL(dev_get_by_name);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
684 struct hlist_node *p;
685 struct net_device *dev;
686 struct hlist_head *head = dev_index_hash(net, ifindex);
688 hlist_for_each_entry(dev, p, head, index_hlist)
689 if (dev->ifindex == ifindex)
694 EXPORT_SYMBOL(__dev_get_by_index);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
709 struct hlist_node *p;
710 struct net_device *dev;
711 struct hlist_head *head = dev_index_hash(net, ifindex);
713 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
714 if (dev->ifindex == ifindex)
719 EXPORT_SYMBOL(dev_get_by_index_rcu);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device *dev_get_by_index(struct net *net, int ifindex)
735 struct net_device *dev;
738 dev = dev_get_by_index_rcu(net, ifindex);
744 EXPORT_SYMBOL(dev_get_by_index);
747 * dev_getbyhwaddr - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device. The caller must hold the
754 * rtnl semaphore. The returned device has not had its ref count increased
755 * and the caller must therefore be careful about locking
758 * If the API was consistent this would be __dev_get_by_hwaddr
761 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
763 struct net_device *dev;
767 for_each_netdev(net, dev)
768 if (dev->type == type &&
769 !memcmp(dev->dev_addr, ha, dev->addr_len))
774 EXPORT_SYMBOL(dev_getbyhwaddr);
776 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
778 struct net_device *dev;
781 for_each_netdev(net, dev)
782 if (dev->type == type)
787 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
789 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
791 struct net_device *dev, *ret = NULL;
794 for_each_netdev_rcu(net, dev)
795 if (dev->type == type) {
803 EXPORT_SYMBOL(dev_getfirstbyhwtype);
806 * dev_get_by_flags - find any device with given flags
807 * @net: the applicable net namespace
808 * @if_flags: IFF_* values
809 * @mask: bitmask of bits in if_flags to check
811 * Search for any interface with the given flags. Returns NULL if a device
812 * is not found or a pointer to the device. The device returned has
813 * had a reference added and the pointer is safe until the user calls
814 * dev_put to indicate they have finished with it.
817 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
820 struct net_device *dev, *ret;
824 for_each_netdev_rcu(net, dev) {
825 if (((dev->flags ^ if_flags) & mask) == 0) {
834 EXPORT_SYMBOL(dev_get_by_flags);
837 * dev_valid_name - check if name is okay for network device
840 * Network device names need to be valid file names to
841 * to allow sysfs to work. We also disallow any kind of
844 int dev_valid_name(const char *name)
848 if (strlen(name) >= IFNAMSIZ)
850 if (!strcmp(name, ".") || !strcmp(name, ".."))
854 if (*name == '/' || isspace(*name))
860 EXPORT_SYMBOL(dev_valid_name);
863 * __dev_alloc_name - allocate a name for a device
864 * @net: network namespace to allocate the device name in
865 * @name: name format string
866 * @buf: scratch buffer and result name string
868 * Passed a format string - eg "lt%d" it will try and find a suitable
869 * id. It scans list of devices to build up a free map, then chooses
870 * the first empty slot. The caller must hold the dev_base or rtnl lock
871 * while allocating the name and adding the device in order to avoid
873 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
874 * Returns the number of the unit assigned or a negative errno code.
877 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
881 const int max_netdevices = 8*PAGE_SIZE;
882 unsigned long *inuse;
883 struct net_device *d;
885 p = strnchr(name, IFNAMSIZ-1, '%');
888 * Verify the string as this thing may have come from
889 * the user. There must be either one "%d" and no other "%"
892 if (p[1] != 'd' || strchr(p + 2, '%'))
895 /* Use one page as a bit array of possible slots */
896 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
900 for_each_netdev(net, d) {
901 if (!sscanf(d->name, name, &i))
903 if (i < 0 || i >= max_netdevices)
906 /* avoid cases where sscanf is not exact inverse of printf */
907 snprintf(buf, IFNAMSIZ, name, i);
908 if (!strncmp(buf, d->name, IFNAMSIZ))
912 i = find_first_zero_bit(inuse, max_netdevices);
913 free_page((unsigned long) inuse);
917 snprintf(buf, IFNAMSIZ, name, i);
918 if (!__dev_get_by_name(net, buf))
921 /* It is possible to run out of possible slots
922 * when the name is long and there isn't enough space left
923 * for the digits, or if all bits are used.
929 * dev_alloc_name - allocate a name for a device
931 * @name: name format string
933 * Passed a format string - eg "lt%d" it will try and find a suitable
934 * id. It scans list of devices to build up a free map, then chooses
935 * the first empty slot. The caller must hold the dev_base or rtnl lock
936 * while allocating the name and adding the device in order to avoid
938 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
939 * Returns the number of the unit assigned or a negative errno code.
942 int dev_alloc_name(struct net_device *dev, const char *name)
948 BUG_ON(!dev_net(dev));
950 ret = __dev_alloc_name(net, name, buf);
952 strlcpy(dev->name, buf, IFNAMSIZ);
955 EXPORT_SYMBOL(dev_alloc_name);
957 static int dev_get_valid_name(struct net *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)
1439 return raw_notifier_call_chain(&netdev_chain, val, dev);
1442 /* When > 0 there are consumers of rx skb time stamps */
1443 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1445 void net_enable_timestamp(void)
1447 atomic_inc(&netstamp_needed);
1449 EXPORT_SYMBOL(net_enable_timestamp);
1451 void net_disable_timestamp(void)
1453 atomic_dec(&netstamp_needed);
1455 EXPORT_SYMBOL(net_disable_timestamp);
1457 static inline void net_timestamp(struct sk_buff *skb)
1459 if (atomic_read(&netstamp_needed))
1460 __net_timestamp(skb);
1462 skb->tstamp.tv64 = 0;
1466 * dev_forward_skb - loopback an skb to another netif
1468 * @dev: destination network device
1469 * @skb: buffer to forward
1472 * NET_RX_SUCCESS (no congestion)
1473 * NET_RX_DROP (packet was dropped)
1475 * dev_forward_skb can be used for injecting an skb from the
1476 * start_xmit function of one device into the receive queue
1477 * of another device.
1479 * The receiving device may be in another namespace, so
1480 * we have to clear all information in the skb that could
1481 * impact namespace isolation.
1483 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1487 if (!(dev->flags & IFF_UP))
1490 if (skb->len > (dev->mtu + dev->hard_header_len))
1493 skb_set_dev(skb, dev);
1494 skb->tstamp.tv64 = 0;
1495 skb->pkt_type = PACKET_HOST;
1496 skb->protocol = eth_type_trans(skb, dev);
1497 return netif_rx(skb);
1499 EXPORT_SYMBOL_GPL(dev_forward_skb);
1502 * Support routine. Sends outgoing frames to any network
1503 * taps currently in use.
1506 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1508 struct packet_type *ptype;
1510 #ifdef CONFIG_NET_CLS_ACT
1511 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1518 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1519 /* Never send packets back to the socket
1520 * they originated from - MvS (miquels@drinkel.ow.org)
1522 if ((ptype->dev == dev || !ptype->dev) &&
1523 (ptype->af_packet_priv == NULL ||
1524 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1525 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1529 /* skb->nh should be correctly
1530 set by sender, so that the second statement is
1531 just protection against buggy protocols.
1533 skb_reset_mac_header(skb2);
1535 if (skb_network_header(skb2) < skb2->data ||
1536 skb2->network_header > skb2->tail) {
1537 if (net_ratelimit())
1538 printk(KERN_CRIT "protocol %04x is "
1540 skb2->protocol, dev->name);
1541 skb_reset_network_header(skb2);
1544 skb2->transport_header = skb2->network_header;
1545 skb2->pkt_type = PACKET_OUTGOING;
1546 ptype->func(skb2, skb->dev, ptype, skb->dev);
1553 static inline void __netif_reschedule(struct Qdisc *q)
1555 struct softnet_data *sd;
1556 unsigned long flags;
1558 local_irq_save(flags);
1559 sd = &__get_cpu_var(softnet_data);
1560 q->next_sched = NULL;
1561 *sd->output_queue_tailp = q;
1562 sd->output_queue_tailp = &q->next_sched;
1563 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1564 local_irq_restore(flags);
1567 void __netif_schedule(struct Qdisc *q)
1569 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1570 __netif_reschedule(q);
1572 EXPORT_SYMBOL(__netif_schedule);
1574 void dev_kfree_skb_irq(struct sk_buff *skb)
1576 if (atomic_dec_and_test(&skb->users)) {
1577 struct softnet_data *sd;
1578 unsigned long flags;
1580 local_irq_save(flags);
1581 sd = &__get_cpu_var(softnet_data);
1582 skb->next = sd->completion_queue;
1583 sd->completion_queue = skb;
1584 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1585 local_irq_restore(flags);
1588 EXPORT_SYMBOL(dev_kfree_skb_irq);
1590 void dev_kfree_skb_any(struct sk_buff *skb)
1592 if (in_irq() || irqs_disabled())
1593 dev_kfree_skb_irq(skb);
1597 EXPORT_SYMBOL(dev_kfree_skb_any);
1601 * netif_device_detach - mark device as removed
1602 * @dev: network device
1604 * Mark device as removed from system and therefore no longer available.
1606 void netif_device_detach(struct net_device *dev)
1608 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1609 netif_running(dev)) {
1610 netif_tx_stop_all_queues(dev);
1613 EXPORT_SYMBOL(netif_device_detach);
1616 * netif_device_attach - mark device as attached
1617 * @dev: network device
1619 * Mark device as attached from system and restart if needed.
1621 void netif_device_attach(struct net_device *dev)
1623 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1624 netif_running(dev)) {
1625 netif_tx_wake_all_queues(dev);
1626 __netdev_watchdog_up(dev);
1629 EXPORT_SYMBOL(netif_device_attach);
1631 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1633 return ((features & NETIF_F_GEN_CSUM) ||
1634 ((features & NETIF_F_IP_CSUM) &&
1635 protocol == htons(ETH_P_IP)) ||
1636 ((features & NETIF_F_IPV6_CSUM) &&
1637 protocol == htons(ETH_P_IPV6)) ||
1638 ((features & NETIF_F_FCOE_CRC) &&
1639 protocol == htons(ETH_P_FCOE)));
1642 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1644 if (can_checksum_protocol(dev->features, skb->protocol))
1647 if (skb->protocol == htons(ETH_P_8021Q)) {
1648 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1649 if (can_checksum_protocol(dev->features & dev->vlan_features,
1650 veh->h_vlan_encapsulated_proto))
1658 * skb_dev_set -- assign a new device to a buffer
1659 * @skb: buffer for the new device
1660 * @dev: network device
1662 * If an skb is owned by a device already, we have to reset
1663 * all data private to the namespace a device belongs to
1664 * before assigning it a new device.
1666 #ifdef CONFIG_NET_NS
1667 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1670 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1673 skb_init_secmark(skb);
1677 skb->ipvs_property = 0;
1678 #ifdef CONFIG_NET_SCHED
1684 EXPORT_SYMBOL(skb_set_dev);
1685 #endif /* CONFIG_NET_NS */
1688 * Invalidate hardware checksum when packet is to be mangled, and
1689 * complete checksum manually on outgoing path.
1691 int skb_checksum_help(struct sk_buff *skb)
1694 int ret = 0, offset;
1696 if (skb->ip_summed == CHECKSUM_COMPLETE)
1697 goto out_set_summed;
1699 if (unlikely(skb_shinfo(skb)->gso_size)) {
1700 /* Let GSO fix up the checksum. */
1701 goto out_set_summed;
1704 offset = skb->csum_start - skb_headroom(skb);
1705 BUG_ON(offset >= skb_headlen(skb));
1706 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1708 offset += skb->csum_offset;
1709 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1711 if (skb_cloned(skb) &&
1712 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1713 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1718 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1720 skb->ip_summed = CHECKSUM_NONE;
1724 EXPORT_SYMBOL(skb_checksum_help);
1727 * skb_gso_segment - Perform segmentation on skb.
1728 * @skb: buffer to segment
1729 * @features: features for the output path (see dev->features)
1731 * This function segments the given skb and returns a list of segments.
1733 * It may return NULL if the skb requires no segmentation. This is
1734 * only possible when GSO is used for verifying header integrity.
1736 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1738 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1739 struct packet_type *ptype;
1740 __be16 type = skb->protocol;
1743 skb_reset_mac_header(skb);
1744 skb->mac_len = skb->network_header - skb->mac_header;
1745 __skb_pull(skb, skb->mac_len);
1747 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1748 struct net_device *dev = skb->dev;
1749 struct ethtool_drvinfo info = {};
1751 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1752 dev->ethtool_ops->get_drvinfo(dev, &info);
1754 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1756 info.driver, dev ? dev->features : 0L,
1757 skb->sk ? skb->sk->sk_route_caps : 0L,
1758 skb->len, skb->data_len, skb->ip_summed);
1760 if (skb_header_cloned(skb) &&
1761 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1762 return ERR_PTR(err);
1766 list_for_each_entry_rcu(ptype,
1767 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1768 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1769 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1770 err = ptype->gso_send_check(skb);
1771 segs = ERR_PTR(err);
1772 if (err || skb_gso_ok(skb, features))
1774 __skb_push(skb, (skb->data -
1775 skb_network_header(skb)));
1777 segs = ptype->gso_segment(skb, features);
1783 __skb_push(skb, skb->data - skb_mac_header(skb));
1787 EXPORT_SYMBOL(skb_gso_segment);
1789 /* Take action when hardware reception checksum errors are detected. */
1791 void netdev_rx_csum_fault(struct net_device *dev)
1793 if (net_ratelimit()) {
1794 printk(KERN_ERR "%s: hw csum failure.\n",
1795 dev ? dev->name : "<unknown>");
1799 EXPORT_SYMBOL(netdev_rx_csum_fault);
1802 /* Actually, we should eliminate this check as soon as we know, that:
1803 * 1. IOMMU is present and allows to map all the memory.
1804 * 2. No high memory really exists on this machine.
1807 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1809 #ifdef CONFIG_HIGHMEM
1811 if (!(dev->features & NETIF_F_HIGHDMA)) {
1812 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1813 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1817 if (PCI_DMA_BUS_IS_PHYS) {
1818 struct device *pdev = dev->dev.parent;
1822 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1823 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1824 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1833 void (*destructor)(struct sk_buff *skb);
1836 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1838 static void dev_gso_skb_destructor(struct sk_buff *skb)
1840 struct dev_gso_cb *cb;
1843 struct sk_buff *nskb = skb->next;
1845 skb->next = nskb->next;
1848 } while (skb->next);
1850 cb = DEV_GSO_CB(skb);
1852 cb->destructor(skb);
1856 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1857 * @skb: buffer to segment
1859 * This function segments the given skb and stores the list of segments
1862 static int dev_gso_segment(struct sk_buff *skb)
1864 struct net_device *dev = skb->dev;
1865 struct sk_buff *segs;
1866 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1869 segs = skb_gso_segment(skb, features);
1871 /* Verifying header integrity only. */
1876 return PTR_ERR(segs);
1879 DEV_GSO_CB(skb)->destructor = skb->destructor;
1880 skb->destructor = dev_gso_skb_destructor;
1886 * Try to orphan skb early, right before transmission by the device.
1887 * We cannot orphan skb if tx timestamp is requested, since
1888 * drivers need to call skb_tstamp_tx() to send the timestamp.
1890 static inline void skb_orphan_try(struct sk_buff *skb)
1892 if (!skb_tx(skb)->flags)
1896 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1897 struct netdev_queue *txq)
1899 const struct net_device_ops *ops = dev->netdev_ops;
1900 int rc = NETDEV_TX_OK;
1902 if (likely(!skb->next)) {
1903 if (!list_empty(&ptype_all))
1904 dev_queue_xmit_nit(skb, dev);
1907 * If device doesnt need skb->dst, release it right now while
1908 * its hot in this cpu cache
1910 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1913 skb_orphan_try(skb);
1915 if (netif_needs_gso(dev, skb)) {
1916 if (unlikely(dev_gso_segment(skb)))
1922 rc = ops->ndo_start_xmit(skb, dev);
1923 if (rc == NETDEV_TX_OK)
1924 txq_trans_update(txq);
1930 struct sk_buff *nskb = skb->next;
1932 skb->next = nskb->next;
1936 * If device doesnt need nskb->dst, release it right now while
1937 * its hot in this cpu cache
1939 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1942 rc = ops->ndo_start_xmit(nskb, dev);
1943 if (unlikely(rc != NETDEV_TX_OK)) {
1944 if (rc & ~NETDEV_TX_MASK)
1945 goto out_kfree_gso_skb;
1946 nskb->next = skb->next;
1950 txq_trans_update(txq);
1951 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1952 return NETDEV_TX_BUSY;
1953 } while (skb->next);
1956 if (likely(skb->next == NULL))
1957 skb->destructor = DEV_GSO_CB(skb)->destructor;
1963 static u32 hashrnd __read_mostly;
1965 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1969 if (skb_rx_queue_recorded(skb)) {
1970 hash = skb_get_rx_queue(skb);
1971 while (unlikely(hash >= dev->real_num_tx_queues))
1972 hash -= dev->real_num_tx_queues;
1976 if (skb->sk && skb->sk->sk_hash)
1977 hash = skb->sk->sk_hash;
1979 hash = (__force u16) skb->protocol;
1981 hash = jhash_1word(hash, hashrnd);
1983 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1985 EXPORT_SYMBOL(skb_tx_hash);
1987 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1989 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1990 if (net_ratelimit()) {
1991 pr_warning("%s selects TX queue %d, but "
1992 "real number of TX queues is %d\n",
1993 dev->name, queue_index, dev->real_num_tx_queues);
2000 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2001 struct sk_buff *skb)
2004 struct sock *sk = skb->sk;
2006 if (sk_tx_queue_recorded(sk)) {
2007 queue_index = sk_tx_queue_get(sk);
2009 const struct net_device_ops *ops = dev->netdev_ops;
2011 if (ops->ndo_select_queue) {
2012 queue_index = ops->ndo_select_queue(dev, skb);
2013 queue_index = dev_cap_txqueue(dev, queue_index);
2016 if (dev->real_num_tx_queues > 1)
2017 queue_index = skb_tx_hash(dev, skb);
2020 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2022 if (dst && skb_dst(skb) == dst)
2023 sk_tx_queue_set(sk, queue_index);
2028 skb_set_queue_mapping(skb, queue_index);
2029 return netdev_get_tx_queue(dev, queue_index);
2032 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2033 struct net_device *dev,
2034 struct netdev_queue *txq)
2036 spinlock_t *root_lock = qdisc_lock(q);
2039 spin_lock(root_lock);
2040 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2043 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2044 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
2046 * This is a work-conserving queue; there are no old skbs
2047 * waiting to be sent out; and the qdisc is not running -
2048 * xmit the skb directly.
2050 __qdisc_update_bstats(q, skb->len);
2051 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
2054 clear_bit(__QDISC_STATE_RUNNING, &q->state);
2056 rc = NET_XMIT_SUCCESS;
2058 rc = qdisc_enqueue_root(skb, q);
2061 spin_unlock(root_lock);
2067 * Returns true if either:
2068 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2069 * 2. skb is fragmented and the device does not support SG, or if
2070 * at least one of fragments is in highmem and device does not
2071 * support DMA from it.
2073 static inline int skb_needs_linearize(struct sk_buff *skb,
2074 struct net_device *dev)
2076 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
2077 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
2078 illegal_highdma(dev, skb)));
2082 * dev_queue_xmit - transmit a buffer
2083 * @skb: buffer to transmit
2085 * Queue a buffer for transmission to a network device. The caller must
2086 * have set the device and priority and built the buffer before calling
2087 * this function. The function can be called from an interrupt.
2089 * A negative errno code is returned on a failure. A success does not
2090 * guarantee the frame will be transmitted as it may be dropped due
2091 * to congestion or traffic shaping.
2093 * -----------------------------------------------------------------------------------
2094 * I notice this method can also return errors from the queue disciplines,
2095 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2098 * Regardless of the return value, the skb is consumed, so it is currently
2099 * difficult to retry a send to this method. (You can bump the ref count
2100 * before sending to hold a reference for retry if you are careful.)
2102 * When calling this method, interrupts MUST be enabled. This is because
2103 * the BH enable code must have IRQs enabled so that it will not deadlock.
2106 int dev_queue_xmit(struct sk_buff *skb)
2108 struct net_device *dev = skb->dev;
2109 struct netdev_queue *txq;
2113 /* GSO will handle the following emulations directly. */
2114 if (netif_needs_gso(dev, skb))
2117 /* Convert a paged skb to linear, if required */
2118 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2121 /* If packet is not checksummed and device does not support
2122 * checksumming for this protocol, complete checksumming here.
2124 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2125 skb_set_transport_header(skb, skb->csum_start -
2127 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2132 /* Disable soft irqs for various locks below. Also
2133 * stops preemption for RCU.
2137 txq = dev_pick_tx(dev, skb);
2138 q = rcu_dereference_bh(txq->qdisc);
2140 #ifdef CONFIG_NET_CLS_ACT
2141 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2144 rc = __dev_xmit_skb(skb, q, dev, txq);
2148 /* The device has no queue. Common case for software devices:
2149 loopback, all the sorts of tunnels...
2151 Really, it is unlikely that netif_tx_lock protection is necessary
2152 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2154 However, it is possible, that they rely on protection
2157 Check this and shot the lock. It is not prone from deadlocks.
2158 Either shot noqueue qdisc, it is even simpler 8)
2160 if (dev->flags & IFF_UP) {
2161 int cpu = smp_processor_id(); /* ok because BHs are off */
2163 if (txq->xmit_lock_owner != cpu) {
2165 HARD_TX_LOCK(dev, txq, cpu);
2167 if (!netif_tx_queue_stopped(txq)) {
2168 rc = dev_hard_start_xmit(skb, dev, txq);
2169 if (dev_xmit_complete(rc)) {
2170 HARD_TX_UNLOCK(dev, txq);
2174 HARD_TX_UNLOCK(dev, txq);
2175 if (net_ratelimit())
2176 printk(KERN_CRIT "Virtual device %s asks to "
2177 "queue packet!\n", dev->name);
2179 /* Recursion is detected! It is possible,
2181 if (net_ratelimit())
2182 printk(KERN_CRIT "Dead loop on virtual device "
2183 "%s, fix it urgently!\n", dev->name);
2188 rcu_read_unlock_bh();
2194 rcu_read_unlock_bh();
2197 EXPORT_SYMBOL(dev_queue_xmit);
2200 /*=======================================================================
2202 =======================================================================*/
2204 int netdev_max_backlog __read_mostly = 1000;
2205 int netdev_budget __read_mostly = 300;
2206 int weight_p __read_mostly = 64; /* old backlog weight */
2208 /* Called with irq disabled */
2209 static inline void ____napi_schedule(struct softnet_data *sd,
2210 struct napi_struct *napi)
2212 list_add_tail(&napi->poll_list, &sd->poll_list);
2213 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2218 /* One global table that all flow-based protocols share. */
2219 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2220 EXPORT_SYMBOL(rps_sock_flow_table);
2223 * get_rps_cpu is called from netif_receive_skb and returns the target
2224 * CPU from the RPS map of the receiving queue for a given skb.
2225 * rcu_read_lock must be held on entry.
2227 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2228 struct rps_dev_flow **rflowp)
2230 struct ipv6hdr *ip6;
2232 struct netdev_rx_queue *rxqueue;
2233 struct rps_map *map;
2234 struct rps_dev_flow_table *flow_table;
2235 struct rps_sock_flow_table *sock_flow_table;
2239 u32 addr1, addr2, ihl;
2245 if (skb_rx_queue_recorded(skb)) {
2246 u16 index = skb_get_rx_queue(skb);
2247 if (unlikely(index >= dev->num_rx_queues)) {
2248 if (net_ratelimit()) {
2249 pr_warning("%s received packet on queue "
2250 "%u, but number of RX queues is %u\n",
2251 dev->name, index, dev->num_rx_queues);
2255 rxqueue = dev->_rx + index;
2259 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2263 goto got_hash; /* Skip hash computation on packet header */
2265 switch (skb->protocol) {
2266 case __constant_htons(ETH_P_IP):
2267 if (!pskb_may_pull(skb, sizeof(*ip)))
2270 ip = (struct iphdr *) skb->data;
2271 ip_proto = ip->protocol;
2272 addr1 = (__force u32) ip->saddr;
2273 addr2 = (__force u32) ip->daddr;
2276 case __constant_htons(ETH_P_IPV6):
2277 if (!pskb_may_pull(skb, sizeof(*ip6)))
2280 ip6 = (struct ipv6hdr *) skb->data;
2281 ip_proto = ip6->nexthdr;
2282 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2283 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2296 case IPPROTO_UDPLITE:
2297 if (pskb_may_pull(skb, (ihl * 4) + 4)) {
2298 ports.v32 = * (__force u32 *) (skb->data + (ihl * 4));
2299 if (ports.v16[1] < ports.v16[0])
2300 swap(ports.v16[0], ports.v16[1]);
2308 /* get a consistent hash (same value on both flow directions) */
2311 skb->rxhash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2316 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2317 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2318 if (flow_table && sock_flow_table) {
2320 struct rps_dev_flow *rflow;
2322 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2325 next_cpu = sock_flow_table->ents[skb->rxhash &
2326 sock_flow_table->mask];
2329 * If the desired CPU (where last recvmsg was done) is
2330 * different from current CPU (one in the rx-queue flow
2331 * table entry), switch if one of the following holds:
2332 * - Current CPU is unset (equal to RPS_NO_CPU).
2333 * - Current CPU is offline.
2334 * - The current CPU's queue tail has advanced beyond the
2335 * last packet that was enqueued using this table entry.
2336 * This guarantees that all previous packets for the flow
2337 * have been dequeued, thus preserving in order delivery.
2339 if (unlikely(tcpu != next_cpu) &&
2340 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2341 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2342 rflow->last_qtail)) >= 0)) {
2343 tcpu = rflow->cpu = next_cpu;
2344 if (tcpu != RPS_NO_CPU)
2345 rflow->last_qtail = per_cpu(softnet_data,
2346 tcpu).input_queue_head;
2348 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2355 map = rcu_dereference(rxqueue->rps_map);
2357 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2359 if (cpu_online(tcpu)) {
2369 /* Called from hardirq (IPI) context */
2370 static void rps_trigger_softirq(void *data)
2372 struct softnet_data *sd = data;
2374 ____napi_schedule(sd, &sd->backlog);
2378 #endif /* CONFIG_RPS */
2381 * Check if this softnet_data structure is another cpu one
2382 * If yes, queue it to our IPI list and return 1
2385 static int rps_ipi_queued(struct softnet_data *sd)
2388 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2391 sd->rps_ipi_next = mysd->rps_ipi_list;
2392 mysd->rps_ipi_list = sd;
2394 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2397 #endif /* CONFIG_RPS */
2402 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2403 * queue (may be a remote CPU queue).
2405 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2406 unsigned int *qtail)
2408 struct softnet_data *sd;
2409 unsigned long flags;
2411 sd = &per_cpu(softnet_data, cpu);
2413 local_irq_save(flags);
2416 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2417 if (skb_queue_len(&sd->input_pkt_queue)) {
2419 __skb_queue_tail(&sd->input_pkt_queue, skb);
2421 *qtail = sd->input_queue_head +
2422 skb_queue_len(&sd->input_pkt_queue);
2425 local_irq_restore(flags);
2426 return NET_RX_SUCCESS;
2429 /* Schedule NAPI for backlog device */
2430 if (napi_schedule_prep(&sd->backlog)) {
2431 if (!rps_ipi_queued(sd))
2432 ____napi_schedule(sd, &sd->backlog);
2440 local_irq_restore(flags);
2447 * netif_rx - post buffer to the network code
2448 * @skb: buffer to post
2450 * This function receives a packet from a device driver and queues it for
2451 * the upper (protocol) levels to process. It always succeeds. The buffer
2452 * may be dropped during processing for congestion control or by the
2456 * NET_RX_SUCCESS (no congestion)
2457 * NET_RX_DROP (packet was dropped)
2461 int netif_rx(struct sk_buff *skb)
2465 /* if netpoll wants it, pretend we never saw it */
2466 if (netpoll_rx(skb))
2469 if (!skb->tstamp.tv64)
2474 struct rps_dev_flow voidflow, *rflow = &voidflow;
2479 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2481 cpu = smp_processor_id();
2483 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2490 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2496 EXPORT_SYMBOL(netif_rx);
2498 int netif_rx_ni(struct sk_buff *skb)
2503 err = netif_rx(skb);
2504 if (local_softirq_pending())
2510 EXPORT_SYMBOL(netif_rx_ni);
2512 static void net_tx_action(struct softirq_action *h)
2514 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2516 if (sd->completion_queue) {
2517 struct sk_buff *clist;
2519 local_irq_disable();
2520 clist = sd->completion_queue;
2521 sd->completion_queue = NULL;
2525 struct sk_buff *skb = clist;
2526 clist = clist->next;
2528 WARN_ON(atomic_read(&skb->users));
2533 if (sd->output_queue) {
2536 local_irq_disable();
2537 head = sd->output_queue;
2538 sd->output_queue = NULL;
2539 sd->output_queue_tailp = &sd->output_queue;
2543 struct Qdisc *q = head;
2544 spinlock_t *root_lock;
2546 head = head->next_sched;
2548 root_lock = qdisc_lock(q);
2549 if (spin_trylock(root_lock)) {
2550 smp_mb__before_clear_bit();
2551 clear_bit(__QDISC_STATE_SCHED,
2554 spin_unlock(root_lock);
2556 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2558 __netif_reschedule(q);
2560 smp_mb__before_clear_bit();
2561 clear_bit(__QDISC_STATE_SCHED,
2569 static inline int deliver_skb(struct sk_buff *skb,
2570 struct packet_type *pt_prev,
2571 struct net_device *orig_dev)
2573 atomic_inc(&skb->users);
2574 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2577 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2579 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2580 /* This hook is defined here for ATM LANE */
2581 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2582 unsigned char *addr) __read_mostly;
2583 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2587 * If bridge module is loaded call bridging hook.
2588 * returns NULL if packet was consumed.
2590 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2591 struct sk_buff *skb) __read_mostly;
2592 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2594 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2595 struct packet_type **pt_prev, int *ret,
2596 struct net_device *orig_dev)
2598 struct net_bridge_port *port;
2600 if (skb->pkt_type == PACKET_LOOPBACK ||
2601 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2605 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2609 return br_handle_frame_hook(port, skb);
2612 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2615 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2616 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2617 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2619 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2620 struct packet_type **pt_prev,
2622 struct net_device *orig_dev)
2624 if (skb->dev->macvlan_port == NULL)
2628 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2631 return macvlan_handle_frame_hook(skb);
2634 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2637 #ifdef CONFIG_NET_CLS_ACT
2638 /* TODO: Maybe we should just force sch_ingress to be compiled in
2639 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2640 * a compare and 2 stores extra right now if we dont have it on
2641 * but have CONFIG_NET_CLS_ACT
2642 * NOTE: This doesnt stop any functionality; if you dont have
2643 * the ingress scheduler, you just cant add policies on ingress.
2646 static int ing_filter(struct sk_buff *skb)
2648 struct net_device *dev = skb->dev;
2649 u32 ttl = G_TC_RTTL(skb->tc_verd);
2650 struct netdev_queue *rxq;
2651 int result = TC_ACT_OK;
2654 if (MAX_RED_LOOP < ttl++) {
2656 "Redir loop detected Dropping packet (%d->%d)\n",
2657 skb->skb_iif, dev->ifindex);
2661 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2662 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2664 rxq = &dev->rx_queue;
2667 if (q != &noop_qdisc) {
2668 spin_lock(qdisc_lock(q));
2669 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2670 result = qdisc_enqueue_root(skb, q);
2671 spin_unlock(qdisc_lock(q));
2677 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2678 struct packet_type **pt_prev,
2679 int *ret, struct net_device *orig_dev)
2681 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2685 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2688 /* Huh? Why does turning on AF_PACKET affect this? */
2689 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2692 switch (ing_filter(skb)) {
2706 * netif_nit_deliver - deliver received packets to network taps
2709 * This function is used to deliver incoming packets to network
2710 * taps. It should be used when the normal netif_receive_skb path
2711 * is bypassed, for example because of VLAN acceleration.
2713 void netif_nit_deliver(struct sk_buff *skb)
2715 struct packet_type *ptype;
2717 if (list_empty(&ptype_all))
2720 skb_reset_network_header(skb);
2721 skb_reset_transport_header(skb);
2722 skb->mac_len = skb->network_header - skb->mac_header;
2725 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2726 if (!ptype->dev || ptype->dev == skb->dev)
2727 deliver_skb(skb, ptype, skb->dev);
2732 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2733 struct net_device *master)
2735 if (skb->pkt_type == PACKET_HOST) {
2736 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2738 memcpy(dest, master->dev_addr, ETH_ALEN);
2742 /* On bonding slaves other than the currently active slave, suppress
2743 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2744 * ARP on active-backup slaves with arp_validate enabled.
2746 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2748 struct net_device *dev = skb->dev;
2750 if (master->priv_flags & IFF_MASTER_ARPMON)
2751 dev->last_rx = jiffies;
2753 if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) {
2754 /* Do address unmangle. The local destination address
2755 * will be always the one master has. Provides the right
2756 * functionality in a bridge.
2758 skb_bond_set_mac_by_master(skb, master);
2761 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2762 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2763 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2766 if (master->priv_flags & IFF_MASTER_ALB) {
2767 if (skb->pkt_type != PACKET_BROADCAST &&
2768 skb->pkt_type != PACKET_MULTICAST)
2771 if (master->priv_flags & IFF_MASTER_8023AD &&
2772 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2779 EXPORT_SYMBOL(__skb_bond_should_drop);
2781 static int __netif_receive_skb(struct sk_buff *skb)
2783 struct packet_type *ptype, *pt_prev;
2784 struct net_device *orig_dev;
2785 struct net_device *master;
2786 struct net_device *null_or_orig;
2787 struct net_device *null_or_bond;
2788 int ret = NET_RX_DROP;
2791 if (!skb->tstamp.tv64)
2794 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2795 return NET_RX_SUCCESS;
2797 /* if we've gotten here through NAPI, check netpoll */
2798 if (netpoll_receive_skb(skb))
2802 skb->skb_iif = skb->dev->ifindex;
2804 null_or_orig = NULL;
2805 orig_dev = skb->dev;
2806 master = ACCESS_ONCE(orig_dev->master);
2808 if (skb_bond_should_drop(skb, master))
2809 null_or_orig = orig_dev; /* deliver only exact match */
2814 __get_cpu_var(softnet_data).processed++;
2816 skb_reset_network_header(skb);
2817 skb_reset_transport_header(skb);
2818 skb->mac_len = skb->network_header - skb->mac_header;
2824 #ifdef CONFIG_NET_CLS_ACT
2825 if (skb->tc_verd & TC_NCLS) {
2826 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2831 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2832 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2833 ptype->dev == orig_dev) {
2835 ret = deliver_skb(skb, pt_prev, orig_dev);
2840 #ifdef CONFIG_NET_CLS_ACT
2841 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2847 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2850 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2855 * Make sure frames received on VLAN interfaces stacked on
2856 * bonding interfaces still make their way to any base bonding
2857 * device that may have registered for a specific ptype. The
2858 * handler may have to adjust skb->dev and orig_dev.
2860 null_or_bond = NULL;
2861 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2862 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2863 null_or_bond = vlan_dev_real_dev(skb->dev);
2866 type = skb->protocol;
2867 list_for_each_entry_rcu(ptype,
2868 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2869 if (ptype->type == type && (ptype->dev == null_or_orig ||
2870 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2871 ptype->dev == null_or_bond)) {
2873 ret = deliver_skb(skb, pt_prev, orig_dev);
2879 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2882 /* Jamal, now you will not able to escape explaining
2883 * me how you were going to use this. :-)
2894 * netif_receive_skb - process receive buffer from network
2895 * @skb: buffer to process
2897 * netif_receive_skb() is the main receive data processing function.
2898 * It always succeeds. The buffer may be dropped during processing
2899 * for congestion control or by the protocol layers.
2901 * This function may only be called from softirq context and interrupts
2902 * should be enabled.
2904 * Return values (usually ignored):
2905 * NET_RX_SUCCESS: no congestion
2906 * NET_RX_DROP: packet was dropped
2908 int netif_receive_skb(struct sk_buff *skb)
2911 struct rps_dev_flow voidflow, *rflow = &voidflow;
2916 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2919 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2923 ret = __netif_receive_skb(skb);
2928 return __netif_receive_skb(skb);
2931 EXPORT_SYMBOL(netif_receive_skb);
2933 /* Network device is going away, flush any packets still pending
2934 * Called with irqs disabled.
2936 static void flush_backlog(void *arg)
2938 struct net_device *dev = arg;
2939 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2940 struct sk_buff *skb, *tmp;
2943 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
2944 if (skb->dev == dev) {
2945 __skb_unlink(skb, &sd->input_pkt_queue);
2947 input_queue_head_add(sd, 1);
2952 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
2953 if (skb->dev == dev) {
2954 __skb_unlink(skb, &sd->process_queue);
2960 static int napi_gro_complete(struct sk_buff *skb)
2962 struct packet_type *ptype;
2963 __be16 type = skb->protocol;
2964 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2967 if (NAPI_GRO_CB(skb)->count == 1) {
2968 skb_shinfo(skb)->gso_size = 0;
2973 list_for_each_entry_rcu(ptype, head, list) {
2974 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2977 err = ptype->gro_complete(skb);
2983 WARN_ON(&ptype->list == head);
2985 return NET_RX_SUCCESS;
2989 return netif_receive_skb(skb);
2992 static void napi_gro_flush(struct napi_struct *napi)
2994 struct sk_buff *skb, *next;
2996 for (skb = napi->gro_list; skb; skb = next) {
2999 napi_gro_complete(skb);
3002 napi->gro_count = 0;
3003 napi->gro_list = NULL;
3006 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3008 struct sk_buff **pp = NULL;
3009 struct packet_type *ptype;
3010 __be16 type = skb->protocol;
3011 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3014 enum gro_result ret;
3016 if (!(skb->dev->features & NETIF_F_GRO))
3019 if (skb_is_gso(skb) || skb_has_frags(skb))
3023 list_for_each_entry_rcu(ptype, head, list) {
3024 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3027 skb_set_network_header(skb, skb_gro_offset(skb));
3028 mac_len = skb->network_header - skb->mac_header;
3029 skb->mac_len = mac_len;
3030 NAPI_GRO_CB(skb)->same_flow = 0;
3031 NAPI_GRO_CB(skb)->flush = 0;
3032 NAPI_GRO_CB(skb)->free = 0;
3034 pp = ptype->gro_receive(&napi->gro_list, skb);
3039 if (&ptype->list == head)
3042 same_flow = NAPI_GRO_CB(skb)->same_flow;
3043 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3046 struct sk_buff *nskb = *pp;
3050 napi_gro_complete(nskb);
3057 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3061 NAPI_GRO_CB(skb)->count = 1;
3062 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3063 skb->next = napi->gro_list;
3064 napi->gro_list = skb;
3068 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3069 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3071 BUG_ON(skb->end - skb->tail < grow);
3073 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3076 skb->data_len -= grow;
3078 skb_shinfo(skb)->frags[0].page_offset += grow;
3079 skb_shinfo(skb)->frags[0].size -= grow;
3081 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3082 put_page(skb_shinfo(skb)->frags[0].page);
3083 memmove(skb_shinfo(skb)->frags,
3084 skb_shinfo(skb)->frags + 1,
3085 --skb_shinfo(skb)->nr_frags);
3096 EXPORT_SYMBOL(dev_gro_receive);
3099 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3103 if (netpoll_rx_on(skb))
3106 for (p = napi->gro_list; p; p = p->next) {
3107 NAPI_GRO_CB(p)->same_flow =
3108 (p->dev == skb->dev) &&
3109 !compare_ether_header(skb_mac_header(p),
3110 skb_gro_mac_header(skb));
3111 NAPI_GRO_CB(p)->flush = 0;
3114 return dev_gro_receive(napi, skb);
3117 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3121 if (netif_receive_skb(skb))
3126 case GRO_MERGED_FREE:
3137 EXPORT_SYMBOL(napi_skb_finish);
3139 void skb_gro_reset_offset(struct sk_buff *skb)
3141 NAPI_GRO_CB(skb)->data_offset = 0;
3142 NAPI_GRO_CB(skb)->frag0 = NULL;
3143 NAPI_GRO_CB(skb)->frag0_len = 0;
3145 if (skb->mac_header == skb->tail &&
3146 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3147 NAPI_GRO_CB(skb)->frag0 =
3148 page_address(skb_shinfo(skb)->frags[0].page) +
3149 skb_shinfo(skb)->frags[0].page_offset;
3150 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3153 EXPORT_SYMBOL(skb_gro_reset_offset);
3155 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3157 skb_gro_reset_offset(skb);
3159 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3161 EXPORT_SYMBOL(napi_gro_receive);
3163 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3165 __skb_pull(skb, skb_headlen(skb));
3166 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3170 EXPORT_SYMBOL(napi_reuse_skb);
3172 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3174 struct sk_buff *skb = napi->skb;
3177 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3183 EXPORT_SYMBOL(napi_get_frags);
3185 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3191 skb->protocol = eth_type_trans(skb, skb->dev);
3193 if (ret == GRO_HELD)
3194 skb_gro_pull(skb, -ETH_HLEN);
3195 else if (netif_receive_skb(skb))
3200 case GRO_MERGED_FREE:
3201 napi_reuse_skb(napi, skb);
3210 EXPORT_SYMBOL(napi_frags_finish);
3212 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3214 struct sk_buff *skb = napi->skb;
3221 skb_reset_mac_header(skb);
3222 skb_gro_reset_offset(skb);
3224 off = skb_gro_offset(skb);
3225 hlen = off + sizeof(*eth);
3226 eth = skb_gro_header_fast(skb, off);
3227 if (skb_gro_header_hard(skb, hlen)) {
3228 eth = skb_gro_header_slow(skb, hlen, off);
3229 if (unlikely(!eth)) {
3230 napi_reuse_skb(napi, skb);
3236 skb_gro_pull(skb, sizeof(*eth));
3239 * This works because the only protocols we care about don't require
3240 * special handling. We'll fix it up properly at the end.
3242 skb->protocol = eth->h_proto;
3247 EXPORT_SYMBOL(napi_frags_skb);
3249 gro_result_t napi_gro_frags(struct napi_struct *napi)
3251 struct sk_buff *skb = napi_frags_skb(napi);
3256 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3258 EXPORT_SYMBOL(napi_gro_frags);
3261 * net_rps_action sends any pending IPI's for rps.
3262 * Note: called with local irq disabled, but exits with local irq enabled.
3264 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3267 struct softnet_data *remsd = sd->rps_ipi_list;
3270 sd->rps_ipi_list = NULL;
3274 /* Send pending IPI's to kick RPS processing on remote cpus. */
3276 struct softnet_data *next = remsd->rps_ipi_next;
3278 if (cpu_online(remsd->cpu))
3279 __smp_call_function_single(remsd->cpu,
3288 static int process_backlog(struct napi_struct *napi, int quota)
3291 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3294 /* Check if we have pending ipi, its better to send them now,
3295 * not waiting net_rx_action() end.
3297 if (sd->rps_ipi_list) {
3298 local_irq_disable();
3299 net_rps_action_and_irq_enable(sd);
3302 napi->weight = weight_p;
3303 local_irq_disable();
3304 while (work < quota) {
3305 struct sk_buff *skb;
3308 while ((skb = __skb_dequeue(&sd->process_queue))) {
3310 __netif_receive_skb(skb);
3311 if (++work >= quota)
3313 local_irq_disable();
3317 qlen = skb_queue_len(&sd->input_pkt_queue);
3319 input_queue_head_add(sd, qlen);
3320 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3321 &sd->process_queue);
3323 if (qlen < quota - work) {
3325 * Inline a custom version of __napi_complete().
3326 * only current cpu owns and manipulates this napi,
3327 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3328 * we can use a plain write instead of clear_bit(),
3329 * and we dont need an smp_mb() memory barrier.
3331 list_del(&napi->poll_list);
3334 quota = work + qlen;
3344 * __napi_schedule - schedule for receive
3345 * @n: entry to schedule
3347 * The entry's receive function will be scheduled to run
3349 void __napi_schedule(struct napi_struct *n)
3351 unsigned long flags;
3353 local_irq_save(flags);
3354 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3355 local_irq_restore(flags);
3357 EXPORT_SYMBOL(__napi_schedule);
3359 void __napi_complete(struct napi_struct *n)
3361 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3362 BUG_ON(n->gro_list);
3364 list_del(&n->poll_list);
3365 smp_mb__before_clear_bit();
3366 clear_bit(NAPI_STATE_SCHED, &n->state);
3368 EXPORT_SYMBOL(__napi_complete);
3370 void napi_complete(struct napi_struct *n)
3372 unsigned long flags;
3375 * don't let napi dequeue from the cpu poll list
3376 * just in case its running on a different cpu
3378 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3382 local_irq_save(flags);
3384 local_irq_restore(flags);
3386 EXPORT_SYMBOL(napi_complete);
3388 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3389 int (*poll)(struct napi_struct *, int), int weight)
3391 INIT_LIST_HEAD(&napi->poll_list);
3392 napi->gro_count = 0;
3393 napi->gro_list = NULL;
3396 napi->weight = weight;
3397 list_add(&napi->dev_list, &dev->napi_list);
3399 #ifdef CONFIG_NETPOLL
3400 spin_lock_init(&napi->poll_lock);
3401 napi->poll_owner = -1;
3403 set_bit(NAPI_STATE_SCHED, &napi->state);
3405 EXPORT_SYMBOL(netif_napi_add);
3407 void netif_napi_del(struct napi_struct *napi)
3409 struct sk_buff *skb, *next;
3411 list_del_init(&napi->dev_list);
3412 napi_free_frags(napi);
3414 for (skb = napi->gro_list; skb; skb = next) {
3420 napi->gro_list = NULL;
3421 napi->gro_count = 0;
3423 EXPORT_SYMBOL(netif_napi_del);
3425 static void net_rx_action(struct softirq_action *h)
3427 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3428 unsigned long time_limit = jiffies + 2;
3429 int budget = netdev_budget;
3432 local_irq_disable();
3434 while (!list_empty(&sd->poll_list)) {
3435 struct napi_struct *n;
3438 /* If softirq window is exhuasted then punt.
3439 * Allow this to run for 2 jiffies since which will allow
3440 * an average latency of 1.5/HZ.
3442 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3447 /* Even though interrupts have been re-enabled, this
3448 * access is safe because interrupts can only add new
3449 * entries to the tail of this list, and only ->poll()
3450 * calls can remove this head entry from the list.
3452 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3454 have = netpoll_poll_lock(n);
3458 /* This NAPI_STATE_SCHED test is for avoiding a race
3459 * with netpoll's poll_napi(). Only the entity which
3460 * obtains the lock and sees NAPI_STATE_SCHED set will
3461 * actually make the ->poll() call. Therefore we avoid
3462 * accidently calling ->poll() when NAPI is not scheduled.
3465 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3466 work = n->poll(n, weight);
3470 WARN_ON_ONCE(work > weight);
3474 local_irq_disable();
3476 /* Drivers must not modify the NAPI state if they
3477 * consume the entire weight. In such cases this code
3478 * still "owns" the NAPI instance and therefore can
3479 * move the instance around on the list at-will.
3481 if (unlikely(work == weight)) {
3482 if (unlikely(napi_disable_pending(n))) {
3485 local_irq_disable();
3487 list_move_tail(&n->poll_list, &sd->poll_list);
3490 netpoll_poll_unlock(have);
3493 net_rps_action_and_irq_enable(sd);
3495 #ifdef CONFIG_NET_DMA
3497 * There may not be any more sk_buffs coming right now, so push
3498 * any pending DMA copies to hardware
3500 dma_issue_pending_all();
3507 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3511 static gifconf_func_t *gifconf_list[NPROTO];
3514 * register_gifconf - register a SIOCGIF handler
3515 * @family: Address family
3516 * @gifconf: Function handler
3518 * Register protocol dependent address dumping routines. The handler
3519 * that is passed must not be freed or reused until it has been replaced
3520 * by another handler.
3522 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3524 if (family >= NPROTO)
3526 gifconf_list[family] = gifconf;
3529 EXPORT_SYMBOL(register_gifconf);
3533 * Map an interface index to its name (SIOCGIFNAME)
3537 * We need this ioctl for efficient implementation of the
3538 * if_indextoname() function required by the IPv6 API. Without
3539 * it, we would have to search all the interfaces to find a
3543 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3545 struct net_device *dev;
3549 * Fetch the caller's info block.
3552 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3556 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3562 strcpy(ifr.ifr_name, dev->name);
3565 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3571 * Perform a SIOCGIFCONF call. This structure will change
3572 * size eventually, and there is nothing I can do about it.
3573 * Thus we will need a 'compatibility mode'.
3576 static int dev_ifconf(struct net *net, char __user *arg)
3579 struct net_device *dev;
3586 * Fetch the caller's info block.
3589 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3596 * Loop over the interfaces, and write an info block for each.
3600 for_each_netdev(net, dev) {
3601 for (i = 0; i < NPROTO; i++) {
3602 if (gifconf_list[i]) {
3605 done = gifconf_list[i](dev, NULL, 0);
3607 done = gifconf_list[i](dev, pos + total,
3617 * All done. Write the updated control block back to the caller.
3619 ifc.ifc_len = total;
3622 * Both BSD and Solaris return 0 here, so we do too.
3624 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3627 #ifdef CONFIG_PROC_FS
3629 * This is invoked by the /proc filesystem handler to display a device
3632 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3635 struct net *net = seq_file_net(seq);
3637 struct net_device *dev;
3641 return SEQ_START_TOKEN;
3644 for_each_netdev_rcu(net, dev)
3651 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3653 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3654 first_net_device(seq_file_net(seq)) :
3655 next_net_device((struct net_device *)v);
3658 return rcu_dereference(dev);
3661 void dev_seq_stop(struct seq_file *seq, void *v)
3667 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3669 const struct net_device_stats *stats = dev_get_stats(dev);
3671 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3672 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3673 dev->name, stats->rx_bytes, stats->rx_packets,
3675 stats->rx_dropped + stats->rx_missed_errors,
3676 stats->rx_fifo_errors,
3677 stats->rx_length_errors + stats->rx_over_errors +
3678 stats->rx_crc_errors + stats->rx_frame_errors,
3679 stats->rx_compressed, stats->multicast,
3680 stats->tx_bytes, stats->tx_packets,
3681 stats->tx_errors, stats->tx_dropped,
3682 stats->tx_fifo_errors, stats->collisions,
3683 stats->tx_carrier_errors +
3684 stats->tx_aborted_errors +
3685 stats->tx_window_errors +
3686 stats->tx_heartbeat_errors,
3687 stats->tx_compressed);
3691 * Called from the PROCfs module. This now uses the new arbitrary sized
3692 * /proc/net interface to create /proc/net/dev
3694 static int dev_seq_show(struct seq_file *seq, void *v)
3696 if (v == SEQ_START_TOKEN)
3697 seq_puts(seq, "Inter-| Receive "
3699 " face |bytes packets errs drop fifo frame "
3700 "compressed multicast|bytes packets errs "
3701 "drop fifo colls carrier compressed\n");
3703 dev_seq_printf_stats(seq, v);
3707 static struct softnet_data *softnet_get_online(loff_t *pos)
3709 struct softnet_data *sd = NULL;
3711 while (*pos < nr_cpu_ids)
3712 if (cpu_online(*pos)) {
3713 sd = &per_cpu(softnet_data, *pos);
3720 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3722 return softnet_get_online(pos);
3725 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3728 return softnet_get_online(pos);
3731 static void softnet_seq_stop(struct seq_file *seq, void *v)
3735 static int softnet_seq_show(struct seq_file *seq, void *v)
3737 struct softnet_data *sd = v;
3739 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3740 sd->processed, sd->dropped, sd->time_squeeze, 0,
3741 0, 0, 0, 0, /* was fastroute */
3742 sd->cpu_collision, sd->received_rps);
3746 static const struct seq_operations dev_seq_ops = {
3747 .start = dev_seq_start,
3748 .next = dev_seq_next,
3749 .stop = dev_seq_stop,
3750 .show = dev_seq_show,
3753 static int dev_seq_open(struct inode *inode, struct file *file)
3755 return seq_open_net(inode, file, &dev_seq_ops,
3756 sizeof(struct seq_net_private));
3759 static const struct file_operations dev_seq_fops = {
3760 .owner = THIS_MODULE,
3761 .open = dev_seq_open,
3763 .llseek = seq_lseek,
3764 .release = seq_release_net,
3767 static const struct seq_operations softnet_seq_ops = {
3768 .start = softnet_seq_start,
3769 .next = softnet_seq_next,
3770 .stop = softnet_seq_stop,
3771 .show = softnet_seq_show,
3774 static int softnet_seq_open(struct inode *inode, struct file *file)
3776 return seq_open(file, &softnet_seq_ops);
3779 static const struct file_operations softnet_seq_fops = {
3780 .owner = THIS_MODULE,
3781 .open = softnet_seq_open,
3783 .llseek = seq_lseek,
3784 .release = seq_release,
3787 static void *ptype_get_idx(loff_t pos)
3789 struct packet_type *pt = NULL;
3793 list_for_each_entry_rcu(pt, &ptype_all, list) {
3799 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3800 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3809 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3813 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3816 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3818 struct packet_type *pt;
3819 struct list_head *nxt;
3823 if (v == SEQ_START_TOKEN)
3824 return ptype_get_idx(0);
3827 nxt = pt->list.next;
3828 if (pt->type == htons(ETH_P_ALL)) {
3829 if (nxt != &ptype_all)
3832 nxt = ptype_base[0].next;
3834 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3836 while (nxt == &ptype_base[hash]) {
3837 if (++hash >= PTYPE_HASH_SIZE)
3839 nxt = ptype_base[hash].next;
3842 return list_entry(nxt, struct packet_type, list);
3845 static void ptype_seq_stop(struct seq_file *seq, void *v)
3851 static int ptype_seq_show(struct seq_file *seq, void *v)
3853 struct packet_type *pt = v;
3855 if (v == SEQ_START_TOKEN)
3856 seq_puts(seq, "Type Device Function\n");
3857 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3858 if (pt->type == htons(ETH_P_ALL))
3859 seq_puts(seq, "ALL ");
3861 seq_printf(seq, "%04x", ntohs(pt->type));
3863 seq_printf(seq, " %-8s %pF\n",
3864 pt->dev ? pt->dev->name : "", pt->func);
3870 static const struct seq_operations ptype_seq_ops = {
3871 .start = ptype_seq_start,
3872 .next = ptype_seq_next,
3873 .stop = ptype_seq_stop,
3874 .show = ptype_seq_show,
3877 static int ptype_seq_open(struct inode *inode, struct file *file)
3879 return seq_open_net(inode, file, &ptype_seq_ops,
3880 sizeof(struct seq_net_private));
3883 static const struct file_operations ptype_seq_fops = {
3884 .owner = THIS_MODULE,
3885 .open = ptype_seq_open,
3887 .llseek = seq_lseek,
3888 .release = seq_release_net,
3892 static int __net_init dev_proc_net_init(struct net *net)
3896 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3898 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3900 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3903 if (wext_proc_init(net))
3909 proc_net_remove(net, "ptype");
3911 proc_net_remove(net, "softnet_stat");
3913 proc_net_remove(net, "dev");
3917 static void __net_exit dev_proc_net_exit(struct net *net)
3919 wext_proc_exit(net);
3921 proc_net_remove(net, "ptype");
3922 proc_net_remove(net, "softnet_stat");
3923 proc_net_remove(net, "dev");
3926 static struct pernet_operations __net_initdata dev_proc_ops = {
3927 .init = dev_proc_net_init,
3928 .exit = dev_proc_net_exit,
3931 static int __init dev_proc_init(void)
3933 return register_pernet_subsys(&dev_proc_ops);
3936 #define dev_proc_init() 0
3937 #endif /* CONFIG_PROC_FS */
3941 * netdev_set_master - set up master/slave pair
3942 * @slave: slave device
3943 * @master: new master device
3945 * Changes the master device of the slave. Pass %NULL to break the
3946 * bonding. The caller must hold the RTNL semaphore. On a failure
3947 * a negative errno code is returned. On success the reference counts
3948 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3949 * function returns zero.
3951 int netdev_set_master(struct net_device *slave, struct net_device *master)
3953 struct net_device *old = slave->master;
3963 slave->master = master;
3970 slave->flags |= IFF_SLAVE;
3972 slave->flags &= ~IFF_SLAVE;
3974 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3977 EXPORT_SYMBOL(netdev_set_master);
3979 static void dev_change_rx_flags(struct net_device *dev, int flags)
3981 const struct net_device_ops *ops = dev->netdev_ops;
3983 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3984 ops->ndo_change_rx_flags(dev, flags);
3987 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3989 unsigned short old_flags = dev->flags;
3995 dev->flags |= IFF_PROMISC;
3996 dev->promiscuity += inc;
3997 if (dev->promiscuity == 0) {
4000 * If inc causes overflow, untouch promisc and return error.
4003 dev->flags &= ~IFF_PROMISC;
4005 dev->promiscuity -= inc;
4006 printk(KERN_WARNING "%s: promiscuity touches roof, "
4007 "set promiscuity failed, promiscuity feature "
4008 "of device might be broken.\n", dev->name);
4012 if (dev->flags != old_flags) {
4013 printk(KERN_INFO "device %s %s promiscuous mode\n",
4014 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4016 if (audit_enabled) {
4017 current_uid_gid(&uid, &gid);
4018 audit_log(current->audit_context, GFP_ATOMIC,
4019 AUDIT_ANOM_PROMISCUOUS,
4020 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4021 dev->name, (dev->flags & IFF_PROMISC),
4022 (old_flags & IFF_PROMISC),
4023 audit_get_loginuid(current),
4025 audit_get_sessionid(current));
4028 dev_change_rx_flags(dev, IFF_PROMISC);
4034 * dev_set_promiscuity - update promiscuity count on a device
4038 * Add or remove promiscuity from a device. While the count in the device
4039 * remains above zero the interface remains promiscuous. Once it hits zero
4040 * the device reverts back to normal filtering operation. A negative inc
4041 * value is used to drop promiscuity on the device.
4042 * Return 0 if successful or a negative errno code on error.
4044 int dev_set_promiscuity(struct net_device *dev, int inc)
4046 unsigned short old_flags = dev->flags;
4049 err = __dev_set_promiscuity(dev, inc);
4052 if (dev->flags != old_flags)
4053 dev_set_rx_mode(dev);
4056 EXPORT_SYMBOL(dev_set_promiscuity);
4059 * dev_set_allmulti - update allmulti count on a device
4063 * Add or remove reception of all multicast frames to a device. While the
4064 * count in the device remains above zero the interface remains listening
4065 * to all interfaces. Once it hits zero the device reverts back to normal
4066 * filtering operation. A negative @inc value is used to drop the counter
4067 * when releasing a resource needing all multicasts.
4068 * Return 0 if successful or a negative errno code on error.
4071 int dev_set_allmulti(struct net_device *dev, int inc)
4073 unsigned short old_flags = dev->flags;
4077 dev->flags |= IFF_ALLMULTI;
4078 dev->allmulti += inc;
4079 if (dev->allmulti == 0) {
4082 * If inc causes overflow, untouch allmulti and return error.
4085 dev->flags &= ~IFF_ALLMULTI;
4087 dev->allmulti -= inc;
4088 printk(KERN_WARNING "%s: allmulti touches roof, "
4089 "set allmulti failed, allmulti feature of "
4090 "device might be broken.\n", dev->name);
4094 if (dev->flags ^ old_flags) {
4095 dev_change_rx_flags(dev, IFF_ALLMULTI);
4096 dev_set_rx_mode(dev);
4100 EXPORT_SYMBOL(dev_set_allmulti);
4103 * Upload unicast and multicast address lists to device and
4104 * configure RX filtering. When the device doesn't support unicast
4105 * filtering it is put in promiscuous mode while unicast addresses
4108 void __dev_set_rx_mode(struct net_device *dev)
4110 const struct net_device_ops *ops = dev->netdev_ops;
4112 /* dev_open will call this function so the list will stay sane. */
4113 if (!(dev->flags&IFF_UP))
4116 if (!netif_device_present(dev))
4119 if (ops->ndo_set_rx_mode)
4120 ops->ndo_set_rx_mode(dev);
4122 /* Unicast addresses changes may only happen under the rtnl,
4123 * therefore calling __dev_set_promiscuity here is safe.
4125 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4126 __dev_set_promiscuity(dev, 1);
4127 dev->uc_promisc = 1;
4128 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4129 __dev_set_promiscuity(dev, -1);
4130 dev->uc_promisc = 0;
4133 if (ops->ndo_set_multicast_list)
4134 ops->ndo_set_multicast_list(dev);
4138 void dev_set_rx_mode(struct net_device *dev)
4140 netif_addr_lock_bh(dev);
4141 __dev_set_rx_mode(dev);
4142 netif_addr_unlock_bh(dev);
4146 * dev_get_flags - get flags reported to userspace
4149 * Get the combination of flag bits exported through APIs to userspace.
4151 unsigned dev_get_flags(const struct net_device *dev)
4155 flags = (dev->flags & ~(IFF_PROMISC |
4160 (dev->gflags & (IFF_PROMISC |
4163 if (netif_running(dev)) {
4164 if (netif_oper_up(dev))
4165 flags |= IFF_RUNNING;
4166 if (netif_carrier_ok(dev))
4167 flags |= IFF_LOWER_UP;
4168 if (netif_dormant(dev))
4169 flags |= IFF_DORMANT;
4174 EXPORT_SYMBOL(dev_get_flags);
4176 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4178 int old_flags = dev->flags;
4184 * Set the flags on our device.
4187 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4188 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4190 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4194 * Load in the correct multicast list now the flags have changed.
4197 if ((old_flags ^ flags) & IFF_MULTICAST)
4198 dev_change_rx_flags(dev, IFF_MULTICAST);
4200 dev_set_rx_mode(dev);
4203 * Have we downed the interface. We handle IFF_UP ourselves
4204 * according to user attempts to set it, rather than blindly
4209 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4210 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4213 dev_set_rx_mode(dev);
4216 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4217 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4219 dev->gflags ^= IFF_PROMISC;
4220 dev_set_promiscuity(dev, inc);
4223 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4224 is important. Some (broken) drivers set IFF_PROMISC, when
4225 IFF_ALLMULTI is requested not asking us and not reporting.
4227 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4228 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4230 dev->gflags ^= IFF_ALLMULTI;
4231 dev_set_allmulti(dev, inc);
4237 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4239 unsigned int changes = dev->flags ^ old_flags;
4241 if (changes & IFF_UP) {
4242 if (dev->flags & IFF_UP)
4243 call_netdevice_notifiers(NETDEV_UP, dev);
4245 call_netdevice_notifiers(NETDEV_DOWN, dev);
4248 if (dev->flags & IFF_UP &&
4249 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4250 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4254 * dev_change_flags - change device settings
4256 * @flags: device state flags
4258 * Change settings on device based state flags. The flags are
4259 * in the userspace exported format.
4261 int dev_change_flags(struct net_device *dev, unsigned flags)
4264 int old_flags = dev->flags;
4266 ret = __dev_change_flags(dev, flags);
4270 changes = old_flags ^ dev->flags;
4272 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4274 __dev_notify_flags(dev, old_flags);
4277 EXPORT_SYMBOL(dev_change_flags);
4280 * dev_set_mtu - Change maximum transfer unit
4282 * @new_mtu: new transfer unit
4284 * Change the maximum transfer size of the network device.
4286 int dev_set_mtu(struct net_device *dev, int new_mtu)
4288 const struct net_device_ops *ops = dev->netdev_ops;
4291 if (new_mtu == dev->mtu)
4294 /* MTU must be positive. */
4298 if (!netif_device_present(dev))
4302 if (ops->ndo_change_mtu)
4303 err = ops->ndo_change_mtu(dev, new_mtu);
4307 if (!err && dev->flags & IFF_UP)
4308 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4311 EXPORT_SYMBOL(dev_set_mtu);
4314 * dev_set_mac_address - Change Media Access Control Address
4318 * Change the hardware (MAC) address of the device
4320 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4322 const struct net_device_ops *ops = dev->netdev_ops;
4325 if (!ops->ndo_set_mac_address)
4327 if (sa->sa_family != dev->type)
4329 if (!netif_device_present(dev))
4331 err = ops->ndo_set_mac_address(dev, sa);
4333 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4336 EXPORT_SYMBOL(dev_set_mac_address);
4339 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4341 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4344 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4350 case SIOCGIFFLAGS: /* Get interface flags */
4351 ifr->ifr_flags = (short) dev_get_flags(dev);
4354 case SIOCGIFMETRIC: /* Get the metric on the interface
4355 (currently unused) */
4356 ifr->ifr_metric = 0;
4359 case SIOCGIFMTU: /* Get the MTU of a device */
4360 ifr->ifr_mtu = dev->mtu;
4365 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4367 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4368 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4369 ifr->ifr_hwaddr.sa_family = dev->type;
4377 ifr->ifr_map.mem_start = dev->mem_start;
4378 ifr->ifr_map.mem_end = dev->mem_end;
4379 ifr->ifr_map.base_addr = dev->base_addr;
4380 ifr->ifr_map.irq = dev->irq;
4381 ifr->ifr_map.dma = dev->dma;
4382 ifr->ifr_map.port = dev->if_port;
4386 ifr->ifr_ifindex = dev->ifindex;
4390 ifr->ifr_qlen = dev->tx_queue_len;
4394 /* dev_ioctl() should ensure this case
4406 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4408 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4411 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4412 const struct net_device_ops *ops;
4417 ops = dev->netdev_ops;
4420 case SIOCSIFFLAGS: /* Set interface flags */
4421 return dev_change_flags(dev, ifr->ifr_flags);
4423 case SIOCSIFMETRIC: /* Set the metric on the interface
4424 (currently unused) */
4427 case SIOCSIFMTU: /* Set the MTU of a device */
4428 return dev_set_mtu(dev, ifr->ifr_mtu);
4431 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4433 case SIOCSIFHWBROADCAST:
4434 if (ifr->ifr_hwaddr.sa_family != dev->type)
4436 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4437 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4438 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4442 if (ops->ndo_set_config) {
4443 if (!netif_device_present(dev))
4445 return ops->ndo_set_config(dev, &ifr->ifr_map);
4450 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4451 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4453 if (!netif_device_present(dev))
4455 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4458 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4459 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4461 if (!netif_device_present(dev))
4463 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4466 if (ifr->ifr_qlen < 0)
4468 dev->tx_queue_len = ifr->ifr_qlen;
4472 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4473 return dev_change_name(dev, ifr->ifr_newname);
4476 * Unknown or private ioctl
4479 if ((cmd >= SIOCDEVPRIVATE &&
4480 cmd <= SIOCDEVPRIVATE + 15) ||
4481 cmd == SIOCBONDENSLAVE ||
4482 cmd == SIOCBONDRELEASE ||
4483 cmd == SIOCBONDSETHWADDR ||
4484 cmd == SIOCBONDSLAVEINFOQUERY ||
4485 cmd == SIOCBONDINFOQUERY ||
4486 cmd == SIOCBONDCHANGEACTIVE ||
4487 cmd == SIOCGMIIPHY ||
4488 cmd == SIOCGMIIREG ||
4489 cmd == SIOCSMIIREG ||
4490 cmd == SIOCBRADDIF ||
4491 cmd == SIOCBRDELIF ||
4492 cmd == SIOCSHWTSTAMP ||
4493 cmd == SIOCWANDEV) {
4495 if (ops->ndo_do_ioctl) {
4496 if (netif_device_present(dev))
4497 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4509 * This function handles all "interface"-type I/O control requests. The actual
4510 * 'doing' part of this is dev_ifsioc above.
4514 * dev_ioctl - network device ioctl
4515 * @net: the applicable net namespace
4516 * @cmd: command to issue
4517 * @arg: pointer to a struct ifreq in user space
4519 * Issue ioctl functions to devices. This is normally called by the
4520 * user space syscall interfaces but can sometimes be useful for
4521 * other purposes. The return value is the return from the syscall if
4522 * positive or a negative errno code on error.
4525 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4531 /* One special case: SIOCGIFCONF takes ifconf argument
4532 and requires shared lock, because it sleeps writing
4536 if (cmd == SIOCGIFCONF) {
4538 ret = dev_ifconf(net, (char __user *) arg);
4542 if (cmd == SIOCGIFNAME)
4543 return dev_ifname(net, (struct ifreq __user *)arg);
4545 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4548 ifr.ifr_name[IFNAMSIZ-1] = 0;
4550 colon = strchr(ifr.ifr_name, ':');
4555 * See which interface the caller is talking about.
4560 * These ioctl calls:
4561 * - can be done by all.
4562 * - atomic and do not require locking.
4573 dev_load(net, ifr.ifr_name);
4575 ret = dev_ifsioc_locked(net, &ifr, cmd);
4580 if (copy_to_user(arg, &ifr,
4581 sizeof(struct ifreq)))
4587 dev_load(net, ifr.ifr_name);
4589 ret = dev_ethtool(net, &ifr);
4594 if (copy_to_user(arg, &ifr,
4595 sizeof(struct ifreq)))
4601 * These ioctl calls:
4602 * - require superuser power.
4603 * - require strict serialization.
4609 if (!capable(CAP_NET_ADMIN))
4611 dev_load(net, ifr.ifr_name);
4613 ret = dev_ifsioc(net, &ifr, cmd);
4618 if (copy_to_user(arg, &ifr,
4619 sizeof(struct ifreq)))
4625 * These ioctl calls:
4626 * - require superuser power.
4627 * - require strict serialization.
4628 * - do not return a value
4638 case SIOCSIFHWBROADCAST:
4641 case SIOCBONDENSLAVE:
4642 case SIOCBONDRELEASE:
4643 case SIOCBONDSETHWADDR:
4644 case SIOCBONDCHANGEACTIVE:
4648 if (!capable(CAP_NET_ADMIN))
4651 case SIOCBONDSLAVEINFOQUERY:
4652 case SIOCBONDINFOQUERY:
4653 dev_load(net, ifr.ifr_name);
4655 ret = dev_ifsioc(net, &ifr, cmd);
4660 /* Get the per device memory space. We can add this but
4661 * currently do not support it */
4663 /* Set the per device memory buffer space.
4664 * Not applicable in our case */
4669 * Unknown or private ioctl.
4672 if (cmd == SIOCWANDEV ||
4673 (cmd >= SIOCDEVPRIVATE &&
4674 cmd <= SIOCDEVPRIVATE + 15)) {
4675 dev_load(net, ifr.ifr_name);
4677 ret = dev_ifsioc(net, &ifr, cmd);
4679 if (!ret && copy_to_user(arg, &ifr,
4680 sizeof(struct ifreq)))
4684 /* Take care of Wireless Extensions */
4685 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4686 return wext_handle_ioctl(net, &ifr, cmd, arg);
4693 * dev_new_index - allocate an ifindex
4694 * @net: the applicable net namespace
4696 * Returns a suitable unique value for a new device interface
4697 * number. The caller must hold the rtnl semaphore or the
4698 * dev_base_lock to be sure it remains unique.
4700 static int dev_new_index(struct net *net)
4706 if (!__dev_get_by_index(net, ifindex))
4711 /* Delayed registration/unregisteration */
4712 static LIST_HEAD(net_todo_list);
4714 static void net_set_todo(struct net_device *dev)
4716 list_add_tail(&dev->todo_list, &net_todo_list);
4719 static void rollback_registered_many(struct list_head *head)
4721 struct net_device *dev, *tmp;
4723 BUG_ON(dev_boot_phase);
4726 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4727 /* Some devices call without registering
4728 * for initialization unwind. Remove those
4729 * devices and proceed with the remaining.
4731 if (dev->reg_state == NETREG_UNINITIALIZED) {
4732 pr_debug("unregister_netdevice: device %s/%p never "
4733 "was registered\n", dev->name, dev);
4736 list_del(&dev->unreg_list);
4740 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4742 /* If device is running, close it first. */
4745 /* And unlink it from device chain. */
4746 unlist_netdevice(dev);
4748 dev->reg_state = NETREG_UNREGISTERING;
4753 list_for_each_entry(dev, head, unreg_list) {
4754 /* Shutdown queueing discipline. */
4758 /* Notify protocols, that we are about to destroy
4759 this device. They should clean all the things.
4761 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4763 if (!dev->rtnl_link_ops ||
4764 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4765 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4768 * Flush the unicast and multicast chains
4773 if (dev->netdev_ops->ndo_uninit)
4774 dev->netdev_ops->ndo_uninit(dev);
4776 /* Notifier chain MUST detach us from master device. */
4777 WARN_ON(dev->master);
4779 /* Remove entries from kobject tree */
4780 netdev_unregister_kobject(dev);
4783 /* Process any work delayed until the end of the batch */
4784 dev = list_first_entry(head, struct net_device, unreg_list);
4785 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4789 list_for_each_entry(dev, head, unreg_list)
4793 static void rollback_registered(struct net_device *dev)
4797 list_add(&dev->unreg_list, &single);
4798 rollback_registered_many(&single);
4801 static void __netdev_init_queue_locks_one(struct net_device *dev,
4802 struct netdev_queue *dev_queue,
4805 spin_lock_init(&dev_queue->_xmit_lock);
4806 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4807 dev_queue->xmit_lock_owner = -1;
4810 static void netdev_init_queue_locks(struct net_device *dev)
4812 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4813 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4816 unsigned long netdev_fix_features(unsigned long features, const char *name)
4818 /* Fix illegal SG+CSUM combinations. */
4819 if ((features & NETIF_F_SG) &&
4820 !(features & NETIF_F_ALL_CSUM)) {
4822 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4823 "checksum feature.\n", name);
4824 features &= ~NETIF_F_SG;
4827 /* TSO requires that SG is present as well. */
4828 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4830 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4831 "SG feature.\n", name);
4832 features &= ~NETIF_F_TSO;
4835 if (features & NETIF_F_UFO) {
4836 if (!(features & NETIF_F_GEN_CSUM)) {
4838 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4839 "since no NETIF_F_HW_CSUM feature.\n",
4841 features &= ~NETIF_F_UFO;
4844 if (!(features & NETIF_F_SG)) {
4846 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4847 "since no NETIF_F_SG feature.\n", name);
4848 features &= ~NETIF_F_UFO;
4854 EXPORT_SYMBOL(netdev_fix_features);
4857 * netif_stacked_transfer_operstate - transfer operstate
4858 * @rootdev: the root or lower level device to transfer state from
4859 * @dev: the device to transfer operstate to
4861 * Transfer operational state from root to device. This is normally
4862 * called when a stacking relationship exists between the root
4863 * device and the device(a leaf device).
4865 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4866 struct net_device *dev)
4868 if (rootdev->operstate == IF_OPER_DORMANT)
4869 netif_dormant_on(dev);
4871 netif_dormant_off(dev);
4873 if (netif_carrier_ok(rootdev)) {
4874 if (!netif_carrier_ok(dev))
4875 netif_carrier_on(dev);
4877 if (netif_carrier_ok(dev))
4878 netif_carrier_off(dev);
4881 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4884 * register_netdevice - register a network device
4885 * @dev: device to register
4887 * Take a completed network device structure and add it to the kernel
4888 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4889 * chain. 0 is returned on success. A negative errno code is returned
4890 * on a failure to set up the device, or if the name is a duplicate.
4892 * Callers must hold the rtnl semaphore. You may want
4893 * register_netdev() instead of this.
4896 * The locking appears insufficient to guarantee two parallel registers
4897 * will not get the same name.
4900 int register_netdevice(struct net_device *dev)
4903 struct net *net = dev_net(dev);
4905 BUG_ON(dev_boot_phase);
4910 /* When net_device's are persistent, this will be fatal. */
4911 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4914 spin_lock_init(&dev->addr_list_lock);
4915 netdev_set_addr_lockdep_class(dev);
4916 netdev_init_queue_locks(dev);
4921 if (!dev->num_rx_queues) {
4923 * Allocate a single RX queue if driver never called
4927 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
4933 dev->_rx->first = dev->_rx;
4934 atomic_set(&dev->_rx->count, 1);
4935 dev->num_rx_queues = 1;
4938 /* Init, if this function is available */
4939 if (dev->netdev_ops->ndo_init) {
4940 ret = dev->netdev_ops->ndo_init(dev);
4948 ret = dev_get_valid_name(net, dev->name, dev->name, 0);
4952 dev->ifindex = dev_new_index(net);
4953 if (dev->iflink == -1)
4954 dev->iflink = dev->ifindex;
4956 /* Fix illegal checksum combinations */
4957 if ((dev->features & NETIF_F_HW_CSUM) &&
4958 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4959 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4961 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4964 if ((dev->features & NETIF_F_NO_CSUM) &&
4965 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4966 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4968 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4971 dev->features = netdev_fix_features(dev->features, dev->name);
4973 /* Enable software GSO if SG is supported. */
4974 if (dev->features & NETIF_F_SG)
4975 dev->features |= NETIF_F_GSO;
4977 netdev_initialize_kobject(dev);
4979 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4980 ret = notifier_to_errno(ret);
4984 ret = netdev_register_kobject(dev);
4987 dev->reg_state = NETREG_REGISTERED;
4990 * Default initial state at registry is that the
4991 * device is present.
4994 set_bit(__LINK_STATE_PRESENT, &dev->state);
4996 dev_init_scheduler(dev);
4998 list_netdevice(dev);
5000 /* Notify protocols, that a new device appeared. */
5001 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5002 ret = notifier_to_errno(ret);
5004 rollback_registered(dev);
5005 dev->reg_state = NETREG_UNREGISTERED;
5008 * Prevent userspace races by waiting until the network
5009 * device is fully setup before sending notifications.
5011 if (!dev->rtnl_link_ops ||
5012 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5013 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5019 if (dev->netdev_ops->ndo_uninit)
5020 dev->netdev_ops->ndo_uninit(dev);
5023 EXPORT_SYMBOL(register_netdevice);
5026 * init_dummy_netdev - init a dummy network device for NAPI
5027 * @dev: device to init
5029 * This takes a network device structure and initialize the minimum
5030 * amount of fields so it can be used to schedule NAPI polls without
5031 * registering a full blown interface. This is to be used by drivers
5032 * that need to tie several hardware interfaces to a single NAPI
5033 * poll scheduler due to HW limitations.
5035 int init_dummy_netdev(struct net_device *dev)
5037 /* Clear everything. Note we don't initialize spinlocks
5038 * are they aren't supposed to be taken by any of the
5039 * NAPI code and this dummy netdev is supposed to be
5040 * only ever used for NAPI polls
5042 memset(dev, 0, sizeof(struct net_device));
5044 /* make sure we BUG if trying to hit standard
5045 * register/unregister code path
5047 dev->reg_state = NETREG_DUMMY;
5049 /* initialize the ref count */
5050 atomic_set(&dev->refcnt, 1);
5052 /* NAPI wants this */
5053 INIT_LIST_HEAD(&dev->napi_list);
5055 /* a dummy interface is started by default */
5056 set_bit(__LINK_STATE_PRESENT, &dev->state);
5057 set_bit(__LINK_STATE_START, &dev->state);
5061 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5065 * register_netdev - register a network device
5066 * @dev: device to register
5068 * Take a completed network device structure and add it to the kernel
5069 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5070 * chain. 0 is returned on success. A negative errno code is returned
5071 * on a failure to set up the device, or if the name is a duplicate.
5073 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5074 * and expands the device name if you passed a format string to
5077 int register_netdev(struct net_device *dev)
5084 * If the name is a format string the caller wants us to do a
5087 if (strchr(dev->name, '%')) {
5088 err = dev_alloc_name(dev, dev->name);
5093 err = register_netdevice(dev);
5098 EXPORT_SYMBOL(register_netdev);
5101 * netdev_wait_allrefs - wait until all references are gone.
5103 * This is called when unregistering network devices.
5105 * Any protocol or device that holds a reference should register
5106 * for netdevice notification, and cleanup and put back the
5107 * reference if they receive an UNREGISTER event.
5108 * We can get stuck here if buggy protocols don't correctly
5111 static void netdev_wait_allrefs(struct net_device *dev)
5113 unsigned long rebroadcast_time, warning_time;
5115 linkwatch_forget_dev(dev);
5117 rebroadcast_time = warning_time = jiffies;
5118 while (atomic_read(&dev->refcnt) != 0) {
5119 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5122 /* Rebroadcast unregister notification */
5123 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5124 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5125 * should have already handle it the first time */
5127 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5129 /* We must not have linkwatch events
5130 * pending on unregister. If this
5131 * happens, we simply run the queue
5132 * unscheduled, resulting in a noop
5135 linkwatch_run_queue();
5140 rebroadcast_time = jiffies;
5145 if (time_after(jiffies, warning_time + 10 * HZ)) {
5146 printk(KERN_EMERG "unregister_netdevice: "
5147 "waiting for %s to become free. Usage "
5149 dev->name, atomic_read(&dev->refcnt));
5150 warning_time = jiffies;
5159 * register_netdevice(x1);
5160 * register_netdevice(x2);
5162 * unregister_netdevice(y1);
5163 * unregister_netdevice(y2);
5169 * We are invoked by rtnl_unlock().
5170 * This allows us to deal with problems:
5171 * 1) We can delete sysfs objects which invoke hotplug
5172 * without deadlocking with linkwatch via keventd.
5173 * 2) Since we run with the RTNL semaphore not held, we can sleep
5174 * safely in order to wait for the netdev refcnt to drop to zero.
5176 * We must not return until all unregister events added during
5177 * the interval the lock was held have been completed.
5179 void netdev_run_todo(void)
5181 struct list_head list;
5183 /* Snapshot list, allow later requests */
5184 list_replace_init(&net_todo_list, &list);
5188 while (!list_empty(&list)) {
5189 struct net_device *dev
5190 = list_first_entry(&list, struct net_device, todo_list);
5191 list_del(&dev->todo_list);
5193 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5194 printk(KERN_ERR "network todo '%s' but state %d\n",
5195 dev->name, dev->reg_state);
5200 dev->reg_state = NETREG_UNREGISTERED;
5202 on_each_cpu(flush_backlog, dev, 1);
5204 netdev_wait_allrefs(dev);
5207 BUG_ON(atomic_read(&dev->refcnt));
5208 WARN_ON(dev->ip_ptr);
5209 WARN_ON(dev->ip6_ptr);
5210 WARN_ON(dev->dn_ptr);
5212 if (dev->destructor)
5213 dev->destructor(dev);
5215 /* Free network device */
5216 kobject_put(&dev->dev.kobj);
5221 * dev_txq_stats_fold - fold tx_queues stats
5222 * @dev: device to get statistics from
5223 * @stats: struct net_device_stats to hold results
5225 void dev_txq_stats_fold(const struct net_device *dev,
5226 struct net_device_stats *stats)
5228 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5230 struct netdev_queue *txq;
5232 for (i = 0; i < dev->num_tx_queues; i++) {
5233 txq = netdev_get_tx_queue(dev, i);
5234 tx_bytes += txq->tx_bytes;
5235 tx_packets += txq->tx_packets;
5236 tx_dropped += txq->tx_dropped;
5238 if (tx_bytes || tx_packets || tx_dropped) {
5239 stats->tx_bytes = tx_bytes;
5240 stats->tx_packets = tx_packets;
5241 stats->tx_dropped = tx_dropped;
5244 EXPORT_SYMBOL(dev_txq_stats_fold);
5247 * dev_get_stats - get network device statistics
5248 * @dev: device to get statistics from
5250 * Get network statistics from device. The device driver may provide
5251 * its own method by setting dev->netdev_ops->get_stats; otherwise
5252 * the internal statistics structure is used.
5254 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5256 const struct net_device_ops *ops = dev->netdev_ops;
5258 if (ops->ndo_get_stats)
5259 return ops->ndo_get_stats(dev);
5261 dev_txq_stats_fold(dev, &dev->stats);
5264 EXPORT_SYMBOL(dev_get_stats);
5266 static void netdev_init_one_queue(struct net_device *dev,
5267 struct netdev_queue *queue,
5273 static void netdev_init_queues(struct net_device *dev)
5275 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5276 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5277 spin_lock_init(&dev->tx_global_lock);
5281 * alloc_netdev_mq - allocate network device
5282 * @sizeof_priv: size of private data to allocate space for
5283 * @name: device name format string
5284 * @setup: callback to initialize device
5285 * @queue_count: the number of subqueues to allocate
5287 * Allocates a struct net_device with private data area for driver use
5288 * and performs basic initialization. Also allocates subquue structs
5289 * for each queue on the device at the end of the netdevice.
5291 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5292 void (*setup)(struct net_device *), unsigned int queue_count)
5294 struct netdev_queue *tx;
5295 struct net_device *dev;
5297 struct net_device *p;
5299 struct netdev_rx_queue *rx;
5303 BUG_ON(strlen(name) >= sizeof(dev->name));
5305 alloc_size = sizeof(struct net_device);
5307 /* ensure 32-byte alignment of private area */
5308 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5309 alloc_size += sizeof_priv;
5311 /* ensure 32-byte alignment of whole construct */
5312 alloc_size += NETDEV_ALIGN - 1;
5314 p = kzalloc(alloc_size, GFP_KERNEL);
5316 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5320 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5322 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5328 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5330 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5335 atomic_set(&rx->count, queue_count);
5338 * Set a pointer to first element in the array which holds the
5341 for (i = 0; i < queue_count; i++)
5345 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5346 dev->padded = (char *)dev - (char *)p;
5348 if (dev_addr_init(dev))
5354 dev_net_set(dev, &init_net);
5357 dev->num_tx_queues = queue_count;
5358 dev->real_num_tx_queues = queue_count;
5362 dev->num_rx_queues = queue_count;
5365 dev->gso_max_size = GSO_MAX_SIZE;
5367 netdev_init_queues(dev);
5369 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5370 dev->ethtool_ntuple_list.count = 0;
5371 INIT_LIST_HEAD(&dev->napi_list);
5372 INIT_LIST_HEAD(&dev->unreg_list);
5373 INIT_LIST_HEAD(&dev->link_watch_list);
5374 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5376 strcpy(dev->name, name);
5389 EXPORT_SYMBOL(alloc_netdev_mq);
5392 * free_netdev - free network device
5395 * This function does the last stage of destroying an allocated device
5396 * interface. The reference to the device object is released.
5397 * If this is the last reference then it will be freed.
5399 void free_netdev(struct net_device *dev)
5401 struct napi_struct *p, *n;
5403 release_net(dev_net(dev));
5407 /* Flush device addresses */
5408 dev_addr_flush(dev);
5410 /* Clear ethtool n-tuple list */
5411 ethtool_ntuple_flush(dev);
5413 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5416 /* Compatibility with error handling in drivers */
5417 if (dev->reg_state == NETREG_UNINITIALIZED) {
5418 kfree((char *)dev - dev->padded);
5422 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5423 dev->reg_state = NETREG_RELEASED;
5425 /* will free via device release */
5426 put_device(&dev->dev);
5428 EXPORT_SYMBOL(free_netdev);
5431 * synchronize_net - Synchronize with packet receive processing
5433 * Wait for packets currently being received to be done.
5434 * Does not block later packets from starting.
5436 void synchronize_net(void)
5441 EXPORT_SYMBOL(synchronize_net);
5444 * unregister_netdevice_queue - remove device from the kernel
5448 * This function shuts down a device interface and removes it
5449 * from the kernel tables.
5450 * If head not NULL, device is queued to be unregistered later.
5452 * Callers must hold the rtnl semaphore. You may want
5453 * unregister_netdev() instead of this.
5456 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5461 list_move_tail(&dev->unreg_list, head);
5463 rollback_registered(dev);
5464 /* Finish processing unregister after unlock */
5468 EXPORT_SYMBOL(unregister_netdevice_queue);
5471 * unregister_netdevice_many - unregister many devices
5472 * @head: list of devices
5474 void unregister_netdevice_many(struct list_head *head)
5476 struct net_device *dev;
5478 if (!list_empty(head)) {
5479 rollback_registered_many(head);
5480 list_for_each_entry(dev, head, unreg_list)
5484 EXPORT_SYMBOL(unregister_netdevice_many);
5487 * unregister_netdev - remove device from the kernel
5490 * This function shuts down a device interface and removes it
5491 * from the kernel tables.
5493 * This is just a wrapper for unregister_netdevice that takes
5494 * the rtnl semaphore. In general you want to use this and not
5495 * unregister_netdevice.
5497 void unregister_netdev(struct net_device *dev)
5500 unregister_netdevice(dev);
5503 EXPORT_SYMBOL(unregister_netdev);
5506 * dev_change_net_namespace - move device to different nethost namespace
5508 * @net: network namespace
5509 * @pat: If not NULL name pattern to try if the current device name
5510 * is already taken in the destination network namespace.
5512 * This function shuts down a device interface and moves it
5513 * to a new network namespace. On success 0 is returned, on
5514 * a failure a netagive errno code is returned.
5516 * Callers must hold the rtnl semaphore.
5519 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5525 /* Don't allow namespace local devices to be moved. */
5527 if (dev->features & NETIF_F_NETNS_LOCAL)
5531 /* Don't allow real devices to be moved when sysfs
5535 if (dev->dev.parent)
5539 /* Ensure the device has been registrered */
5541 if (dev->reg_state != NETREG_REGISTERED)
5544 /* Get out if there is nothing todo */
5546 if (net_eq(dev_net(dev), net))
5549 /* Pick the destination device name, and ensure
5550 * we can use it in the destination network namespace.
5553 if (__dev_get_by_name(net, dev->name)) {
5554 /* We get here if we can't use the current device name */
5557 if (dev_get_valid_name(net, pat, dev->name, 1))
5562 * And now a mini version of register_netdevice unregister_netdevice.
5565 /* If device is running close it first. */
5568 /* And unlink it from device chain */
5570 unlist_netdevice(dev);
5574 /* Shutdown queueing discipline. */
5577 /* Notify protocols, that we are about to destroy
5578 this device. They should clean all the things.
5580 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5581 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5584 * Flush the unicast and multicast chains
5589 netdev_unregister_kobject(dev);
5591 /* Actually switch the network namespace */
5592 dev_net_set(dev, net);
5594 /* If there is an ifindex conflict assign a new one */
5595 if (__dev_get_by_index(net, dev->ifindex)) {
5596 int iflink = (dev->iflink == dev->ifindex);
5597 dev->ifindex = dev_new_index(net);
5599 dev->iflink = dev->ifindex;
5602 /* Fixup kobjects */
5603 err = netdev_register_kobject(dev);
5606 /* Add the device back in the hashes */
5607 list_netdevice(dev);
5609 /* Notify protocols, that a new device appeared. */
5610 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5613 * Prevent userspace races by waiting until the network
5614 * device is fully setup before sending notifications.
5616 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5623 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5625 static int dev_cpu_callback(struct notifier_block *nfb,
5626 unsigned long action,
5629 struct sk_buff **list_skb;
5630 struct sk_buff *skb;
5631 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5632 struct softnet_data *sd, *oldsd;
5634 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5637 local_irq_disable();
5638 cpu = smp_processor_id();
5639 sd = &per_cpu(softnet_data, cpu);
5640 oldsd = &per_cpu(softnet_data, oldcpu);
5642 /* Find end of our completion_queue. */
5643 list_skb = &sd->completion_queue;
5645 list_skb = &(*list_skb)->next;
5646 /* Append completion queue from offline CPU. */
5647 *list_skb = oldsd->completion_queue;
5648 oldsd->completion_queue = NULL;
5650 /* Append output queue from offline CPU. */
5651 if (oldsd->output_queue) {
5652 *sd->output_queue_tailp = oldsd->output_queue;
5653 sd->output_queue_tailp = oldsd->output_queue_tailp;
5654 oldsd->output_queue = NULL;
5655 oldsd->output_queue_tailp = &oldsd->output_queue;
5658 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5661 /* Process offline CPU's input_pkt_queue */
5662 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5664 input_queue_head_add(oldsd, 1);
5666 while ((skb = __skb_dequeue(&oldsd->process_queue)))
5674 * netdev_increment_features - increment feature set by one
5675 * @all: current feature set
5676 * @one: new feature set
5677 * @mask: mask feature set
5679 * Computes a new feature set after adding a device with feature set
5680 * @one to the master device with current feature set @all. Will not
5681 * enable anything that is off in @mask. Returns the new feature set.
5683 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5686 /* If device needs checksumming, downgrade to it. */
5687 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5688 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5689 else if (mask & NETIF_F_ALL_CSUM) {
5690 /* If one device supports v4/v6 checksumming, set for all. */
5691 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5692 !(all & NETIF_F_GEN_CSUM)) {
5693 all &= ~NETIF_F_ALL_CSUM;
5694 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5697 /* If one device supports hw checksumming, set for all. */
5698 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5699 all &= ~NETIF_F_ALL_CSUM;
5700 all |= NETIF_F_HW_CSUM;
5704 one |= NETIF_F_ALL_CSUM;
5706 one |= all & NETIF_F_ONE_FOR_ALL;
5707 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5708 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5712 EXPORT_SYMBOL(netdev_increment_features);
5714 static struct hlist_head *netdev_create_hash(void)
5717 struct hlist_head *hash;
5719 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5721 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5722 INIT_HLIST_HEAD(&hash[i]);
5727 /* Initialize per network namespace state */
5728 static int __net_init netdev_init(struct net *net)
5730 INIT_LIST_HEAD(&net->dev_base_head);
5732 net->dev_name_head = netdev_create_hash();
5733 if (net->dev_name_head == NULL)
5736 net->dev_index_head = netdev_create_hash();
5737 if (net->dev_index_head == NULL)
5743 kfree(net->dev_name_head);
5749 * netdev_drivername - network driver for the device
5750 * @dev: network device
5751 * @buffer: buffer for resulting name
5752 * @len: size of buffer
5754 * Determine network driver for device.
5756 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5758 const struct device_driver *driver;
5759 const struct device *parent;
5761 if (len <= 0 || !buffer)
5765 parent = dev->dev.parent;
5770 driver = parent->driver;
5771 if (driver && driver->name)
5772 strlcpy(buffer, driver->name, len);
5776 static void __net_exit netdev_exit(struct net *net)
5778 kfree(net->dev_name_head);
5779 kfree(net->dev_index_head);
5782 static struct pernet_operations __net_initdata netdev_net_ops = {
5783 .init = netdev_init,
5784 .exit = netdev_exit,
5787 static void __net_exit default_device_exit(struct net *net)
5789 struct net_device *dev, *aux;
5791 * Push all migratable network devices back to the
5792 * initial network namespace
5795 for_each_netdev_safe(net, dev, aux) {
5797 char fb_name[IFNAMSIZ];
5799 /* Ignore unmoveable devices (i.e. loopback) */
5800 if (dev->features & NETIF_F_NETNS_LOCAL)
5803 /* Leave virtual devices for the generic cleanup */
5804 if (dev->rtnl_link_ops)
5807 /* Push remaing network devices to init_net */
5808 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5809 err = dev_change_net_namespace(dev, &init_net, fb_name);
5811 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5812 __func__, dev->name, err);
5819 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5821 /* At exit all network devices most be removed from a network
5822 * namespace. Do this in the reverse order of registeration.
5823 * Do this across as many network namespaces as possible to
5824 * improve batching efficiency.
5826 struct net_device *dev;
5828 LIST_HEAD(dev_kill_list);
5831 list_for_each_entry(net, net_list, exit_list) {
5832 for_each_netdev_reverse(net, dev) {
5833 if (dev->rtnl_link_ops)
5834 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5836 unregister_netdevice_queue(dev, &dev_kill_list);
5839 unregister_netdevice_many(&dev_kill_list);
5843 static struct pernet_operations __net_initdata default_device_ops = {
5844 .exit = default_device_exit,
5845 .exit_batch = default_device_exit_batch,
5849 * Initialize the DEV module. At boot time this walks the device list and
5850 * unhooks any devices that fail to initialise (normally hardware not
5851 * present) and leaves us with a valid list of present and active devices.
5856 * This is called single threaded during boot, so no need
5857 * to take the rtnl semaphore.
5859 static int __init net_dev_init(void)
5861 int i, rc = -ENOMEM;
5863 BUG_ON(!dev_boot_phase);
5865 if (dev_proc_init())
5868 if (netdev_kobject_init())
5871 INIT_LIST_HEAD(&ptype_all);
5872 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5873 INIT_LIST_HEAD(&ptype_base[i]);
5875 if (register_pernet_subsys(&netdev_net_ops))
5879 * Initialise the packet receive queues.
5882 for_each_possible_cpu(i) {
5883 struct softnet_data *sd = &per_cpu(softnet_data, i);
5885 memset(sd, 0, sizeof(*sd));
5886 skb_queue_head_init(&sd->input_pkt_queue);
5887 skb_queue_head_init(&sd->process_queue);
5888 sd->completion_queue = NULL;
5889 INIT_LIST_HEAD(&sd->poll_list);
5890 sd->output_queue = NULL;
5891 sd->output_queue_tailp = &sd->output_queue;
5893 sd->csd.func = rps_trigger_softirq;
5899 sd->backlog.poll = process_backlog;
5900 sd->backlog.weight = weight_p;
5901 sd->backlog.gro_list = NULL;
5902 sd->backlog.gro_count = 0;
5907 /* The loopback device is special if any other network devices
5908 * is present in a network namespace the loopback device must
5909 * be present. Since we now dynamically allocate and free the
5910 * loopback device ensure this invariant is maintained by
5911 * keeping the loopback device as the first device on the
5912 * list of network devices. Ensuring the loopback devices
5913 * is the first device that appears and the last network device
5916 if (register_pernet_device(&loopback_net_ops))
5919 if (register_pernet_device(&default_device_ops))
5922 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5923 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5925 hotcpu_notifier(dev_cpu_callback, 0);
5933 subsys_initcall(net_dev_init);
5935 static int __init initialize_hashrnd(void)
5937 get_random_bytes(&hashrnd, sizeof(hashrnd));
5941 late_initcall_sync(initialize_hashrnd);