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_set(struct sk_buff *skb)
1459 if (atomic_read(&netstamp_needed))
1460 __net_timestamp(skb);
1462 skb->tstamp.tv64 = 0;
1465 static inline void net_timestamp_check(struct sk_buff *skb)
1467 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1468 __net_timestamp(skb);
1472 * dev_forward_skb - loopback an skb to another netif
1474 * @dev: destination network device
1475 * @skb: buffer to forward
1478 * NET_RX_SUCCESS (no congestion)
1479 * NET_RX_DROP (packet was dropped, but freed)
1481 * dev_forward_skb can be used for injecting an skb from the
1482 * start_xmit function of one device into the receive queue
1483 * of another device.
1485 * The receiving device may be in another namespace, so
1486 * we have to clear all information in the skb that could
1487 * impact namespace isolation.
1489 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1493 if (!(dev->flags & IFF_UP) ||
1494 (skb->len > (dev->mtu + dev->hard_header_len))) {
1498 skb_set_dev(skb, dev);
1499 skb->tstamp.tv64 = 0;
1500 skb->pkt_type = PACKET_HOST;
1501 skb->protocol = eth_type_trans(skb, dev);
1502 return netif_rx(skb);
1504 EXPORT_SYMBOL_GPL(dev_forward_skb);
1507 * Support routine. Sends outgoing frames to any network
1508 * taps currently in use.
1511 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1513 struct packet_type *ptype;
1515 #ifdef CONFIG_NET_CLS_ACT
1516 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1517 net_timestamp_set(skb);
1519 net_timestamp_set(skb);
1523 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1524 /* Never send packets back to the socket
1525 * they originated from - MvS (miquels@drinkel.ow.org)
1527 if ((ptype->dev == dev || !ptype->dev) &&
1528 (ptype->af_packet_priv == NULL ||
1529 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1530 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1534 /* skb->nh should be correctly
1535 set by sender, so that the second statement is
1536 just protection against buggy protocols.
1538 skb_reset_mac_header(skb2);
1540 if (skb_network_header(skb2) < skb2->data ||
1541 skb2->network_header > skb2->tail) {
1542 if (net_ratelimit())
1543 printk(KERN_CRIT "protocol %04x is "
1545 skb2->protocol, dev->name);
1546 skb_reset_network_header(skb2);
1549 skb2->transport_header = skb2->network_header;
1550 skb2->pkt_type = PACKET_OUTGOING;
1551 ptype->func(skb2, skb->dev, ptype, skb->dev);
1558 static inline void __netif_reschedule(struct Qdisc *q)
1560 struct softnet_data *sd;
1561 unsigned long flags;
1563 local_irq_save(flags);
1564 sd = &__get_cpu_var(softnet_data);
1565 q->next_sched = NULL;
1566 *sd->output_queue_tailp = q;
1567 sd->output_queue_tailp = &q->next_sched;
1568 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1569 local_irq_restore(flags);
1572 void __netif_schedule(struct Qdisc *q)
1574 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1575 __netif_reschedule(q);
1577 EXPORT_SYMBOL(__netif_schedule);
1579 void dev_kfree_skb_irq(struct sk_buff *skb)
1581 if (atomic_dec_and_test(&skb->users)) {
1582 struct softnet_data *sd;
1583 unsigned long flags;
1585 local_irq_save(flags);
1586 sd = &__get_cpu_var(softnet_data);
1587 skb->next = sd->completion_queue;
1588 sd->completion_queue = skb;
1589 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1590 local_irq_restore(flags);
1593 EXPORT_SYMBOL(dev_kfree_skb_irq);
1595 void dev_kfree_skb_any(struct sk_buff *skb)
1597 if (in_irq() || irqs_disabled())
1598 dev_kfree_skb_irq(skb);
1602 EXPORT_SYMBOL(dev_kfree_skb_any);
1606 * netif_device_detach - mark device as removed
1607 * @dev: network device
1609 * Mark device as removed from system and therefore no longer available.
1611 void netif_device_detach(struct net_device *dev)
1613 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1614 netif_running(dev)) {
1615 netif_tx_stop_all_queues(dev);
1618 EXPORT_SYMBOL(netif_device_detach);
1621 * netif_device_attach - mark device as attached
1622 * @dev: network device
1624 * Mark device as attached from system and restart if needed.
1626 void netif_device_attach(struct net_device *dev)
1628 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1629 netif_running(dev)) {
1630 netif_tx_wake_all_queues(dev);
1631 __netdev_watchdog_up(dev);
1634 EXPORT_SYMBOL(netif_device_attach);
1636 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1638 return ((features & NETIF_F_GEN_CSUM) ||
1639 ((features & NETIF_F_IP_CSUM) &&
1640 protocol == htons(ETH_P_IP)) ||
1641 ((features & NETIF_F_IPV6_CSUM) &&
1642 protocol == htons(ETH_P_IPV6)) ||
1643 ((features & NETIF_F_FCOE_CRC) &&
1644 protocol == htons(ETH_P_FCOE)));
1647 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1649 if (can_checksum_protocol(dev->features, skb->protocol))
1652 if (skb->protocol == htons(ETH_P_8021Q)) {
1653 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1654 if (can_checksum_protocol(dev->features & dev->vlan_features,
1655 veh->h_vlan_encapsulated_proto))
1663 * skb_dev_set -- assign a new device to a buffer
1664 * @skb: buffer for the new device
1665 * @dev: network device
1667 * If an skb is owned by a device already, we have to reset
1668 * all data private to the namespace a device belongs to
1669 * before assigning it a new device.
1671 #ifdef CONFIG_NET_NS
1672 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1675 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1678 skb_init_secmark(skb);
1682 skb->ipvs_property = 0;
1683 #ifdef CONFIG_NET_SCHED
1689 EXPORT_SYMBOL(skb_set_dev);
1690 #endif /* CONFIG_NET_NS */
1693 * Invalidate hardware checksum when packet is to be mangled, and
1694 * complete checksum manually on outgoing path.
1696 int skb_checksum_help(struct sk_buff *skb)
1699 int ret = 0, offset;
1701 if (skb->ip_summed == CHECKSUM_COMPLETE)
1702 goto out_set_summed;
1704 if (unlikely(skb_shinfo(skb)->gso_size)) {
1705 /* Let GSO fix up the checksum. */
1706 goto out_set_summed;
1709 offset = skb->csum_start - skb_headroom(skb);
1710 BUG_ON(offset >= skb_headlen(skb));
1711 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1713 offset += skb->csum_offset;
1714 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1716 if (skb_cloned(skb) &&
1717 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1718 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1723 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1725 skb->ip_summed = CHECKSUM_NONE;
1729 EXPORT_SYMBOL(skb_checksum_help);
1732 * skb_gso_segment - Perform segmentation on skb.
1733 * @skb: buffer to segment
1734 * @features: features for the output path (see dev->features)
1736 * This function segments the given skb and returns a list of segments.
1738 * It may return NULL if the skb requires no segmentation. This is
1739 * only possible when GSO is used for verifying header integrity.
1741 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1743 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1744 struct packet_type *ptype;
1745 __be16 type = skb->protocol;
1748 skb_reset_mac_header(skb);
1749 skb->mac_len = skb->network_header - skb->mac_header;
1750 __skb_pull(skb, skb->mac_len);
1752 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1753 struct net_device *dev = skb->dev;
1754 struct ethtool_drvinfo info = {};
1756 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1757 dev->ethtool_ops->get_drvinfo(dev, &info);
1759 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1761 info.driver, dev ? dev->features : 0L,
1762 skb->sk ? skb->sk->sk_route_caps : 0L,
1763 skb->len, skb->data_len, skb->ip_summed);
1765 if (skb_header_cloned(skb) &&
1766 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1767 return ERR_PTR(err);
1771 list_for_each_entry_rcu(ptype,
1772 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1773 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1774 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1775 err = ptype->gso_send_check(skb);
1776 segs = ERR_PTR(err);
1777 if (err || skb_gso_ok(skb, features))
1779 __skb_push(skb, (skb->data -
1780 skb_network_header(skb)));
1782 segs = ptype->gso_segment(skb, features);
1788 __skb_push(skb, skb->data - skb_mac_header(skb));
1792 EXPORT_SYMBOL(skb_gso_segment);
1794 /* Take action when hardware reception checksum errors are detected. */
1796 void netdev_rx_csum_fault(struct net_device *dev)
1798 if (net_ratelimit()) {
1799 printk(KERN_ERR "%s: hw csum failure.\n",
1800 dev ? dev->name : "<unknown>");
1804 EXPORT_SYMBOL(netdev_rx_csum_fault);
1807 /* Actually, we should eliminate this check as soon as we know, that:
1808 * 1. IOMMU is present and allows to map all the memory.
1809 * 2. No high memory really exists on this machine.
1812 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1814 #ifdef CONFIG_HIGHMEM
1816 if (!(dev->features & NETIF_F_HIGHDMA)) {
1817 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1818 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1822 if (PCI_DMA_BUS_IS_PHYS) {
1823 struct device *pdev = dev->dev.parent;
1827 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1828 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1829 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1838 void (*destructor)(struct sk_buff *skb);
1841 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1843 static void dev_gso_skb_destructor(struct sk_buff *skb)
1845 struct dev_gso_cb *cb;
1848 struct sk_buff *nskb = skb->next;
1850 skb->next = nskb->next;
1853 } while (skb->next);
1855 cb = DEV_GSO_CB(skb);
1857 cb->destructor(skb);
1861 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1862 * @skb: buffer to segment
1864 * This function segments the given skb and stores the list of segments
1867 static int dev_gso_segment(struct sk_buff *skb)
1869 struct net_device *dev = skb->dev;
1870 struct sk_buff *segs;
1871 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1874 segs = skb_gso_segment(skb, features);
1876 /* Verifying header integrity only. */
1881 return PTR_ERR(segs);
1884 DEV_GSO_CB(skb)->destructor = skb->destructor;
1885 skb->destructor = dev_gso_skb_destructor;
1891 * Try to orphan skb early, right before transmission by the device.
1892 * We cannot orphan skb if tx timestamp is requested, since
1893 * drivers need to call skb_tstamp_tx() to send the timestamp.
1895 static inline void skb_orphan_try(struct sk_buff *skb)
1897 if (!skb_tx(skb)->flags)
1901 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1902 struct netdev_queue *txq)
1904 const struct net_device_ops *ops = dev->netdev_ops;
1905 int rc = NETDEV_TX_OK;
1907 if (likely(!skb->next)) {
1908 if (!list_empty(&ptype_all))
1909 dev_queue_xmit_nit(skb, dev);
1912 * If device doesnt need skb->dst, release it right now while
1913 * its hot in this cpu cache
1915 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1918 skb_orphan_try(skb);
1920 if (netif_needs_gso(dev, skb)) {
1921 if (unlikely(dev_gso_segment(skb)))
1927 rc = ops->ndo_start_xmit(skb, dev);
1928 if (rc == NETDEV_TX_OK)
1929 txq_trans_update(txq);
1935 struct sk_buff *nskb = skb->next;
1937 skb->next = nskb->next;
1941 * If device doesnt need nskb->dst, release it right now while
1942 * its hot in this cpu cache
1944 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1947 rc = ops->ndo_start_xmit(nskb, dev);
1948 if (unlikely(rc != NETDEV_TX_OK)) {
1949 if (rc & ~NETDEV_TX_MASK)
1950 goto out_kfree_gso_skb;
1951 nskb->next = skb->next;
1955 txq_trans_update(txq);
1956 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1957 return NETDEV_TX_BUSY;
1958 } while (skb->next);
1961 if (likely(skb->next == NULL))
1962 skb->destructor = DEV_GSO_CB(skb)->destructor;
1968 static u32 hashrnd __read_mostly;
1970 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1974 if (skb_rx_queue_recorded(skb)) {
1975 hash = skb_get_rx_queue(skb);
1976 while (unlikely(hash >= dev->real_num_tx_queues))
1977 hash -= dev->real_num_tx_queues;
1981 if (skb->sk && skb->sk->sk_hash)
1982 hash = skb->sk->sk_hash;
1984 hash = (__force u16) skb->protocol;
1986 hash = jhash_1word(hash, hashrnd);
1988 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1990 EXPORT_SYMBOL(skb_tx_hash);
1992 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1994 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1995 if (net_ratelimit()) {
1996 pr_warning("%s selects TX queue %d, but "
1997 "real number of TX queues is %d\n",
1998 dev->name, queue_index, dev->real_num_tx_queues);
2005 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2006 struct sk_buff *skb)
2009 struct sock *sk = skb->sk;
2011 if (sk_tx_queue_recorded(sk)) {
2012 queue_index = sk_tx_queue_get(sk);
2014 const struct net_device_ops *ops = dev->netdev_ops;
2016 if (ops->ndo_select_queue) {
2017 queue_index = ops->ndo_select_queue(dev, skb);
2018 queue_index = dev_cap_txqueue(dev, queue_index);
2021 if (dev->real_num_tx_queues > 1)
2022 queue_index = skb_tx_hash(dev, skb);
2025 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2027 if (dst && skb_dst(skb) == dst)
2028 sk_tx_queue_set(sk, queue_index);
2033 skb_set_queue_mapping(skb, queue_index);
2034 return netdev_get_tx_queue(dev, queue_index);
2037 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2038 struct net_device *dev,
2039 struct netdev_queue *txq)
2041 spinlock_t *root_lock = qdisc_lock(q);
2044 spin_lock(root_lock);
2045 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2048 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2049 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
2051 * This is a work-conserving queue; there are no old skbs
2052 * waiting to be sent out; and the qdisc is not running -
2053 * xmit the skb directly.
2055 __qdisc_update_bstats(q, skb->len);
2056 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
2059 clear_bit(__QDISC_STATE_RUNNING, &q->state);
2061 rc = NET_XMIT_SUCCESS;
2063 rc = qdisc_enqueue_root(skb, q);
2066 spin_unlock(root_lock);
2072 * Returns true if either:
2073 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2074 * 2. skb is fragmented and the device does not support SG, or if
2075 * at least one of fragments is in highmem and device does not
2076 * support DMA from it.
2078 static inline int skb_needs_linearize(struct sk_buff *skb,
2079 struct net_device *dev)
2081 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
2082 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
2083 illegal_highdma(dev, skb)));
2087 * dev_queue_xmit - transmit a buffer
2088 * @skb: buffer to transmit
2090 * Queue a buffer for transmission to a network device. The caller must
2091 * have set the device and priority and built the buffer before calling
2092 * this function. The function can be called from an interrupt.
2094 * A negative errno code is returned on a failure. A success does not
2095 * guarantee the frame will be transmitted as it may be dropped due
2096 * to congestion or traffic shaping.
2098 * -----------------------------------------------------------------------------------
2099 * I notice this method can also return errors from the queue disciplines,
2100 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2103 * Regardless of the return value, the skb is consumed, so it is currently
2104 * difficult to retry a send to this method. (You can bump the ref count
2105 * before sending to hold a reference for retry if you are careful.)
2107 * When calling this method, interrupts MUST be enabled. This is because
2108 * the BH enable code must have IRQs enabled so that it will not deadlock.
2111 int dev_queue_xmit(struct sk_buff *skb)
2113 struct net_device *dev = skb->dev;
2114 struct netdev_queue *txq;
2118 /* GSO will handle the following emulations directly. */
2119 if (netif_needs_gso(dev, skb))
2122 /* Convert a paged skb to linear, if required */
2123 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2126 /* If packet is not checksummed and device does not support
2127 * checksumming for this protocol, complete checksumming here.
2129 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2130 skb_set_transport_header(skb, skb->csum_start -
2132 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2137 /* Disable soft irqs for various locks below. Also
2138 * stops preemption for RCU.
2142 txq = dev_pick_tx(dev, skb);
2143 q = rcu_dereference_bh(txq->qdisc);
2145 #ifdef CONFIG_NET_CLS_ACT
2146 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2149 rc = __dev_xmit_skb(skb, q, dev, txq);
2153 /* The device has no queue. Common case for software devices:
2154 loopback, all the sorts of tunnels...
2156 Really, it is unlikely that netif_tx_lock protection is necessary
2157 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2159 However, it is possible, that they rely on protection
2162 Check this and shot the lock. It is not prone from deadlocks.
2163 Either shot noqueue qdisc, it is even simpler 8)
2165 if (dev->flags & IFF_UP) {
2166 int cpu = smp_processor_id(); /* ok because BHs are off */
2168 if (txq->xmit_lock_owner != cpu) {
2170 HARD_TX_LOCK(dev, txq, cpu);
2172 if (!netif_tx_queue_stopped(txq)) {
2173 rc = dev_hard_start_xmit(skb, dev, txq);
2174 if (dev_xmit_complete(rc)) {
2175 HARD_TX_UNLOCK(dev, txq);
2179 HARD_TX_UNLOCK(dev, txq);
2180 if (net_ratelimit())
2181 printk(KERN_CRIT "Virtual device %s asks to "
2182 "queue packet!\n", dev->name);
2184 /* Recursion is detected! It is possible,
2186 if (net_ratelimit())
2187 printk(KERN_CRIT "Dead loop on virtual device "
2188 "%s, fix it urgently!\n", dev->name);
2193 rcu_read_unlock_bh();
2199 rcu_read_unlock_bh();
2202 EXPORT_SYMBOL(dev_queue_xmit);
2205 /*=======================================================================
2207 =======================================================================*/
2209 int netdev_max_backlog __read_mostly = 1000;
2210 int netdev_tstamp_prequeue __read_mostly = 1;
2211 int netdev_budget __read_mostly = 300;
2212 int weight_p __read_mostly = 64; /* old backlog weight */
2214 /* Called with irq disabled */
2215 static inline void ____napi_schedule(struct softnet_data *sd,
2216 struct napi_struct *napi)
2218 list_add_tail(&napi->poll_list, &sd->poll_list);
2219 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2224 /* One global table that all flow-based protocols share. */
2225 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2226 EXPORT_SYMBOL(rps_sock_flow_table);
2229 * get_rps_cpu is called from netif_receive_skb and returns the target
2230 * CPU from the RPS map of the receiving queue for a given skb.
2231 * rcu_read_lock must be held on entry.
2233 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2234 struct rps_dev_flow **rflowp)
2236 struct ipv6hdr *ip6;
2238 struct netdev_rx_queue *rxqueue;
2239 struct rps_map *map;
2240 struct rps_dev_flow_table *flow_table;
2241 struct rps_sock_flow_table *sock_flow_table;
2245 u32 addr1, addr2, ihl;
2251 if (skb_rx_queue_recorded(skb)) {
2252 u16 index = skb_get_rx_queue(skb);
2253 if (unlikely(index >= dev->num_rx_queues)) {
2254 if (net_ratelimit()) {
2255 pr_warning("%s received packet on queue "
2256 "%u, but number of RX queues is %u\n",
2257 dev->name, index, dev->num_rx_queues);
2261 rxqueue = dev->_rx + index;
2265 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2269 goto got_hash; /* Skip hash computation on packet header */
2271 switch (skb->protocol) {
2272 case __constant_htons(ETH_P_IP):
2273 if (!pskb_may_pull(skb, sizeof(*ip)))
2276 ip = (struct iphdr *) skb->data;
2277 ip_proto = ip->protocol;
2278 addr1 = (__force u32) ip->saddr;
2279 addr2 = (__force u32) ip->daddr;
2282 case __constant_htons(ETH_P_IPV6):
2283 if (!pskb_may_pull(skb, sizeof(*ip6)))
2286 ip6 = (struct ipv6hdr *) skb->data;
2287 ip_proto = ip6->nexthdr;
2288 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2289 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2302 case IPPROTO_UDPLITE:
2303 if (pskb_may_pull(skb, (ihl * 4) + 4)) {
2304 ports.v32 = * (__force u32 *) (skb->data + (ihl * 4));
2305 if (ports.v16[1] < ports.v16[0])
2306 swap(ports.v16[0], ports.v16[1]);
2314 /* get a consistent hash (same value on both flow directions) */
2317 skb->rxhash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2322 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2323 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2324 if (flow_table && sock_flow_table) {
2326 struct rps_dev_flow *rflow;
2328 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2331 next_cpu = sock_flow_table->ents[skb->rxhash &
2332 sock_flow_table->mask];
2335 * If the desired CPU (where last recvmsg was done) is
2336 * different from current CPU (one in the rx-queue flow
2337 * table entry), switch if one of the following holds:
2338 * - Current CPU is unset (equal to RPS_NO_CPU).
2339 * - Current CPU is offline.
2340 * - The current CPU's queue tail has advanced beyond the
2341 * last packet that was enqueued using this table entry.
2342 * This guarantees that all previous packets for the flow
2343 * have been dequeued, thus preserving in order delivery.
2345 if (unlikely(tcpu != next_cpu) &&
2346 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2347 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2348 rflow->last_qtail)) >= 0)) {
2349 tcpu = rflow->cpu = next_cpu;
2350 if (tcpu != RPS_NO_CPU)
2351 rflow->last_qtail = per_cpu(softnet_data,
2352 tcpu).input_queue_head;
2354 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2361 map = rcu_dereference(rxqueue->rps_map);
2363 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2365 if (cpu_online(tcpu)) {
2375 /* Called from hardirq (IPI) context */
2376 static void rps_trigger_softirq(void *data)
2378 struct softnet_data *sd = data;
2380 ____napi_schedule(sd, &sd->backlog);
2384 #endif /* CONFIG_RPS */
2387 * Check if this softnet_data structure is another cpu one
2388 * If yes, queue it to our IPI list and return 1
2391 static int rps_ipi_queued(struct softnet_data *sd)
2394 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2397 sd->rps_ipi_next = mysd->rps_ipi_list;
2398 mysd->rps_ipi_list = sd;
2400 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2403 #endif /* CONFIG_RPS */
2408 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2409 * queue (may be a remote CPU queue).
2411 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2412 unsigned int *qtail)
2414 struct softnet_data *sd;
2415 unsigned long flags;
2417 sd = &per_cpu(softnet_data, cpu);
2419 local_irq_save(flags);
2422 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2423 if (skb_queue_len(&sd->input_pkt_queue)) {
2425 __skb_queue_tail(&sd->input_pkt_queue, skb);
2427 *qtail = sd->input_queue_head +
2428 skb_queue_len(&sd->input_pkt_queue);
2431 local_irq_restore(flags);
2432 return NET_RX_SUCCESS;
2435 /* Schedule NAPI for backlog device
2436 * We can use non atomic operation since we own the queue lock
2438 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2439 if (!rps_ipi_queued(sd))
2440 ____napi_schedule(sd, &sd->backlog);
2448 local_irq_restore(flags);
2455 * netif_rx - post buffer to the network code
2456 * @skb: buffer to post
2458 * This function receives a packet from a device driver and queues it for
2459 * the upper (protocol) levels to process. It always succeeds. The buffer
2460 * may be dropped during processing for congestion control or by the
2464 * NET_RX_SUCCESS (no congestion)
2465 * NET_RX_DROP (packet was dropped)
2469 int netif_rx(struct sk_buff *skb)
2473 /* if netpoll wants it, pretend we never saw it */
2474 if (netpoll_rx(skb))
2477 if (netdev_tstamp_prequeue)
2478 net_timestamp_check(skb);
2482 struct rps_dev_flow voidflow, *rflow = &voidflow;
2487 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2489 cpu = smp_processor_id();
2491 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2498 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2504 EXPORT_SYMBOL(netif_rx);
2506 int netif_rx_ni(struct sk_buff *skb)
2511 err = netif_rx(skb);
2512 if (local_softirq_pending())
2518 EXPORT_SYMBOL(netif_rx_ni);
2520 static void net_tx_action(struct softirq_action *h)
2522 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2524 if (sd->completion_queue) {
2525 struct sk_buff *clist;
2527 local_irq_disable();
2528 clist = sd->completion_queue;
2529 sd->completion_queue = NULL;
2533 struct sk_buff *skb = clist;
2534 clist = clist->next;
2536 WARN_ON(atomic_read(&skb->users));
2541 if (sd->output_queue) {
2544 local_irq_disable();
2545 head = sd->output_queue;
2546 sd->output_queue = NULL;
2547 sd->output_queue_tailp = &sd->output_queue;
2551 struct Qdisc *q = head;
2552 spinlock_t *root_lock;
2554 head = head->next_sched;
2556 root_lock = qdisc_lock(q);
2557 if (spin_trylock(root_lock)) {
2558 smp_mb__before_clear_bit();
2559 clear_bit(__QDISC_STATE_SCHED,
2562 spin_unlock(root_lock);
2564 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2566 __netif_reschedule(q);
2568 smp_mb__before_clear_bit();
2569 clear_bit(__QDISC_STATE_SCHED,
2577 static inline int deliver_skb(struct sk_buff *skb,
2578 struct packet_type *pt_prev,
2579 struct net_device *orig_dev)
2581 atomic_inc(&skb->users);
2582 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2585 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2587 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2588 /* This hook is defined here for ATM LANE */
2589 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2590 unsigned char *addr) __read_mostly;
2591 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2595 * If bridge module is loaded call bridging hook.
2596 * returns NULL if packet was consumed.
2598 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2599 struct sk_buff *skb) __read_mostly;
2600 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2602 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2603 struct packet_type **pt_prev, int *ret,
2604 struct net_device *orig_dev)
2606 struct net_bridge_port *port;
2608 if (skb->pkt_type == PACKET_LOOPBACK ||
2609 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2613 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2617 return br_handle_frame_hook(port, skb);
2620 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2623 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2624 struct sk_buff *(*macvlan_handle_frame_hook)(struct macvlan_port *p,
2625 struct sk_buff *skb) __read_mostly;
2626 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2628 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2629 struct packet_type **pt_prev,
2631 struct net_device *orig_dev)
2633 struct macvlan_port *port;
2635 port = rcu_dereference(skb->dev->macvlan_port);
2640 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2643 return macvlan_handle_frame_hook(port, skb);
2646 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2649 #ifdef CONFIG_NET_CLS_ACT
2650 /* TODO: Maybe we should just force sch_ingress to be compiled in
2651 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2652 * a compare and 2 stores extra right now if we dont have it on
2653 * but have CONFIG_NET_CLS_ACT
2654 * NOTE: This doesnt stop any functionality; if you dont have
2655 * the ingress scheduler, you just cant add policies on ingress.
2658 static int ing_filter(struct sk_buff *skb)
2660 struct net_device *dev = skb->dev;
2661 u32 ttl = G_TC_RTTL(skb->tc_verd);
2662 struct netdev_queue *rxq;
2663 int result = TC_ACT_OK;
2666 if (MAX_RED_LOOP < ttl++) {
2668 "Redir loop detected Dropping packet (%d->%d)\n",
2669 skb->skb_iif, dev->ifindex);
2673 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2674 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2676 rxq = &dev->rx_queue;
2679 if (q != &noop_qdisc) {
2680 spin_lock(qdisc_lock(q));
2681 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2682 result = qdisc_enqueue_root(skb, q);
2683 spin_unlock(qdisc_lock(q));
2689 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2690 struct packet_type **pt_prev,
2691 int *ret, struct net_device *orig_dev)
2693 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2697 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2700 /* Huh? Why does turning on AF_PACKET affect this? */
2701 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2704 switch (ing_filter(skb)) {
2718 * netif_nit_deliver - deliver received packets to network taps
2721 * This function is used to deliver incoming packets to network
2722 * taps. It should be used when the normal netif_receive_skb path
2723 * is bypassed, for example because of VLAN acceleration.
2725 void netif_nit_deliver(struct sk_buff *skb)
2727 struct packet_type *ptype;
2729 if (list_empty(&ptype_all))
2732 skb_reset_network_header(skb);
2733 skb_reset_transport_header(skb);
2734 skb->mac_len = skb->network_header - skb->mac_header;
2737 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2738 if (!ptype->dev || ptype->dev == skb->dev)
2739 deliver_skb(skb, ptype, skb->dev);
2744 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2745 struct net_device *master)
2747 if (skb->pkt_type == PACKET_HOST) {
2748 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2750 memcpy(dest, master->dev_addr, ETH_ALEN);
2754 /* On bonding slaves other than the currently active slave, suppress
2755 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2756 * ARP on active-backup slaves with arp_validate enabled.
2758 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2760 struct net_device *dev = skb->dev;
2762 if (master->priv_flags & IFF_MASTER_ARPMON)
2763 dev->last_rx = jiffies;
2765 if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) {
2766 /* Do address unmangle. The local destination address
2767 * will be always the one master has. Provides the right
2768 * functionality in a bridge.
2770 skb_bond_set_mac_by_master(skb, master);
2773 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2774 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2775 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2778 if (master->priv_flags & IFF_MASTER_ALB) {
2779 if (skb->pkt_type != PACKET_BROADCAST &&
2780 skb->pkt_type != PACKET_MULTICAST)
2783 if (master->priv_flags & IFF_MASTER_8023AD &&
2784 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2791 EXPORT_SYMBOL(__skb_bond_should_drop);
2793 static int __netif_receive_skb(struct sk_buff *skb)
2795 struct packet_type *ptype, *pt_prev;
2796 struct net_device *orig_dev;
2797 struct net_device *master;
2798 struct net_device *null_or_orig;
2799 struct net_device *null_or_bond;
2800 int ret = NET_RX_DROP;
2803 if (!netdev_tstamp_prequeue)
2804 net_timestamp_check(skb);
2806 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2807 return NET_RX_SUCCESS;
2809 /* if we've gotten here through NAPI, check netpoll */
2810 if (netpoll_receive_skb(skb))
2814 skb->skb_iif = skb->dev->ifindex;
2816 null_or_orig = NULL;
2817 orig_dev = skb->dev;
2818 master = ACCESS_ONCE(orig_dev->master);
2820 if (skb_bond_should_drop(skb, master))
2821 null_or_orig = orig_dev; /* deliver only exact match */
2826 __get_cpu_var(softnet_data).processed++;
2828 skb_reset_network_header(skb);
2829 skb_reset_transport_header(skb);
2830 skb->mac_len = skb->network_header - skb->mac_header;
2836 #ifdef CONFIG_NET_CLS_ACT
2837 if (skb->tc_verd & TC_NCLS) {
2838 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2843 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2844 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2845 ptype->dev == orig_dev) {
2847 ret = deliver_skb(skb, pt_prev, orig_dev);
2852 #ifdef CONFIG_NET_CLS_ACT
2853 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2859 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2862 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2867 * Make sure frames received on VLAN interfaces stacked on
2868 * bonding interfaces still make their way to any base bonding
2869 * device that may have registered for a specific ptype. The
2870 * handler may have to adjust skb->dev and orig_dev.
2872 null_or_bond = NULL;
2873 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2874 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2875 null_or_bond = vlan_dev_real_dev(skb->dev);
2878 type = skb->protocol;
2879 list_for_each_entry_rcu(ptype,
2880 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2881 if (ptype->type == type && (ptype->dev == null_or_orig ||
2882 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2883 ptype->dev == null_or_bond)) {
2885 ret = deliver_skb(skb, pt_prev, orig_dev);
2891 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2894 /* Jamal, now you will not able to escape explaining
2895 * me how you were going to use this. :-)
2906 * netif_receive_skb - process receive buffer from network
2907 * @skb: buffer to process
2909 * netif_receive_skb() is the main receive data processing function.
2910 * It always succeeds. The buffer may be dropped during processing
2911 * for congestion control or by the protocol layers.
2913 * This function may only be called from softirq context and interrupts
2914 * should be enabled.
2916 * Return values (usually ignored):
2917 * NET_RX_SUCCESS: no congestion
2918 * NET_RX_DROP: packet was dropped
2920 int netif_receive_skb(struct sk_buff *skb)
2922 if (netdev_tstamp_prequeue)
2923 net_timestamp_check(skb);
2927 struct rps_dev_flow voidflow, *rflow = &voidflow;
2932 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2935 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2939 ret = __netif_receive_skb(skb);
2945 return __netif_receive_skb(skb);
2948 EXPORT_SYMBOL(netif_receive_skb);
2950 /* Network device is going away, flush any packets still pending
2951 * Called with irqs disabled.
2953 static void flush_backlog(void *arg)
2955 struct net_device *dev = arg;
2956 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2957 struct sk_buff *skb, *tmp;
2960 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
2961 if (skb->dev == dev) {
2962 __skb_unlink(skb, &sd->input_pkt_queue);
2964 input_queue_head_add(sd, 1);
2969 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
2970 if (skb->dev == dev) {
2971 __skb_unlink(skb, &sd->process_queue);
2977 static int napi_gro_complete(struct sk_buff *skb)
2979 struct packet_type *ptype;
2980 __be16 type = skb->protocol;
2981 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2984 if (NAPI_GRO_CB(skb)->count == 1) {
2985 skb_shinfo(skb)->gso_size = 0;
2990 list_for_each_entry_rcu(ptype, head, list) {
2991 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2994 err = ptype->gro_complete(skb);
3000 WARN_ON(&ptype->list == head);
3002 return NET_RX_SUCCESS;
3006 return netif_receive_skb(skb);
3009 static void napi_gro_flush(struct napi_struct *napi)
3011 struct sk_buff *skb, *next;
3013 for (skb = napi->gro_list; skb; skb = next) {
3016 napi_gro_complete(skb);
3019 napi->gro_count = 0;
3020 napi->gro_list = NULL;
3023 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3025 struct sk_buff **pp = NULL;
3026 struct packet_type *ptype;
3027 __be16 type = skb->protocol;
3028 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3031 enum gro_result ret;
3033 if (!(skb->dev->features & NETIF_F_GRO))
3036 if (skb_is_gso(skb) || skb_has_frags(skb))
3040 list_for_each_entry_rcu(ptype, head, list) {
3041 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3044 skb_set_network_header(skb, skb_gro_offset(skb));
3045 mac_len = skb->network_header - skb->mac_header;
3046 skb->mac_len = mac_len;
3047 NAPI_GRO_CB(skb)->same_flow = 0;
3048 NAPI_GRO_CB(skb)->flush = 0;
3049 NAPI_GRO_CB(skb)->free = 0;
3051 pp = ptype->gro_receive(&napi->gro_list, skb);
3056 if (&ptype->list == head)
3059 same_flow = NAPI_GRO_CB(skb)->same_flow;
3060 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3063 struct sk_buff *nskb = *pp;
3067 napi_gro_complete(nskb);
3074 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3078 NAPI_GRO_CB(skb)->count = 1;
3079 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3080 skb->next = napi->gro_list;
3081 napi->gro_list = skb;
3085 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3086 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3088 BUG_ON(skb->end - skb->tail < grow);
3090 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3093 skb->data_len -= grow;
3095 skb_shinfo(skb)->frags[0].page_offset += grow;
3096 skb_shinfo(skb)->frags[0].size -= grow;
3098 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3099 put_page(skb_shinfo(skb)->frags[0].page);
3100 memmove(skb_shinfo(skb)->frags,
3101 skb_shinfo(skb)->frags + 1,
3102 --skb_shinfo(skb)->nr_frags);
3113 EXPORT_SYMBOL(dev_gro_receive);
3116 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3120 if (netpoll_rx_on(skb))
3123 for (p = napi->gro_list; p; p = p->next) {
3124 NAPI_GRO_CB(p)->same_flow =
3125 (p->dev == skb->dev) &&
3126 !compare_ether_header(skb_mac_header(p),
3127 skb_gro_mac_header(skb));
3128 NAPI_GRO_CB(p)->flush = 0;
3131 return dev_gro_receive(napi, skb);
3134 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3138 if (netif_receive_skb(skb))
3143 case GRO_MERGED_FREE:
3154 EXPORT_SYMBOL(napi_skb_finish);
3156 void skb_gro_reset_offset(struct sk_buff *skb)
3158 NAPI_GRO_CB(skb)->data_offset = 0;
3159 NAPI_GRO_CB(skb)->frag0 = NULL;
3160 NAPI_GRO_CB(skb)->frag0_len = 0;
3162 if (skb->mac_header == skb->tail &&
3163 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3164 NAPI_GRO_CB(skb)->frag0 =
3165 page_address(skb_shinfo(skb)->frags[0].page) +
3166 skb_shinfo(skb)->frags[0].page_offset;
3167 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3170 EXPORT_SYMBOL(skb_gro_reset_offset);
3172 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3174 skb_gro_reset_offset(skb);
3176 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3178 EXPORT_SYMBOL(napi_gro_receive);
3180 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3182 __skb_pull(skb, skb_headlen(skb));
3183 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3187 EXPORT_SYMBOL(napi_reuse_skb);
3189 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3191 struct sk_buff *skb = napi->skb;
3194 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3200 EXPORT_SYMBOL(napi_get_frags);
3202 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3208 skb->protocol = eth_type_trans(skb, skb->dev);
3210 if (ret == GRO_HELD)
3211 skb_gro_pull(skb, -ETH_HLEN);
3212 else if (netif_receive_skb(skb))
3217 case GRO_MERGED_FREE:
3218 napi_reuse_skb(napi, skb);
3227 EXPORT_SYMBOL(napi_frags_finish);
3229 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3231 struct sk_buff *skb = napi->skb;
3238 skb_reset_mac_header(skb);
3239 skb_gro_reset_offset(skb);
3241 off = skb_gro_offset(skb);
3242 hlen = off + sizeof(*eth);
3243 eth = skb_gro_header_fast(skb, off);
3244 if (skb_gro_header_hard(skb, hlen)) {
3245 eth = skb_gro_header_slow(skb, hlen, off);
3246 if (unlikely(!eth)) {
3247 napi_reuse_skb(napi, skb);
3253 skb_gro_pull(skb, sizeof(*eth));
3256 * This works because the only protocols we care about don't require
3257 * special handling. We'll fix it up properly at the end.
3259 skb->protocol = eth->h_proto;
3264 EXPORT_SYMBOL(napi_frags_skb);
3266 gro_result_t napi_gro_frags(struct napi_struct *napi)
3268 struct sk_buff *skb = napi_frags_skb(napi);
3273 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3275 EXPORT_SYMBOL(napi_gro_frags);
3278 * net_rps_action sends any pending IPI's for rps.
3279 * Note: called with local irq disabled, but exits with local irq enabled.
3281 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3284 struct softnet_data *remsd = sd->rps_ipi_list;
3287 sd->rps_ipi_list = NULL;
3291 /* Send pending IPI's to kick RPS processing on remote cpus. */
3293 struct softnet_data *next = remsd->rps_ipi_next;
3295 if (cpu_online(remsd->cpu))
3296 __smp_call_function_single(remsd->cpu,
3305 static int process_backlog(struct napi_struct *napi, int quota)
3308 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3311 /* Check if we have pending ipi, its better to send them now,
3312 * not waiting net_rx_action() end.
3314 if (sd->rps_ipi_list) {
3315 local_irq_disable();
3316 net_rps_action_and_irq_enable(sd);
3319 napi->weight = weight_p;
3320 local_irq_disable();
3321 while (work < quota) {
3322 struct sk_buff *skb;
3325 while ((skb = __skb_dequeue(&sd->process_queue))) {
3327 __netif_receive_skb(skb);
3328 if (++work >= quota)
3330 local_irq_disable();
3334 qlen = skb_queue_len(&sd->input_pkt_queue);
3336 input_queue_head_add(sd, qlen);
3337 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3338 &sd->process_queue);
3340 if (qlen < quota - work) {
3342 * Inline a custom version of __napi_complete().
3343 * only current cpu owns and manipulates this napi,
3344 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3345 * we can use a plain write instead of clear_bit(),
3346 * and we dont need an smp_mb() memory barrier.
3348 list_del(&napi->poll_list);
3351 quota = work + qlen;
3361 * __napi_schedule - schedule for receive
3362 * @n: entry to schedule
3364 * The entry's receive function will be scheduled to run
3366 void __napi_schedule(struct napi_struct *n)
3368 unsigned long flags;
3370 local_irq_save(flags);
3371 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3372 local_irq_restore(flags);
3374 EXPORT_SYMBOL(__napi_schedule);
3376 void __napi_complete(struct napi_struct *n)
3378 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3379 BUG_ON(n->gro_list);
3381 list_del(&n->poll_list);
3382 smp_mb__before_clear_bit();
3383 clear_bit(NAPI_STATE_SCHED, &n->state);
3385 EXPORT_SYMBOL(__napi_complete);
3387 void napi_complete(struct napi_struct *n)
3389 unsigned long flags;
3392 * don't let napi dequeue from the cpu poll list
3393 * just in case its running on a different cpu
3395 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3399 local_irq_save(flags);
3401 local_irq_restore(flags);
3403 EXPORT_SYMBOL(napi_complete);
3405 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3406 int (*poll)(struct napi_struct *, int), int weight)
3408 INIT_LIST_HEAD(&napi->poll_list);
3409 napi->gro_count = 0;
3410 napi->gro_list = NULL;
3413 napi->weight = weight;
3414 list_add(&napi->dev_list, &dev->napi_list);
3416 #ifdef CONFIG_NETPOLL
3417 spin_lock_init(&napi->poll_lock);
3418 napi->poll_owner = -1;
3420 set_bit(NAPI_STATE_SCHED, &napi->state);
3422 EXPORT_SYMBOL(netif_napi_add);
3424 void netif_napi_del(struct napi_struct *napi)
3426 struct sk_buff *skb, *next;
3428 list_del_init(&napi->dev_list);
3429 napi_free_frags(napi);
3431 for (skb = napi->gro_list; skb; skb = next) {
3437 napi->gro_list = NULL;
3438 napi->gro_count = 0;
3440 EXPORT_SYMBOL(netif_napi_del);
3442 static void net_rx_action(struct softirq_action *h)
3444 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3445 unsigned long time_limit = jiffies + 2;
3446 int budget = netdev_budget;
3449 local_irq_disable();
3451 while (!list_empty(&sd->poll_list)) {
3452 struct napi_struct *n;
3455 /* If softirq window is exhuasted then punt.
3456 * Allow this to run for 2 jiffies since which will allow
3457 * an average latency of 1.5/HZ.
3459 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3464 /* Even though interrupts have been re-enabled, this
3465 * access is safe because interrupts can only add new
3466 * entries to the tail of this list, and only ->poll()
3467 * calls can remove this head entry from the list.
3469 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3471 have = netpoll_poll_lock(n);
3475 /* This NAPI_STATE_SCHED test is for avoiding a race
3476 * with netpoll's poll_napi(). Only the entity which
3477 * obtains the lock and sees NAPI_STATE_SCHED set will
3478 * actually make the ->poll() call. Therefore we avoid
3479 * accidently calling ->poll() when NAPI is not scheduled.
3482 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3483 work = n->poll(n, weight);
3487 WARN_ON_ONCE(work > weight);
3491 local_irq_disable();
3493 /* Drivers must not modify the NAPI state if they
3494 * consume the entire weight. In such cases this code
3495 * still "owns" the NAPI instance and therefore can
3496 * move the instance around on the list at-will.
3498 if (unlikely(work == weight)) {
3499 if (unlikely(napi_disable_pending(n))) {
3502 local_irq_disable();
3504 list_move_tail(&n->poll_list, &sd->poll_list);
3507 netpoll_poll_unlock(have);
3510 net_rps_action_and_irq_enable(sd);
3512 #ifdef CONFIG_NET_DMA
3514 * There may not be any more sk_buffs coming right now, so push
3515 * any pending DMA copies to hardware
3517 dma_issue_pending_all();
3524 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3528 static gifconf_func_t *gifconf_list[NPROTO];
3531 * register_gifconf - register a SIOCGIF handler
3532 * @family: Address family
3533 * @gifconf: Function handler
3535 * Register protocol dependent address dumping routines. The handler
3536 * that is passed must not be freed or reused until it has been replaced
3537 * by another handler.
3539 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3541 if (family >= NPROTO)
3543 gifconf_list[family] = gifconf;
3546 EXPORT_SYMBOL(register_gifconf);
3550 * Map an interface index to its name (SIOCGIFNAME)
3554 * We need this ioctl for efficient implementation of the
3555 * if_indextoname() function required by the IPv6 API. Without
3556 * it, we would have to search all the interfaces to find a
3560 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3562 struct net_device *dev;
3566 * Fetch the caller's info block.
3569 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3573 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3579 strcpy(ifr.ifr_name, dev->name);
3582 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3588 * Perform a SIOCGIFCONF call. This structure will change
3589 * size eventually, and there is nothing I can do about it.
3590 * Thus we will need a 'compatibility mode'.
3593 static int dev_ifconf(struct net *net, char __user *arg)
3596 struct net_device *dev;
3603 * Fetch the caller's info block.
3606 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3613 * Loop over the interfaces, and write an info block for each.
3617 for_each_netdev(net, dev) {
3618 for (i = 0; i < NPROTO; i++) {
3619 if (gifconf_list[i]) {
3622 done = gifconf_list[i](dev, NULL, 0);
3624 done = gifconf_list[i](dev, pos + total,
3634 * All done. Write the updated control block back to the caller.
3636 ifc.ifc_len = total;
3639 * Both BSD and Solaris return 0 here, so we do too.
3641 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3644 #ifdef CONFIG_PROC_FS
3646 * This is invoked by the /proc filesystem handler to display a device
3649 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3652 struct net *net = seq_file_net(seq);
3654 struct net_device *dev;
3658 return SEQ_START_TOKEN;
3661 for_each_netdev_rcu(net, dev)
3668 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3670 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3671 first_net_device(seq_file_net(seq)) :
3672 next_net_device((struct net_device *)v);
3675 return rcu_dereference(dev);
3678 void dev_seq_stop(struct seq_file *seq, void *v)
3684 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3686 const struct net_device_stats *stats = dev_get_stats(dev);
3688 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3689 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3690 dev->name, stats->rx_bytes, stats->rx_packets,
3692 stats->rx_dropped + stats->rx_missed_errors,
3693 stats->rx_fifo_errors,
3694 stats->rx_length_errors + stats->rx_over_errors +
3695 stats->rx_crc_errors + stats->rx_frame_errors,
3696 stats->rx_compressed, stats->multicast,
3697 stats->tx_bytes, stats->tx_packets,
3698 stats->tx_errors, stats->tx_dropped,
3699 stats->tx_fifo_errors, stats->collisions,
3700 stats->tx_carrier_errors +
3701 stats->tx_aborted_errors +
3702 stats->tx_window_errors +
3703 stats->tx_heartbeat_errors,
3704 stats->tx_compressed);
3708 * Called from the PROCfs module. This now uses the new arbitrary sized
3709 * /proc/net interface to create /proc/net/dev
3711 static int dev_seq_show(struct seq_file *seq, void *v)
3713 if (v == SEQ_START_TOKEN)
3714 seq_puts(seq, "Inter-| Receive "
3716 " face |bytes packets errs drop fifo frame "
3717 "compressed multicast|bytes packets errs "
3718 "drop fifo colls carrier compressed\n");
3720 dev_seq_printf_stats(seq, v);
3724 static struct softnet_data *softnet_get_online(loff_t *pos)
3726 struct softnet_data *sd = NULL;
3728 while (*pos < nr_cpu_ids)
3729 if (cpu_online(*pos)) {
3730 sd = &per_cpu(softnet_data, *pos);
3737 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3739 return softnet_get_online(pos);
3742 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3745 return softnet_get_online(pos);
3748 static void softnet_seq_stop(struct seq_file *seq, void *v)
3752 static int softnet_seq_show(struct seq_file *seq, void *v)
3754 struct softnet_data *sd = v;
3756 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3757 sd->processed, sd->dropped, sd->time_squeeze, 0,
3758 0, 0, 0, 0, /* was fastroute */
3759 sd->cpu_collision, sd->received_rps);
3763 static const struct seq_operations dev_seq_ops = {
3764 .start = dev_seq_start,
3765 .next = dev_seq_next,
3766 .stop = dev_seq_stop,
3767 .show = dev_seq_show,
3770 static int dev_seq_open(struct inode *inode, struct file *file)
3772 return seq_open_net(inode, file, &dev_seq_ops,
3773 sizeof(struct seq_net_private));
3776 static const struct file_operations dev_seq_fops = {
3777 .owner = THIS_MODULE,
3778 .open = dev_seq_open,
3780 .llseek = seq_lseek,
3781 .release = seq_release_net,
3784 static const struct seq_operations softnet_seq_ops = {
3785 .start = softnet_seq_start,
3786 .next = softnet_seq_next,
3787 .stop = softnet_seq_stop,
3788 .show = softnet_seq_show,
3791 static int softnet_seq_open(struct inode *inode, struct file *file)
3793 return seq_open(file, &softnet_seq_ops);
3796 static const struct file_operations softnet_seq_fops = {
3797 .owner = THIS_MODULE,
3798 .open = softnet_seq_open,
3800 .llseek = seq_lseek,
3801 .release = seq_release,
3804 static void *ptype_get_idx(loff_t pos)
3806 struct packet_type *pt = NULL;
3810 list_for_each_entry_rcu(pt, &ptype_all, list) {
3816 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3817 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3826 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3830 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3833 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3835 struct packet_type *pt;
3836 struct list_head *nxt;
3840 if (v == SEQ_START_TOKEN)
3841 return ptype_get_idx(0);
3844 nxt = pt->list.next;
3845 if (pt->type == htons(ETH_P_ALL)) {
3846 if (nxt != &ptype_all)
3849 nxt = ptype_base[0].next;
3851 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3853 while (nxt == &ptype_base[hash]) {
3854 if (++hash >= PTYPE_HASH_SIZE)
3856 nxt = ptype_base[hash].next;
3859 return list_entry(nxt, struct packet_type, list);
3862 static void ptype_seq_stop(struct seq_file *seq, void *v)
3868 static int ptype_seq_show(struct seq_file *seq, void *v)
3870 struct packet_type *pt = v;
3872 if (v == SEQ_START_TOKEN)
3873 seq_puts(seq, "Type Device Function\n");
3874 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3875 if (pt->type == htons(ETH_P_ALL))
3876 seq_puts(seq, "ALL ");
3878 seq_printf(seq, "%04x", ntohs(pt->type));
3880 seq_printf(seq, " %-8s %pF\n",
3881 pt->dev ? pt->dev->name : "", pt->func);
3887 static const struct seq_operations ptype_seq_ops = {
3888 .start = ptype_seq_start,
3889 .next = ptype_seq_next,
3890 .stop = ptype_seq_stop,
3891 .show = ptype_seq_show,
3894 static int ptype_seq_open(struct inode *inode, struct file *file)
3896 return seq_open_net(inode, file, &ptype_seq_ops,
3897 sizeof(struct seq_net_private));
3900 static const struct file_operations ptype_seq_fops = {
3901 .owner = THIS_MODULE,
3902 .open = ptype_seq_open,
3904 .llseek = seq_lseek,
3905 .release = seq_release_net,
3909 static int __net_init dev_proc_net_init(struct net *net)
3913 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3915 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3917 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3920 if (wext_proc_init(net))
3926 proc_net_remove(net, "ptype");
3928 proc_net_remove(net, "softnet_stat");
3930 proc_net_remove(net, "dev");
3934 static void __net_exit dev_proc_net_exit(struct net *net)
3936 wext_proc_exit(net);
3938 proc_net_remove(net, "ptype");
3939 proc_net_remove(net, "softnet_stat");
3940 proc_net_remove(net, "dev");
3943 static struct pernet_operations __net_initdata dev_proc_ops = {
3944 .init = dev_proc_net_init,
3945 .exit = dev_proc_net_exit,
3948 static int __init dev_proc_init(void)
3950 return register_pernet_subsys(&dev_proc_ops);
3953 #define dev_proc_init() 0
3954 #endif /* CONFIG_PROC_FS */
3958 * netdev_set_master - set up master/slave pair
3959 * @slave: slave device
3960 * @master: new master device
3962 * Changes the master device of the slave. Pass %NULL to break the
3963 * bonding. The caller must hold the RTNL semaphore. On a failure
3964 * a negative errno code is returned. On success the reference counts
3965 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3966 * function returns zero.
3968 int netdev_set_master(struct net_device *slave, struct net_device *master)
3970 struct net_device *old = slave->master;
3980 slave->master = master;
3987 slave->flags |= IFF_SLAVE;
3989 slave->flags &= ~IFF_SLAVE;
3991 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3994 EXPORT_SYMBOL(netdev_set_master);
3996 static void dev_change_rx_flags(struct net_device *dev, int flags)
3998 const struct net_device_ops *ops = dev->netdev_ops;
4000 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4001 ops->ndo_change_rx_flags(dev, flags);
4004 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4006 unsigned short old_flags = dev->flags;
4012 dev->flags |= IFF_PROMISC;
4013 dev->promiscuity += inc;
4014 if (dev->promiscuity == 0) {
4017 * If inc causes overflow, untouch promisc and return error.
4020 dev->flags &= ~IFF_PROMISC;
4022 dev->promiscuity -= inc;
4023 printk(KERN_WARNING "%s: promiscuity touches roof, "
4024 "set promiscuity failed, promiscuity feature "
4025 "of device might be broken.\n", dev->name);
4029 if (dev->flags != old_flags) {
4030 printk(KERN_INFO "device %s %s promiscuous mode\n",
4031 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4033 if (audit_enabled) {
4034 current_uid_gid(&uid, &gid);
4035 audit_log(current->audit_context, GFP_ATOMIC,
4036 AUDIT_ANOM_PROMISCUOUS,
4037 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4038 dev->name, (dev->flags & IFF_PROMISC),
4039 (old_flags & IFF_PROMISC),
4040 audit_get_loginuid(current),
4042 audit_get_sessionid(current));
4045 dev_change_rx_flags(dev, IFF_PROMISC);
4051 * dev_set_promiscuity - update promiscuity count on a device
4055 * Add or remove promiscuity from a device. While the count in the device
4056 * remains above zero the interface remains promiscuous. Once it hits zero
4057 * the device reverts back to normal filtering operation. A negative inc
4058 * value is used to drop promiscuity on the device.
4059 * Return 0 if successful or a negative errno code on error.
4061 int dev_set_promiscuity(struct net_device *dev, int inc)
4063 unsigned short old_flags = dev->flags;
4066 err = __dev_set_promiscuity(dev, inc);
4069 if (dev->flags != old_flags)
4070 dev_set_rx_mode(dev);
4073 EXPORT_SYMBOL(dev_set_promiscuity);
4076 * dev_set_allmulti - update allmulti count on a device
4080 * Add or remove reception of all multicast frames to a device. While the
4081 * count in the device remains above zero the interface remains listening
4082 * to all interfaces. Once it hits zero the device reverts back to normal
4083 * filtering operation. A negative @inc value is used to drop the counter
4084 * when releasing a resource needing all multicasts.
4085 * Return 0 if successful or a negative errno code on error.
4088 int dev_set_allmulti(struct net_device *dev, int inc)
4090 unsigned short old_flags = dev->flags;
4094 dev->flags |= IFF_ALLMULTI;
4095 dev->allmulti += inc;
4096 if (dev->allmulti == 0) {
4099 * If inc causes overflow, untouch allmulti and return error.
4102 dev->flags &= ~IFF_ALLMULTI;
4104 dev->allmulti -= inc;
4105 printk(KERN_WARNING "%s: allmulti touches roof, "
4106 "set allmulti failed, allmulti feature of "
4107 "device might be broken.\n", dev->name);
4111 if (dev->flags ^ old_flags) {
4112 dev_change_rx_flags(dev, IFF_ALLMULTI);
4113 dev_set_rx_mode(dev);
4117 EXPORT_SYMBOL(dev_set_allmulti);
4120 * Upload unicast and multicast address lists to device and
4121 * configure RX filtering. When the device doesn't support unicast
4122 * filtering it is put in promiscuous mode while unicast addresses
4125 void __dev_set_rx_mode(struct net_device *dev)
4127 const struct net_device_ops *ops = dev->netdev_ops;
4129 /* dev_open will call this function so the list will stay sane. */
4130 if (!(dev->flags&IFF_UP))
4133 if (!netif_device_present(dev))
4136 if (ops->ndo_set_rx_mode)
4137 ops->ndo_set_rx_mode(dev);
4139 /* Unicast addresses changes may only happen under the rtnl,
4140 * therefore calling __dev_set_promiscuity here is safe.
4142 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4143 __dev_set_promiscuity(dev, 1);
4144 dev->uc_promisc = 1;
4145 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4146 __dev_set_promiscuity(dev, -1);
4147 dev->uc_promisc = 0;
4150 if (ops->ndo_set_multicast_list)
4151 ops->ndo_set_multicast_list(dev);
4155 void dev_set_rx_mode(struct net_device *dev)
4157 netif_addr_lock_bh(dev);
4158 __dev_set_rx_mode(dev);
4159 netif_addr_unlock_bh(dev);
4163 * dev_get_flags - get flags reported to userspace
4166 * Get the combination of flag bits exported through APIs to userspace.
4168 unsigned dev_get_flags(const struct net_device *dev)
4172 flags = (dev->flags & ~(IFF_PROMISC |
4177 (dev->gflags & (IFF_PROMISC |
4180 if (netif_running(dev)) {
4181 if (netif_oper_up(dev))
4182 flags |= IFF_RUNNING;
4183 if (netif_carrier_ok(dev))
4184 flags |= IFF_LOWER_UP;
4185 if (netif_dormant(dev))
4186 flags |= IFF_DORMANT;
4191 EXPORT_SYMBOL(dev_get_flags);
4193 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4195 int old_flags = dev->flags;
4201 * Set the flags on our device.
4204 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4205 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4207 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4211 * Load in the correct multicast list now the flags have changed.
4214 if ((old_flags ^ flags) & IFF_MULTICAST)
4215 dev_change_rx_flags(dev, IFF_MULTICAST);
4217 dev_set_rx_mode(dev);
4220 * Have we downed the interface. We handle IFF_UP ourselves
4221 * according to user attempts to set it, rather than blindly
4226 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4227 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4230 dev_set_rx_mode(dev);
4233 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4234 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4236 dev->gflags ^= IFF_PROMISC;
4237 dev_set_promiscuity(dev, inc);
4240 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4241 is important. Some (broken) drivers set IFF_PROMISC, when
4242 IFF_ALLMULTI is requested not asking us and not reporting.
4244 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4245 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4247 dev->gflags ^= IFF_ALLMULTI;
4248 dev_set_allmulti(dev, inc);
4254 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4256 unsigned int changes = dev->flags ^ old_flags;
4258 if (changes & IFF_UP) {
4259 if (dev->flags & IFF_UP)
4260 call_netdevice_notifiers(NETDEV_UP, dev);
4262 call_netdevice_notifiers(NETDEV_DOWN, dev);
4265 if (dev->flags & IFF_UP &&
4266 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4267 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4271 * dev_change_flags - change device settings
4273 * @flags: device state flags
4275 * Change settings on device based state flags. The flags are
4276 * in the userspace exported format.
4278 int dev_change_flags(struct net_device *dev, unsigned flags)
4281 int old_flags = dev->flags;
4283 ret = __dev_change_flags(dev, flags);
4287 changes = old_flags ^ dev->flags;
4289 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4291 __dev_notify_flags(dev, old_flags);
4294 EXPORT_SYMBOL(dev_change_flags);
4297 * dev_set_mtu - Change maximum transfer unit
4299 * @new_mtu: new transfer unit
4301 * Change the maximum transfer size of the network device.
4303 int dev_set_mtu(struct net_device *dev, int new_mtu)
4305 const struct net_device_ops *ops = dev->netdev_ops;
4308 if (new_mtu == dev->mtu)
4311 /* MTU must be positive. */
4315 if (!netif_device_present(dev))
4319 if (ops->ndo_change_mtu)
4320 err = ops->ndo_change_mtu(dev, new_mtu);
4324 if (!err && dev->flags & IFF_UP)
4325 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4328 EXPORT_SYMBOL(dev_set_mtu);
4331 * dev_set_mac_address - Change Media Access Control Address
4335 * Change the hardware (MAC) address of the device
4337 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4339 const struct net_device_ops *ops = dev->netdev_ops;
4342 if (!ops->ndo_set_mac_address)
4344 if (sa->sa_family != dev->type)
4346 if (!netif_device_present(dev))
4348 err = ops->ndo_set_mac_address(dev, sa);
4350 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4353 EXPORT_SYMBOL(dev_set_mac_address);
4356 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4358 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4361 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4367 case SIOCGIFFLAGS: /* Get interface flags */
4368 ifr->ifr_flags = (short) dev_get_flags(dev);
4371 case SIOCGIFMETRIC: /* Get the metric on the interface
4372 (currently unused) */
4373 ifr->ifr_metric = 0;
4376 case SIOCGIFMTU: /* Get the MTU of a device */
4377 ifr->ifr_mtu = dev->mtu;
4382 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4384 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4385 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4386 ifr->ifr_hwaddr.sa_family = dev->type;
4394 ifr->ifr_map.mem_start = dev->mem_start;
4395 ifr->ifr_map.mem_end = dev->mem_end;
4396 ifr->ifr_map.base_addr = dev->base_addr;
4397 ifr->ifr_map.irq = dev->irq;
4398 ifr->ifr_map.dma = dev->dma;
4399 ifr->ifr_map.port = dev->if_port;
4403 ifr->ifr_ifindex = dev->ifindex;
4407 ifr->ifr_qlen = dev->tx_queue_len;
4411 /* dev_ioctl() should ensure this case
4423 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4425 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4428 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4429 const struct net_device_ops *ops;
4434 ops = dev->netdev_ops;
4437 case SIOCSIFFLAGS: /* Set interface flags */
4438 return dev_change_flags(dev, ifr->ifr_flags);
4440 case SIOCSIFMETRIC: /* Set the metric on the interface
4441 (currently unused) */
4444 case SIOCSIFMTU: /* Set the MTU of a device */
4445 return dev_set_mtu(dev, ifr->ifr_mtu);
4448 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4450 case SIOCSIFHWBROADCAST:
4451 if (ifr->ifr_hwaddr.sa_family != dev->type)
4453 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4454 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4455 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4459 if (ops->ndo_set_config) {
4460 if (!netif_device_present(dev))
4462 return ops->ndo_set_config(dev, &ifr->ifr_map);
4467 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4468 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4470 if (!netif_device_present(dev))
4472 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4475 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4476 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4478 if (!netif_device_present(dev))
4480 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4483 if (ifr->ifr_qlen < 0)
4485 dev->tx_queue_len = ifr->ifr_qlen;
4489 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4490 return dev_change_name(dev, ifr->ifr_newname);
4493 * Unknown or private ioctl
4496 if ((cmd >= SIOCDEVPRIVATE &&
4497 cmd <= SIOCDEVPRIVATE + 15) ||
4498 cmd == SIOCBONDENSLAVE ||
4499 cmd == SIOCBONDRELEASE ||
4500 cmd == SIOCBONDSETHWADDR ||
4501 cmd == SIOCBONDSLAVEINFOQUERY ||
4502 cmd == SIOCBONDINFOQUERY ||
4503 cmd == SIOCBONDCHANGEACTIVE ||
4504 cmd == SIOCGMIIPHY ||
4505 cmd == SIOCGMIIREG ||
4506 cmd == SIOCSMIIREG ||
4507 cmd == SIOCBRADDIF ||
4508 cmd == SIOCBRDELIF ||
4509 cmd == SIOCSHWTSTAMP ||
4510 cmd == SIOCWANDEV) {
4512 if (ops->ndo_do_ioctl) {
4513 if (netif_device_present(dev))
4514 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4526 * This function handles all "interface"-type I/O control requests. The actual
4527 * 'doing' part of this is dev_ifsioc above.
4531 * dev_ioctl - network device ioctl
4532 * @net: the applicable net namespace
4533 * @cmd: command to issue
4534 * @arg: pointer to a struct ifreq in user space
4536 * Issue ioctl functions to devices. This is normally called by the
4537 * user space syscall interfaces but can sometimes be useful for
4538 * other purposes. The return value is the return from the syscall if
4539 * positive or a negative errno code on error.
4542 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4548 /* One special case: SIOCGIFCONF takes ifconf argument
4549 and requires shared lock, because it sleeps writing
4553 if (cmd == SIOCGIFCONF) {
4555 ret = dev_ifconf(net, (char __user *) arg);
4559 if (cmd == SIOCGIFNAME)
4560 return dev_ifname(net, (struct ifreq __user *)arg);
4562 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4565 ifr.ifr_name[IFNAMSIZ-1] = 0;
4567 colon = strchr(ifr.ifr_name, ':');
4572 * See which interface the caller is talking about.
4577 * These ioctl calls:
4578 * - can be done by all.
4579 * - atomic and do not require locking.
4590 dev_load(net, ifr.ifr_name);
4592 ret = dev_ifsioc_locked(net, &ifr, cmd);
4597 if (copy_to_user(arg, &ifr,
4598 sizeof(struct ifreq)))
4604 dev_load(net, ifr.ifr_name);
4606 ret = dev_ethtool(net, &ifr);
4611 if (copy_to_user(arg, &ifr,
4612 sizeof(struct ifreq)))
4618 * These ioctl calls:
4619 * - require superuser power.
4620 * - require strict serialization.
4626 if (!capable(CAP_NET_ADMIN))
4628 dev_load(net, ifr.ifr_name);
4630 ret = dev_ifsioc(net, &ifr, cmd);
4635 if (copy_to_user(arg, &ifr,
4636 sizeof(struct ifreq)))
4642 * These ioctl calls:
4643 * - require superuser power.
4644 * - require strict serialization.
4645 * - do not return a value
4655 case SIOCSIFHWBROADCAST:
4658 case SIOCBONDENSLAVE:
4659 case SIOCBONDRELEASE:
4660 case SIOCBONDSETHWADDR:
4661 case SIOCBONDCHANGEACTIVE:
4665 if (!capable(CAP_NET_ADMIN))
4668 case SIOCBONDSLAVEINFOQUERY:
4669 case SIOCBONDINFOQUERY:
4670 dev_load(net, ifr.ifr_name);
4672 ret = dev_ifsioc(net, &ifr, cmd);
4677 /* Get the per device memory space. We can add this but
4678 * currently do not support it */
4680 /* Set the per device memory buffer space.
4681 * Not applicable in our case */
4686 * Unknown or private ioctl.
4689 if (cmd == SIOCWANDEV ||
4690 (cmd >= SIOCDEVPRIVATE &&
4691 cmd <= SIOCDEVPRIVATE + 15)) {
4692 dev_load(net, ifr.ifr_name);
4694 ret = dev_ifsioc(net, &ifr, cmd);
4696 if (!ret && copy_to_user(arg, &ifr,
4697 sizeof(struct ifreq)))
4701 /* Take care of Wireless Extensions */
4702 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4703 return wext_handle_ioctl(net, &ifr, cmd, arg);
4710 * dev_new_index - allocate an ifindex
4711 * @net: the applicable net namespace
4713 * Returns a suitable unique value for a new device interface
4714 * number. The caller must hold the rtnl semaphore or the
4715 * dev_base_lock to be sure it remains unique.
4717 static int dev_new_index(struct net *net)
4723 if (!__dev_get_by_index(net, ifindex))
4728 /* Delayed registration/unregisteration */
4729 static LIST_HEAD(net_todo_list);
4731 static void net_set_todo(struct net_device *dev)
4733 list_add_tail(&dev->todo_list, &net_todo_list);
4736 static void rollback_registered_many(struct list_head *head)
4738 struct net_device *dev, *tmp;
4740 BUG_ON(dev_boot_phase);
4743 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4744 /* Some devices call without registering
4745 * for initialization unwind. Remove those
4746 * devices and proceed with the remaining.
4748 if (dev->reg_state == NETREG_UNINITIALIZED) {
4749 pr_debug("unregister_netdevice: device %s/%p never "
4750 "was registered\n", dev->name, dev);
4753 list_del(&dev->unreg_list);
4757 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4759 /* If device is running, close it first. */
4762 /* And unlink it from device chain. */
4763 unlist_netdevice(dev);
4765 dev->reg_state = NETREG_UNREGISTERING;
4770 list_for_each_entry(dev, head, unreg_list) {
4771 /* Shutdown queueing discipline. */
4775 /* Notify protocols, that we are about to destroy
4776 this device. They should clean all the things.
4778 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4780 if (!dev->rtnl_link_ops ||
4781 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4782 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4785 * Flush the unicast and multicast chains
4790 if (dev->netdev_ops->ndo_uninit)
4791 dev->netdev_ops->ndo_uninit(dev);
4793 /* Notifier chain MUST detach us from master device. */
4794 WARN_ON(dev->master);
4796 /* Remove entries from kobject tree */
4797 netdev_unregister_kobject(dev);
4800 /* Process any work delayed until the end of the batch */
4801 dev = list_first_entry(head, struct net_device, unreg_list);
4802 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4806 list_for_each_entry(dev, head, unreg_list)
4810 static void rollback_registered(struct net_device *dev)
4814 list_add(&dev->unreg_list, &single);
4815 rollback_registered_many(&single);
4818 static void __netdev_init_queue_locks_one(struct net_device *dev,
4819 struct netdev_queue *dev_queue,
4822 spin_lock_init(&dev_queue->_xmit_lock);
4823 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4824 dev_queue->xmit_lock_owner = -1;
4827 static void netdev_init_queue_locks(struct net_device *dev)
4829 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4830 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4833 unsigned long netdev_fix_features(unsigned long features, const char *name)
4835 /* Fix illegal SG+CSUM combinations. */
4836 if ((features & NETIF_F_SG) &&
4837 !(features & NETIF_F_ALL_CSUM)) {
4839 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4840 "checksum feature.\n", name);
4841 features &= ~NETIF_F_SG;
4844 /* TSO requires that SG is present as well. */
4845 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4847 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4848 "SG feature.\n", name);
4849 features &= ~NETIF_F_TSO;
4852 if (features & NETIF_F_UFO) {
4853 if (!(features & NETIF_F_GEN_CSUM)) {
4855 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4856 "since no NETIF_F_HW_CSUM feature.\n",
4858 features &= ~NETIF_F_UFO;
4861 if (!(features & NETIF_F_SG)) {
4863 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4864 "since no NETIF_F_SG feature.\n", name);
4865 features &= ~NETIF_F_UFO;
4871 EXPORT_SYMBOL(netdev_fix_features);
4874 * netif_stacked_transfer_operstate - transfer operstate
4875 * @rootdev: the root or lower level device to transfer state from
4876 * @dev: the device to transfer operstate to
4878 * Transfer operational state from root to device. This is normally
4879 * called when a stacking relationship exists between the root
4880 * device and the device(a leaf device).
4882 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4883 struct net_device *dev)
4885 if (rootdev->operstate == IF_OPER_DORMANT)
4886 netif_dormant_on(dev);
4888 netif_dormant_off(dev);
4890 if (netif_carrier_ok(rootdev)) {
4891 if (!netif_carrier_ok(dev))
4892 netif_carrier_on(dev);
4894 if (netif_carrier_ok(dev))
4895 netif_carrier_off(dev);
4898 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4901 * register_netdevice - register a network device
4902 * @dev: device to register
4904 * Take a completed network device structure and add it to the kernel
4905 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4906 * chain. 0 is returned on success. A negative errno code is returned
4907 * on a failure to set up the device, or if the name is a duplicate.
4909 * Callers must hold the rtnl semaphore. You may want
4910 * register_netdev() instead of this.
4913 * The locking appears insufficient to guarantee two parallel registers
4914 * will not get the same name.
4917 int register_netdevice(struct net_device *dev)
4920 struct net *net = dev_net(dev);
4922 BUG_ON(dev_boot_phase);
4927 /* When net_device's are persistent, this will be fatal. */
4928 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4931 spin_lock_init(&dev->addr_list_lock);
4932 netdev_set_addr_lockdep_class(dev);
4933 netdev_init_queue_locks(dev);
4938 if (!dev->num_rx_queues) {
4940 * Allocate a single RX queue if driver never called
4944 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
4950 dev->_rx->first = dev->_rx;
4951 atomic_set(&dev->_rx->count, 1);
4952 dev->num_rx_queues = 1;
4955 /* Init, if this function is available */
4956 if (dev->netdev_ops->ndo_init) {
4957 ret = dev->netdev_ops->ndo_init(dev);
4965 ret = dev_get_valid_name(net, dev->name, dev->name, 0);
4969 dev->ifindex = dev_new_index(net);
4970 if (dev->iflink == -1)
4971 dev->iflink = dev->ifindex;
4973 /* Fix illegal checksum combinations */
4974 if ((dev->features & NETIF_F_HW_CSUM) &&
4975 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4976 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4978 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4981 if ((dev->features & NETIF_F_NO_CSUM) &&
4982 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4983 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4985 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4988 dev->features = netdev_fix_features(dev->features, dev->name);
4990 /* Enable software GSO if SG is supported. */
4991 if (dev->features & NETIF_F_SG)
4992 dev->features |= NETIF_F_GSO;
4994 netdev_initialize_kobject(dev);
4996 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4997 ret = notifier_to_errno(ret);
5001 ret = netdev_register_kobject(dev);
5004 dev->reg_state = NETREG_REGISTERED;
5007 * Default initial state at registry is that the
5008 * device is present.
5011 set_bit(__LINK_STATE_PRESENT, &dev->state);
5013 dev_init_scheduler(dev);
5015 list_netdevice(dev);
5017 /* Notify protocols, that a new device appeared. */
5018 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5019 ret = notifier_to_errno(ret);
5021 rollback_registered(dev);
5022 dev->reg_state = NETREG_UNREGISTERED;
5025 * Prevent userspace races by waiting until the network
5026 * device is fully setup before sending notifications.
5028 if (!dev->rtnl_link_ops ||
5029 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5030 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5036 if (dev->netdev_ops->ndo_uninit)
5037 dev->netdev_ops->ndo_uninit(dev);
5040 EXPORT_SYMBOL(register_netdevice);
5043 * init_dummy_netdev - init a dummy network device for NAPI
5044 * @dev: device to init
5046 * This takes a network device structure and initialize the minimum
5047 * amount of fields so it can be used to schedule NAPI polls without
5048 * registering a full blown interface. This is to be used by drivers
5049 * that need to tie several hardware interfaces to a single NAPI
5050 * poll scheduler due to HW limitations.
5052 int init_dummy_netdev(struct net_device *dev)
5054 /* Clear everything. Note we don't initialize spinlocks
5055 * are they aren't supposed to be taken by any of the
5056 * NAPI code and this dummy netdev is supposed to be
5057 * only ever used for NAPI polls
5059 memset(dev, 0, sizeof(struct net_device));
5061 /* make sure we BUG if trying to hit standard
5062 * register/unregister code path
5064 dev->reg_state = NETREG_DUMMY;
5066 /* initialize the ref count */
5067 atomic_set(&dev->refcnt, 1);
5069 /* NAPI wants this */
5070 INIT_LIST_HEAD(&dev->napi_list);
5072 /* a dummy interface is started by default */
5073 set_bit(__LINK_STATE_PRESENT, &dev->state);
5074 set_bit(__LINK_STATE_START, &dev->state);
5078 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5082 * register_netdev - register a network device
5083 * @dev: device to register
5085 * Take a completed network device structure and add it to the kernel
5086 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5087 * chain. 0 is returned on success. A negative errno code is returned
5088 * on a failure to set up the device, or if the name is a duplicate.
5090 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5091 * and expands the device name if you passed a format string to
5094 int register_netdev(struct net_device *dev)
5101 * If the name is a format string the caller wants us to do a
5104 if (strchr(dev->name, '%')) {
5105 err = dev_alloc_name(dev, dev->name);
5110 err = register_netdevice(dev);
5115 EXPORT_SYMBOL(register_netdev);
5118 * netdev_wait_allrefs - wait until all references are gone.
5120 * This is called when unregistering network devices.
5122 * Any protocol or device that holds a reference should register
5123 * for netdevice notification, and cleanup and put back the
5124 * reference if they receive an UNREGISTER event.
5125 * We can get stuck here if buggy protocols don't correctly
5128 static void netdev_wait_allrefs(struct net_device *dev)
5130 unsigned long rebroadcast_time, warning_time;
5132 linkwatch_forget_dev(dev);
5134 rebroadcast_time = warning_time = jiffies;
5135 while (atomic_read(&dev->refcnt) != 0) {
5136 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5139 /* Rebroadcast unregister notification */
5140 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5141 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5142 * should have already handle it the first time */
5144 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5146 /* We must not have linkwatch events
5147 * pending on unregister. If this
5148 * happens, we simply run the queue
5149 * unscheduled, resulting in a noop
5152 linkwatch_run_queue();
5157 rebroadcast_time = jiffies;
5162 if (time_after(jiffies, warning_time + 10 * HZ)) {
5163 printk(KERN_EMERG "unregister_netdevice: "
5164 "waiting for %s to become free. Usage "
5166 dev->name, atomic_read(&dev->refcnt));
5167 warning_time = jiffies;
5176 * register_netdevice(x1);
5177 * register_netdevice(x2);
5179 * unregister_netdevice(y1);
5180 * unregister_netdevice(y2);
5186 * We are invoked by rtnl_unlock().
5187 * This allows us to deal with problems:
5188 * 1) We can delete sysfs objects which invoke hotplug
5189 * without deadlocking with linkwatch via keventd.
5190 * 2) Since we run with the RTNL semaphore not held, we can sleep
5191 * safely in order to wait for the netdev refcnt to drop to zero.
5193 * We must not return until all unregister events added during
5194 * the interval the lock was held have been completed.
5196 void netdev_run_todo(void)
5198 struct list_head list;
5200 /* Snapshot list, allow later requests */
5201 list_replace_init(&net_todo_list, &list);
5205 while (!list_empty(&list)) {
5206 struct net_device *dev
5207 = list_first_entry(&list, struct net_device, todo_list);
5208 list_del(&dev->todo_list);
5210 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5211 printk(KERN_ERR "network todo '%s' but state %d\n",
5212 dev->name, dev->reg_state);
5217 dev->reg_state = NETREG_UNREGISTERED;
5219 on_each_cpu(flush_backlog, dev, 1);
5221 netdev_wait_allrefs(dev);
5224 BUG_ON(atomic_read(&dev->refcnt));
5225 WARN_ON(dev->ip_ptr);
5226 WARN_ON(dev->ip6_ptr);
5227 WARN_ON(dev->dn_ptr);
5229 if (dev->destructor)
5230 dev->destructor(dev);
5232 /* Free network device */
5233 kobject_put(&dev->dev.kobj);
5238 * dev_txq_stats_fold - fold tx_queues stats
5239 * @dev: device to get statistics from
5240 * @stats: struct net_device_stats to hold results
5242 void dev_txq_stats_fold(const struct net_device *dev,
5243 struct net_device_stats *stats)
5245 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5247 struct netdev_queue *txq;
5249 for (i = 0; i < dev->num_tx_queues; i++) {
5250 txq = netdev_get_tx_queue(dev, i);
5251 tx_bytes += txq->tx_bytes;
5252 tx_packets += txq->tx_packets;
5253 tx_dropped += txq->tx_dropped;
5255 if (tx_bytes || tx_packets || tx_dropped) {
5256 stats->tx_bytes = tx_bytes;
5257 stats->tx_packets = tx_packets;
5258 stats->tx_dropped = tx_dropped;
5261 EXPORT_SYMBOL(dev_txq_stats_fold);
5264 * dev_get_stats - get network device statistics
5265 * @dev: device to get statistics from
5267 * Get network statistics from device. The device driver may provide
5268 * its own method by setting dev->netdev_ops->get_stats; otherwise
5269 * the internal statistics structure is used.
5271 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5273 const struct net_device_ops *ops = dev->netdev_ops;
5275 if (ops->ndo_get_stats)
5276 return ops->ndo_get_stats(dev);
5278 dev_txq_stats_fold(dev, &dev->stats);
5281 EXPORT_SYMBOL(dev_get_stats);
5283 static void netdev_init_one_queue(struct net_device *dev,
5284 struct netdev_queue *queue,
5290 static void netdev_init_queues(struct net_device *dev)
5292 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5293 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5294 spin_lock_init(&dev->tx_global_lock);
5298 * alloc_netdev_mq - allocate network device
5299 * @sizeof_priv: size of private data to allocate space for
5300 * @name: device name format string
5301 * @setup: callback to initialize device
5302 * @queue_count: the number of subqueues to allocate
5304 * Allocates a struct net_device with private data area for driver use
5305 * and performs basic initialization. Also allocates subquue structs
5306 * for each queue on the device at the end of the netdevice.
5308 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5309 void (*setup)(struct net_device *), unsigned int queue_count)
5311 struct netdev_queue *tx;
5312 struct net_device *dev;
5314 struct net_device *p;
5316 struct netdev_rx_queue *rx;
5320 BUG_ON(strlen(name) >= sizeof(dev->name));
5322 alloc_size = sizeof(struct net_device);
5324 /* ensure 32-byte alignment of private area */
5325 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5326 alloc_size += sizeof_priv;
5328 /* ensure 32-byte alignment of whole construct */
5329 alloc_size += NETDEV_ALIGN - 1;
5331 p = kzalloc(alloc_size, GFP_KERNEL);
5333 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5337 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5339 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5345 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5347 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5352 atomic_set(&rx->count, queue_count);
5355 * Set a pointer to first element in the array which holds the
5358 for (i = 0; i < queue_count; i++)
5362 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5363 dev->padded = (char *)dev - (char *)p;
5365 if (dev_addr_init(dev))
5371 dev_net_set(dev, &init_net);
5374 dev->num_tx_queues = queue_count;
5375 dev->real_num_tx_queues = queue_count;
5379 dev->num_rx_queues = queue_count;
5382 dev->gso_max_size = GSO_MAX_SIZE;
5384 netdev_init_queues(dev);
5386 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5387 dev->ethtool_ntuple_list.count = 0;
5388 INIT_LIST_HEAD(&dev->napi_list);
5389 INIT_LIST_HEAD(&dev->unreg_list);
5390 INIT_LIST_HEAD(&dev->link_watch_list);
5391 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5393 strcpy(dev->name, name);
5406 EXPORT_SYMBOL(alloc_netdev_mq);
5409 * free_netdev - free network device
5412 * This function does the last stage of destroying an allocated device
5413 * interface. The reference to the device object is released.
5414 * If this is the last reference then it will be freed.
5416 void free_netdev(struct net_device *dev)
5418 struct napi_struct *p, *n;
5420 release_net(dev_net(dev));
5424 /* Flush device addresses */
5425 dev_addr_flush(dev);
5427 /* Clear ethtool n-tuple list */
5428 ethtool_ntuple_flush(dev);
5430 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5433 /* Compatibility with error handling in drivers */
5434 if (dev->reg_state == NETREG_UNINITIALIZED) {
5435 kfree((char *)dev - dev->padded);
5439 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5440 dev->reg_state = NETREG_RELEASED;
5442 /* will free via device release */
5443 put_device(&dev->dev);
5445 EXPORT_SYMBOL(free_netdev);
5448 * synchronize_net - Synchronize with packet receive processing
5450 * Wait for packets currently being received to be done.
5451 * Does not block later packets from starting.
5453 void synchronize_net(void)
5458 EXPORT_SYMBOL(synchronize_net);
5461 * unregister_netdevice_queue - remove device from the kernel
5465 * This function shuts down a device interface and removes it
5466 * from the kernel tables.
5467 * If head not NULL, device is queued to be unregistered later.
5469 * Callers must hold the rtnl semaphore. You may want
5470 * unregister_netdev() instead of this.
5473 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5478 list_move_tail(&dev->unreg_list, head);
5480 rollback_registered(dev);
5481 /* Finish processing unregister after unlock */
5485 EXPORT_SYMBOL(unregister_netdevice_queue);
5488 * unregister_netdevice_many - unregister many devices
5489 * @head: list of devices
5491 void unregister_netdevice_many(struct list_head *head)
5493 struct net_device *dev;
5495 if (!list_empty(head)) {
5496 rollback_registered_many(head);
5497 list_for_each_entry(dev, head, unreg_list)
5501 EXPORT_SYMBOL(unregister_netdevice_many);
5504 * unregister_netdev - remove device from the kernel
5507 * This function shuts down a device interface and removes it
5508 * from the kernel tables.
5510 * This is just a wrapper for unregister_netdevice that takes
5511 * the rtnl semaphore. In general you want to use this and not
5512 * unregister_netdevice.
5514 void unregister_netdev(struct net_device *dev)
5517 unregister_netdevice(dev);
5520 EXPORT_SYMBOL(unregister_netdev);
5523 * dev_change_net_namespace - move device to different nethost namespace
5525 * @net: network namespace
5526 * @pat: If not NULL name pattern to try if the current device name
5527 * is already taken in the destination network namespace.
5529 * This function shuts down a device interface and moves it
5530 * to a new network namespace. On success 0 is returned, on
5531 * a failure a netagive errno code is returned.
5533 * Callers must hold the rtnl semaphore.
5536 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5542 /* Don't allow namespace local devices to be moved. */
5544 if (dev->features & NETIF_F_NETNS_LOCAL)
5548 /* Don't allow real devices to be moved when sysfs
5552 if (dev->dev.parent)
5556 /* Ensure the device has been registrered */
5558 if (dev->reg_state != NETREG_REGISTERED)
5561 /* Get out if there is nothing todo */
5563 if (net_eq(dev_net(dev), net))
5566 /* Pick the destination device name, and ensure
5567 * we can use it in the destination network namespace.
5570 if (__dev_get_by_name(net, dev->name)) {
5571 /* We get here if we can't use the current device name */
5574 if (dev_get_valid_name(net, pat, dev->name, 1))
5579 * And now a mini version of register_netdevice unregister_netdevice.
5582 /* If device is running close it first. */
5585 /* And unlink it from device chain */
5587 unlist_netdevice(dev);
5591 /* Shutdown queueing discipline. */
5594 /* Notify protocols, that we are about to destroy
5595 this device. They should clean all the things.
5597 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5598 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5601 * Flush the unicast and multicast chains
5606 netdev_unregister_kobject(dev);
5608 /* Actually switch the network namespace */
5609 dev_net_set(dev, net);
5611 /* If there is an ifindex conflict assign a new one */
5612 if (__dev_get_by_index(net, dev->ifindex)) {
5613 int iflink = (dev->iflink == dev->ifindex);
5614 dev->ifindex = dev_new_index(net);
5616 dev->iflink = dev->ifindex;
5619 /* Fixup kobjects */
5620 err = netdev_register_kobject(dev);
5623 /* Add the device back in the hashes */
5624 list_netdevice(dev);
5626 /* Notify protocols, that a new device appeared. */
5627 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5630 * Prevent userspace races by waiting until the network
5631 * device is fully setup before sending notifications.
5633 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5640 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5642 static int dev_cpu_callback(struct notifier_block *nfb,
5643 unsigned long action,
5646 struct sk_buff **list_skb;
5647 struct sk_buff *skb;
5648 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5649 struct softnet_data *sd, *oldsd;
5651 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5654 local_irq_disable();
5655 cpu = smp_processor_id();
5656 sd = &per_cpu(softnet_data, cpu);
5657 oldsd = &per_cpu(softnet_data, oldcpu);
5659 /* Find end of our completion_queue. */
5660 list_skb = &sd->completion_queue;
5662 list_skb = &(*list_skb)->next;
5663 /* Append completion queue from offline CPU. */
5664 *list_skb = oldsd->completion_queue;
5665 oldsd->completion_queue = NULL;
5667 /* Append output queue from offline CPU. */
5668 if (oldsd->output_queue) {
5669 *sd->output_queue_tailp = oldsd->output_queue;
5670 sd->output_queue_tailp = oldsd->output_queue_tailp;
5671 oldsd->output_queue = NULL;
5672 oldsd->output_queue_tailp = &oldsd->output_queue;
5675 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5678 /* Process offline CPU's input_pkt_queue */
5679 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5681 input_queue_head_add(oldsd, 1);
5683 while ((skb = __skb_dequeue(&oldsd->process_queue)))
5691 * netdev_increment_features - increment feature set by one
5692 * @all: current feature set
5693 * @one: new feature set
5694 * @mask: mask feature set
5696 * Computes a new feature set after adding a device with feature set
5697 * @one to the master device with current feature set @all. Will not
5698 * enable anything that is off in @mask. Returns the new feature set.
5700 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5703 /* If device needs checksumming, downgrade to it. */
5704 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5705 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5706 else if (mask & NETIF_F_ALL_CSUM) {
5707 /* If one device supports v4/v6 checksumming, set for all. */
5708 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5709 !(all & NETIF_F_GEN_CSUM)) {
5710 all &= ~NETIF_F_ALL_CSUM;
5711 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5714 /* If one device supports hw checksumming, set for all. */
5715 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5716 all &= ~NETIF_F_ALL_CSUM;
5717 all |= NETIF_F_HW_CSUM;
5721 one |= NETIF_F_ALL_CSUM;
5723 one |= all & NETIF_F_ONE_FOR_ALL;
5724 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5725 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5729 EXPORT_SYMBOL(netdev_increment_features);
5731 static struct hlist_head *netdev_create_hash(void)
5734 struct hlist_head *hash;
5736 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5738 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5739 INIT_HLIST_HEAD(&hash[i]);
5744 /* Initialize per network namespace state */
5745 static int __net_init netdev_init(struct net *net)
5747 INIT_LIST_HEAD(&net->dev_base_head);
5749 net->dev_name_head = netdev_create_hash();
5750 if (net->dev_name_head == NULL)
5753 net->dev_index_head = netdev_create_hash();
5754 if (net->dev_index_head == NULL)
5760 kfree(net->dev_name_head);
5766 * netdev_drivername - network driver for the device
5767 * @dev: network device
5768 * @buffer: buffer for resulting name
5769 * @len: size of buffer
5771 * Determine network driver for device.
5773 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5775 const struct device_driver *driver;
5776 const struct device *parent;
5778 if (len <= 0 || !buffer)
5782 parent = dev->dev.parent;
5787 driver = parent->driver;
5788 if (driver && driver->name)
5789 strlcpy(buffer, driver->name, len);
5793 static void __net_exit netdev_exit(struct net *net)
5795 kfree(net->dev_name_head);
5796 kfree(net->dev_index_head);
5799 static struct pernet_operations __net_initdata netdev_net_ops = {
5800 .init = netdev_init,
5801 .exit = netdev_exit,
5804 static void __net_exit default_device_exit(struct net *net)
5806 struct net_device *dev, *aux;
5808 * Push all migratable network devices back to the
5809 * initial network namespace
5812 for_each_netdev_safe(net, dev, aux) {
5814 char fb_name[IFNAMSIZ];
5816 /* Ignore unmoveable devices (i.e. loopback) */
5817 if (dev->features & NETIF_F_NETNS_LOCAL)
5820 /* Leave virtual devices for the generic cleanup */
5821 if (dev->rtnl_link_ops)
5824 /* Push remaing network devices to init_net */
5825 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5826 err = dev_change_net_namespace(dev, &init_net, fb_name);
5828 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5829 __func__, dev->name, err);
5836 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5838 /* At exit all network devices most be removed from a network
5839 * namespace. Do this in the reverse order of registeration.
5840 * Do this across as many network namespaces as possible to
5841 * improve batching efficiency.
5843 struct net_device *dev;
5845 LIST_HEAD(dev_kill_list);
5848 list_for_each_entry(net, net_list, exit_list) {
5849 for_each_netdev_reverse(net, dev) {
5850 if (dev->rtnl_link_ops)
5851 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5853 unregister_netdevice_queue(dev, &dev_kill_list);
5856 unregister_netdevice_many(&dev_kill_list);
5860 static struct pernet_operations __net_initdata default_device_ops = {
5861 .exit = default_device_exit,
5862 .exit_batch = default_device_exit_batch,
5866 * Initialize the DEV module. At boot time this walks the device list and
5867 * unhooks any devices that fail to initialise (normally hardware not
5868 * present) and leaves us with a valid list of present and active devices.
5873 * This is called single threaded during boot, so no need
5874 * to take the rtnl semaphore.
5876 static int __init net_dev_init(void)
5878 int i, rc = -ENOMEM;
5880 BUG_ON(!dev_boot_phase);
5882 if (dev_proc_init())
5885 if (netdev_kobject_init())
5888 INIT_LIST_HEAD(&ptype_all);
5889 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5890 INIT_LIST_HEAD(&ptype_base[i]);
5892 if (register_pernet_subsys(&netdev_net_ops))
5896 * Initialise the packet receive queues.
5899 for_each_possible_cpu(i) {
5900 struct softnet_data *sd = &per_cpu(softnet_data, i);
5902 memset(sd, 0, sizeof(*sd));
5903 skb_queue_head_init(&sd->input_pkt_queue);
5904 skb_queue_head_init(&sd->process_queue);
5905 sd->completion_queue = NULL;
5906 INIT_LIST_HEAD(&sd->poll_list);
5907 sd->output_queue = NULL;
5908 sd->output_queue_tailp = &sd->output_queue;
5910 sd->csd.func = rps_trigger_softirq;
5916 sd->backlog.poll = process_backlog;
5917 sd->backlog.weight = weight_p;
5918 sd->backlog.gro_list = NULL;
5919 sd->backlog.gro_count = 0;
5924 /* The loopback device is special if any other network devices
5925 * is present in a network namespace the loopback device must
5926 * be present. Since we now dynamically allocate and free the
5927 * loopback device ensure this invariant is maintained by
5928 * keeping the loopback device as the first device on the
5929 * list of network devices. Ensuring the loopback devices
5930 * is the first device that appears and the last network device
5933 if (register_pernet_device(&loopback_net_ops))
5936 if (register_pernet_device(&default_device_ops))
5939 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5940 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5942 hotcpu_notifier(dev_cpu_callback, 0);
5950 subsys_initcall(net_dev_init);
5952 static int __init initialize_hashrnd(void)
5954 get_random_bytes(&hashrnd, sizeof(hashrnd));
5958 late_initcall_sync(initialize_hashrnd);