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)
1438 return raw_notifier_call_chain(&netdev_chain, val, dev);
1441 /* When > 0 there are consumers of rx skb time stamps */
1442 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1444 void net_enable_timestamp(void)
1446 atomic_inc(&netstamp_needed);
1448 EXPORT_SYMBOL(net_enable_timestamp);
1450 void net_disable_timestamp(void)
1452 atomic_dec(&netstamp_needed);
1454 EXPORT_SYMBOL(net_disable_timestamp);
1456 static inline void net_timestamp(struct sk_buff *skb)
1458 if (atomic_read(&netstamp_needed))
1459 __net_timestamp(skb);
1461 skb->tstamp.tv64 = 0;
1465 * dev_forward_skb - loopback an skb to another netif
1467 * @dev: destination network device
1468 * @skb: buffer to forward
1471 * NET_RX_SUCCESS (no congestion)
1472 * NET_RX_DROP (packet was dropped)
1474 * dev_forward_skb can be used for injecting an skb from the
1475 * start_xmit function of one device into the receive queue
1476 * of another device.
1478 * The receiving device may be in another namespace, so
1479 * we have to clear all information in the skb that could
1480 * impact namespace isolation.
1482 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1486 if (!(dev->flags & IFF_UP))
1489 if (skb->len > (dev->mtu + dev->hard_header_len))
1492 skb_set_dev(skb, dev);
1493 skb->tstamp.tv64 = 0;
1494 skb->pkt_type = PACKET_HOST;
1495 skb->protocol = eth_type_trans(skb, dev);
1496 return netif_rx(skb);
1498 EXPORT_SYMBOL_GPL(dev_forward_skb);
1501 * Support routine. Sends outgoing frames to any network
1502 * taps currently in use.
1505 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1507 struct packet_type *ptype;
1509 #ifdef CONFIG_NET_CLS_ACT
1510 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1517 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1518 /* Never send packets back to the socket
1519 * they originated from - MvS (miquels@drinkel.ow.org)
1521 if ((ptype->dev == dev || !ptype->dev) &&
1522 (ptype->af_packet_priv == NULL ||
1523 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1524 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1528 /* skb->nh should be correctly
1529 set by sender, so that the second statement is
1530 just protection against buggy protocols.
1532 skb_reset_mac_header(skb2);
1534 if (skb_network_header(skb2) < skb2->data ||
1535 skb2->network_header > skb2->tail) {
1536 if (net_ratelimit())
1537 printk(KERN_CRIT "protocol %04x is "
1539 skb2->protocol, dev->name);
1540 skb_reset_network_header(skb2);
1543 skb2->transport_header = skb2->network_header;
1544 skb2->pkt_type = PACKET_OUTGOING;
1545 ptype->func(skb2, skb->dev, ptype, skb->dev);
1552 static inline void __netif_reschedule(struct Qdisc *q)
1554 struct softnet_data *sd;
1555 unsigned long flags;
1557 local_irq_save(flags);
1558 sd = &__get_cpu_var(softnet_data);
1559 q->next_sched = sd->output_queue;
1560 sd->output_queue = q;
1561 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1562 local_irq_restore(flags);
1565 void __netif_schedule(struct Qdisc *q)
1567 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1568 __netif_reschedule(q);
1570 EXPORT_SYMBOL(__netif_schedule);
1572 void dev_kfree_skb_irq(struct sk_buff *skb)
1574 if (atomic_dec_and_test(&skb->users)) {
1575 struct softnet_data *sd;
1576 unsigned long flags;
1578 local_irq_save(flags);
1579 sd = &__get_cpu_var(softnet_data);
1580 skb->next = sd->completion_queue;
1581 sd->completion_queue = skb;
1582 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1583 local_irq_restore(flags);
1586 EXPORT_SYMBOL(dev_kfree_skb_irq);
1588 void dev_kfree_skb_any(struct sk_buff *skb)
1590 if (in_irq() || irqs_disabled())
1591 dev_kfree_skb_irq(skb);
1595 EXPORT_SYMBOL(dev_kfree_skb_any);
1599 * netif_device_detach - mark device as removed
1600 * @dev: network device
1602 * Mark device as removed from system and therefore no longer available.
1604 void netif_device_detach(struct net_device *dev)
1606 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1607 netif_running(dev)) {
1608 netif_tx_stop_all_queues(dev);
1611 EXPORT_SYMBOL(netif_device_detach);
1614 * netif_device_attach - mark device as attached
1615 * @dev: network device
1617 * Mark device as attached from system and restart if needed.
1619 void netif_device_attach(struct net_device *dev)
1621 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1622 netif_running(dev)) {
1623 netif_tx_wake_all_queues(dev);
1624 __netdev_watchdog_up(dev);
1627 EXPORT_SYMBOL(netif_device_attach);
1629 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1631 return ((features & NETIF_F_GEN_CSUM) ||
1632 ((features & NETIF_F_IP_CSUM) &&
1633 protocol == htons(ETH_P_IP)) ||
1634 ((features & NETIF_F_IPV6_CSUM) &&
1635 protocol == htons(ETH_P_IPV6)) ||
1636 ((features & NETIF_F_FCOE_CRC) &&
1637 protocol == htons(ETH_P_FCOE)));
1640 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1642 if (can_checksum_protocol(dev->features, skb->protocol))
1645 if (skb->protocol == htons(ETH_P_8021Q)) {
1646 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1647 if (can_checksum_protocol(dev->features & dev->vlan_features,
1648 veh->h_vlan_encapsulated_proto))
1656 * skb_dev_set -- assign a new device to a buffer
1657 * @skb: buffer for the new device
1658 * @dev: network device
1660 * If an skb is owned by a device already, we have to reset
1661 * all data private to the namespace a device belongs to
1662 * before assigning it a new device.
1664 #ifdef CONFIG_NET_NS
1665 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1668 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1671 skb_init_secmark(skb);
1675 skb->ipvs_property = 0;
1676 #ifdef CONFIG_NET_SCHED
1682 EXPORT_SYMBOL(skb_set_dev);
1683 #endif /* CONFIG_NET_NS */
1686 * Invalidate hardware checksum when packet is to be mangled, and
1687 * complete checksum manually on outgoing path.
1689 int skb_checksum_help(struct sk_buff *skb)
1692 int ret = 0, offset;
1694 if (skb->ip_summed == CHECKSUM_COMPLETE)
1695 goto out_set_summed;
1697 if (unlikely(skb_shinfo(skb)->gso_size)) {
1698 /* Let GSO fix up the checksum. */
1699 goto out_set_summed;
1702 offset = skb->csum_start - skb_headroom(skb);
1703 BUG_ON(offset >= skb_headlen(skb));
1704 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1706 offset += skb->csum_offset;
1707 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1709 if (skb_cloned(skb) &&
1710 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1711 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1716 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1718 skb->ip_summed = CHECKSUM_NONE;
1722 EXPORT_SYMBOL(skb_checksum_help);
1725 * skb_gso_segment - Perform segmentation on skb.
1726 * @skb: buffer to segment
1727 * @features: features for the output path (see dev->features)
1729 * This function segments the given skb and returns a list of segments.
1731 * It may return NULL if the skb requires no segmentation. This is
1732 * only possible when GSO is used for verifying header integrity.
1734 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1736 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1737 struct packet_type *ptype;
1738 __be16 type = skb->protocol;
1741 skb_reset_mac_header(skb);
1742 skb->mac_len = skb->network_header - skb->mac_header;
1743 __skb_pull(skb, skb->mac_len);
1745 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1746 struct net_device *dev = skb->dev;
1747 struct ethtool_drvinfo info = {};
1749 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1750 dev->ethtool_ops->get_drvinfo(dev, &info);
1752 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1754 info.driver, dev ? dev->features : 0L,
1755 skb->sk ? skb->sk->sk_route_caps : 0L,
1756 skb->len, skb->data_len, skb->ip_summed);
1758 if (skb_header_cloned(skb) &&
1759 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1760 return ERR_PTR(err);
1764 list_for_each_entry_rcu(ptype,
1765 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1766 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1767 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1768 err = ptype->gso_send_check(skb);
1769 segs = ERR_PTR(err);
1770 if (err || skb_gso_ok(skb, features))
1772 __skb_push(skb, (skb->data -
1773 skb_network_header(skb)));
1775 segs = ptype->gso_segment(skb, features);
1781 __skb_push(skb, skb->data - skb_mac_header(skb));
1785 EXPORT_SYMBOL(skb_gso_segment);
1787 /* Take action when hardware reception checksum errors are detected. */
1789 void netdev_rx_csum_fault(struct net_device *dev)
1791 if (net_ratelimit()) {
1792 printk(KERN_ERR "%s: hw csum failure.\n",
1793 dev ? dev->name : "<unknown>");
1797 EXPORT_SYMBOL(netdev_rx_csum_fault);
1800 /* Actually, we should eliminate this check as soon as we know, that:
1801 * 1. IOMMU is present and allows to map all the memory.
1802 * 2. No high memory really exists on this machine.
1805 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1807 #ifdef CONFIG_HIGHMEM
1809 if (!(dev->features & NETIF_F_HIGHDMA)) {
1810 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1811 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1815 if (PCI_DMA_BUS_IS_PHYS) {
1816 struct device *pdev = dev->dev.parent;
1820 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1821 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1822 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1831 void (*destructor)(struct sk_buff *skb);
1834 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1836 static void dev_gso_skb_destructor(struct sk_buff *skb)
1838 struct dev_gso_cb *cb;
1841 struct sk_buff *nskb = skb->next;
1843 skb->next = nskb->next;
1846 } while (skb->next);
1848 cb = DEV_GSO_CB(skb);
1850 cb->destructor(skb);
1854 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1855 * @skb: buffer to segment
1857 * This function segments the given skb and stores the list of segments
1860 static int dev_gso_segment(struct sk_buff *skb)
1862 struct net_device *dev = skb->dev;
1863 struct sk_buff *segs;
1864 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1867 segs = skb_gso_segment(skb, features);
1869 /* Verifying header integrity only. */
1874 return PTR_ERR(segs);
1877 DEV_GSO_CB(skb)->destructor = skb->destructor;
1878 skb->destructor = dev_gso_skb_destructor;
1884 * Try to orphan skb early, right before transmission by the device.
1885 * We cannot orphan skb if tx timestamp is requested, since
1886 * drivers need to call skb_tstamp_tx() to send the timestamp.
1888 static inline void skb_orphan_try(struct sk_buff *skb)
1890 if (!skb_tx(skb)->flags)
1894 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1895 struct netdev_queue *txq)
1897 const struct net_device_ops *ops = dev->netdev_ops;
1898 int rc = NETDEV_TX_OK;
1900 if (likely(!skb->next)) {
1901 if (!list_empty(&ptype_all))
1902 dev_queue_xmit_nit(skb, dev);
1904 if (netif_needs_gso(dev, skb)) {
1905 if (unlikely(dev_gso_segment(skb)))
1912 * If device doesnt need skb->dst, release it right now while
1913 * its hot in this cpu cache
1915 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1918 skb_orphan_try(skb);
1919 rc = ops->ndo_start_xmit(skb, dev);
1920 if (rc == NETDEV_TX_OK)
1921 txq_trans_update(txq);
1927 struct sk_buff *nskb = skb->next;
1929 skb->next = nskb->next;
1933 * If device doesnt need nskb->dst, release it right now while
1934 * its hot in this cpu cache
1936 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1939 skb_orphan_try(nskb);
1940 rc = ops->ndo_start_xmit(nskb, dev);
1941 if (unlikely(rc != NETDEV_TX_OK)) {
1942 if (rc & ~NETDEV_TX_MASK)
1943 goto out_kfree_gso_skb;
1944 nskb->next = skb->next;
1948 txq_trans_update(txq);
1949 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1950 return NETDEV_TX_BUSY;
1951 } while (skb->next);
1954 if (likely(skb->next == NULL))
1955 skb->destructor = DEV_GSO_CB(skb)->destructor;
1961 static u32 hashrnd __read_mostly;
1963 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1967 if (skb_rx_queue_recorded(skb)) {
1968 hash = skb_get_rx_queue(skb);
1969 while (unlikely(hash >= dev->real_num_tx_queues))
1970 hash -= dev->real_num_tx_queues;
1974 if (skb->sk && skb->sk->sk_hash)
1975 hash = skb->sk->sk_hash;
1977 hash = skb->protocol;
1979 hash = jhash_1word(hash, hashrnd);
1981 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1983 EXPORT_SYMBOL(skb_tx_hash);
1985 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1987 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1988 if (net_ratelimit()) {
1989 pr_warning("%s selects TX queue %d, but "
1990 "real number of TX queues is %d\n",
1991 dev->name, queue_index, dev->real_num_tx_queues);
1998 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1999 struct sk_buff *skb)
2002 struct sock *sk = skb->sk;
2004 if (sk_tx_queue_recorded(sk)) {
2005 queue_index = sk_tx_queue_get(sk);
2007 const struct net_device_ops *ops = dev->netdev_ops;
2009 if (ops->ndo_select_queue) {
2010 queue_index = ops->ndo_select_queue(dev, skb);
2011 queue_index = dev_cap_txqueue(dev, queue_index);
2014 if (dev->real_num_tx_queues > 1)
2015 queue_index = skb_tx_hash(dev, skb);
2017 if (sk && rcu_dereference_check(sk->sk_dst_cache, 1))
2018 sk_tx_queue_set(sk, queue_index);
2022 skb_set_queue_mapping(skb, queue_index);
2023 return netdev_get_tx_queue(dev, queue_index);
2026 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2027 struct net_device *dev,
2028 struct netdev_queue *txq)
2030 spinlock_t *root_lock = qdisc_lock(q);
2033 spin_lock(root_lock);
2034 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2037 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2038 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
2040 * This is a work-conserving queue; there are no old skbs
2041 * waiting to be sent out; and the qdisc is not running -
2042 * xmit the skb directly.
2044 __qdisc_update_bstats(q, skb->len);
2045 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
2048 clear_bit(__QDISC_STATE_RUNNING, &q->state);
2050 rc = NET_XMIT_SUCCESS;
2052 rc = qdisc_enqueue_root(skb, q);
2055 spin_unlock(root_lock);
2061 * Returns true if either:
2062 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2063 * 2. skb is fragmented and the device does not support SG, or if
2064 * at least one of fragments is in highmem and device does not
2065 * support DMA from it.
2067 static inline int skb_needs_linearize(struct sk_buff *skb,
2068 struct net_device *dev)
2070 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
2071 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
2072 illegal_highdma(dev, skb)));
2076 * dev_queue_xmit - transmit a buffer
2077 * @skb: buffer to transmit
2079 * Queue a buffer for transmission to a network device. The caller must
2080 * have set the device and priority and built the buffer before calling
2081 * this function. The function can be called from an interrupt.
2083 * A negative errno code is returned on a failure. A success does not
2084 * guarantee the frame will be transmitted as it may be dropped due
2085 * to congestion or traffic shaping.
2087 * -----------------------------------------------------------------------------------
2088 * I notice this method can also return errors from the queue disciplines,
2089 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2092 * Regardless of the return value, the skb is consumed, so it is currently
2093 * difficult to retry a send to this method. (You can bump the ref count
2094 * before sending to hold a reference for retry if you are careful.)
2096 * When calling this method, interrupts MUST be enabled. This is because
2097 * the BH enable code must have IRQs enabled so that it will not deadlock.
2100 int dev_queue_xmit(struct sk_buff *skb)
2102 struct net_device *dev = skb->dev;
2103 struct netdev_queue *txq;
2107 /* GSO will handle the following emulations directly. */
2108 if (netif_needs_gso(dev, skb))
2111 /* Convert a paged skb to linear, if required */
2112 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2115 /* If packet is not checksummed and device does not support
2116 * checksumming for this protocol, complete checksumming here.
2118 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2119 skb_set_transport_header(skb, skb->csum_start -
2121 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2126 /* Disable soft irqs for various locks below. Also
2127 * stops preemption for RCU.
2131 txq = dev_pick_tx(dev, skb);
2132 q = rcu_dereference_bh(txq->qdisc);
2134 #ifdef CONFIG_NET_CLS_ACT
2135 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2138 rc = __dev_xmit_skb(skb, q, dev, txq);
2142 /* The device has no queue. Common case for software devices:
2143 loopback, all the sorts of tunnels...
2145 Really, it is unlikely that netif_tx_lock protection is necessary
2146 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2148 However, it is possible, that they rely on protection
2151 Check this and shot the lock. It is not prone from deadlocks.
2152 Either shot noqueue qdisc, it is even simpler 8)
2154 if (dev->flags & IFF_UP) {
2155 int cpu = smp_processor_id(); /* ok because BHs are off */
2157 if (txq->xmit_lock_owner != cpu) {
2159 HARD_TX_LOCK(dev, txq, cpu);
2161 if (!netif_tx_queue_stopped(txq)) {
2162 rc = dev_hard_start_xmit(skb, dev, txq);
2163 if (dev_xmit_complete(rc)) {
2164 HARD_TX_UNLOCK(dev, txq);
2168 HARD_TX_UNLOCK(dev, txq);
2169 if (net_ratelimit())
2170 printk(KERN_CRIT "Virtual device %s asks to "
2171 "queue packet!\n", dev->name);
2173 /* Recursion is detected! It is possible,
2175 if (net_ratelimit())
2176 printk(KERN_CRIT "Dead loop on virtual device "
2177 "%s, fix it urgently!\n", dev->name);
2182 rcu_read_unlock_bh();
2188 rcu_read_unlock_bh();
2191 EXPORT_SYMBOL(dev_queue_xmit);
2194 /*=======================================================================
2196 =======================================================================*/
2198 int netdev_max_backlog __read_mostly = 1000;
2199 int netdev_budget __read_mostly = 300;
2200 int weight_p __read_mostly = 64; /* old backlog weight */
2202 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2206 /* One global table that all flow-based protocols share. */
2207 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2208 EXPORT_SYMBOL(rps_sock_flow_table);
2211 * get_rps_cpu is called from netif_receive_skb and returns the target
2212 * CPU from the RPS map of the receiving queue for a given skb.
2213 * rcu_read_lock must be held on entry.
2215 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2216 struct rps_dev_flow **rflowp)
2218 struct ipv6hdr *ip6;
2220 struct netdev_rx_queue *rxqueue;
2221 struct rps_map *map;
2222 struct rps_dev_flow_table *flow_table;
2223 struct rps_sock_flow_table *sock_flow_table;
2227 u32 addr1, addr2, ports, ihl;
2229 if (skb_rx_queue_recorded(skb)) {
2230 u16 index = skb_get_rx_queue(skb);
2231 if (unlikely(index >= dev->num_rx_queues)) {
2232 if (net_ratelimit()) {
2233 pr_warning("%s received packet on queue "
2234 "%u, but number of RX queues is %u\n",
2235 dev->name, index, dev->num_rx_queues);
2239 rxqueue = dev->_rx + index;
2243 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2247 goto got_hash; /* Skip hash computation on packet header */
2249 switch (skb->protocol) {
2250 case __constant_htons(ETH_P_IP):
2251 if (!pskb_may_pull(skb, sizeof(*ip)))
2254 ip = (struct iphdr *) skb->data;
2255 ip_proto = ip->protocol;
2260 case __constant_htons(ETH_P_IPV6):
2261 if (!pskb_may_pull(skb, sizeof(*ip6)))
2264 ip6 = (struct ipv6hdr *) skb->data;
2265 ip_proto = ip6->nexthdr;
2266 addr1 = ip6->saddr.s6_addr32[3];
2267 addr2 = ip6->daddr.s6_addr32[3];
2281 case IPPROTO_UDPLITE:
2282 if (pskb_may_pull(skb, (ihl * 4) + 4))
2283 ports = *((u32 *) (skb->data + (ihl * 4)));
2290 skb->rxhash = jhash_3words(addr1, addr2, ports, hashrnd);
2295 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2296 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2297 if (flow_table && sock_flow_table) {
2299 struct rps_dev_flow *rflow;
2301 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2304 next_cpu = sock_flow_table->ents[skb->rxhash &
2305 sock_flow_table->mask];
2308 * If the desired CPU (where last recvmsg was done) is
2309 * different from current CPU (one in the rx-queue flow
2310 * table entry), switch if one of the following holds:
2311 * - Current CPU is unset (equal to RPS_NO_CPU).
2312 * - Current CPU is offline.
2313 * - The current CPU's queue tail has advanced beyond the
2314 * last packet that was enqueued using this table entry.
2315 * This guarantees that all previous packets for the flow
2316 * have been dequeued, thus preserving in order delivery.
2318 if (unlikely(tcpu != next_cpu) &&
2319 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2320 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2321 rflow->last_qtail)) >= 0)) {
2322 tcpu = rflow->cpu = next_cpu;
2323 if (tcpu != RPS_NO_CPU)
2324 rflow->last_qtail = per_cpu(softnet_data,
2325 tcpu).input_queue_head;
2327 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2334 map = rcu_dereference(rxqueue->rps_map);
2336 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2338 if (cpu_online(tcpu)) {
2348 /* Called from hardirq (IPI) context */
2349 static void rps_trigger_softirq(void *data)
2351 struct softnet_data *sd = data;
2353 __napi_schedule(&sd->backlog);
2354 __get_cpu_var(netdev_rx_stat).received_rps++;
2357 #endif /* CONFIG_RPS */
2360 * Check if this softnet_data structure is another cpu one
2361 * If yes, queue it to our IPI list and return 1
2364 static int rps_ipi_queued(struct softnet_data *sd)
2367 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2370 sd->rps_ipi_next = mysd->rps_ipi_list;
2371 mysd->rps_ipi_list = sd;
2373 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2376 #endif /* CONFIG_RPS */
2381 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2382 * queue (may be a remote CPU queue).
2384 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2385 unsigned int *qtail)
2387 struct softnet_data *sd;
2388 unsigned long flags;
2390 sd = &per_cpu(softnet_data, cpu);
2392 local_irq_save(flags);
2393 __get_cpu_var(netdev_rx_stat).total++;
2396 if (sd->input_pkt_queue.qlen <= netdev_max_backlog) {
2397 if (sd->input_pkt_queue.qlen) {
2399 __skb_queue_tail(&sd->input_pkt_queue, skb);
2401 *qtail = sd->input_queue_head + sd->input_pkt_queue.qlen;
2404 local_irq_restore(flags);
2405 return NET_RX_SUCCESS;
2408 /* Schedule NAPI for backlog device */
2409 if (napi_schedule_prep(&sd->backlog)) {
2410 if (!rps_ipi_queued(sd))
2411 __napi_schedule(&sd->backlog);
2418 __get_cpu_var(netdev_rx_stat).dropped++;
2419 local_irq_restore(flags);
2426 * netif_rx - post buffer to the network code
2427 * @skb: buffer to post
2429 * This function receives a packet from a device driver and queues it for
2430 * the upper (protocol) levels to process. It always succeeds. The buffer
2431 * may be dropped during processing for congestion control or by the
2435 * NET_RX_SUCCESS (no congestion)
2436 * NET_RX_DROP (packet was dropped)
2440 int netif_rx(struct sk_buff *skb)
2444 /* if netpoll wants it, pretend we never saw it */
2445 if (netpoll_rx(skb))
2448 if (!skb->tstamp.tv64)
2453 struct rps_dev_flow voidflow, *rflow = &voidflow;
2458 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2460 cpu = smp_processor_id();
2462 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2469 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2475 EXPORT_SYMBOL(netif_rx);
2477 int netif_rx_ni(struct sk_buff *skb)
2482 err = netif_rx(skb);
2483 if (local_softirq_pending())
2489 EXPORT_SYMBOL(netif_rx_ni);
2491 static void net_tx_action(struct softirq_action *h)
2493 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2495 if (sd->completion_queue) {
2496 struct sk_buff *clist;
2498 local_irq_disable();
2499 clist = sd->completion_queue;
2500 sd->completion_queue = NULL;
2504 struct sk_buff *skb = clist;
2505 clist = clist->next;
2507 WARN_ON(atomic_read(&skb->users));
2512 if (sd->output_queue) {
2515 local_irq_disable();
2516 head = sd->output_queue;
2517 sd->output_queue = NULL;
2521 struct Qdisc *q = head;
2522 spinlock_t *root_lock;
2524 head = head->next_sched;
2526 root_lock = qdisc_lock(q);
2527 if (spin_trylock(root_lock)) {
2528 smp_mb__before_clear_bit();
2529 clear_bit(__QDISC_STATE_SCHED,
2532 spin_unlock(root_lock);
2534 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2536 __netif_reschedule(q);
2538 smp_mb__before_clear_bit();
2539 clear_bit(__QDISC_STATE_SCHED,
2547 static inline int deliver_skb(struct sk_buff *skb,
2548 struct packet_type *pt_prev,
2549 struct net_device *orig_dev)
2551 atomic_inc(&skb->users);
2552 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2555 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2557 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2558 /* This hook is defined here for ATM LANE */
2559 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2560 unsigned char *addr) __read_mostly;
2561 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2565 * If bridge module is loaded call bridging hook.
2566 * returns NULL if packet was consumed.
2568 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2569 struct sk_buff *skb) __read_mostly;
2570 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2572 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2573 struct packet_type **pt_prev, int *ret,
2574 struct net_device *orig_dev)
2576 struct net_bridge_port *port;
2578 if (skb->pkt_type == PACKET_LOOPBACK ||
2579 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2583 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2587 return br_handle_frame_hook(port, skb);
2590 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2593 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2594 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2595 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2597 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2598 struct packet_type **pt_prev,
2600 struct net_device *orig_dev)
2602 if (skb->dev->macvlan_port == NULL)
2606 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2609 return macvlan_handle_frame_hook(skb);
2612 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2615 #ifdef CONFIG_NET_CLS_ACT
2616 /* TODO: Maybe we should just force sch_ingress to be compiled in
2617 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2618 * a compare and 2 stores extra right now if we dont have it on
2619 * but have CONFIG_NET_CLS_ACT
2620 * NOTE: This doesnt stop any functionality; if you dont have
2621 * the ingress scheduler, you just cant add policies on ingress.
2624 static int ing_filter(struct sk_buff *skb)
2626 struct net_device *dev = skb->dev;
2627 u32 ttl = G_TC_RTTL(skb->tc_verd);
2628 struct netdev_queue *rxq;
2629 int result = TC_ACT_OK;
2632 if (MAX_RED_LOOP < ttl++) {
2634 "Redir loop detected Dropping packet (%d->%d)\n",
2635 skb->skb_iif, dev->ifindex);
2639 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2640 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2642 rxq = &dev->rx_queue;
2645 if (q != &noop_qdisc) {
2646 spin_lock(qdisc_lock(q));
2647 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2648 result = qdisc_enqueue_root(skb, q);
2649 spin_unlock(qdisc_lock(q));
2655 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2656 struct packet_type **pt_prev,
2657 int *ret, struct net_device *orig_dev)
2659 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2663 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2666 /* Huh? Why does turning on AF_PACKET affect this? */
2667 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2670 switch (ing_filter(skb)) {
2684 * netif_nit_deliver - deliver received packets to network taps
2687 * This function is used to deliver incoming packets to network
2688 * taps. It should be used when the normal netif_receive_skb path
2689 * is bypassed, for example because of VLAN acceleration.
2691 void netif_nit_deliver(struct sk_buff *skb)
2693 struct packet_type *ptype;
2695 if (list_empty(&ptype_all))
2698 skb_reset_network_header(skb);
2699 skb_reset_transport_header(skb);
2700 skb->mac_len = skb->network_header - skb->mac_header;
2703 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2704 if (!ptype->dev || ptype->dev == skb->dev)
2705 deliver_skb(skb, ptype, skb->dev);
2710 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2711 struct net_device *master)
2713 if (skb->pkt_type == PACKET_HOST) {
2714 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2716 memcpy(dest, master->dev_addr, ETH_ALEN);
2720 /* On bonding slaves other than the currently active slave, suppress
2721 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2722 * ARP on active-backup slaves with arp_validate enabled.
2724 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2726 struct net_device *dev = skb->dev;
2728 if (master->priv_flags & IFF_MASTER_ARPMON)
2729 dev->last_rx = jiffies;
2731 if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) {
2732 /* Do address unmangle. The local destination address
2733 * will be always the one master has. Provides the right
2734 * functionality in a bridge.
2736 skb_bond_set_mac_by_master(skb, master);
2739 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2740 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2741 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2744 if (master->priv_flags & IFF_MASTER_ALB) {
2745 if (skb->pkt_type != PACKET_BROADCAST &&
2746 skb->pkt_type != PACKET_MULTICAST)
2749 if (master->priv_flags & IFF_MASTER_8023AD &&
2750 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2757 EXPORT_SYMBOL(__skb_bond_should_drop);
2759 static int __netif_receive_skb(struct sk_buff *skb)
2761 struct packet_type *ptype, *pt_prev;
2762 struct net_device *orig_dev;
2763 struct net_device *master;
2764 struct net_device *null_or_orig;
2765 struct net_device *null_or_bond;
2766 int ret = NET_RX_DROP;
2769 if (!skb->tstamp.tv64)
2772 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2773 return NET_RX_SUCCESS;
2775 /* if we've gotten here through NAPI, check netpoll */
2776 if (netpoll_receive_skb(skb))
2780 skb->skb_iif = skb->dev->ifindex;
2782 null_or_orig = NULL;
2783 orig_dev = skb->dev;
2784 master = ACCESS_ONCE(orig_dev->master);
2786 if (skb_bond_should_drop(skb, master))
2787 null_or_orig = orig_dev; /* deliver only exact match */
2792 __get_cpu_var(netdev_rx_stat).total++;
2794 skb_reset_network_header(skb);
2795 skb_reset_transport_header(skb);
2796 skb->mac_len = skb->network_header - skb->mac_header;
2802 #ifdef CONFIG_NET_CLS_ACT
2803 if (skb->tc_verd & TC_NCLS) {
2804 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2809 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2810 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2811 ptype->dev == orig_dev) {
2813 ret = deliver_skb(skb, pt_prev, orig_dev);
2818 #ifdef CONFIG_NET_CLS_ACT
2819 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2825 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2828 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2833 * Make sure frames received on VLAN interfaces stacked on
2834 * bonding interfaces still make their way to any base bonding
2835 * device that may have registered for a specific ptype. The
2836 * handler may have to adjust skb->dev and orig_dev.
2838 null_or_bond = NULL;
2839 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2840 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2841 null_or_bond = vlan_dev_real_dev(skb->dev);
2844 type = skb->protocol;
2845 list_for_each_entry_rcu(ptype,
2846 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2847 if (ptype->type == type && (ptype->dev == null_or_orig ||
2848 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2849 ptype->dev == null_or_bond)) {
2851 ret = deliver_skb(skb, pt_prev, orig_dev);
2857 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2860 /* Jamal, now you will not able to escape explaining
2861 * me how you were going to use this. :-)
2872 * netif_receive_skb - process receive buffer from network
2873 * @skb: buffer to process
2875 * netif_receive_skb() is the main receive data processing function.
2876 * It always succeeds. The buffer may be dropped during processing
2877 * for congestion control or by the protocol layers.
2879 * This function may only be called from softirq context and interrupts
2880 * should be enabled.
2882 * Return values (usually ignored):
2883 * NET_RX_SUCCESS: no congestion
2884 * NET_RX_DROP: packet was dropped
2886 int netif_receive_skb(struct sk_buff *skb)
2889 struct rps_dev_flow voidflow, *rflow = &voidflow;
2894 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2897 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2901 ret = __netif_receive_skb(skb);
2906 return __netif_receive_skb(skb);
2909 EXPORT_SYMBOL(netif_receive_skb);
2911 /* Network device is going away, flush any packets still pending
2912 * Called with irqs disabled.
2914 static void flush_backlog(void *arg)
2916 struct net_device *dev = arg;
2917 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2918 struct sk_buff *skb, *tmp;
2921 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp)
2922 if (skb->dev == dev) {
2923 __skb_unlink(skb, &sd->input_pkt_queue);
2925 input_queue_head_incr(sd);
2930 static int napi_gro_complete(struct sk_buff *skb)
2932 struct packet_type *ptype;
2933 __be16 type = skb->protocol;
2934 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2937 if (NAPI_GRO_CB(skb)->count == 1) {
2938 skb_shinfo(skb)->gso_size = 0;
2943 list_for_each_entry_rcu(ptype, head, list) {
2944 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2947 err = ptype->gro_complete(skb);
2953 WARN_ON(&ptype->list == head);
2955 return NET_RX_SUCCESS;
2959 return netif_receive_skb(skb);
2962 static void napi_gro_flush(struct napi_struct *napi)
2964 struct sk_buff *skb, *next;
2966 for (skb = napi->gro_list; skb; skb = next) {
2969 napi_gro_complete(skb);
2972 napi->gro_count = 0;
2973 napi->gro_list = NULL;
2976 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2978 struct sk_buff **pp = NULL;
2979 struct packet_type *ptype;
2980 __be16 type = skb->protocol;
2981 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2984 enum gro_result ret;
2986 if (!(skb->dev->features & NETIF_F_GRO))
2989 if (skb_is_gso(skb) || skb_has_frags(skb))
2993 list_for_each_entry_rcu(ptype, head, list) {
2994 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2997 skb_set_network_header(skb, skb_gro_offset(skb));
2998 mac_len = skb->network_header - skb->mac_header;
2999 skb->mac_len = mac_len;
3000 NAPI_GRO_CB(skb)->same_flow = 0;
3001 NAPI_GRO_CB(skb)->flush = 0;
3002 NAPI_GRO_CB(skb)->free = 0;
3004 pp = ptype->gro_receive(&napi->gro_list, skb);
3009 if (&ptype->list == head)
3012 same_flow = NAPI_GRO_CB(skb)->same_flow;
3013 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3016 struct sk_buff *nskb = *pp;
3020 napi_gro_complete(nskb);
3027 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3031 NAPI_GRO_CB(skb)->count = 1;
3032 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3033 skb->next = napi->gro_list;
3034 napi->gro_list = skb;
3038 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3039 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3041 BUG_ON(skb->end - skb->tail < grow);
3043 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3046 skb->data_len -= grow;
3048 skb_shinfo(skb)->frags[0].page_offset += grow;
3049 skb_shinfo(skb)->frags[0].size -= grow;
3051 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3052 put_page(skb_shinfo(skb)->frags[0].page);
3053 memmove(skb_shinfo(skb)->frags,
3054 skb_shinfo(skb)->frags + 1,
3055 --skb_shinfo(skb)->nr_frags);
3066 EXPORT_SYMBOL(dev_gro_receive);
3069 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3073 if (netpoll_rx_on(skb))
3076 for (p = napi->gro_list; p; p = p->next) {
3077 NAPI_GRO_CB(p)->same_flow =
3078 (p->dev == skb->dev) &&
3079 !compare_ether_header(skb_mac_header(p),
3080 skb_gro_mac_header(skb));
3081 NAPI_GRO_CB(p)->flush = 0;
3084 return dev_gro_receive(napi, skb);
3087 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3091 if (netif_receive_skb(skb))
3096 case GRO_MERGED_FREE:
3107 EXPORT_SYMBOL(napi_skb_finish);
3109 void skb_gro_reset_offset(struct sk_buff *skb)
3111 NAPI_GRO_CB(skb)->data_offset = 0;
3112 NAPI_GRO_CB(skb)->frag0 = NULL;
3113 NAPI_GRO_CB(skb)->frag0_len = 0;
3115 if (skb->mac_header == skb->tail &&
3116 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3117 NAPI_GRO_CB(skb)->frag0 =
3118 page_address(skb_shinfo(skb)->frags[0].page) +
3119 skb_shinfo(skb)->frags[0].page_offset;
3120 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3123 EXPORT_SYMBOL(skb_gro_reset_offset);
3125 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3127 skb_gro_reset_offset(skb);
3129 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3131 EXPORT_SYMBOL(napi_gro_receive);
3133 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3135 __skb_pull(skb, skb_headlen(skb));
3136 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3140 EXPORT_SYMBOL(napi_reuse_skb);
3142 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3144 struct sk_buff *skb = napi->skb;
3147 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3153 EXPORT_SYMBOL(napi_get_frags);
3155 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3161 skb->protocol = eth_type_trans(skb, skb->dev);
3163 if (ret == GRO_HELD)
3164 skb_gro_pull(skb, -ETH_HLEN);
3165 else if (netif_receive_skb(skb))
3170 case GRO_MERGED_FREE:
3171 napi_reuse_skb(napi, skb);
3180 EXPORT_SYMBOL(napi_frags_finish);
3182 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3184 struct sk_buff *skb = napi->skb;
3191 skb_reset_mac_header(skb);
3192 skb_gro_reset_offset(skb);
3194 off = skb_gro_offset(skb);
3195 hlen = off + sizeof(*eth);
3196 eth = skb_gro_header_fast(skb, off);
3197 if (skb_gro_header_hard(skb, hlen)) {
3198 eth = skb_gro_header_slow(skb, hlen, off);
3199 if (unlikely(!eth)) {
3200 napi_reuse_skb(napi, skb);
3206 skb_gro_pull(skb, sizeof(*eth));
3209 * This works because the only protocols we care about don't require
3210 * special handling. We'll fix it up properly at the end.
3212 skb->protocol = eth->h_proto;
3217 EXPORT_SYMBOL(napi_frags_skb);
3219 gro_result_t napi_gro_frags(struct napi_struct *napi)
3221 struct sk_buff *skb = napi_frags_skb(napi);
3226 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3228 EXPORT_SYMBOL(napi_gro_frags);
3230 static int process_backlog(struct napi_struct *napi, int quota)
3233 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3235 napi->weight = weight_p;
3237 struct sk_buff *skb;
3239 local_irq_disable();
3241 skb = __skb_dequeue(&sd->input_pkt_queue);
3243 __napi_complete(napi);
3248 input_queue_head_incr(sd);
3252 __netif_receive_skb(skb);
3253 } while (++work < quota);
3259 * __napi_schedule - schedule for receive
3260 * @n: entry to schedule
3262 * The entry's receive function will be scheduled to run
3264 void __napi_schedule(struct napi_struct *n)
3266 unsigned long flags;
3268 local_irq_save(flags);
3269 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
3270 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3271 local_irq_restore(flags);
3273 EXPORT_SYMBOL(__napi_schedule);
3275 void __napi_complete(struct napi_struct *n)
3277 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3278 BUG_ON(n->gro_list);
3280 list_del(&n->poll_list);
3281 smp_mb__before_clear_bit();
3282 clear_bit(NAPI_STATE_SCHED, &n->state);
3284 EXPORT_SYMBOL(__napi_complete);
3286 void napi_complete(struct napi_struct *n)
3288 unsigned long flags;
3291 * don't let napi dequeue from the cpu poll list
3292 * just in case its running on a different cpu
3294 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3298 local_irq_save(flags);
3300 local_irq_restore(flags);
3302 EXPORT_SYMBOL(napi_complete);
3304 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3305 int (*poll)(struct napi_struct *, int), int weight)
3307 INIT_LIST_HEAD(&napi->poll_list);
3308 napi->gro_count = 0;
3309 napi->gro_list = NULL;
3312 napi->weight = weight;
3313 list_add(&napi->dev_list, &dev->napi_list);
3315 #ifdef CONFIG_NETPOLL
3316 spin_lock_init(&napi->poll_lock);
3317 napi->poll_owner = -1;
3319 set_bit(NAPI_STATE_SCHED, &napi->state);
3321 EXPORT_SYMBOL(netif_napi_add);
3323 void netif_napi_del(struct napi_struct *napi)
3325 struct sk_buff *skb, *next;
3327 list_del_init(&napi->dev_list);
3328 napi_free_frags(napi);
3330 for (skb = napi->gro_list; skb; skb = next) {
3336 napi->gro_list = NULL;
3337 napi->gro_count = 0;
3339 EXPORT_SYMBOL(netif_napi_del);
3342 * net_rps_action sends any pending IPI's for rps.
3343 * Note: called with local irq disabled, but exits with local irq enabled.
3345 static void net_rps_action_and_irq_disable(void)
3348 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3349 struct softnet_data *remsd = sd->rps_ipi_list;
3352 sd->rps_ipi_list = NULL;
3356 /* Send pending IPI's to kick RPS processing on remote cpus. */
3358 struct softnet_data *next = remsd->rps_ipi_next;
3360 if (cpu_online(remsd->cpu))
3361 __smp_call_function_single(remsd->cpu,
3370 static void net_rx_action(struct softirq_action *h)
3372 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
3373 unsigned long time_limit = jiffies + 2;
3374 int budget = netdev_budget;
3377 local_irq_disable();
3379 while (!list_empty(list)) {
3380 struct napi_struct *n;
3383 /* If softirq window is exhuasted then punt.
3384 * Allow this to run for 2 jiffies since which will allow
3385 * an average latency of 1.5/HZ.
3387 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3392 /* Even though interrupts have been re-enabled, this
3393 * access is safe because interrupts can only add new
3394 * entries to the tail of this list, and only ->poll()
3395 * calls can remove this head entry from the list.
3397 n = list_first_entry(list, struct napi_struct, poll_list);
3399 have = netpoll_poll_lock(n);
3403 /* This NAPI_STATE_SCHED test is for avoiding a race
3404 * with netpoll's poll_napi(). Only the entity which
3405 * obtains the lock and sees NAPI_STATE_SCHED set will
3406 * actually make the ->poll() call. Therefore we avoid
3407 * accidently calling ->poll() when NAPI is not scheduled.
3410 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3411 work = n->poll(n, weight);
3415 WARN_ON_ONCE(work > weight);
3419 local_irq_disable();
3421 /* Drivers must not modify the NAPI state if they
3422 * consume the entire weight. In such cases this code
3423 * still "owns" the NAPI instance and therefore can
3424 * move the instance around on the list at-will.
3426 if (unlikely(work == weight)) {
3427 if (unlikely(napi_disable_pending(n))) {
3430 local_irq_disable();
3432 list_move_tail(&n->poll_list, list);
3435 netpoll_poll_unlock(have);
3438 net_rps_action_and_irq_disable();
3440 #ifdef CONFIG_NET_DMA
3442 * There may not be any more sk_buffs coming right now, so push
3443 * any pending DMA copies to hardware
3445 dma_issue_pending_all();
3451 __get_cpu_var(netdev_rx_stat).time_squeeze++;
3452 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3456 static gifconf_func_t *gifconf_list[NPROTO];
3459 * register_gifconf - register a SIOCGIF handler
3460 * @family: Address family
3461 * @gifconf: Function handler
3463 * Register protocol dependent address dumping routines. The handler
3464 * that is passed must not be freed or reused until it has been replaced
3465 * by another handler.
3467 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3469 if (family >= NPROTO)
3471 gifconf_list[family] = gifconf;
3474 EXPORT_SYMBOL(register_gifconf);
3478 * Map an interface index to its name (SIOCGIFNAME)
3482 * We need this ioctl for efficient implementation of the
3483 * if_indextoname() function required by the IPv6 API. Without
3484 * it, we would have to search all the interfaces to find a
3488 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3490 struct net_device *dev;
3494 * Fetch the caller's info block.
3497 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3501 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3507 strcpy(ifr.ifr_name, dev->name);
3510 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3516 * Perform a SIOCGIFCONF call. This structure will change
3517 * size eventually, and there is nothing I can do about it.
3518 * Thus we will need a 'compatibility mode'.
3521 static int dev_ifconf(struct net *net, char __user *arg)
3524 struct net_device *dev;
3531 * Fetch the caller's info block.
3534 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3541 * Loop over the interfaces, and write an info block for each.
3545 for_each_netdev(net, dev) {
3546 for (i = 0; i < NPROTO; i++) {
3547 if (gifconf_list[i]) {
3550 done = gifconf_list[i](dev, NULL, 0);
3552 done = gifconf_list[i](dev, pos + total,
3562 * All done. Write the updated control block back to the caller.
3564 ifc.ifc_len = total;
3567 * Both BSD and Solaris return 0 here, so we do too.
3569 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3572 #ifdef CONFIG_PROC_FS
3574 * This is invoked by the /proc filesystem handler to display a device
3577 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3580 struct net *net = seq_file_net(seq);
3582 struct net_device *dev;
3586 return SEQ_START_TOKEN;
3589 for_each_netdev_rcu(net, dev)
3596 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3598 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3599 first_net_device(seq_file_net(seq)) :
3600 next_net_device((struct net_device *)v);
3603 return rcu_dereference(dev);
3606 void dev_seq_stop(struct seq_file *seq, void *v)
3612 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3614 const struct net_device_stats *stats = dev_get_stats(dev);
3616 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3617 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3618 dev->name, stats->rx_bytes, stats->rx_packets,
3620 stats->rx_dropped + stats->rx_missed_errors,
3621 stats->rx_fifo_errors,
3622 stats->rx_length_errors + stats->rx_over_errors +
3623 stats->rx_crc_errors + stats->rx_frame_errors,
3624 stats->rx_compressed, stats->multicast,
3625 stats->tx_bytes, stats->tx_packets,
3626 stats->tx_errors, stats->tx_dropped,
3627 stats->tx_fifo_errors, stats->collisions,
3628 stats->tx_carrier_errors +
3629 stats->tx_aborted_errors +
3630 stats->tx_window_errors +
3631 stats->tx_heartbeat_errors,
3632 stats->tx_compressed);
3636 * Called from the PROCfs module. This now uses the new arbitrary sized
3637 * /proc/net interface to create /proc/net/dev
3639 static int dev_seq_show(struct seq_file *seq, void *v)
3641 if (v == SEQ_START_TOKEN)
3642 seq_puts(seq, "Inter-| Receive "
3644 " face |bytes packets errs drop fifo frame "
3645 "compressed multicast|bytes packets errs "
3646 "drop fifo colls carrier compressed\n");
3648 dev_seq_printf_stats(seq, v);
3652 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3654 struct netif_rx_stats *rc = NULL;
3656 while (*pos < nr_cpu_ids)
3657 if (cpu_online(*pos)) {
3658 rc = &per_cpu(netdev_rx_stat, *pos);
3665 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3667 return softnet_get_online(pos);
3670 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3673 return softnet_get_online(pos);
3676 static void softnet_seq_stop(struct seq_file *seq, void *v)
3680 static int softnet_seq_show(struct seq_file *seq, void *v)
3682 struct netif_rx_stats *s = v;
3684 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3685 s->total, s->dropped, s->time_squeeze, 0,
3686 0, 0, 0, 0, /* was fastroute */
3687 s->cpu_collision, s->received_rps);
3691 static const struct seq_operations dev_seq_ops = {
3692 .start = dev_seq_start,
3693 .next = dev_seq_next,
3694 .stop = dev_seq_stop,
3695 .show = dev_seq_show,
3698 static int dev_seq_open(struct inode *inode, struct file *file)
3700 return seq_open_net(inode, file, &dev_seq_ops,
3701 sizeof(struct seq_net_private));
3704 static const struct file_operations dev_seq_fops = {
3705 .owner = THIS_MODULE,
3706 .open = dev_seq_open,
3708 .llseek = seq_lseek,
3709 .release = seq_release_net,
3712 static const struct seq_operations softnet_seq_ops = {
3713 .start = softnet_seq_start,
3714 .next = softnet_seq_next,
3715 .stop = softnet_seq_stop,
3716 .show = softnet_seq_show,
3719 static int softnet_seq_open(struct inode *inode, struct file *file)
3721 return seq_open(file, &softnet_seq_ops);
3724 static const struct file_operations softnet_seq_fops = {
3725 .owner = THIS_MODULE,
3726 .open = softnet_seq_open,
3728 .llseek = seq_lseek,
3729 .release = seq_release,
3732 static void *ptype_get_idx(loff_t pos)
3734 struct packet_type *pt = NULL;
3738 list_for_each_entry_rcu(pt, &ptype_all, list) {
3744 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3745 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3754 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3758 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3761 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3763 struct packet_type *pt;
3764 struct list_head *nxt;
3768 if (v == SEQ_START_TOKEN)
3769 return ptype_get_idx(0);
3772 nxt = pt->list.next;
3773 if (pt->type == htons(ETH_P_ALL)) {
3774 if (nxt != &ptype_all)
3777 nxt = ptype_base[0].next;
3779 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3781 while (nxt == &ptype_base[hash]) {
3782 if (++hash >= PTYPE_HASH_SIZE)
3784 nxt = ptype_base[hash].next;
3787 return list_entry(nxt, struct packet_type, list);
3790 static void ptype_seq_stop(struct seq_file *seq, void *v)
3796 static int ptype_seq_show(struct seq_file *seq, void *v)
3798 struct packet_type *pt = v;
3800 if (v == SEQ_START_TOKEN)
3801 seq_puts(seq, "Type Device Function\n");
3802 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3803 if (pt->type == htons(ETH_P_ALL))
3804 seq_puts(seq, "ALL ");
3806 seq_printf(seq, "%04x", ntohs(pt->type));
3808 seq_printf(seq, " %-8s %pF\n",
3809 pt->dev ? pt->dev->name : "", pt->func);
3815 static const struct seq_operations ptype_seq_ops = {
3816 .start = ptype_seq_start,
3817 .next = ptype_seq_next,
3818 .stop = ptype_seq_stop,
3819 .show = ptype_seq_show,
3822 static int ptype_seq_open(struct inode *inode, struct file *file)
3824 return seq_open_net(inode, file, &ptype_seq_ops,
3825 sizeof(struct seq_net_private));
3828 static const struct file_operations ptype_seq_fops = {
3829 .owner = THIS_MODULE,
3830 .open = ptype_seq_open,
3832 .llseek = seq_lseek,
3833 .release = seq_release_net,
3837 static int __net_init dev_proc_net_init(struct net *net)
3841 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3843 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3845 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3848 if (wext_proc_init(net))
3854 proc_net_remove(net, "ptype");
3856 proc_net_remove(net, "softnet_stat");
3858 proc_net_remove(net, "dev");
3862 static void __net_exit dev_proc_net_exit(struct net *net)
3864 wext_proc_exit(net);
3866 proc_net_remove(net, "ptype");
3867 proc_net_remove(net, "softnet_stat");
3868 proc_net_remove(net, "dev");
3871 static struct pernet_operations __net_initdata dev_proc_ops = {
3872 .init = dev_proc_net_init,
3873 .exit = dev_proc_net_exit,
3876 static int __init dev_proc_init(void)
3878 return register_pernet_subsys(&dev_proc_ops);
3881 #define dev_proc_init() 0
3882 #endif /* CONFIG_PROC_FS */
3886 * netdev_set_master - set up master/slave pair
3887 * @slave: slave device
3888 * @master: new master device
3890 * Changes the master device of the slave. Pass %NULL to break the
3891 * bonding. The caller must hold the RTNL semaphore. On a failure
3892 * a negative errno code is returned. On success the reference counts
3893 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3894 * function returns zero.
3896 int netdev_set_master(struct net_device *slave, struct net_device *master)
3898 struct net_device *old = slave->master;
3908 slave->master = master;
3915 slave->flags |= IFF_SLAVE;
3917 slave->flags &= ~IFF_SLAVE;
3919 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3922 EXPORT_SYMBOL(netdev_set_master);
3924 static void dev_change_rx_flags(struct net_device *dev, int flags)
3926 const struct net_device_ops *ops = dev->netdev_ops;
3928 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3929 ops->ndo_change_rx_flags(dev, flags);
3932 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3934 unsigned short old_flags = dev->flags;
3940 dev->flags |= IFF_PROMISC;
3941 dev->promiscuity += inc;
3942 if (dev->promiscuity == 0) {
3945 * If inc causes overflow, untouch promisc and return error.
3948 dev->flags &= ~IFF_PROMISC;
3950 dev->promiscuity -= inc;
3951 printk(KERN_WARNING "%s: promiscuity touches roof, "
3952 "set promiscuity failed, promiscuity feature "
3953 "of device might be broken.\n", dev->name);
3957 if (dev->flags != old_flags) {
3958 printk(KERN_INFO "device %s %s promiscuous mode\n",
3959 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3961 if (audit_enabled) {
3962 current_uid_gid(&uid, &gid);
3963 audit_log(current->audit_context, GFP_ATOMIC,
3964 AUDIT_ANOM_PROMISCUOUS,
3965 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3966 dev->name, (dev->flags & IFF_PROMISC),
3967 (old_flags & IFF_PROMISC),
3968 audit_get_loginuid(current),
3970 audit_get_sessionid(current));
3973 dev_change_rx_flags(dev, IFF_PROMISC);
3979 * dev_set_promiscuity - update promiscuity count on a device
3983 * Add or remove promiscuity from a device. While the count in the device
3984 * remains above zero the interface remains promiscuous. Once it hits zero
3985 * the device reverts back to normal filtering operation. A negative inc
3986 * value is used to drop promiscuity on the device.
3987 * Return 0 if successful or a negative errno code on error.
3989 int dev_set_promiscuity(struct net_device *dev, int inc)
3991 unsigned short old_flags = dev->flags;
3994 err = __dev_set_promiscuity(dev, inc);
3997 if (dev->flags != old_flags)
3998 dev_set_rx_mode(dev);
4001 EXPORT_SYMBOL(dev_set_promiscuity);
4004 * dev_set_allmulti - update allmulti count on a device
4008 * Add or remove reception of all multicast frames to a device. While the
4009 * count in the device remains above zero the interface remains listening
4010 * to all interfaces. Once it hits zero the device reverts back to normal
4011 * filtering operation. A negative @inc value is used to drop the counter
4012 * when releasing a resource needing all multicasts.
4013 * Return 0 if successful or a negative errno code on error.
4016 int dev_set_allmulti(struct net_device *dev, int inc)
4018 unsigned short old_flags = dev->flags;
4022 dev->flags |= IFF_ALLMULTI;
4023 dev->allmulti += inc;
4024 if (dev->allmulti == 0) {
4027 * If inc causes overflow, untouch allmulti and return error.
4030 dev->flags &= ~IFF_ALLMULTI;
4032 dev->allmulti -= inc;
4033 printk(KERN_WARNING "%s: allmulti touches roof, "
4034 "set allmulti failed, allmulti feature of "
4035 "device might be broken.\n", dev->name);
4039 if (dev->flags ^ old_flags) {
4040 dev_change_rx_flags(dev, IFF_ALLMULTI);
4041 dev_set_rx_mode(dev);
4045 EXPORT_SYMBOL(dev_set_allmulti);
4048 * Upload unicast and multicast address lists to device and
4049 * configure RX filtering. When the device doesn't support unicast
4050 * filtering it is put in promiscuous mode while unicast addresses
4053 void __dev_set_rx_mode(struct net_device *dev)
4055 const struct net_device_ops *ops = dev->netdev_ops;
4057 /* dev_open will call this function so the list will stay sane. */
4058 if (!(dev->flags&IFF_UP))
4061 if (!netif_device_present(dev))
4064 if (ops->ndo_set_rx_mode)
4065 ops->ndo_set_rx_mode(dev);
4067 /* Unicast addresses changes may only happen under the rtnl,
4068 * therefore calling __dev_set_promiscuity here is safe.
4070 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4071 __dev_set_promiscuity(dev, 1);
4072 dev->uc_promisc = 1;
4073 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4074 __dev_set_promiscuity(dev, -1);
4075 dev->uc_promisc = 0;
4078 if (ops->ndo_set_multicast_list)
4079 ops->ndo_set_multicast_list(dev);
4083 void dev_set_rx_mode(struct net_device *dev)
4085 netif_addr_lock_bh(dev);
4086 __dev_set_rx_mode(dev);
4087 netif_addr_unlock_bh(dev);
4091 * dev_get_flags - get flags reported to userspace
4094 * Get the combination of flag bits exported through APIs to userspace.
4096 unsigned dev_get_flags(const struct net_device *dev)
4100 flags = (dev->flags & ~(IFF_PROMISC |
4105 (dev->gflags & (IFF_PROMISC |
4108 if (netif_running(dev)) {
4109 if (netif_oper_up(dev))
4110 flags |= IFF_RUNNING;
4111 if (netif_carrier_ok(dev))
4112 flags |= IFF_LOWER_UP;
4113 if (netif_dormant(dev))
4114 flags |= IFF_DORMANT;
4119 EXPORT_SYMBOL(dev_get_flags);
4121 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4123 int old_flags = dev->flags;
4129 * Set the flags on our device.
4132 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4133 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4135 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4139 * Load in the correct multicast list now the flags have changed.
4142 if ((old_flags ^ flags) & IFF_MULTICAST)
4143 dev_change_rx_flags(dev, IFF_MULTICAST);
4145 dev_set_rx_mode(dev);
4148 * Have we downed the interface. We handle IFF_UP ourselves
4149 * according to user attempts to set it, rather than blindly
4154 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4155 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4158 dev_set_rx_mode(dev);
4161 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4162 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4164 dev->gflags ^= IFF_PROMISC;
4165 dev_set_promiscuity(dev, inc);
4168 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4169 is important. Some (broken) drivers set IFF_PROMISC, when
4170 IFF_ALLMULTI is requested not asking us and not reporting.
4172 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4173 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4175 dev->gflags ^= IFF_ALLMULTI;
4176 dev_set_allmulti(dev, inc);
4182 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4184 unsigned int changes = dev->flags ^ old_flags;
4186 if (changes & IFF_UP) {
4187 if (dev->flags & IFF_UP)
4188 call_netdevice_notifiers(NETDEV_UP, dev);
4190 call_netdevice_notifiers(NETDEV_DOWN, dev);
4193 if (dev->flags & IFF_UP &&
4194 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4195 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4199 * dev_change_flags - change device settings
4201 * @flags: device state flags
4203 * Change settings on device based state flags. The flags are
4204 * in the userspace exported format.
4206 int dev_change_flags(struct net_device *dev, unsigned flags)
4209 int old_flags = dev->flags;
4211 ret = __dev_change_flags(dev, flags);
4215 changes = old_flags ^ dev->flags;
4217 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4219 __dev_notify_flags(dev, old_flags);
4222 EXPORT_SYMBOL(dev_change_flags);
4225 * dev_set_mtu - Change maximum transfer unit
4227 * @new_mtu: new transfer unit
4229 * Change the maximum transfer size of the network device.
4231 int dev_set_mtu(struct net_device *dev, int new_mtu)
4233 const struct net_device_ops *ops = dev->netdev_ops;
4236 if (new_mtu == dev->mtu)
4239 /* MTU must be positive. */
4243 if (!netif_device_present(dev))
4247 if (ops->ndo_change_mtu)
4248 err = ops->ndo_change_mtu(dev, new_mtu);
4252 if (!err && dev->flags & IFF_UP)
4253 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4256 EXPORT_SYMBOL(dev_set_mtu);
4259 * dev_set_mac_address - Change Media Access Control Address
4263 * Change the hardware (MAC) address of the device
4265 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4267 const struct net_device_ops *ops = dev->netdev_ops;
4270 if (!ops->ndo_set_mac_address)
4272 if (sa->sa_family != dev->type)
4274 if (!netif_device_present(dev))
4276 err = ops->ndo_set_mac_address(dev, sa);
4278 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4281 EXPORT_SYMBOL(dev_set_mac_address);
4284 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4286 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4289 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4295 case SIOCGIFFLAGS: /* Get interface flags */
4296 ifr->ifr_flags = (short) dev_get_flags(dev);
4299 case SIOCGIFMETRIC: /* Get the metric on the interface
4300 (currently unused) */
4301 ifr->ifr_metric = 0;
4304 case SIOCGIFMTU: /* Get the MTU of a device */
4305 ifr->ifr_mtu = dev->mtu;
4310 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4312 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4313 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4314 ifr->ifr_hwaddr.sa_family = dev->type;
4322 ifr->ifr_map.mem_start = dev->mem_start;
4323 ifr->ifr_map.mem_end = dev->mem_end;
4324 ifr->ifr_map.base_addr = dev->base_addr;
4325 ifr->ifr_map.irq = dev->irq;
4326 ifr->ifr_map.dma = dev->dma;
4327 ifr->ifr_map.port = dev->if_port;
4331 ifr->ifr_ifindex = dev->ifindex;
4335 ifr->ifr_qlen = dev->tx_queue_len;
4339 /* dev_ioctl() should ensure this case
4351 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4353 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4356 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4357 const struct net_device_ops *ops;
4362 ops = dev->netdev_ops;
4365 case SIOCSIFFLAGS: /* Set interface flags */
4366 return dev_change_flags(dev, ifr->ifr_flags);
4368 case SIOCSIFMETRIC: /* Set the metric on the interface
4369 (currently unused) */
4372 case SIOCSIFMTU: /* Set the MTU of a device */
4373 return dev_set_mtu(dev, ifr->ifr_mtu);
4376 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4378 case SIOCSIFHWBROADCAST:
4379 if (ifr->ifr_hwaddr.sa_family != dev->type)
4381 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4382 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4383 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4387 if (ops->ndo_set_config) {
4388 if (!netif_device_present(dev))
4390 return ops->ndo_set_config(dev, &ifr->ifr_map);
4395 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4396 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4398 if (!netif_device_present(dev))
4400 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4403 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4404 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4406 if (!netif_device_present(dev))
4408 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4411 if (ifr->ifr_qlen < 0)
4413 dev->tx_queue_len = ifr->ifr_qlen;
4417 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4418 return dev_change_name(dev, ifr->ifr_newname);
4421 * Unknown or private ioctl
4424 if ((cmd >= SIOCDEVPRIVATE &&
4425 cmd <= SIOCDEVPRIVATE + 15) ||
4426 cmd == SIOCBONDENSLAVE ||
4427 cmd == SIOCBONDRELEASE ||
4428 cmd == SIOCBONDSETHWADDR ||
4429 cmd == SIOCBONDSLAVEINFOQUERY ||
4430 cmd == SIOCBONDINFOQUERY ||
4431 cmd == SIOCBONDCHANGEACTIVE ||
4432 cmd == SIOCGMIIPHY ||
4433 cmd == SIOCGMIIREG ||
4434 cmd == SIOCSMIIREG ||
4435 cmd == SIOCBRADDIF ||
4436 cmd == SIOCBRDELIF ||
4437 cmd == SIOCSHWTSTAMP ||
4438 cmd == SIOCWANDEV) {
4440 if (ops->ndo_do_ioctl) {
4441 if (netif_device_present(dev))
4442 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4454 * This function handles all "interface"-type I/O control requests. The actual
4455 * 'doing' part of this is dev_ifsioc above.
4459 * dev_ioctl - network device ioctl
4460 * @net: the applicable net namespace
4461 * @cmd: command to issue
4462 * @arg: pointer to a struct ifreq in user space
4464 * Issue ioctl functions to devices. This is normally called by the
4465 * user space syscall interfaces but can sometimes be useful for
4466 * other purposes. The return value is the return from the syscall if
4467 * positive or a negative errno code on error.
4470 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4476 /* One special case: SIOCGIFCONF takes ifconf argument
4477 and requires shared lock, because it sleeps writing
4481 if (cmd == SIOCGIFCONF) {
4483 ret = dev_ifconf(net, (char __user *) arg);
4487 if (cmd == SIOCGIFNAME)
4488 return dev_ifname(net, (struct ifreq __user *)arg);
4490 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4493 ifr.ifr_name[IFNAMSIZ-1] = 0;
4495 colon = strchr(ifr.ifr_name, ':');
4500 * See which interface the caller is talking about.
4505 * These ioctl calls:
4506 * - can be done by all.
4507 * - atomic and do not require locking.
4518 dev_load(net, ifr.ifr_name);
4520 ret = dev_ifsioc_locked(net, &ifr, cmd);
4525 if (copy_to_user(arg, &ifr,
4526 sizeof(struct ifreq)))
4532 dev_load(net, ifr.ifr_name);
4534 ret = dev_ethtool(net, &ifr);
4539 if (copy_to_user(arg, &ifr,
4540 sizeof(struct ifreq)))
4546 * These ioctl calls:
4547 * - require superuser power.
4548 * - require strict serialization.
4554 if (!capable(CAP_NET_ADMIN))
4556 dev_load(net, ifr.ifr_name);
4558 ret = dev_ifsioc(net, &ifr, cmd);
4563 if (copy_to_user(arg, &ifr,
4564 sizeof(struct ifreq)))
4570 * These ioctl calls:
4571 * - require superuser power.
4572 * - require strict serialization.
4573 * - do not return a value
4583 case SIOCSIFHWBROADCAST:
4586 case SIOCBONDENSLAVE:
4587 case SIOCBONDRELEASE:
4588 case SIOCBONDSETHWADDR:
4589 case SIOCBONDCHANGEACTIVE:
4593 if (!capable(CAP_NET_ADMIN))
4596 case SIOCBONDSLAVEINFOQUERY:
4597 case SIOCBONDINFOQUERY:
4598 dev_load(net, ifr.ifr_name);
4600 ret = dev_ifsioc(net, &ifr, cmd);
4605 /* Get the per device memory space. We can add this but
4606 * currently do not support it */
4608 /* Set the per device memory buffer space.
4609 * Not applicable in our case */
4614 * Unknown or private ioctl.
4617 if (cmd == SIOCWANDEV ||
4618 (cmd >= SIOCDEVPRIVATE &&
4619 cmd <= SIOCDEVPRIVATE + 15)) {
4620 dev_load(net, ifr.ifr_name);
4622 ret = dev_ifsioc(net, &ifr, cmd);
4624 if (!ret && copy_to_user(arg, &ifr,
4625 sizeof(struct ifreq)))
4629 /* Take care of Wireless Extensions */
4630 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4631 return wext_handle_ioctl(net, &ifr, cmd, arg);
4638 * dev_new_index - allocate an ifindex
4639 * @net: the applicable net namespace
4641 * Returns a suitable unique value for a new device interface
4642 * number. The caller must hold the rtnl semaphore or the
4643 * dev_base_lock to be sure it remains unique.
4645 static int dev_new_index(struct net *net)
4651 if (!__dev_get_by_index(net, ifindex))
4656 /* Delayed registration/unregisteration */
4657 static LIST_HEAD(net_todo_list);
4659 static void net_set_todo(struct net_device *dev)
4661 list_add_tail(&dev->todo_list, &net_todo_list);
4664 static void rollback_registered_many(struct list_head *head)
4666 struct net_device *dev, *tmp;
4668 BUG_ON(dev_boot_phase);
4671 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4672 /* Some devices call without registering
4673 * for initialization unwind. Remove those
4674 * devices and proceed with the remaining.
4676 if (dev->reg_state == NETREG_UNINITIALIZED) {
4677 pr_debug("unregister_netdevice: device %s/%p never "
4678 "was registered\n", dev->name, dev);
4681 list_del(&dev->unreg_list);
4685 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4687 /* If device is running, close it first. */
4690 /* And unlink it from device chain. */
4691 unlist_netdevice(dev);
4693 dev->reg_state = NETREG_UNREGISTERING;
4698 list_for_each_entry(dev, head, unreg_list) {
4699 /* Shutdown queueing discipline. */
4703 /* Notify protocols, that we are about to destroy
4704 this device. They should clean all the things.
4706 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4708 if (!dev->rtnl_link_ops ||
4709 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4710 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4713 * Flush the unicast and multicast chains
4718 if (dev->netdev_ops->ndo_uninit)
4719 dev->netdev_ops->ndo_uninit(dev);
4721 /* Notifier chain MUST detach us from master device. */
4722 WARN_ON(dev->master);
4724 /* Remove entries from kobject tree */
4725 netdev_unregister_kobject(dev);
4728 /* Process any work delayed until the end of the batch */
4729 dev = list_first_entry(head, struct net_device, unreg_list);
4730 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4734 list_for_each_entry(dev, head, unreg_list)
4738 static void rollback_registered(struct net_device *dev)
4742 list_add(&dev->unreg_list, &single);
4743 rollback_registered_many(&single);
4746 static void __netdev_init_queue_locks_one(struct net_device *dev,
4747 struct netdev_queue *dev_queue,
4750 spin_lock_init(&dev_queue->_xmit_lock);
4751 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4752 dev_queue->xmit_lock_owner = -1;
4755 static void netdev_init_queue_locks(struct net_device *dev)
4757 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4758 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4761 unsigned long netdev_fix_features(unsigned long features, const char *name)
4763 /* Fix illegal SG+CSUM combinations. */
4764 if ((features & NETIF_F_SG) &&
4765 !(features & NETIF_F_ALL_CSUM)) {
4767 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4768 "checksum feature.\n", name);
4769 features &= ~NETIF_F_SG;
4772 /* TSO requires that SG is present as well. */
4773 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4775 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4776 "SG feature.\n", name);
4777 features &= ~NETIF_F_TSO;
4780 if (features & NETIF_F_UFO) {
4781 if (!(features & NETIF_F_GEN_CSUM)) {
4783 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4784 "since no NETIF_F_HW_CSUM feature.\n",
4786 features &= ~NETIF_F_UFO;
4789 if (!(features & NETIF_F_SG)) {
4791 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4792 "since no NETIF_F_SG feature.\n", name);
4793 features &= ~NETIF_F_UFO;
4799 EXPORT_SYMBOL(netdev_fix_features);
4802 * netif_stacked_transfer_operstate - transfer operstate
4803 * @rootdev: the root or lower level device to transfer state from
4804 * @dev: the device to transfer operstate to
4806 * Transfer operational state from root to device. This is normally
4807 * called when a stacking relationship exists between the root
4808 * device and the device(a leaf device).
4810 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4811 struct net_device *dev)
4813 if (rootdev->operstate == IF_OPER_DORMANT)
4814 netif_dormant_on(dev);
4816 netif_dormant_off(dev);
4818 if (netif_carrier_ok(rootdev)) {
4819 if (!netif_carrier_ok(dev))
4820 netif_carrier_on(dev);
4822 if (netif_carrier_ok(dev))
4823 netif_carrier_off(dev);
4826 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4829 * register_netdevice - register a network device
4830 * @dev: device to register
4832 * Take a completed network device structure and add it to the kernel
4833 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4834 * chain. 0 is returned on success. A negative errno code is returned
4835 * on a failure to set up the device, or if the name is a duplicate.
4837 * Callers must hold the rtnl semaphore. You may want
4838 * register_netdev() instead of this.
4841 * The locking appears insufficient to guarantee two parallel registers
4842 * will not get the same name.
4845 int register_netdevice(struct net_device *dev)
4848 struct net *net = dev_net(dev);
4850 BUG_ON(dev_boot_phase);
4855 /* When net_device's are persistent, this will be fatal. */
4856 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4859 spin_lock_init(&dev->addr_list_lock);
4860 netdev_set_addr_lockdep_class(dev);
4861 netdev_init_queue_locks(dev);
4866 if (!dev->num_rx_queues) {
4868 * Allocate a single RX queue if driver never called
4872 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
4878 dev->_rx->first = dev->_rx;
4879 atomic_set(&dev->_rx->count, 1);
4880 dev->num_rx_queues = 1;
4883 /* Init, if this function is available */
4884 if (dev->netdev_ops->ndo_init) {
4885 ret = dev->netdev_ops->ndo_init(dev);
4893 ret = dev_get_valid_name(net, dev->name, dev->name, 0);
4897 dev->ifindex = dev_new_index(net);
4898 if (dev->iflink == -1)
4899 dev->iflink = dev->ifindex;
4901 /* Fix illegal checksum combinations */
4902 if ((dev->features & NETIF_F_HW_CSUM) &&
4903 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4904 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4906 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4909 if ((dev->features & NETIF_F_NO_CSUM) &&
4910 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4911 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4913 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4916 dev->features = netdev_fix_features(dev->features, dev->name);
4918 /* Enable software GSO if SG is supported. */
4919 if (dev->features & NETIF_F_SG)
4920 dev->features |= NETIF_F_GSO;
4922 netdev_initialize_kobject(dev);
4924 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4925 ret = notifier_to_errno(ret);
4929 ret = netdev_register_kobject(dev);
4932 dev->reg_state = NETREG_REGISTERED;
4935 * Default initial state at registry is that the
4936 * device is present.
4939 set_bit(__LINK_STATE_PRESENT, &dev->state);
4941 dev_init_scheduler(dev);
4943 list_netdevice(dev);
4945 /* Notify protocols, that a new device appeared. */
4946 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4947 ret = notifier_to_errno(ret);
4949 rollback_registered(dev);
4950 dev->reg_state = NETREG_UNREGISTERED;
4953 * Prevent userspace races by waiting until the network
4954 * device is fully setup before sending notifications.
4956 if (!dev->rtnl_link_ops ||
4957 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4958 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
4964 if (dev->netdev_ops->ndo_uninit)
4965 dev->netdev_ops->ndo_uninit(dev);
4968 EXPORT_SYMBOL(register_netdevice);
4971 * init_dummy_netdev - init a dummy network device for NAPI
4972 * @dev: device to init
4974 * This takes a network device structure and initialize the minimum
4975 * amount of fields so it can be used to schedule NAPI polls without
4976 * registering a full blown interface. This is to be used by drivers
4977 * that need to tie several hardware interfaces to a single NAPI
4978 * poll scheduler due to HW limitations.
4980 int init_dummy_netdev(struct net_device *dev)
4982 /* Clear everything. Note we don't initialize spinlocks
4983 * are they aren't supposed to be taken by any of the
4984 * NAPI code and this dummy netdev is supposed to be
4985 * only ever used for NAPI polls
4987 memset(dev, 0, sizeof(struct net_device));
4989 /* make sure we BUG if trying to hit standard
4990 * register/unregister code path
4992 dev->reg_state = NETREG_DUMMY;
4994 /* initialize the ref count */
4995 atomic_set(&dev->refcnt, 1);
4997 /* NAPI wants this */
4998 INIT_LIST_HEAD(&dev->napi_list);
5000 /* a dummy interface is started by default */
5001 set_bit(__LINK_STATE_PRESENT, &dev->state);
5002 set_bit(__LINK_STATE_START, &dev->state);
5006 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5010 * register_netdev - register a network device
5011 * @dev: device to register
5013 * Take a completed network device structure and add it to the kernel
5014 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5015 * chain. 0 is returned on success. A negative errno code is returned
5016 * on a failure to set up the device, or if the name is a duplicate.
5018 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5019 * and expands the device name if you passed a format string to
5022 int register_netdev(struct net_device *dev)
5029 * If the name is a format string the caller wants us to do a
5032 if (strchr(dev->name, '%')) {
5033 err = dev_alloc_name(dev, dev->name);
5038 err = register_netdevice(dev);
5043 EXPORT_SYMBOL(register_netdev);
5046 * netdev_wait_allrefs - wait until all references are gone.
5048 * This is called when unregistering network devices.
5050 * Any protocol or device that holds a reference should register
5051 * for netdevice notification, and cleanup and put back the
5052 * reference if they receive an UNREGISTER event.
5053 * We can get stuck here if buggy protocols don't correctly
5056 static void netdev_wait_allrefs(struct net_device *dev)
5058 unsigned long rebroadcast_time, warning_time;
5060 linkwatch_forget_dev(dev);
5062 rebroadcast_time = warning_time = jiffies;
5063 while (atomic_read(&dev->refcnt) != 0) {
5064 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5067 /* Rebroadcast unregister notification */
5068 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5069 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5070 * should have already handle it the first time */
5072 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5074 /* We must not have linkwatch events
5075 * pending on unregister. If this
5076 * happens, we simply run the queue
5077 * unscheduled, resulting in a noop
5080 linkwatch_run_queue();
5085 rebroadcast_time = jiffies;
5090 if (time_after(jiffies, warning_time + 10 * HZ)) {
5091 printk(KERN_EMERG "unregister_netdevice: "
5092 "waiting for %s to become free. Usage "
5094 dev->name, atomic_read(&dev->refcnt));
5095 warning_time = jiffies;
5104 * register_netdevice(x1);
5105 * register_netdevice(x2);
5107 * unregister_netdevice(y1);
5108 * unregister_netdevice(y2);
5114 * We are invoked by rtnl_unlock().
5115 * This allows us to deal with problems:
5116 * 1) We can delete sysfs objects which invoke hotplug
5117 * without deadlocking with linkwatch via keventd.
5118 * 2) Since we run with the RTNL semaphore not held, we can sleep
5119 * safely in order to wait for the netdev refcnt to drop to zero.
5121 * We must not return until all unregister events added during
5122 * the interval the lock was held have been completed.
5124 void netdev_run_todo(void)
5126 struct list_head list;
5128 /* Snapshot list, allow later requests */
5129 list_replace_init(&net_todo_list, &list);
5133 while (!list_empty(&list)) {
5134 struct net_device *dev
5135 = list_first_entry(&list, struct net_device, todo_list);
5136 list_del(&dev->todo_list);
5138 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5139 printk(KERN_ERR "network todo '%s' but state %d\n",
5140 dev->name, dev->reg_state);
5145 dev->reg_state = NETREG_UNREGISTERED;
5147 on_each_cpu(flush_backlog, dev, 1);
5149 netdev_wait_allrefs(dev);
5152 BUG_ON(atomic_read(&dev->refcnt));
5153 WARN_ON(dev->ip_ptr);
5154 WARN_ON(dev->ip6_ptr);
5155 WARN_ON(dev->dn_ptr);
5157 if (dev->destructor)
5158 dev->destructor(dev);
5160 /* Free network device */
5161 kobject_put(&dev->dev.kobj);
5166 * dev_txq_stats_fold - fold tx_queues stats
5167 * @dev: device to get statistics from
5168 * @stats: struct net_device_stats to hold results
5170 void dev_txq_stats_fold(const struct net_device *dev,
5171 struct net_device_stats *stats)
5173 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5175 struct netdev_queue *txq;
5177 for (i = 0; i < dev->num_tx_queues; i++) {
5178 txq = netdev_get_tx_queue(dev, i);
5179 tx_bytes += txq->tx_bytes;
5180 tx_packets += txq->tx_packets;
5181 tx_dropped += txq->tx_dropped;
5183 if (tx_bytes || tx_packets || tx_dropped) {
5184 stats->tx_bytes = tx_bytes;
5185 stats->tx_packets = tx_packets;
5186 stats->tx_dropped = tx_dropped;
5189 EXPORT_SYMBOL(dev_txq_stats_fold);
5192 * dev_get_stats - get network device statistics
5193 * @dev: device to get statistics from
5195 * Get network statistics from device. The device driver may provide
5196 * its own method by setting dev->netdev_ops->get_stats; otherwise
5197 * the internal statistics structure is used.
5199 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5201 const struct net_device_ops *ops = dev->netdev_ops;
5203 if (ops->ndo_get_stats)
5204 return ops->ndo_get_stats(dev);
5206 dev_txq_stats_fold(dev, &dev->stats);
5209 EXPORT_SYMBOL(dev_get_stats);
5211 static void netdev_init_one_queue(struct net_device *dev,
5212 struct netdev_queue *queue,
5218 static void netdev_init_queues(struct net_device *dev)
5220 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5221 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5222 spin_lock_init(&dev->tx_global_lock);
5226 * alloc_netdev_mq - allocate network device
5227 * @sizeof_priv: size of private data to allocate space for
5228 * @name: device name format string
5229 * @setup: callback to initialize device
5230 * @queue_count: the number of subqueues to allocate
5232 * Allocates a struct net_device with private data area for driver use
5233 * and performs basic initialization. Also allocates subquue structs
5234 * for each queue on the device at the end of the netdevice.
5236 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5237 void (*setup)(struct net_device *), unsigned int queue_count)
5239 struct netdev_queue *tx;
5240 struct net_device *dev;
5242 struct net_device *p;
5244 struct netdev_rx_queue *rx;
5248 BUG_ON(strlen(name) >= sizeof(dev->name));
5250 alloc_size = sizeof(struct net_device);
5252 /* ensure 32-byte alignment of private area */
5253 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5254 alloc_size += sizeof_priv;
5256 /* ensure 32-byte alignment of whole construct */
5257 alloc_size += NETDEV_ALIGN - 1;
5259 p = kzalloc(alloc_size, GFP_KERNEL);
5261 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5265 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5267 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5273 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5275 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5280 atomic_set(&rx->count, queue_count);
5283 * Set a pointer to first element in the array which holds the
5286 for (i = 0; i < queue_count; i++)
5290 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5291 dev->padded = (char *)dev - (char *)p;
5293 if (dev_addr_init(dev))
5299 dev_net_set(dev, &init_net);
5302 dev->num_tx_queues = queue_count;
5303 dev->real_num_tx_queues = queue_count;
5307 dev->num_rx_queues = queue_count;
5310 dev->gso_max_size = GSO_MAX_SIZE;
5312 netdev_init_queues(dev);
5314 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5315 dev->ethtool_ntuple_list.count = 0;
5316 INIT_LIST_HEAD(&dev->napi_list);
5317 INIT_LIST_HEAD(&dev->unreg_list);
5318 INIT_LIST_HEAD(&dev->link_watch_list);
5319 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5321 strcpy(dev->name, name);
5334 EXPORT_SYMBOL(alloc_netdev_mq);
5337 * free_netdev - free network device
5340 * This function does the last stage of destroying an allocated device
5341 * interface. The reference to the device object is released.
5342 * If this is the last reference then it will be freed.
5344 void free_netdev(struct net_device *dev)
5346 struct napi_struct *p, *n;
5348 release_net(dev_net(dev));
5352 /* Flush device addresses */
5353 dev_addr_flush(dev);
5355 /* Clear ethtool n-tuple list */
5356 ethtool_ntuple_flush(dev);
5358 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5361 /* Compatibility with error handling in drivers */
5362 if (dev->reg_state == NETREG_UNINITIALIZED) {
5363 kfree((char *)dev - dev->padded);
5367 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5368 dev->reg_state = NETREG_RELEASED;
5370 /* will free via device release */
5371 put_device(&dev->dev);
5373 EXPORT_SYMBOL(free_netdev);
5376 * synchronize_net - Synchronize with packet receive processing
5378 * Wait for packets currently being received to be done.
5379 * Does not block later packets from starting.
5381 void synchronize_net(void)
5386 EXPORT_SYMBOL(synchronize_net);
5389 * unregister_netdevice_queue - remove device from the kernel
5393 * This function shuts down a device interface and removes it
5394 * from the kernel tables.
5395 * If head not NULL, device is queued to be unregistered later.
5397 * Callers must hold the rtnl semaphore. You may want
5398 * unregister_netdev() instead of this.
5401 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5406 list_move_tail(&dev->unreg_list, head);
5408 rollback_registered(dev);
5409 /* Finish processing unregister after unlock */
5413 EXPORT_SYMBOL(unregister_netdevice_queue);
5416 * unregister_netdevice_many - unregister many devices
5417 * @head: list of devices
5419 void unregister_netdevice_many(struct list_head *head)
5421 struct net_device *dev;
5423 if (!list_empty(head)) {
5424 rollback_registered_many(head);
5425 list_for_each_entry(dev, head, unreg_list)
5429 EXPORT_SYMBOL(unregister_netdevice_many);
5432 * unregister_netdev - remove device from the kernel
5435 * This function shuts down a device interface and removes it
5436 * from the kernel tables.
5438 * This is just a wrapper for unregister_netdevice that takes
5439 * the rtnl semaphore. In general you want to use this and not
5440 * unregister_netdevice.
5442 void unregister_netdev(struct net_device *dev)
5445 unregister_netdevice(dev);
5448 EXPORT_SYMBOL(unregister_netdev);
5451 * dev_change_net_namespace - move device to different nethost namespace
5453 * @net: network namespace
5454 * @pat: If not NULL name pattern to try if the current device name
5455 * is already taken in the destination network namespace.
5457 * This function shuts down a device interface and moves it
5458 * to a new network namespace. On success 0 is returned, on
5459 * a failure a netagive errno code is returned.
5461 * Callers must hold the rtnl semaphore.
5464 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5470 /* Don't allow namespace local devices to be moved. */
5472 if (dev->features & NETIF_F_NETNS_LOCAL)
5476 /* Don't allow real devices to be moved when sysfs
5480 if (dev->dev.parent)
5484 /* Ensure the device has been registrered */
5486 if (dev->reg_state != NETREG_REGISTERED)
5489 /* Get out if there is nothing todo */
5491 if (net_eq(dev_net(dev), net))
5494 /* Pick the destination device name, and ensure
5495 * we can use it in the destination network namespace.
5498 if (__dev_get_by_name(net, dev->name)) {
5499 /* We get here if we can't use the current device name */
5502 if (dev_get_valid_name(net, pat, dev->name, 1))
5507 * And now a mini version of register_netdevice unregister_netdevice.
5510 /* If device is running close it first. */
5513 /* And unlink it from device chain */
5515 unlist_netdevice(dev);
5519 /* Shutdown queueing discipline. */
5522 /* Notify protocols, that we are about to destroy
5523 this device. They should clean all the things.
5525 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5526 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5529 * Flush the unicast and multicast chains
5534 netdev_unregister_kobject(dev);
5536 /* Actually switch the network namespace */
5537 dev_net_set(dev, net);
5539 /* If there is an ifindex conflict assign a new one */
5540 if (__dev_get_by_index(net, dev->ifindex)) {
5541 int iflink = (dev->iflink == dev->ifindex);
5542 dev->ifindex = dev_new_index(net);
5544 dev->iflink = dev->ifindex;
5547 /* Fixup kobjects */
5548 err = netdev_register_kobject(dev);
5551 /* Add the device back in the hashes */
5552 list_netdevice(dev);
5554 /* Notify protocols, that a new device appeared. */
5555 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5558 * Prevent userspace races by waiting until the network
5559 * device is fully setup before sending notifications.
5561 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5568 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5570 static int dev_cpu_callback(struct notifier_block *nfb,
5571 unsigned long action,
5574 struct sk_buff **list_skb;
5575 struct Qdisc **list_net;
5576 struct sk_buff *skb;
5577 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5578 struct softnet_data *sd, *oldsd;
5580 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5583 local_irq_disable();
5584 cpu = smp_processor_id();
5585 sd = &per_cpu(softnet_data, cpu);
5586 oldsd = &per_cpu(softnet_data, oldcpu);
5588 /* Find end of our completion_queue. */
5589 list_skb = &sd->completion_queue;
5591 list_skb = &(*list_skb)->next;
5592 /* Append completion queue from offline CPU. */
5593 *list_skb = oldsd->completion_queue;
5594 oldsd->completion_queue = NULL;
5596 /* Find end of our output_queue. */
5597 list_net = &sd->output_queue;
5599 list_net = &(*list_net)->next_sched;
5600 /* Append output queue from offline CPU. */
5601 *list_net = oldsd->output_queue;
5602 oldsd->output_queue = NULL;
5604 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5607 /* Process offline CPU's input_pkt_queue */
5608 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5610 input_queue_head_incr(oldsd);
5618 * netdev_increment_features - increment feature set by one
5619 * @all: current feature set
5620 * @one: new feature set
5621 * @mask: mask feature set
5623 * Computes a new feature set after adding a device with feature set
5624 * @one to the master device with current feature set @all. Will not
5625 * enable anything that is off in @mask. Returns the new feature set.
5627 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5630 /* If device needs checksumming, downgrade to it. */
5631 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5632 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5633 else if (mask & NETIF_F_ALL_CSUM) {
5634 /* If one device supports v4/v6 checksumming, set for all. */
5635 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5636 !(all & NETIF_F_GEN_CSUM)) {
5637 all &= ~NETIF_F_ALL_CSUM;
5638 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5641 /* If one device supports hw checksumming, set for all. */
5642 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5643 all &= ~NETIF_F_ALL_CSUM;
5644 all |= NETIF_F_HW_CSUM;
5648 one |= NETIF_F_ALL_CSUM;
5650 one |= all & NETIF_F_ONE_FOR_ALL;
5651 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5652 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5656 EXPORT_SYMBOL(netdev_increment_features);
5658 static struct hlist_head *netdev_create_hash(void)
5661 struct hlist_head *hash;
5663 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5665 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5666 INIT_HLIST_HEAD(&hash[i]);
5671 /* Initialize per network namespace state */
5672 static int __net_init netdev_init(struct net *net)
5674 INIT_LIST_HEAD(&net->dev_base_head);
5676 net->dev_name_head = netdev_create_hash();
5677 if (net->dev_name_head == NULL)
5680 net->dev_index_head = netdev_create_hash();
5681 if (net->dev_index_head == NULL)
5687 kfree(net->dev_name_head);
5693 * netdev_drivername - network driver for the device
5694 * @dev: network device
5695 * @buffer: buffer for resulting name
5696 * @len: size of buffer
5698 * Determine network driver for device.
5700 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5702 const struct device_driver *driver;
5703 const struct device *parent;
5705 if (len <= 0 || !buffer)
5709 parent = dev->dev.parent;
5714 driver = parent->driver;
5715 if (driver && driver->name)
5716 strlcpy(buffer, driver->name, len);
5720 static void __net_exit netdev_exit(struct net *net)
5722 kfree(net->dev_name_head);
5723 kfree(net->dev_index_head);
5726 static struct pernet_operations __net_initdata netdev_net_ops = {
5727 .init = netdev_init,
5728 .exit = netdev_exit,
5731 static void __net_exit default_device_exit(struct net *net)
5733 struct net_device *dev, *aux;
5735 * Push all migratable network devices back to the
5736 * initial network namespace
5739 for_each_netdev_safe(net, dev, aux) {
5741 char fb_name[IFNAMSIZ];
5743 /* Ignore unmoveable devices (i.e. loopback) */
5744 if (dev->features & NETIF_F_NETNS_LOCAL)
5747 /* Leave virtual devices for the generic cleanup */
5748 if (dev->rtnl_link_ops)
5751 /* Push remaing network devices to init_net */
5752 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5753 err = dev_change_net_namespace(dev, &init_net, fb_name);
5755 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5756 __func__, dev->name, err);
5763 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5765 /* At exit all network devices most be removed from a network
5766 * namespace. Do this in the reverse order of registeration.
5767 * Do this across as many network namespaces as possible to
5768 * improve batching efficiency.
5770 struct net_device *dev;
5772 LIST_HEAD(dev_kill_list);
5775 list_for_each_entry(net, net_list, exit_list) {
5776 for_each_netdev_reverse(net, dev) {
5777 if (dev->rtnl_link_ops)
5778 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5780 unregister_netdevice_queue(dev, &dev_kill_list);
5783 unregister_netdevice_many(&dev_kill_list);
5787 static struct pernet_operations __net_initdata default_device_ops = {
5788 .exit = default_device_exit,
5789 .exit_batch = default_device_exit_batch,
5793 * Initialize the DEV module. At boot time this walks the device list and
5794 * unhooks any devices that fail to initialise (normally hardware not
5795 * present) and leaves us with a valid list of present and active devices.
5800 * This is called single threaded during boot, so no need
5801 * to take the rtnl semaphore.
5803 static int __init net_dev_init(void)
5805 int i, rc = -ENOMEM;
5807 BUG_ON(!dev_boot_phase);
5809 if (dev_proc_init())
5812 if (netdev_kobject_init())
5815 INIT_LIST_HEAD(&ptype_all);
5816 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5817 INIT_LIST_HEAD(&ptype_base[i]);
5819 if (register_pernet_subsys(&netdev_net_ops))
5823 * Initialise the packet receive queues.
5826 for_each_possible_cpu(i) {
5827 struct softnet_data *sd = &per_cpu(softnet_data, i);
5829 skb_queue_head_init(&sd->input_pkt_queue);
5830 sd->completion_queue = NULL;
5831 INIT_LIST_HEAD(&sd->poll_list);
5834 sd->csd.func = rps_trigger_softirq;
5840 sd->backlog.poll = process_backlog;
5841 sd->backlog.weight = weight_p;
5842 sd->backlog.gro_list = NULL;
5843 sd->backlog.gro_count = 0;
5848 /* The loopback device is special if any other network devices
5849 * is present in a network namespace the loopback device must
5850 * be present. Since we now dynamically allocate and free the
5851 * loopback device ensure this invariant is maintained by
5852 * keeping the loopback device as the first device on the
5853 * list of network devices. Ensuring the loopback devices
5854 * is the first device that appears and the last network device
5857 if (register_pernet_device(&loopback_net_ops))
5860 if (register_pernet_device(&default_device_ops))
5863 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5864 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5866 hotcpu_notifier(dev_cpu_callback, 0);
5874 subsys_initcall(net_dev_init);
5876 static int __init initialize_hashrnd(void)
5878 get_random_bytes(&hashrnd, sizeof(hashrnd));
5882 late_initcall_sync(initialize_hashrnd);