2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
104 #include <linux/if_bridge.h>
105 #include <linux/if_macvlan.h>
107 #include <net/pkt_sched.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/kmod.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/wext.h>
118 #include <net/iw_handler.h>
119 #include <asm/current.h>
120 #include <linux/audit.h>
121 #include <linux/dmaengine.h>
122 #include <linux/err.h>
123 #include <linux/ctype.h>
124 #include <linux/if_arp.h>
125 #include <linux/if_vlan.h>
126 #include <linux/ip.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
133 #include <linux/pci.h>
135 #include "net-sysfs.h"
137 /* Instead of increasing this, you should create a hash table. */
138 #define MAX_GRO_SKBS 8
140 /* This should be increased if a protocol with a bigger head is added. */
141 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 * The list of packet types we will receive (as opposed to discard)
145 * and the routines to invoke.
147 * Why 16. Because with 16 the only overlap we get on a hash of the
148 * low nibble of the protocol value is RARP/SNAP/X.25.
150 * NOTE: That is no longer true with the addition of VLAN tags. Not
151 * sure which should go first, but I bet it won't make much
152 * difference if we are running VLANs. The good news is that
153 * this protocol won't be in the list unless compiled in, so
154 * the average user (w/out VLANs) will not be adversely affected.
171 #define PTYPE_HASH_SIZE (16)
172 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
174 static DEFINE_SPINLOCK(ptype_lock);
175 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
176 static struct list_head ptype_all __read_mostly; /* Taps */
179 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
182 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
184 * Writers must hold the rtnl semaphore while they loop through the
185 * dev_base_head list, and hold dev_base_lock for writing when they do the
186 * actual updates. This allows pure readers to access the list even
187 * while a writer is preparing to update it.
189 * To put it another way, dev_base_lock is held for writing only to
190 * protect against pure readers; the rtnl semaphore provides the
191 * protection against other writers.
193 * See, for example usages, register_netdevice() and
194 * unregister_netdevice(), which must be called with the rtnl
197 DEFINE_RWLOCK(dev_base_lock);
198 EXPORT_SYMBOL(dev_base_lock);
200 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
202 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
203 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
206 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
208 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
211 static inline void rps_lock(struct softnet_data *sd)
214 spin_lock(&sd->input_pkt_queue.lock);
218 static inline void rps_unlock(struct softnet_data *sd)
221 spin_unlock(&sd->input_pkt_queue.lock);
225 /* Device list insertion */
226 static int list_netdevice(struct net_device *dev)
228 struct net *net = dev_net(dev);
232 write_lock_bh(&dev_base_lock);
233 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
234 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
235 hlist_add_head_rcu(&dev->index_hlist,
236 dev_index_hash(net, dev->ifindex));
237 write_unlock_bh(&dev_base_lock);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device *dev)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock);
250 list_del_rcu(&dev->dev_list);
251 hlist_del_rcu(&dev->name_hlist);
252 hlist_del_rcu(&dev->index_hlist);
253 write_unlock_bh(&dev_base_lock);
260 static RAW_NOTIFIER_HEAD(netdev_chain);
263 * Device drivers call our routines to queue packets here. We empty the
264 * queue in the local softnet handler.
267 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
268 EXPORT_PER_CPU_SYMBOL(softnet_data);
270 #ifdef CONFIG_LOCKDEP
272 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
273 * according to dev->type
275 static const unsigned short netdev_lock_type[] =
276 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
277 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
278 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
279 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
280 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
281 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
282 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
283 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
284 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
285 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
286 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
287 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
288 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
289 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
290 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
291 ARPHRD_VOID, ARPHRD_NONE};
293 static const char *const netdev_lock_name[] =
294 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
295 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
296 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
297 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
298 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
299 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
300 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
301 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
302 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
303 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
304 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
305 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
306 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
307 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
308 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
309 "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
318 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
319 if (netdev_lock_type[i] == dev_type)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
326 unsigned short dev_type)
330 i = netdev_lock_pos(dev_type);
331 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
332 netdev_lock_name[i]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
339 i = netdev_lock_pos(dev->type);
340 lockdep_set_class_and_name(&dev->addr_list_lock,
341 &netdev_addr_lock_key[i],
342 netdev_lock_name[i]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
346 unsigned short dev_type)
349 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
377 * dev_add_pack - add packet handler
378 * @pt: packet type declaration
380 * Add a protocol handler to the networking stack. The passed &packet_type
381 * is linked into kernel lists and may not be freed until it has been
382 * removed from the kernel lists.
384 * This call does not sleep therefore it can not
385 * guarantee all CPU's that are in middle of receiving packets
386 * will see the new packet type (until the next received packet).
389 void dev_add_pack(struct packet_type *pt)
393 spin_lock_bh(&ptype_lock);
394 if (pt->type == htons(ETH_P_ALL))
395 list_add_rcu(&pt->list, &ptype_all);
397 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
398 list_add_rcu(&pt->list, &ptype_base[hash]);
400 spin_unlock_bh(&ptype_lock);
402 EXPORT_SYMBOL(dev_add_pack);
405 * __dev_remove_pack - remove packet handler
406 * @pt: packet type declaration
408 * Remove a protocol handler that was previously added to the kernel
409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
410 * from the kernel lists and can be freed or reused once this function
413 * The packet type might still be in use by receivers
414 * and must not be freed until after all the CPU's have gone
415 * through a quiescent state.
417 void __dev_remove_pack(struct packet_type *pt)
419 struct list_head *head;
420 struct packet_type *pt1;
422 spin_lock_bh(&ptype_lock);
424 if (pt->type == htons(ETH_P_ALL))
427 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
429 list_for_each_entry(pt1, head, list) {
431 list_del_rcu(&pt->list);
436 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
438 spin_unlock_bh(&ptype_lock);
440 EXPORT_SYMBOL(__dev_remove_pack);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type *pt)
456 __dev_remove_pack(pt);
460 EXPORT_SYMBOL(dev_remove_pack);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name, struct ifmap *map)
482 struct netdev_boot_setup *s;
486 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
487 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
488 memset(s[i].name, 0, sizeof(s[i].name));
489 strlcpy(s[i].name, name, IFNAMSIZ);
490 memcpy(&s[i].map, map, sizeof(s[i].map));
495 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device *dev)
509 struct netdev_boot_setup *s = dev_boot_setup;
512 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
513 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
514 !strcmp(dev->name, s[i].name)) {
515 dev->irq = s[i].map.irq;
516 dev->base_addr = s[i].map.base_addr;
517 dev->mem_start = s[i].map.mem_start;
518 dev->mem_end = s[i].map.mem_end;
524 EXPORT_SYMBOL(netdev_boot_setup_check);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix, int unit)
539 const struct netdev_boot_setup *s = dev_boot_setup;
543 sprintf(name, "%s%d", prefix, unit);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net, name))
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
553 if (!strcmp(name, s[i].name))
554 return s[i].map.base_addr;
559 * Saves at boot time configured settings for any netdevice.
561 int __init netdev_boot_setup(char *str)
566 str = get_options(str, ARRAY_SIZE(ints), ints);
571 memset(&map, 0, sizeof(map));
575 map.base_addr = ints[2];
577 map.mem_start = ints[3];
579 map.mem_end = ints[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str, &map);
585 __setup("netdev=", netdev_boot_setup);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device *__dev_get_by_name(struct net *net, const char *name)
607 struct hlist_node *p;
608 struct net_device *dev;
609 struct hlist_head *head = dev_name_hash(net, name);
611 hlist_for_each_entry(dev, p, head, name_hlist)
612 if (!strncmp(dev->name, name, IFNAMSIZ))
617 EXPORT_SYMBOL(__dev_get_by_name);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
633 struct hlist_node *p;
634 struct net_device *dev;
635 struct hlist_head *head = dev_name_hash(net, name);
637 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
638 if (!strncmp(dev->name, name, IFNAMSIZ))
643 EXPORT_SYMBOL(dev_get_by_name_rcu);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device *dev_get_by_name(struct net *net, const char *name)
659 struct net_device *dev;
662 dev = dev_get_by_name_rcu(net, name);
668 EXPORT_SYMBOL(dev_get_by_name);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
684 struct hlist_node *p;
685 struct net_device *dev;
686 struct hlist_head *head = dev_index_hash(net, ifindex);
688 hlist_for_each_entry(dev, p, head, index_hlist)
689 if (dev->ifindex == ifindex)
694 EXPORT_SYMBOL(__dev_get_by_index);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
709 struct hlist_node *p;
710 struct net_device *dev;
711 struct hlist_head *head = dev_index_hash(net, ifindex);
713 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
714 if (dev->ifindex == ifindex)
719 EXPORT_SYMBOL(dev_get_by_index_rcu);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device *dev_get_by_index(struct net *net, int ifindex)
735 struct net_device *dev;
738 dev = dev_get_by_index_rcu(net, ifindex);
744 EXPORT_SYMBOL(dev_get_by_index);
747 * dev_getbyhwaddr - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device. The caller must hold the
754 * rtnl semaphore. The returned device has not had its ref count increased
755 * and the caller must therefore be careful about locking
758 * If the API was consistent this would be __dev_get_by_hwaddr
761 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
763 struct net_device *dev;
767 for_each_netdev(net, dev)
768 if (dev->type == type &&
769 !memcmp(dev->dev_addr, ha, dev->addr_len))
774 EXPORT_SYMBOL(dev_getbyhwaddr);
776 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
778 struct net_device *dev;
781 for_each_netdev(net, dev)
782 if (dev->type == type)
787 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
789 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
791 struct net_device *dev, *ret = NULL;
794 for_each_netdev_rcu(net, dev)
795 if (dev->type == type) {
803 EXPORT_SYMBOL(dev_getfirstbyhwtype);
806 * dev_get_by_flags - find any device with given flags
807 * @net: the applicable net namespace
808 * @if_flags: IFF_* values
809 * @mask: bitmask of bits in if_flags to check
811 * Search for any interface with the given flags. Returns NULL if a device
812 * is not found or a pointer to the device. The device returned has
813 * had a reference added and the pointer is safe until the user calls
814 * dev_put to indicate they have finished with it.
817 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
820 struct net_device *dev, *ret;
824 for_each_netdev_rcu(net, dev) {
825 if (((dev->flags ^ if_flags) & mask) == 0) {
834 EXPORT_SYMBOL(dev_get_by_flags);
837 * dev_valid_name - check if name is okay for network device
840 * Network device names need to be valid file names to
841 * to allow sysfs to work. We also disallow any kind of
844 int dev_valid_name(const char *name)
848 if (strlen(name) >= IFNAMSIZ)
850 if (!strcmp(name, ".") || !strcmp(name, ".."))
854 if (*name == '/' || isspace(*name))
860 EXPORT_SYMBOL(dev_valid_name);
863 * __dev_alloc_name - allocate a name for a device
864 * @net: network namespace to allocate the device name in
865 * @name: name format string
866 * @buf: scratch buffer and result name string
868 * Passed a format string - eg "lt%d" it will try and find a suitable
869 * id. It scans list of devices to build up a free map, then chooses
870 * the first empty slot. The caller must hold the dev_base or rtnl lock
871 * while allocating the name and adding the device in order to avoid
873 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
874 * Returns the number of the unit assigned or a negative errno code.
877 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
881 const int max_netdevices = 8*PAGE_SIZE;
882 unsigned long *inuse;
883 struct net_device *d;
885 p = strnchr(name, IFNAMSIZ-1, '%');
888 * Verify the string as this thing may have come from
889 * the user. There must be either one "%d" and no other "%"
892 if (p[1] != 'd' || strchr(p + 2, '%'))
895 /* Use one page as a bit array of possible slots */
896 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
900 for_each_netdev(net, d) {
901 if (!sscanf(d->name, name, &i))
903 if (i < 0 || i >= max_netdevices)
906 /* avoid cases where sscanf is not exact inverse of printf */
907 snprintf(buf, IFNAMSIZ, name, i);
908 if (!strncmp(buf, d->name, IFNAMSIZ))
912 i = find_first_zero_bit(inuse, max_netdevices);
913 free_page((unsigned long) inuse);
917 snprintf(buf, IFNAMSIZ, name, i);
918 if (!__dev_get_by_name(net, buf))
921 /* It is possible to run out of possible slots
922 * when the name is long and there isn't enough space left
923 * for the digits, or if all bits are used.
929 * dev_alloc_name - allocate a name for a device
931 * @name: name format string
933 * Passed a format string - eg "lt%d" it will try and find a suitable
934 * id. It scans list of devices to build up a free map, then chooses
935 * the first empty slot. The caller must hold the dev_base or rtnl lock
936 * while allocating the name and adding the device in order to avoid
938 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
939 * Returns the number of the unit assigned or a negative errno code.
942 int dev_alloc_name(struct net_device *dev, const char *name)
948 BUG_ON(!dev_net(dev));
950 ret = __dev_alloc_name(net, name, buf);
952 strlcpy(dev->name, buf, IFNAMSIZ);
955 EXPORT_SYMBOL(dev_alloc_name);
957 static int dev_get_valid_name(struct net *net, const char *name, char *buf,
960 if (!dev_valid_name(name))
963 if (fmt && strchr(name, '%'))
964 return __dev_alloc_name(net, name, buf);
965 else if (__dev_get_by_name(net, name))
967 else if (buf != name)
968 strlcpy(buf, name, IFNAMSIZ);
974 * dev_change_name - change name of a device
976 * @newname: name (or format string) must be at least IFNAMSIZ
978 * Change name of a device, can pass format strings "eth%d".
981 int dev_change_name(struct net_device *dev, const char *newname)
983 char oldname[IFNAMSIZ];
989 BUG_ON(!dev_net(dev));
992 if (dev->flags & IFF_UP)
995 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
998 memcpy(oldname, dev->name, IFNAMSIZ);
1000 err = dev_get_valid_name(net, newname, dev->name, 1);
1005 /* For now only devices in the initial network namespace
1008 if (net_eq(net, &init_net)) {
1009 ret = device_rename(&dev->dev, dev->name);
1011 memcpy(dev->name, oldname, IFNAMSIZ);
1016 write_lock_bh(&dev_base_lock);
1017 hlist_del(&dev->name_hlist);
1018 write_unlock_bh(&dev_base_lock);
1022 write_lock_bh(&dev_base_lock);
1023 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1024 write_unlock_bh(&dev_base_lock);
1026 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1027 ret = notifier_to_errno(ret);
1030 /* err >= 0 after dev_alloc_name() or stores the first errno */
1033 memcpy(dev->name, oldname, IFNAMSIZ);
1037 "%s: name change rollback failed: %d.\n",
1046 * dev_set_alias - change ifalias of a device
1048 * @alias: name up to IFALIASZ
1049 * @len: limit of bytes to copy from info
1051 * Set ifalias for a device,
1053 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1057 if (len >= IFALIASZ)
1062 kfree(dev->ifalias);
1063 dev->ifalias = NULL;
1068 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1072 strlcpy(dev->ifalias, alias, len+1);
1078 * netdev_features_change - device changes features
1079 * @dev: device to cause notification
1081 * Called to indicate a device has changed features.
1083 void netdev_features_change(struct net_device *dev)
1085 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1087 EXPORT_SYMBOL(netdev_features_change);
1090 * netdev_state_change - device changes state
1091 * @dev: device to cause notification
1093 * Called to indicate a device has changed state. This function calls
1094 * the notifier chains for netdev_chain and sends a NEWLINK message
1095 * to the routing socket.
1097 void netdev_state_change(struct net_device *dev)
1099 if (dev->flags & IFF_UP) {
1100 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1101 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1104 EXPORT_SYMBOL(netdev_state_change);
1106 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1108 return call_netdevice_notifiers(event, dev);
1110 EXPORT_SYMBOL(netdev_bonding_change);
1113 * dev_load - load a network module
1114 * @net: the applicable net namespace
1115 * @name: name of interface
1117 * If a network interface is not present and the process has suitable
1118 * privileges this function loads the module. If module loading is not
1119 * available in this kernel then it becomes a nop.
1122 void dev_load(struct net *net, const char *name)
1124 struct net_device *dev;
1127 dev = dev_get_by_name_rcu(net, name);
1130 if (!dev && capable(CAP_NET_ADMIN))
1131 request_module("%s", name);
1133 EXPORT_SYMBOL(dev_load);
1135 static int __dev_open(struct net_device *dev)
1137 const struct net_device_ops *ops = dev->netdev_ops;
1143 * Is it even present?
1145 if (!netif_device_present(dev))
1148 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1149 ret = notifier_to_errno(ret);
1154 * Call device private open method
1156 set_bit(__LINK_STATE_START, &dev->state);
1158 if (ops->ndo_validate_addr)
1159 ret = ops->ndo_validate_addr(dev);
1161 if (!ret && ops->ndo_open)
1162 ret = ops->ndo_open(dev);
1165 * If it went open OK then:
1169 clear_bit(__LINK_STATE_START, &dev->state);
1174 dev->flags |= IFF_UP;
1179 net_dmaengine_get();
1182 * Initialize multicasting status
1184 dev_set_rx_mode(dev);
1187 * Wakeup transmit queue engine
1196 * dev_open - prepare an interface for use.
1197 * @dev: device to open
1199 * Takes a device from down to up state. The device's private open
1200 * function is invoked and then the multicast lists are loaded. Finally
1201 * the device is moved into the up state and a %NETDEV_UP message is
1202 * sent to the netdev notifier chain.
1204 * Calling this function on an active interface is a nop. On a failure
1205 * a negative errno code is returned.
1207 int dev_open(struct net_device *dev)
1214 if (dev->flags & IFF_UP)
1220 ret = __dev_open(dev);
1225 * ... and announce new interface.
1227 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1228 call_netdevice_notifiers(NETDEV_UP, dev);
1232 EXPORT_SYMBOL(dev_open);
1234 static int __dev_close(struct net_device *dev)
1236 const struct net_device_ops *ops = dev->netdev_ops;
1242 * Tell people we are going down, so that they can
1243 * prepare to death, when device is still operating.
1245 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1247 clear_bit(__LINK_STATE_START, &dev->state);
1249 /* Synchronize to scheduled poll. We cannot touch poll list,
1250 * it can be even on different cpu. So just clear netif_running().
1252 * dev->stop() will invoke napi_disable() on all of it's
1253 * napi_struct instances on this device.
1255 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1257 dev_deactivate(dev);
1260 * Call the device specific close. This cannot fail.
1261 * Only if device is UP
1263 * We allow it to be called even after a DETACH hot-plug
1270 * Device is now down.
1273 dev->flags &= ~IFF_UP;
1278 net_dmaengine_put();
1284 * dev_close - shutdown an interface.
1285 * @dev: device to shutdown
1287 * This function moves an active device into down state. A
1288 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1289 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1292 int dev_close(struct net_device *dev)
1294 if (!(dev->flags & IFF_UP))
1300 * Tell people we are down
1302 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1303 call_netdevice_notifiers(NETDEV_DOWN, dev);
1307 EXPORT_SYMBOL(dev_close);
1311 * dev_disable_lro - disable Large Receive Offload on a device
1314 * Disable Large Receive Offload (LRO) on a net device. Must be
1315 * called under RTNL. This is needed if received packets may be
1316 * forwarded to another interface.
1318 void dev_disable_lro(struct net_device *dev)
1320 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1321 dev->ethtool_ops->set_flags) {
1322 u32 flags = dev->ethtool_ops->get_flags(dev);
1323 if (flags & ETH_FLAG_LRO) {
1324 flags &= ~ETH_FLAG_LRO;
1325 dev->ethtool_ops->set_flags(dev, flags);
1328 WARN_ON(dev->features & NETIF_F_LRO);
1330 EXPORT_SYMBOL(dev_disable_lro);
1333 static int dev_boot_phase = 1;
1336 * Device change register/unregister. These are not inline or static
1337 * as we export them to the world.
1341 * register_netdevice_notifier - register a network notifier block
1344 * Register a notifier to be called when network device events occur.
1345 * The notifier passed is linked into the kernel structures and must
1346 * not be reused until it has been unregistered. A negative errno code
1347 * is returned on a failure.
1349 * When registered all registration and up events are replayed
1350 * to the new notifier to allow device to have a race free
1351 * view of the network device list.
1354 int register_netdevice_notifier(struct notifier_block *nb)
1356 struct net_device *dev;
1357 struct net_device *last;
1362 err = raw_notifier_chain_register(&netdev_chain, nb);
1368 for_each_netdev(net, dev) {
1369 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1370 err = notifier_to_errno(err);
1374 if (!(dev->flags & IFF_UP))
1377 nb->notifier_call(nb, NETDEV_UP, dev);
1388 for_each_netdev(net, dev) {
1392 if (dev->flags & IFF_UP) {
1393 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1394 nb->notifier_call(nb, NETDEV_DOWN, dev);
1396 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1397 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1401 raw_notifier_chain_unregister(&netdev_chain, nb);
1404 EXPORT_SYMBOL(register_netdevice_notifier);
1407 * unregister_netdevice_notifier - unregister a network notifier block
1410 * Unregister a notifier previously registered by
1411 * register_netdevice_notifier(). The notifier is unlinked into the
1412 * kernel structures and may then be reused. A negative errno code
1413 * is returned on a failure.
1416 int unregister_netdevice_notifier(struct notifier_block *nb)
1421 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1425 EXPORT_SYMBOL(unregister_netdevice_notifier);
1428 * call_netdevice_notifiers - call all network notifier blocks
1429 * @val: value passed unmodified to notifier function
1430 * @dev: net_device pointer passed unmodified to notifier function
1432 * Call all network notifier blocks. Parameters and return value
1433 * are as for raw_notifier_call_chain().
1436 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1439 return raw_notifier_call_chain(&netdev_chain, val, dev);
1442 /* When > 0 there are consumers of rx skb time stamps */
1443 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1445 void net_enable_timestamp(void)
1447 atomic_inc(&netstamp_needed);
1449 EXPORT_SYMBOL(net_enable_timestamp);
1451 void net_disable_timestamp(void)
1453 atomic_dec(&netstamp_needed);
1455 EXPORT_SYMBOL(net_disable_timestamp);
1457 static inline void net_timestamp(struct sk_buff *skb)
1459 if (atomic_read(&netstamp_needed))
1460 __net_timestamp(skb);
1462 skb->tstamp.tv64 = 0;
1466 * dev_forward_skb - loopback an skb to another netif
1468 * @dev: destination network device
1469 * @skb: buffer to forward
1472 * NET_RX_SUCCESS (no congestion)
1473 * NET_RX_DROP (packet was dropped, but freed)
1475 * dev_forward_skb can be used for injecting an skb from the
1476 * start_xmit function of one device into the receive queue
1477 * of another device.
1479 * The receiving device may be in another namespace, so
1480 * we have to clear all information in the skb that could
1481 * impact namespace isolation.
1483 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1487 if (!(dev->flags & IFF_UP) ||
1488 (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 = NULL;
1560 *sd->output_queue_tailp = q;
1561 sd->output_queue_tailp = &q->next_sched;
1562 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1563 local_irq_restore(flags);
1566 void __netif_schedule(struct Qdisc *q)
1568 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1569 __netif_reschedule(q);
1571 EXPORT_SYMBOL(__netif_schedule);
1573 void dev_kfree_skb_irq(struct sk_buff *skb)
1575 if (atomic_dec_and_test(&skb->users)) {
1576 struct softnet_data *sd;
1577 unsigned long flags;
1579 local_irq_save(flags);
1580 sd = &__get_cpu_var(softnet_data);
1581 skb->next = sd->completion_queue;
1582 sd->completion_queue = skb;
1583 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1584 local_irq_restore(flags);
1587 EXPORT_SYMBOL(dev_kfree_skb_irq);
1589 void dev_kfree_skb_any(struct sk_buff *skb)
1591 if (in_irq() || irqs_disabled())
1592 dev_kfree_skb_irq(skb);
1596 EXPORT_SYMBOL(dev_kfree_skb_any);
1600 * netif_device_detach - mark device as removed
1601 * @dev: network device
1603 * Mark device as removed from system and therefore no longer available.
1605 void netif_device_detach(struct net_device *dev)
1607 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1608 netif_running(dev)) {
1609 netif_tx_stop_all_queues(dev);
1612 EXPORT_SYMBOL(netif_device_detach);
1615 * netif_device_attach - mark device as attached
1616 * @dev: network device
1618 * Mark device as attached from system and restart if needed.
1620 void netif_device_attach(struct net_device *dev)
1622 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1623 netif_running(dev)) {
1624 netif_tx_wake_all_queues(dev);
1625 __netdev_watchdog_up(dev);
1628 EXPORT_SYMBOL(netif_device_attach);
1630 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1632 return ((features & NETIF_F_GEN_CSUM) ||
1633 ((features & NETIF_F_IP_CSUM) &&
1634 protocol == htons(ETH_P_IP)) ||
1635 ((features & NETIF_F_IPV6_CSUM) &&
1636 protocol == htons(ETH_P_IPV6)) ||
1637 ((features & NETIF_F_FCOE_CRC) &&
1638 protocol == htons(ETH_P_FCOE)));
1641 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1643 if (can_checksum_protocol(dev->features, skb->protocol))
1646 if (skb->protocol == htons(ETH_P_8021Q)) {
1647 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1648 if (can_checksum_protocol(dev->features & dev->vlan_features,
1649 veh->h_vlan_encapsulated_proto))
1657 * skb_dev_set -- assign a new device to a buffer
1658 * @skb: buffer for the new device
1659 * @dev: network device
1661 * If an skb is owned by a device already, we have to reset
1662 * all data private to the namespace a device belongs to
1663 * before assigning it a new device.
1665 #ifdef CONFIG_NET_NS
1666 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1669 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1672 skb_init_secmark(skb);
1676 skb->ipvs_property = 0;
1677 #ifdef CONFIG_NET_SCHED
1683 EXPORT_SYMBOL(skb_set_dev);
1684 #endif /* CONFIG_NET_NS */
1687 * Invalidate hardware checksum when packet is to be mangled, and
1688 * complete checksum manually on outgoing path.
1690 int skb_checksum_help(struct sk_buff *skb)
1693 int ret = 0, offset;
1695 if (skb->ip_summed == CHECKSUM_COMPLETE)
1696 goto out_set_summed;
1698 if (unlikely(skb_shinfo(skb)->gso_size)) {
1699 /* Let GSO fix up the checksum. */
1700 goto out_set_summed;
1703 offset = skb->csum_start - skb_headroom(skb);
1704 BUG_ON(offset >= skb_headlen(skb));
1705 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1707 offset += skb->csum_offset;
1708 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1710 if (skb_cloned(skb) &&
1711 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1712 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1717 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1719 skb->ip_summed = CHECKSUM_NONE;
1723 EXPORT_SYMBOL(skb_checksum_help);
1726 * skb_gso_segment - Perform segmentation on skb.
1727 * @skb: buffer to segment
1728 * @features: features for the output path (see dev->features)
1730 * This function segments the given skb and returns a list of segments.
1732 * It may return NULL if the skb requires no segmentation. This is
1733 * only possible when GSO is used for verifying header integrity.
1735 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1737 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1738 struct packet_type *ptype;
1739 __be16 type = skb->protocol;
1742 skb_reset_mac_header(skb);
1743 skb->mac_len = skb->network_header - skb->mac_header;
1744 __skb_pull(skb, skb->mac_len);
1746 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1747 struct net_device *dev = skb->dev;
1748 struct ethtool_drvinfo info = {};
1750 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1751 dev->ethtool_ops->get_drvinfo(dev, &info);
1753 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1755 info.driver, dev ? dev->features : 0L,
1756 skb->sk ? skb->sk->sk_route_caps : 0L,
1757 skb->len, skb->data_len, skb->ip_summed);
1759 if (skb_header_cloned(skb) &&
1760 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1761 return ERR_PTR(err);
1765 list_for_each_entry_rcu(ptype,
1766 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1767 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1768 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1769 err = ptype->gso_send_check(skb);
1770 segs = ERR_PTR(err);
1771 if (err || skb_gso_ok(skb, features))
1773 __skb_push(skb, (skb->data -
1774 skb_network_header(skb)));
1776 segs = ptype->gso_segment(skb, features);
1782 __skb_push(skb, skb->data - skb_mac_header(skb));
1786 EXPORT_SYMBOL(skb_gso_segment);
1788 /* Take action when hardware reception checksum errors are detected. */
1790 void netdev_rx_csum_fault(struct net_device *dev)
1792 if (net_ratelimit()) {
1793 printk(KERN_ERR "%s: hw csum failure.\n",
1794 dev ? dev->name : "<unknown>");
1798 EXPORT_SYMBOL(netdev_rx_csum_fault);
1801 /* Actually, we should eliminate this check as soon as we know, that:
1802 * 1. IOMMU is present and allows to map all the memory.
1803 * 2. No high memory really exists on this machine.
1806 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1808 #ifdef CONFIG_HIGHMEM
1810 if (!(dev->features & NETIF_F_HIGHDMA)) {
1811 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1812 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1816 if (PCI_DMA_BUS_IS_PHYS) {
1817 struct device *pdev = dev->dev.parent;
1821 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1822 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1823 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1832 void (*destructor)(struct sk_buff *skb);
1835 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1837 static void dev_gso_skb_destructor(struct sk_buff *skb)
1839 struct dev_gso_cb *cb;
1842 struct sk_buff *nskb = skb->next;
1844 skb->next = nskb->next;
1847 } while (skb->next);
1849 cb = DEV_GSO_CB(skb);
1851 cb->destructor(skb);
1855 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1856 * @skb: buffer to segment
1858 * This function segments the given skb and stores the list of segments
1861 static int dev_gso_segment(struct sk_buff *skb)
1863 struct net_device *dev = skb->dev;
1864 struct sk_buff *segs;
1865 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1868 segs = skb_gso_segment(skb, features);
1870 /* Verifying header integrity only. */
1875 return PTR_ERR(segs);
1878 DEV_GSO_CB(skb)->destructor = skb->destructor;
1879 skb->destructor = dev_gso_skb_destructor;
1885 * Try to orphan skb early, right before transmission by the device.
1886 * We cannot orphan skb if tx timestamp is requested, since
1887 * drivers need to call skb_tstamp_tx() to send the timestamp.
1889 static inline void skb_orphan_try(struct sk_buff *skb)
1891 if (!skb_tx(skb)->flags)
1895 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1896 struct netdev_queue *txq)
1898 const struct net_device_ops *ops = dev->netdev_ops;
1899 int rc = NETDEV_TX_OK;
1901 if (likely(!skb->next)) {
1902 if (!list_empty(&ptype_all))
1903 dev_queue_xmit_nit(skb, dev);
1906 * If device doesnt need skb->dst, release it right now while
1907 * its hot in this cpu cache
1909 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1912 skb_orphan_try(skb);
1914 if (netif_needs_gso(dev, skb)) {
1915 if (unlikely(dev_gso_segment(skb)))
1921 rc = ops->ndo_start_xmit(skb, dev);
1922 if (rc == NETDEV_TX_OK)
1923 txq_trans_update(txq);
1929 struct sk_buff *nskb = skb->next;
1931 skb->next = nskb->next;
1935 * If device doesnt need nskb->dst, release it right now while
1936 * its hot in this cpu cache
1938 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1941 rc = ops->ndo_start_xmit(nskb, dev);
1942 if (unlikely(rc != NETDEV_TX_OK)) {
1943 if (rc & ~NETDEV_TX_MASK)
1944 goto out_kfree_gso_skb;
1945 nskb->next = skb->next;
1949 txq_trans_update(txq);
1950 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1951 return NETDEV_TX_BUSY;
1952 } while (skb->next);
1955 if (likely(skb->next == NULL))
1956 skb->destructor = DEV_GSO_CB(skb)->destructor;
1962 static u32 hashrnd __read_mostly;
1964 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1968 if (skb_rx_queue_recorded(skb)) {
1969 hash = skb_get_rx_queue(skb);
1970 while (unlikely(hash >= dev->real_num_tx_queues))
1971 hash -= dev->real_num_tx_queues;
1975 if (skb->sk && skb->sk->sk_hash)
1976 hash = skb->sk->sk_hash;
1978 hash = (__force u16) skb->protocol;
1980 hash = jhash_1word(hash, hashrnd);
1982 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1984 EXPORT_SYMBOL(skb_tx_hash);
1986 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1988 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1989 if (net_ratelimit()) {
1990 pr_warning("%s selects TX queue %d, but "
1991 "real number of TX queues is %d\n",
1992 dev->name, queue_index, dev->real_num_tx_queues);
1999 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2000 struct sk_buff *skb)
2003 struct sock *sk = skb->sk;
2005 if (sk_tx_queue_recorded(sk)) {
2006 queue_index = sk_tx_queue_get(sk);
2008 const struct net_device_ops *ops = dev->netdev_ops;
2010 if (ops->ndo_select_queue) {
2011 queue_index = ops->ndo_select_queue(dev, skb);
2012 queue_index = dev_cap_txqueue(dev, queue_index);
2015 if (dev->real_num_tx_queues > 1)
2016 queue_index = skb_tx_hash(dev, skb);
2019 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2021 if (dst && skb_dst(skb) == dst)
2022 sk_tx_queue_set(sk, queue_index);
2027 skb_set_queue_mapping(skb, queue_index);
2028 return netdev_get_tx_queue(dev, queue_index);
2031 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2032 struct net_device *dev,
2033 struct netdev_queue *txq)
2035 spinlock_t *root_lock = qdisc_lock(q);
2038 spin_lock(root_lock);
2039 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2042 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2043 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
2045 * This is a work-conserving queue; there are no old skbs
2046 * waiting to be sent out; and the qdisc is not running -
2047 * xmit the skb directly.
2049 __qdisc_update_bstats(q, skb->len);
2050 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
2053 clear_bit(__QDISC_STATE_RUNNING, &q->state);
2055 rc = NET_XMIT_SUCCESS;
2057 rc = qdisc_enqueue_root(skb, q);
2060 spin_unlock(root_lock);
2066 * Returns true if either:
2067 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2068 * 2. skb is fragmented and the device does not support SG, or if
2069 * at least one of fragments is in highmem and device does not
2070 * support DMA from it.
2072 static inline int skb_needs_linearize(struct sk_buff *skb,
2073 struct net_device *dev)
2075 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
2076 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
2077 illegal_highdma(dev, skb)));
2081 * dev_queue_xmit - transmit a buffer
2082 * @skb: buffer to transmit
2084 * Queue a buffer for transmission to a network device. The caller must
2085 * have set the device and priority and built the buffer before calling
2086 * this function. The function can be called from an interrupt.
2088 * A negative errno code is returned on a failure. A success does not
2089 * guarantee the frame will be transmitted as it may be dropped due
2090 * to congestion or traffic shaping.
2092 * -----------------------------------------------------------------------------------
2093 * I notice this method can also return errors from the queue disciplines,
2094 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2097 * Regardless of the return value, the skb is consumed, so it is currently
2098 * difficult to retry a send to this method. (You can bump the ref count
2099 * before sending to hold a reference for retry if you are careful.)
2101 * When calling this method, interrupts MUST be enabled. This is because
2102 * the BH enable code must have IRQs enabled so that it will not deadlock.
2105 int dev_queue_xmit(struct sk_buff *skb)
2107 struct net_device *dev = skb->dev;
2108 struct netdev_queue *txq;
2112 /* GSO will handle the following emulations directly. */
2113 if (netif_needs_gso(dev, skb))
2116 /* Convert a paged skb to linear, if required */
2117 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2120 /* If packet is not checksummed and device does not support
2121 * checksumming for this protocol, complete checksumming here.
2123 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2124 skb_set_transport_header(skb, skb->csum_start -
2126 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2131 /* Disable soft irqs for various locks below. Also
2132 * stops preemption for RCU.
2136 txq = dev_pick_tx(dev, skb);
2137 q = rcu_dereference_bh(txq->qdisc);
2139 #ifdef CONFIG_NET_CLS_ACT
2140 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2143 rc = __dev_xmit_skb(skb, q, dev, txq);
2147 /* The device has no queue. Common case for software devices:
2148 loopback, all the sorts of tunnels...
2150 Really, it is unlikely that netif_tx_lock protection is necessary
2151 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2153 However, it is possible, that they rely on protection
2156 Check this and shot the lock. It is not prone from deadlocks.
2157 Either shot noqueue qdisc, it is even simpler 8)
2159 if (dev->flags & IFF_UP) {
2160 int cpu = smp_processor_id(); /* ok because BHs are off */
2162 if (txq->xmit_lock_owner != cpu) {
2164 HARD_TX_LOCK(dev, txq, cpu);
2166 if (!netif_tx_queue_stopped(txq)) {
2167 rc = dev_hard_start_xmit(skb, dev, txq);
2168 if (dev_xmit_complete(rc)) {
2169 HARD_TX_UNLOCK(dev, txq);
2173 HARD_TX_UNLOCK(dev, txq);
2174 if (net_ratelimit())
2175 printk(KERN_CRIT "Virtual device %s asks to "
2176 "queue packet!\n", dev->name);
2178 /* Recursion is detected! It is possible,
2180 if (net_ratelimit())
2181 printk(KERN_CRIT "Dead loop on virtual device "
2182 "%s, fix it urgently!\n", dev->name);
2187 rcu_read_unlock_bh();
2193 rcu_read_unlock_bh();
2196 EXPORT_SYMBOL(dev_queue_xmit);
2199 /*=======================================================================
2201 =======================================================================*/
2203 int netdev_max_backlog __read_mostly = 1000;
2204 int netdev_budget __read_mostly = 300;
2205 int weight_p __read_mostly = 64; /* old backlog weight */
2207 /* Called with irq disabled */
2208 static inline void ____napi_schedule(struct softnet_data *sd,
2209 struct napi_struct *napi)
2211 list_add_tail(&napi->poll_list, &sd->poll_list);
2212 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2217 /* One global table that all flow-based protocols share. */
2218 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2219 EXPORT_SYMBOL(rps_sock_flow_table);
2222 * get_rps_cpu is called from netif_receive_skb and returns the target
2223 * CPU from the RPS map of the receiving queue for a given skb.
2224 * rcu_read_lock must be held on entry.
2226 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2227 struct rps_dev_flow **rflowp)
2229 struct ipv6hdr *ip6;
2231 struct netdev_rx_queue *rxqueue;
2232 struct rps_map *map;
2233 struct rps_dev_flow_table *flow_table;
2234 struct rps_sock_flow_table *sock_flow_table;
2238 u32 addr1, addr2, ihl;
2244 if (skb_rx_queue_recorded(skb)) {
2245 u16 index = skb_get_rx_queue(skb);
2246 if (unlikely(index >= dev->num_rx_queues)) {
2247 if (net_ratelimit()) {
2248 pr_warning("%s received packet on queue "
2249 "%u, but number of RX queues is %u\n",
2250 dev->name, index, dev->num_rx_queues);
2254 rxqueue = dev->_rx + index;
2258 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2262 goto got_hash; /* Skip hash computation on packet header */
2264 switch (skb->protocol) {
2265 case __constant_htons(ETH_P_IP):
2266 if (!pskb_may_pull(skb, sizeof(*ip)))
2269 ip = (struct iphdr *) skb->data;
2270 ip_proto = ip->protocol;
2271 addr1 = (__force u32) ip->saddr;
2272 addr2 = (__force u32) ip->daddr;
2275 case __constant_htons(ETH_P_IPV6):
2276 if (!pskb_may_pull(skb, sizeof(*ip6)))
2279 ip6 = (struct ipv6hdr *) skb->data;
2280 ip_proto = ip6->nexthdr;
2281 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2282 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2295 case IPPROTO_UDPLITE:
2296 if (pskb_may_pull(skb, (ihl * 4) + 4)) {
2297 ports.v32 = * (__force u32 *) (skb->data + (ihl * 4));
2298 if (ports.v16[1] < ports.v16[0])
2299 swap(ports.v16[0], ports.v16[1]);
2307 /* get a consistent hash (same value on both flow directions) */
2310 skb->rxhash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2315 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2316 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2317 if (flow_table && sock_flow_table) {
2319 struct rps_dev_flow *rflow;
2321 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2324 next_cpu = sock_flow_table->ents[skb->rxhash &
2325 sock_flow_table->mask];
2328 * If the desired CPU (where last recvmsg was done) is
2329 * different from current CPU (one in the rx-queue flow
2330 * table entry), switch if one of the following holds:
2331 * - Current CPU is unset (equal to RPS_NO_CPU).
2332 * - Current CPU is offline.
2333 * - The current CPU's queue tail has advanced beyond the
2334 * last packet that was enqueued using this table entry.
2335 * This guarantees that all previous packets for the flow
2336 * have been dequeued, thus preserving in order delivery.
2338 if (unlikely(tcpu != next_cpu) &&
2339 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2340 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2341 rflow->last_qtail)) >= 0)) {
2342 tcpu = rflow->cpu = next_cpu;
2343 if (tcpu != RPS_NO_CPU)
2344 rflow->last_qtail = per_cpu(softnet_data,
2345 tcpu).input_queue_head;
2347 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2354 map = rcu_dereference(rxqueue->rps_map);
2356 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2358 if (cpu_online(tcpu)) {
2368 /* Called from hardirq (IPI) context */
2369 static void rps_trigger_softirq(void *data)
2371 struct softnet_data *sd = data;
2373 ____napi_schedule(sd, &sd->backlog);
2377 #endif /* CONFIG_RPS */
2380 * Check if this softnet_data structure is another cpu one
2381 * If yes, queue it to our IPI list and return 1
2384 static int rps_ipi_queued(struct softnet_data *sd)
2387 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2390 sd->rps_ipi_next = mysd->rps_ipi_list;
2391 mysd->rps_ipi_list = sd;
2393 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2396 #endif /* CONFIG_RPS */
2401 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2402 * queue (may be a remote CPU queue).
2404 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2405 unsigned int *qtail)
2407 struct softnet_data *sd;
2408 unsigned long flags;
2410 sd = &per_cpu(softnet_data, cpu);
2412 local_irq_save(flags);
2415 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2416 if (skb_queue_len(&sd->input_pkt_queue)) {
2418 __skb_queue_tail(&sd->input_pkt_queue, skb);
2420 *qtail = sd->input_queue_head +
2421 skb_queue_len(&sd->input_pkt_queue);
2424 local_irq_restore(flags);
2425 return NET_RX_SUCCESS;
2428 /* Schedule NAPI for backlog device */
2429 if (napi_schedule_prep(&sd->backlog)) {
2430 if (!rps_ipi_queued(sd))
2431 ____napi_schedule(sd, &sd->backlog);
2439 local_irq_restore(flags);
2446 * netif_rx - post buffer to the network code
2447 * @skb: buffer to post
2449 * This function receives a packet from a device driver and queues it for
2450 * the upper (protocol) levels to process. It always succeeds. The buffer
2451 * may be dropped during processing for congestion control or by the
2455 * NET_RX_SUCCESS (no congestion)
2456 * NET_RX_DROP (packet was dropped)
2460 int netif_rx(struct sk_buff *skb)
2464 /* if netpoll wants it, pretend we never saw it */
2465 if (netpoll_rx(skb))
2468 if (!skb->tstamp.tv64)
2473 struct rps_dev_flow voidflow, *rflow = &voidflow;
2478 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2480 cpu = smp_processor_id();
2482 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2489 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2495 EXPORT_SYMBOL(netif_rx);
2497 int netif_rx_ni(struct sk_buff *skb)
2502 err = netif_rx(skb);
2503 if (local_softirq_pending())
2509 EXPORT_SYMBOL(netif_rx_ni);
2511 static void net_tx_action(struct softirq_action *h)
2513 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2515 if (sd->completion_queue) {
2516 struct sk_buff *clist;
2518 local_irq_disable();
2519 clist = sd->completion_queue;
2520 sd->completion_queue = NULL;
2524 struct sk_buff *skb = clist;
2525 clist = clist->next;
2527 WARN_ON(atomic_read(&skb->users));
2532 if (sd->output_queue) {
2535 local_irq_disable();
2536 head = sd->output_queue;
2537 sd->output_queue = NULL;
2538 sd->output_queue_tailp = &sd->output_queue;
2542 struct Qdisc *q = head;
2543 spinlock_t *root_lock;
2545 head = head->next_sched;
2547 root_lock = qdisc_lock(q);
2548 if (spin_trylock(root_lock)) {
2549 smp_mb__before_clear_bit();
2550 clear_bit(__QDISC_STATE_SCHED,
2553 spin_unlock(root_lock);
2555 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2557 __netif_reschedule(q);
2559 smp_mb__before_clear_bit();
2560 clear_bit(__QDISC_STATE_SCHED,
2568 static inline int deliver_skb(struct sk_buff *skb,
2569 struct packet_type *pt_prev,
2570 struct net_device *orig_dev)
2572 atomic_inc(&skb->users);
2573 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2576 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2578 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2579 /* This hook is defined here for ATM LANE */
2580 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2581 unsigned char *addr) __read_mostly;
2582 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2586 * If bridge module is loaded call bridging hook.
2587 * returns NULL if packet was consumed.
2589 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2590 struct sk_buff *skb) __read_mostly;
2591 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2593 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2594 struct packet_type **pt_prev, int *ret,
2595 struct net_device *orig_dev)
2597 struct net_bridge_port *port;
2599 if (skb->pkt_type == PACKET_LOOPBACK ||
2600 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2604 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2608 return br_handle_frame_hook(port, skb);
2611 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2614 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2615 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2616 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2618 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2619 struct packet_type **pt_prev,
2621 struct net_device *orig_dev)
2623 if (skb->dev->macvlan_port == NULL)
2627 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2630 return macvlan_handle_frame_hook(skb);
2633 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2636 #ifdef CONFIG_NET_CLS_ACT
2637 /* TODO: Maybe we should just force sch_ingress to be compiled in
2638 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2639 * a compare and 2 stores extra right now if we dont have it on
2640 * but have CONFIG_NET_CLS_ACT
2641 * NOTE: This doesnt stop any functionality; if you dont have
2642 * the ingress scheduler, you just cant add policies on ingress.
2645 static int ing_filter(struct sk_buff *skb)
2647 struct net_device *dev = skb->dev;
2648 u32 ttl = G_TC_RTTL(skb->tc_verd);
2649 struct netdev_queue *rxq;
2650 int result = TC_ACT_OK;
2653 if (MAX_RED_LOOP < ttl++) {
2655 "Redir loop detected Dropping packet (%d->%d)\n",
2656 skb->skb_iif, dev->ifindex);
2660 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2661 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2663 rxq = &dev->rx_queue;
2666 if (q != &noop_qdisc) {
2667 spin_lock(qdisc_lock(q));
2668 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2669 result = qdisc_enqueue_root(skb, q);
2670 spin_unlock(qdisc_lock(q));
2676 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2677 struct packet_type **pt_prev,
2678 int *ret, struct net_device *orig_dev)
2680 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2684 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2687 /* Huh? Why does turning on AF_PACKET affect this? */
2688 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2691 switch (ing_filter(skb)) {
2705 * netif_nit_deliver - deliver received packets to network taps
2708 * This function is used to deliver incoming packets to network
2709 * taps. It should be used when the normal netif_receive_skb path
2710 * is bypassed, for example because of VLAN acceleration.
2712 void netif_nit_deliver(struct sk_buff *skb)
2714 struct packet_type *ptype;
2716 if (list_empty(&ptype_all))
2719 skb_reset_network_header(skb);
2720 skb_reset_transport_header(skb);
2721 skb->mac_len = skb->network_header - skb->mac_header;
2724 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2725 if (!ptype->dev || ptype->dev == skb->dev)
2726 deliver_skb(skb, ptype, skb->dev);
2731 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2732 struct net_device *master)
2734 if (skb->pkt_type == PACKET_HOST) {
2735 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2737 memcpy(dest, master->dev_addr, ETH_ALEN);
2741 /* On bonding slaves other than the currently active slave, suppress
2742 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2743 * ARP on active-backup slaves with arp_validate enabled.
2745 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2747 struct net_device *dev = skb->dev;
2749 if (master->priv_flags & IFF_MASTER_ARPMON)
2750 dev->last_rx = jiffies;
2752 if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) {
2753 /* Do address unmangle. The local destination address
2754 * will be always the one master has. Provides the right
2755 * functionality in a bridge.
2757 skb_bond_set_mac_by_master(skb, master);
2760 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2761 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2762 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2765 if (master->priv_flags & IFF_MASTER_ALB) {
2766 if (skb->pkt_type != PACKET_BROADCAST &&
2767 skb->pkt_type != PACKET_MULTICAST)
2770 if (master->priv_flags & IFF_MASTER_8023AD &&
2771 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2778 EXPORT_SYMBOL(__skb_bond_should_drop);
2780 static int __netif_receive_skb(struct sk_buff *skb)
2782 struct packet_type *ptype, *pt_prev;
2783 struct net_device *orig_dev;
2784 struct net_device *master;
2785 struct net_device *null_or_orig;
2786 struct net_device *null_or_bond;
2787 int ret = NET_RX_DROP;
2790 if (!skb->tstamp.tv64)
2793 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2794 return NET_RX_SUCCESS;
2796 /* if we've gotten here through NAPI, check netpoll */
2797 if (netpoll_receive_skb(skb))
2801 skb->skb_iif = skb->dev->ifindex;
2803 null_or_orig = NULL;
2804 orig_dev = skb->dev;
2805 master = ACCESS_ONCE(orig_dev->master);
2807 if (skb_bond_should_drop(skb, master))
2808 null_or_orig = orig_dev; /* deliver only exact match */
2813 __get_cpu_var(softnet_data).processed++;
2815 skb_reset_network_header(skb);
2816 skb_reset_transport_header(skb);
2817 skb->mac_len = skb->network_header - skb->mac_header;
2823 #ifdef CONFIG_NET_CLS_ACT
2824 if (skb->tc_verd & TC_NCLS) {
2825 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2830 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2831 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2832 ptype->dev == orig_dev) {
2834 ret = deliver_skb(skb, pt_prev, orig_dev);
2839 #ifdef CONFIG_NET_CLS_ACT
2840 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2846 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2849 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2854 * Make sure frames received on VLAN interfaces stacked on
2855 * bonding interfaces still make their way to any base bonding
2856 * device that may have registered for a specific ptype. The
2857 * handler may have to adjust skb->dev and orig_dev.
2859 null_or_bond = NULL;
2860 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2861 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2862 null_or_bond = vlan_dev_real_dev(skb->dev);
2865 type = skb->protocol;
2866 list_for_each_entry_rcu(ptype,
2867 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2868 if (ptype->type == type && (ptype->dev == null_or_orig ||
2869 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2870 ptype->dev == null_or_bond)) {
2872 ret = deliver_skb(skb, pt_prev, orig_dev);
2878 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2881 /* Jamal, now you will not able to escape explaining
2882 * me how you were going to use this. :-)
2893 * netif_receive_skb - process receive buffer from network
2894 * @skb: buffer to process
2896 * netif_receive_skb() is the main receive data processing function.
2897 * It always succeeds. The buffer may be dropped during processing
2898 * for congestion control or by the protocol layers.
2900 * This function may only be called from softirq context and interrupts
2901 * should be enabled.
2903 * Return values (usually ignored):
2904 * NET_RX_SUCCESS: no congestion
2905 * NET_RX_DROP: packet was dropped
2907 int netif_receive_skb(struct sk_buff *skb)
2910 struct rps_dev_flow voidflow, *rflow = &voidflow;
2915 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2918 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2922 ret = __netif_receive_skb(skb);
2927 return __netif_receive_skb(skb);
2930 EXPORT_SYMBOL(netif_receive_skb);
2932 /* Network device is going away, flush any packets still pending
2933 * Called with irqs disabled.
2935 static void flush_backlog(void *arg)
2937 struct net_device *dev = arg;
2938 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2939 struct sk_buff *skb, *tmp;
2942 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
2943 if (skb->dev == dev) {
2944 __skb_unlink(skb, &sd->input_pkt_queue);
2946 input_queue_head_add(sd, 1);
2951 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
2952 if (skb->dev == dev) {
2953 __skb_unlink(skb, &sd->process_queue);
2959 static int napi_gro_complete(struct sk_buff *skb)
2961 struct packet_type *ptype;
2962 __be16 type = skb->protocol;
2963 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2966 if (NAPI_GRO_CB(skb)->count == 1) {
2967 skb_shinfo(skb)->gso_size = 0;
2972 list_for_each_entry_rcu(ptype, head, list) {
2973 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2976 err = ptype->gro_complete(skb);
2982 WARN_ON(&ptype->list == head);
2984 return NET_RX_SUCCESS;
2988 return netif_receive_skb(skb);
2991 static void napi_gro_flush(struct napi_struct *napi)
2993 struct sk_buff *skb, *next;
2995 for (skb = napi->gro_list; skb; skb = next) {
2998 napi_gro_complete(skb);
3001 napi->gro_count = 0;
3002 napi->gro_list = NULL;
3005 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3007 struct sk_buff **pp = NULL;
3008 struct packet_type *ptype;
3009 __be16 type = skb->protocol;
3010 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3013 enum gro_result ret;
3015 if (!(skb->dev->features & NETIF_F_GRO))
3018 if (skb_is_gso(skb) || skb_has_frags(skb))
3022 list_for_each_entry_rcu(ptype, head, list) {
3023 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3026 skb_set_network_header(skb, skb_gro_offset(skb));
3027 mac_len = skb->network_header - skb->mac_header;
3028 skb->mac_len = mac_len;
3029 NAPI_GRO_CB(skb)->same_flow = 0;
3030 NAPI_GRO_CB(skb)->flush = 0;
3031 NAPI_GRO_CB(skb)->free = 0;
3033 pp = ptype->gro_receive(&napi->gro_list, skb);
3038 if (&ptype->list == head)
3041 same_flow = NAPI_GRO_CB(skb)->same_flow;
3042 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3045 struct sk_buff *nskb = *pp;
3049 napi_gro_complete(nskb);
3056 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3060 NAPI_GRO_CB(skb)->count = 1;
3061 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3062 skb->next = napi->gro_list;
3063 napi->gro_list = skb;
3067 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3068 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3070 BUG_ON(skb->end - skb->tail < grow);
3072 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3075 skb->data_len -= grow;
3077 skb_shinfo(skb)->frags[0].page_offset += grow;
3078 skb_shinfo(skb)->frags[0].size -= grow;
3080 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3081 put_page(skb_shinfo(skb)->frags[0].page);
3082 memmove(skb_shinfo(skb)->frags,
3083 skb_shinfo(skb)->frags + 1,
3084 --skb_shinfo(skb)->nr_frags);
3095 EXPORT_SYMBOL(dev_gro_receive);
3098 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3102 if (netpoll_rx_on(skb))
3105 for (p = napi->gro_list; p; p = p->next) {
3106 NAPI_GRO_CB(p)->same_flow =
3107 (p->dev == skb->dev) &&
3108 !compare_ether_header(skb_mac_header(p),
3109 skb_gro_mac_header(skb));
3110 NAPI_GRO_CB(p)->flush = 0;
3113 return dev_gro_receive(napi, skb);
3116 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3120 if (netif_receive_skb(skb))
3125 case GRO_MERGED_FREE:
3136 EXPORT_SYMBOL(napi_skb_finish);
3138 void skb_gro_reset_offset(struct sk_buff *skb)
3140 NAPI_GRO_CB(skb)->data_offset = 0;
3141 NAPI_GRO_CB(skb)->frag0 = NULL;
3142 NAPI_GRO_CB(skb)->frag0_len = 0;
3144 if (skb->mac_header == skb->tail &&
3145 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3146 NAPI_GRO_CB(skb)->frag0 =
3147 page_address(skb_shinfo(skb)->frags[0].page) +
3148 skb_shinfo(skb)->frags[0].page_offset;
3149 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3152 EXPORT_SYMBOL(skb_gro_reset_offset);
3154 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3156 skb_gro_reset_offset(skb);
3158 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3160 EXPORT_SYMBOL(napi_gro_receive);
3162 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3164 __skb_pull(skb, skb_headlen(skb));
3165 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3169 EXPORT_SYMBOL(napi_reuse_skb);
3171 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3173 struct sk_buff *skb = napi->skb;
3176 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3182 EXPORT_SYMBOL(napi_get_frags);
3184 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3190 skb->protocol = eth_type_trans(skb, skb->dev);
3192 if (ret == GRO_HELD)
3193 skb_gro_pull(skb, -ETH_HLEN);
3194 else if (netif_receive_skb(skb))
3199 case GRO_MERGED_FREE:
3200 napi_reuse_skb(napi, skb);
3209 EXPORT_SYMBOL(napi_frags_finish);
3211 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3213 struct sk_buff *skb = napi->skb;
3220 skb_reset_mac_header(skb);
3221 skb_gro_reset_offset(skb);
3223 off = skb_gro_offset(skb);
3224 hlen = off + sizeof(*eth);
3225 eth = skb_gro_header_fast(skb, off);
3226 if (skb_gro_header_hard(skb, hlen)) {
3227 eth = skb_gro_header_slow(skb, hlen, off);
3228 if (unlikely(!eth)) {
3229 napi_reuse_skb(napi, skb);
3235 skb_gro_pull(skb, sizeof(*eth));
3238 * This works because the only protocols we care about don't require
3239 * special handling. We'll fix it up properly at the end.
3241 skb->protocol = eth->h_proto;
3246 EXPORT_SYMBOL(napi_frags_skb);
3248 gro_result_t napi_gro_frags(struct napi_struct *napi)
3250 struct sk_buff *skb = napi_frags_skb(napi);
3255 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3257 EXPORT_SYMBOL(napi_gro_frags);
3260 * net_rps_action sends any pending IPI's for rps.
3261 * Note: called with local irq disabled, but exits with local irq enabled.
3263 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3266 struct softnet_data *remsd = sd->rps_ipi_list;
3269 sd->rps_ipi_list = NULL;
3273 /* Send pending IPI's to kick RPS processing on remote cpus. */
3275 struct softnet_data *next = remsd->rps_ipi_next;
3277 if (cpu_online(remsd->cpu))
3278 __smp_call_function_single(remsd->cpu,
3287 static int process_backlog(struct napi_struct *napi, int quota)
3290 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3293 /* Check if we have pending ipi, its better to send them now,
3294 * not waiting net_rx_action() end.
3296 if (sd->rps_ipi_list) {
3297 local_irq_disable();
3298 net_rps_action_and_irq_enable(sd);
3301 napi->weight = weight_p;
3302 local_irq_disable();
3303 while (work < quota) {
3304 struct sk_buff *skb;
3307 while ((skb = __skb_dequeue(&sd->process_queue))) {
3309 __netif_receive_skb(skb);
3310 if (++work >= quota)
3312 local_irq_disable();
3316 qlen = skb_queue_len(&sd->input_pkt_queue);
3318 input_queue_head_add(sd, qlen);
3319 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3320 &sd->process_queue);
3322 if (qlen < quota - work) {
3324 * Inline a custom version of __napi_complete().
3325 * only current cpu owns and manipulates this napi,
3326 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3327 * we can use a plain write instead of clear_bit(),
3328 * and we dont need an smp_mb() memory barrier.
3330 list_del(&napi->poll_list);
3333 quota = work + qlen;
3343 * __napi_schedule - schedule for receive
3344 * @n: entry to schedule
3346 * The entry's receive function will be scheduled to run
3348 void __napi_schedule(struct napi_struct *n)
3350 unsigned long flags;
3352 local_irq_save(flags);
3353 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3354 local_irq_restore(flags);
3356 EXPORT_SYMBOL(__napi_schedule);
3358 void __napi_complete(struct napi_struct *n)
3360 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3361 BUG_ON(n->gro_list);
3363 list_del(&n->poll_list);
3364 smp_mb__before_clear_bit();
3365 clear_bit(NAPI_STATE_SCHED, &n->state);
3367 EXPORT_SYMBOL(__napi_complete);
3369 void napi_complete(struct napi_struct *n)
3371 unsigned long flags;
3374 * don't let napi dequeue from the cpu poll list
3375 * just in case its running on a different cpu
3377 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3381 local_irq_save(flags);
3383 local_irq_restore(flags);
3385 EXPORT_SYMBOL(napi_complete);
3387 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3388 int (*poll)(struct napi_struct *, int), int weight)
3390 INIT_LIST_HEAD(&napi->poll_list);
3391 napi->gro_count = 0;
3392 napi->gro_list = NULL;
3395 napi->weight = weight;
3396 list_add(&napi->dev_list, &dev->napi_list);
3398 #ifdef CONFIG_NETPOLL
3399 spin_lock_init(&napi->poll_lock);
3400 napi->poll_owner = -1;
3402 set_bit(NAPI_STATE_SCHED, &napi->state);
3404 EXPORT_SYMBOL(netif_napi_add);
3406 void netif_napi_del(struct napi_struct *napi)
3408 struct sk_buff *skb, *next;
3410 list_del_init(&napi->dev_list);
3411 napi_free_frags(napi);
3413 for (skb = napi->gro_list; skb; skb = next) {
3419 napi->gro_list = NULL;
3420 napi->gro_count = 0;
3422 EXPORT_SYMBOL(netif_napi_del);
3424 static void net_rx_action(struct softirq_action *h)
3426 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3427 unsigned long time_limit = jiffies + 2;
3428 int budget = netdev_budget;
3431 local_irq_disable();
3433 while (!list_empty(&sd->poll_list)) {
3434 struct napi_struct *n;
3437 /* If softirq window is exhuasted then punt.
3438 * Allow this to run for 2 jiffies since which will allow
3439 * an average latency of 1.5/HZ.
3441 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3446 /* Even though interrupts have been re-enabled, this
3447 * access is safe because interrupts can only add new
3448 * entries to the tail of this list, and only ->poll()
3449 * calls can remove this head entry from the list.
3451 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3453 have = netpoll_poll_lock(n);
3457 /* This NAPI_STATE_SCHED test is for avoiding a race
3458 * with netpoll's poll_napi(). Only the entity which
3459 * obtains the lock and sees NAPI_STATE_SCHED set will
3460 * actually make the ->poll() call. Therefore we avoid
3461 * accidently calling ->poll() when NAPI is not scheduled.
3464 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3465 work = n->poll(n, weight);
3469 WARN_ON_ONCE(work > weight);
3473 local_irq_disable();
3475 /* Drivers must not modify the NAPI state if they
3476 * consume the entire weight. In such cases this code
3477 * still "owns" the NAPI instance and therefore can
3478 * move the instance around on the list at-will.
3480 if (unlikely(work == weight)) {
3481 if (unlikely(napi_disable_pending(n))) {
3484 local_irq_disable();
3486 list_move_tail(&n->poll_list, &sd->poll_list);
3489 netpoll_poll_unlock(have);
3492 net_rps_action_and_irq_enable(sd);
3494 #ifdef CONFIG_NET_DMA
3496 * There may not be any more sk_buffs coming right now, so push
3497 * any pending DMA copies to hardware
3499 dma_issue_pending_all();
3506 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3510 static gifconf_func_t *gifconf_list[NPROTO];
3513 * register_gifconf - register a SIOCGIF handler
3514 * @family: Address family
3515 * @gifconf: Function handler
3517 * Register protocol dependent address dumping routines. The handler
3518 * that is passed must not be freed or reused until it has been replaced
3519 * by another handler.
3521 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3523 if (family >= NPROTO)
3525 gifconf_list[family] = gifconf;
3528 EXPORT_SYMBOL(register_gifconf);
3532 * Map an interface index to its name (SIOCGIFNAME)
3536 * We need this ioctl for efficient implementation of the
3537 * if_indextoname() function required by the IPv6 API. Without
3538 * it, we would have to search all the interfaces to find a
3542 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3544 struct net_device *dev;
3548 * Fetch the caller's info block.
3551 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3555 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3561 strcpy(ifr.ifr_name, dev->name);
3564 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3570 * Perform a SIOCGIFCONF call. This structure will change
3571 * size eventually, and there is nothing I can do about it.
3572 * Thus we will need a 'compatibility mode'.
3575 static int dev_ifconf(struct net *net, char __user *arg)
3578 struct net_device *dev;
3585 * Fetch the caller's info block.
3588 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3595 * Loop over the interfaces, and write an info block for each.
3599 for_each_netdev(net, dev) {
3600 for (i = 0; i < NPROTO; i++) {
3601 if (gifconf_list[i]) {
3604 done = gifconf_list[i](dev, NULL, 0);
3606 done = gifconf_list[i](dev, pos + total,
3616 * All done. Write the updated control block back to the caller.
3618 ifc.ifc_len = total;
3621 * Both BSD and Solaris return 0 here, so we do too.
3623 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3626 #ifdef CONFIG_PROC_FS
3628 * This is invoked by the /proc filesystem handler to display a device
3631 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3634 struct net *net = seq_file_net(seq);
3636 struct net_device *dev;
3640 return SEQ_START_TOKEN;
3643 for_each_netdev_rcu(net, dev)
3650 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3652 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3653 first_net_device(seq_file_net(seq)) :
3654 next_net_device((struct net_device *)v);
3657 return rcu_dereference(dev);
3660 void dev_seq_stop(struct seq_file *seq, void *v)
3666 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3668 const struct net_device_stats *stats = dev_get_stats(dev);
3670 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3671 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3672 dev->name, stats->rx_bytes, stats->rx_packets,
3674 stats->rx_dropped + stats->rx_missed_errors,
3675 stats->rx_fifo_errors,
3676 stats->rx_length_errors + stats->rx_over_errors +
3677 stats->rx_crc_errors + stats->rx_frame_errors,
3678 stats->rx_compressed, stats->multicast,
3679 stats->tx_bytes, stats->tx_packets,
3680 stats->tx_errors, stats->tx_dropped,
3681 stats->tx_fifo_errors, stats->collisions,
3682 stats->tx_carrier_errors +
3683 stats->tx_aborted_errors +
3684 stats->tx_window_errors +
3685 stats->tx_heartbeat_errors,
3686 stats->tx_compressed);
3690 * Called from the PROCfs module. This now uses the new arbitrary sized
3691 * /proc/net interface to create /proc/net/dev
3693 static int dev_seq_show(struct seq_file *seq, void *v)
3695 if (v == SEQ_START_TOKEN)
3696 seq_puts(seq, "Inter-| Receive "
3698 " face |bytes packets errs drop fifo frame "
3699 "compressed multicast|bytes packets errs "
3700 "drop fifo colls carrier compressed\n");
3702 dev_seq_printf_stats(seq, v);
3706 static struct softnet_data *softnet_get_online(loff_t *pos)
3708 struct softnet_data *sd = NULL;
3710 while (*pos < nr_cpu_ids)
3711 if (cpu_online(*pos)) {
3712 sd = &per_cpu(softnet_data, *pos);
3719 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3721 return softnet_get_online(pos);
3724 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3727 return softnet_get_online(pos);
3730 static void softnet_seq_stop(struct seq_file *seq, void *v)
3734 static int softnet_seq_show(struct seq_file *seq, void *v)
3736 struct softnet_data *sd = v;
3738 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3739 sd->processed, sd->dropped, sd->time_squeeze, 0,
3740 0, 0, 0, 0, /* was fastroute */
3741 sd->cpu_collision, sd->received_rps);
3745 static const struct seq_operations dev_seq_ops = {
3746 .start = dev_seq_start,
3747 .next = dev_seq_next,
3748 .stop = dev_seq_stop,
3749 .show = dev_seq_show,
3752 static int dev_seq_open(struct inode *inode, struct file *file)
3754 return seq_open_net(inode, file, &dev_seq_ops,
3755 sizeof(struct seq_net_private));
3758 static const struct file_operations dev_seq_fops = {
3759 .owner = THIS_MODULE,
3760 .open = dev_seq_open,
3762 .llseek = seq_lseek,
3763 .release = seq_release_net,
3766 static const struct seq_operations softnet_seq_ops = {
3767 .start = softnet_seq_start,
3768 .next = softnet_seq_next,
3769 .stop = softnet_seq_stop,
3770 .show = softnet_seq_show,
3773 static int softnet_seq_open(struct inode *inode, struct file *file)
3775 return seq_open(file, &softnet_seq_ops);
3778 static const struct file_operations softnet_seq_fops = {
3779 .owner = THIS_MODULE,
3780 .open = softnet_seq_open,
3782 .llseek = seq_lseek,
3783 .release = seq_release,
3786 static void *ptype_get_idx(loff_t pos)
3788 struct packet_type *pt = NULL;
3792 list_for_each_entry_rcu(pt, &ptype_all, list) {
3798 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3799 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3808 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3812 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3815 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3817 struct packet_type *pt;
3818 struct list_head *nxt;
3822 if (v == SEQ_START_TOKEN)
3823 return ptype_get_idx(0);
3826 nxt = pt->list.next;
3827 if (pt->type == htons(ETH_P_ALL)) {
3828 if (nxt != &ptype_all)
3831 nxt = ptype_base[0].next;
3833 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3835 while (nxt == &ptype_base[hash]) {
3836 if (++hash >= PTYPE_HASH_SIZE)
3838 nxt = ptype_base[hash].next;
3841 return list_entry(nxt, struct packet_type, list);
3844 static void ptype_seq_stop(struct seq_file *seq, void *v)
3850 static int ptype_seq_show(struct seq_file *seq, void *v)
3852 struct packet_type *pt = v;
3854 if (v == SEQ_START_TOKEN)
3855 seq_puts(seq, "Type Device Function\n");
3856 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3857 if (pt->type == htons(ETH_P_ALL))
3858 seq_puts(seq, "ALL ");
3860 seq_printf(seq, "%04x", ntohs(pt->type));
3862 seq_printf(seq, " %-8s %pF\n",
3863 pt->dev ? pt->dev->name : "", pt->func);
3869 static const struct seq_operations ptype_seq_ops = {
3870 .start = ptype_seq_start,
3871 .next = ptype_seq_next,
3872 .stop = ptype_seq_stop,
3873 .show = ptype_seq_show,
3876 static int ptype_seq_open(struct inode *inode, struct file *file)
3878 return seq_open_net(inode, file, &ptype_seq_ops,
3879 sizeof(struct seq_net_private));
3882 static const struct file_operations ptype_seq_fops = {
3883 .owner = THIS_MODULE,
3884 .open = ptype_seq_open,
3886 .llseek = seq_lseek,
3887 .release = seq_release_net,
3891 static int __net_init dev_proc_net_init(struct net *net)
3895 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3897 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3899 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3902 if (wext_proc_init(net))
3908 proc_net_remove(net, "ptype");
3910 proc_net_remove(net, "softnet_stat");
3912 proc_net_remove(net, "dev");
3916 static void __net_exit dev_proc_net_exit(struct net *net)
3918 wext_proc_exit(net);
3920 proc_net_remove(net, "ptype");
3921 proc_net_remove(net, "softnet_stat");
3922 proc_net_remove(net, "dev");
3925 static struct pernet_operations __net_initdata dev_proc_ops = {
3926 .init = dev_proc_net_init,
3927 .exit = dev_proc_net_exit,
3930 static int __init dev_proc_init(void)
3932 return register_pernet_subsys(&dev_proc_ops);
3935 #define dev_proc_init() 0
3936 #endif /* CONFIG_PROC_FS */
3940 * netdev_set_master - set up master/slave pair
3941 * @slave: slave device
3942 * @master: new master device
3944 * Changes the master device of the slave. Pass %NULL to break the
3945 * bonding. The caller must hold the RTNL semaphore. On a failure
3946 * a negative errno code is returned. On success the reference counts
3947 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3948 * function returns zero.
3950 int netdev_set_master(struct net_device *slave, struct net_device *master)
3952 struct net_device *old = slave->master;
3962 slave->master = master;
3969 slave->flags |= IFF_SLAVE;
3971 slave->flags &= ~IFF_SLAVE;
3973 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3976 EXPORT_SYMBOL(netdev_set_master);
3978 static void dev_change_rx_flags(struct net_device *dev, int flags)
3980 const struct net_device_ops *ops = dev->netdev_ops;
3982 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3983 ops->ndo_change_rx_flags(dev, flags);
3986 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3988 unsigned short old_flags = dev->flags;
3994 dev->flags |= IFF_PROMISC;
3995 dev->promiscuity += inc;
3996 if (dev->promiscuity == 0) {
3999 * If inc causes overflow, untouch promisc and return error.
4002 dev->flags &= ~IFF_PROMISC;
4004 dev->promiscuity -= inc;
4005 printk(KERN_WARNING "%s: promiscuity touches roof, "
4006 "set promiscuity failed, promiscuity feature "
4007 "of device might be broken.\n", dev->name);
4011 if (dev->flags != old_flags) {
4012 printk(KERN_INFO "device %s %s promiscuous mode\n",
4013 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4015 if (audit_enabled) {
4016 current_uid_gid(&uid, &gid);
4017 audit_log(current->audit_context, GFP_ATOMIC,
4018 AUDIT_ANOM_PROMISCUOUS,
4019 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4020 dev->name, (dev->flags & IFF_PROMISC),
4021 (old_flags & IFF_PROMISC),
4022 audit_get_loginuid(current),
4024 audit_get_sessionid(current));
4027 dev_change_rx_flags(dev, IFF_PROMISC);
4033 * dev_set_promiscuity - update promiscuity count on a device
4037 * Add or remove promiscuity from a device. While the count in the device
4038 * remains above zero the interface remains promiscuous. Once it hits zero
4039 * the device reverts back to normal filtering operation. A negative inc
4040 * value is used to drop promiscuity on the device.
4041 * Return 0 if successful or a negative errno code on error.
4043 int dev_set_promiscuity(struct net_device *dev, int inc)
4045 unsigned short old_flags = dev->flags;
4048 err = __dev_set_promiscuity(dev, inc);
4051 if (dev->flags != old_flags)
4052 dev_set_rx_mode(dev);
4055 EXPORT_SYMBOL(dev_set_promiscuity);
4058 * dev_set_allmulti - update allmulti count on a device
4062 * Add or remove reception of all multicast frames to a device. While the
4063 * count in the device remains above zero the interface remains listening
4064 * to all interfaces. Once it hits zero the device reverts back to normal
4065 * filtering operation. A negative @inc value is used to drop the counter
4066 * when releasing a resource needing all multicasts.
4067 * Return 0 if successful or a negative errno code on error.
4070 int dev_set_allmulti(struct net_device *dev, int inc)
4072 unsigned short old_flags = dev->flags;
4076 dev->flags |= IFF_ALLMULTI;
4077 dev->allmulti += inc;
4078 if (dev->allmulti == 0) {
4081 * If inc causes overflow, untouch allmulti and return error.
4084 dev->flags &= ~IFF_ALLMULTI;
4086 dev->allmulti -= inc;
4087 printk(KERN_WARNING "%s: allmulti touches roof, "
4088 "set allmulti failed, allmulti feature of "
4089 "device might be broken.\n", dev->name);
4093 if (dev->flags ^ old_flags) {
4094 dev_change_rx_flags(dev, IFF_ALLMULTI);
4095 dev_set_rx_mode(dev);
4099 EXPORT_SYMBOL(dev_set_allmulti);
4102 * Upload unicast and multicast address lists to device and
4103 * configure RX filtering. When the device doesn't support unicast
4104 * filtering it is put in promiscuous mode while unicast addresses
4107 void __dev_set_rx_mode(struct net_device *dev)
4109 const struct net_device_ops *ops = dev->netdev_ops;
4111 /* dev_open will call this function so the list will stay sane. */
4112 if (!(dev->flags&IFF_UP))
4115 if (!netif_device_present(dev))
4118 if (ops->ndo_set_rx_mode)
4119 ops->ndo_set_rx_mode(dev);
4121 /* Unicast addresses changes may only happen under the rtnl,
4122 * therefore calling __dev_set_promiscuity here is safe.
4124 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4125 __dev_set_promiscuity(dev, 1);
4126 dev->uc_promisc = 1;
4127 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4128 __dev_set_promiscuity(dev, -1);
4129 dev->uc_promisc = 0;
4132 if (ops->ndo_set_multicast_list)
4133 ops->ndo_set_multicast_list(dev);
4137 void dev_set_rx_mode(struct net_device *dev)
4139 netif_addr_lock_bh(dev);
4140 __dev_set_rx_mode(dev);
4141 netif_addr_unlock_bh(dev);
4145 * dev_get_flags - get flags reported to userspace
4148 * Get the combination of flag bits exported through APIs to userspace.
4150 unsigned dev_get_flags(const struct net_device *dev)
4154 flags = (dev->flags & ~(IFF_PROMISC |
4159 (dev->gflags & (IFF_PROMISC |
4162 if (netif_running(dev)) {
4163 if (netif_oper_up(dev))
4164 flags |= IFF_RUNNING;
4165 if (netif_carrier_ok(dev))
4166 flags |= IFF_LOWER_UP;
4167 if (netif_dormant(dev))
4168 flags |= IFF_DORMANT;
4173 EXPORT_SYMBOL(dev_get_flags);
4175 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4177 int old_flags = dev->flags;
4183 * Set the flags on our device.
4186 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4187 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4189 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4193 * Load in the correct multicast list now the flags have changed.
4196 if ((old_flags ^ flags) & IFF_MULTICAST)
4197 dev_change_rx_flags(dev, IFF_MULTICAST);
4199 dev_set_rx_mode(dev);
4202 * Have we downed the interface. We handle IFF_UP ourselves
4203 * according to user attempts to set it, rather than blindly
4208 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4209 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4212 dev_set_rx_mode(dev);
4215 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4216 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4218 dev->gflags ^= IFF_PROMISC;
4219 dev_set_promiscuity(dev, inc);
4222 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4223 is important. Some (broken) drivers set IFF_PROMISC, when
4224 IFF_ALLMULTI is requested not asking us and not reporting.
4226 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4227 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4229 dev->gflags ^= IFF_ALLMULTI;
4230 dev_set_allmulti(dev, inc);
4236 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4238 unsigned int changes = dev->flags ^ old_flags;
4240 if (changes & IFF_UP) {
4241 if (dev->flags & IFF_UP)
4242 call_netdevice_notifiers(NETDEV_UP, dev);
4244 call_netdevice_notifiers(NETDEV_DOWN, dev);
4247 if (dev->flags & IFF_UP &&
4248 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4249 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4253 * dev_change_flags - change device settings
4255 * @flags: device state flags
4257 * Change settings on device based state flags. The flags are
4258 * in the userspace exported format.
4260 int dev_change_flags(struct net_device *dev, unsigned flags)
4263 int old_flags = dev->flags;
4265 ret = __dev_change_flags(dev, flags);
4269 changes = old_flags ^ dev->flags;
4271 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4273 __dev_notify_flags(dev, old_flags);
4276 EXPORT_SYMBOL(dev_change_flags);
4279 * dev_set_mtu - Change maximum transfer unit
4281 * @new_mtu: new transfer unit
4283 * Change the maximum transfer size of the network device.
4285 int dev_set_mtu(struct net_device *dev, int new_mtu)
4287 const struct net_device_ops *ops = dev->netdev_ops;
4290 if (new_mtu == dev->mtu)
4293 /* MTU must be positive. */
4297 if (!netif_device_present(dev))
4301 if (ops->ndo_change_mtu)
4302 err = ops->ndo_change_mtu(dev, new_mtu);
4306 if (!err && dev->flags & IFF_UP)
4307 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4310 EXPORT_SYMBOL(dev_set_mtu);
4313 * dev_set_mac_address - Change Media Access Control Address
4317 * Change the hardware (MAC) address of the device
4319 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4321 const struct net_device_ops *ops = dev->netdev_ops;
4324 if (!ops->ndo_set_mac_address)
4326 if (sa->sa_family != dev->type)
4328 if (!netif_device_present(dev))
4330 err = ops->ndo_set_mac_address(dev, sa);
4332 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4335 EXPORT_SYMBOL(dev_set_mac_address);
4338 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4340 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4343 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4349 case SIOCGIFFLAGS: /* Get interface flags */
4350 ifr->ifr_flags = (short) dev_get_flags(dev);
4353 case SIOCGIFMETRIC: /* Get the metric on the interface
4354 (currently unused) */
4355 ifr->ifr_metric = 0;
4358 case SIOCGIFMTU: /* Get the MTU of a device */
4359 ifr->ifr_mtu = dev->mtu;
4364 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4366 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4367 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4368 ifr->ifr_hwaddr.sa_family = dev->type;
4376 ifr->ifr_map.mem_start = dev->mem_start;
4377 ifr->ifr_map.mem_end = dev->mem_end;
4378 ifr->ifr_map.base_addr = dev->base_addr;
4379 ifr->ifr_map.irq = dev->irq;
4380 ifr->ifr_map.dma = dev->dma;
4381 ifr->ifr_map.port = dev->if_port;
4385 ifr->ifr_ifindex = dev->ifindex;
4389 ifr->ifr_qlen = dev->tx_queue_len;
4393 /* dev_ioctl() should ensure this case
4405 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4407 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4410 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4411 const struct net_device_ops *ops;
4416 ops = dev->netdev_ops;
4419 case SIOCSIFFLAGS: /* Set interface flags */
4420 return dev_change_flags(dev, ifr->ifr_flags);
4422 case SIOCSIFMETRIC: /* Set the metric on the interface
4423 (currently unused) */
4426 case SIOCSIFMTU: /* Set the MTU of a device */
4427 return dev_set_mtu(dev, ifr->ifr_mtu);
4430 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4432 case SIOCSIFHWBROADCAST:
4433 if (ifr->ifr_hwaddr.sa_family != dev->type)
4435 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4436 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4437 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4441 if (ops->ndo_set_config) {
4442 if (!netif_device_present(dev))
4444 return ops->ndo_set_config(dev, &ifr->ifr_map);
4449 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4450 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4452 if (!netif_device_present(dev))
4454 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4457 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4458 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4460 if (!netif_device_present(dev))
4462 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4465 if (ifr->ifr_qlen < 0)
4467 dev->tx_queue_len = ifr->ifr_qlen;
4471 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4472 return dev_change_name(dev, ifr->ifr_newname);
4475 * Unknown or private ioctl
4478 if ((cmd >= SIOCDEVPRIVATE &&
4479 cmd <= SIOCDEVPRIVATE + 15) ||
4480 cmd == SIOCBONDENSLAVE ||
4481 cmd == SIOCBONDRELEASE ||
4482 cmd == SIOCBONDSETHWADDR ||
4483 cmd == SIOCBONDSLAVEINFOQUERY ||
4484 cmd == SIOCBONDINFOQUERY ||
4485 cmd == SIOCBONDCHANGEACTIVE ||
4486 cmd == SIOCGMIIPHY ||
4487 cmd == SIOCGMIIREG ||
4488 cmd == SIOCSMIIREG ||
4489 cmd == SIOCBRADDIF ||
4490 cmd == SIOCBRDELIF ||
4491 cmd == SIOCSHWTSTAMP ||
4492 cmd == SIOCWANDEV) {
4494 if (ops->ndo_do_ioctl) {
4495 if (netif_device_present(dev))
4496 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4508 * This function handles all "interface"-type I/O control requests. The actual
4509 * 'doing' part of this is dev_ifsioc above.
4513 * dev_ioctl - network device ioctl
4514 * @net: the applicable net namespace
4515 * @cmd: command to issue
4516 * @arg: pointer to a struct ifreq in user space
4518 * Issue ioctl functions to devices. This is normally called by the
4519 * user space syscall interfaces but can sometimes be useful for
4520 * other purposes. The return value is the return from the syscall if
4521 * positive or a negative errno code on error.
4524 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4530 /* One special case: SIOCGIFCONF takes ifconf argument
4531 and requires shared lock, because it sleeps writing
4535 if (cmd == SIOCGIFCONF) {
4537 ret = dev_ifconf(net, (char __user *) arg);
4541 if (cmd == SIOCGIFNAME)
4542 return dev_ifname(net, (struct ifreq __user *)arg);
4544 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4547 ifr.ifr_name[IFNAMSIZ-1] = 0;
4549 colon = strchr(ifr.ifr_name, ':');
4554 * See which interface the caller is talking about.
4559 * These ioctl calls:
4560 * - can be done by all.
4561 * - atomic and do not require locking.
4572 dev_load(net, ifr.ifr_name);
4574 ret = dev_ifsioc_locked(net, &ifr, cmd);
4579 if (copy_to_user(arg, &ifr,
4580 sizeof(struct ifreq)))
4586 dev_load(net, ifr.ifr_name);
4588 ret = dev_ethtool(net, &ifr);
4593 if (copy_to_user(arg, &ifr,
4594 sizeof(struct ifreq)))
4600 * These ioctl calls:
4601 * - require superuser power.
4602 * - require strict serialization.
4608 if (!capable(CAP_NET_ADMIN))
4610 dev_load(net, ifr.ifr_name);
4612 ret = dev_ifsioc(net, &ifr, cmd);
4617 if (copy_to_user(arg, &ifr,
4618 sizeof(struct ifreq)))
4624 * These ioctl calls:
4625 * - require superuser power.
4626 * - require strict serialization.
4627 * - do not return a value
4637 case SIOCSIFHWBROADCAST:
4640 case SIOCBONDENSLAVE:
4641 case SIOCBONDRELEASE:
4642 case SIOCBONDSETHWADDR:
4643 case SIOCBONDCHANGEACTIVE:
4647 if (!capable(CAP_NET_ADMIN))
4650 case SIOCBONDSLAVEINFOQUERY:
4651 case SIOCBONDINFOQUERY:
4652 dev_load(net, ifr.ifr_name);
4654 ret = dev_ifsioc(net, &ifr, cmd);
4659 /* Get the per device memory space. We can add this but
4660 * currently do not support it */
4662 /* Set the per device memory buffer space.
4663 * Not applicable in our case */
4668 * Unknown or private ioctl.
4671 if (cmd == SIOCWANDEV ||
4672 (cmd >= SIOCDEVPRIVATE &&
4673 cmd <= SIOCDEVPRIVATE + 15)) {
4674 dev_load(net, ifr.ifr_name);
4676 ret = dev_ifsioc(net, &ifr, cmd);
4678 if (!ret && copy_to_user(arg, &ifr,
4679 sizeof(struct ifreq)))
4683 /* Take care of Wireless Extensions */
4684 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4685 return wext_handle_ioctl(net, &ifr, cmd, arg);
4692 * dev_new_index - allocate an ifindex
4693 * @net: the applicable net namespace
4695 * Returns a suitable unique value for a new device interface
4696 * number. The caller must hold the rtnl semaphore or the
4697 * dev_base_lock to be sure it remains unique.
4699 static int dev_new_index(struct net *net)
4705 if (!__dev_get_by_index(net, ifindex))
4710 /* Delayed registration/unregisteration */
4711 static LIST_HEAD(net_todo_list);
4713 static void net_set_todo(struct net_device *dev)
4715 list_add_tail(&dev->todo_list, &net_todo_list);
4718 static void rollback_registered_many(struct list_head *head)
4720 struct net_device *dev, *tmp;
4722 BUG_ON(dev_boot_phase);
4725 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4726 /* Some devices call without registering
4727 * for initialization unwind. Remove those
4728 * devices and proceed with the remaining.
4730 if (dev->reg_state == NETREG_UNINITIALIZED) {
4731 pr_debug("unregister_netdevice: device %s/%p never "
4732 "was registered\n", dev->name, dev);
4735 list_del(&dev->unreg_list);
4739 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4741 /* If device is running, close it first. */
4744 /* And unlink it from device chain. */
4745 unlist_netdevice(dev);
4747 dev->reg_state = NETREG_UNREGISTERING;
4752 list_for_each_entry(dev, head, unreg_list) {
4753 /* Shutdown queueing discipline. */
4757 /* Notify protocols, that we are about to destroy
4758 this device. They should clean all the things.
4760 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4762 if (!dev->rtnl_link_ops ||
4763 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4764 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4767 * Flush the unicast and multicast chains
4772 if (dev->netdev_ops->ndo_uninit)
4773 dev->netdev_ops->ndo_uninit(dev);
4775 /* Notifier chain MUST detach us from master device. */
4776 WARN_ON(dev->master);
4778 /* Remove entries from kobject tree */
4779 netdev_unregister_kobject(dev);
4782 /* Process any work delayed until the end of the batch */
4783 dev = list_first_entry(head, struct net_device, unreg_list);
4784 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4788 list_for_each_entry(dev, head, unreg_list)
4792 static void rollback_registered(struct net_device *dev)
4796 list_add(&dev->unreg_list, &single);
4797 rollback_registered_many(&single);
4800 static void __netdev_init_queue_locks_one(struct net_device *dev,
4801 struct netdev_queue *dev_queue,
4804 spin_lock_init(&dev_queue->_xmit_lock);
4805 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4806 dev_queue->xmit_lock_owner = -1;
4809 static void netdev_init_queue_locks(struct net_device *dev)
4811 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4812 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4815 unsigned long netdev_fix_features(unsigned long features, const char *name)
4817 /* Fix illegal SG+CSUM combinations. */
4818 if ((features & NETIF_F_SG) &&
4819 !(features & NETIF_F_ALL_CSUM)) {
4821 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4822 "checksum feature.\n", name);
4823 features &= ~NETIF_F_SG;
4826 /* TSO requires that SG is present as well. */
4827 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4829 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4830 "SG feature.\n", name);
4831 features &= ~NETIF_F_TSO;
4834 if (features & NETIF_F_UFO) {
4835 if (!(features & NETIF_F_GEN_CSUM)) {
4837 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4838 "since no NETIF_F_HW_CSUM feature.\n",
4840 features &= ~NETIF_F_UFO;
4843 if (!(features & NETIF_F_SG)) {
4845 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4846 "since no NETIF_F_SG feature.\n", name);
4847 features &= ~NETIF_F_UFO;
4853 EXPORT_SYMBOL(netdev_fix_features);
4856 * netif_stacked_transfer_operstate - transfer operstate
4857 * @rootdev: the root or lower level device to transfer state from
4858 * @dev: the device to transfer operstate to
4860 * Transfer operational state from root to device. This is normally
4861 * called when a stacking relationship exists between the root
4862 * device and the device(a leaf device).
4864 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4865 struct net_device *dev)
4867 if (rootdev->operstate == IF_OPER_DORMANT)
4868 netif_dormant_on(dev);
4870 netif_dormant_off(dev);
4872 if (netif_carrier_ok(rootdev)) {
4873 if (!netif_carrier_ok(dev))
4874 netif_carrier_on(dev);
4876 if (netif_carrier_ok(dev))
4877 netif_carrier_off(dev);
4880 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4883 * register_netdevice - register a network device
4884 * @dev: device to register
4886 * Take a completed network device structure and add it to the kernel
4887 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4888 * chain. 0 is returned on success. A negative errno code is returned
4889 * on a failure to set up the device, or if the name is a duplicate.
4891 * Callers must hold the rtnl semaphore. You may want
4892 * register_netdev() instead of this.
4895 * The locking appears insufficient to guarantee two parallel registers
4896 * will not get the same name.
4899 int register_netdevice(struct net_device *dev)
4902 struct net *net = dev_net(dev);
4904 BUG_ON(dev_boot_phase);
4909 /* When net_device's are persistent, this will be fatal. */
4910 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4913 spin_lock_init(&dev->addr_list_lock);
4914 netdev_set_addr_lockdep_class(dev);
4915 netdev_init_queue_locks(dev);
4920 if (!dev->num_rx_queues) {
4922 * Allocate a single RX queue if driver never called
4926 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
4932 dev->_rx->first = dev->_rx;
4933 atomic_set(&dev->_rx->count, 1);
4934 dev->num_rx_queues = 1;
4937 /* Init, if this function is available */
4938 if (dev->netdev_ops->ndo_init) {
4939 ret = dev->netdev_ops->ndo_init(dev);
4947 ret = dev_get_valid_name(net, dev->name, dev->name, 0);
4951 dev->ifindex = dev_new_index(net);
4952 if (dev->iflink == -1)
4953 dev->iflink = dev->ifindex;
4955 /* Fix illegal checksum combinations */
4956 if ((dev->features & NETIF_F_HW_CSUM) &&
4957 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4958 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4960 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4963 if ((dev->features & NETIF_F_NO_CSUM) &&
4964 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4965 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4967 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4970 dev->features = netdev_fix_features(dev->features, dev->name);
4972 /* Enable software GSO if SG is supported. */
4973 if (dev->features & NETIF_F_SG)
4974 dev->features |= NETIF_F_GSO;
4976 netdev_initialize_kobject(dev);
4978 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4979 ret = notifier_to_errno(ret);
4983 ret = netdev_register_kobject(dev);
4986 dev->reg_state = NETREG_REGISTERED;
4989 * Default initial state at registry is that the
4990 * device is present.
4993 set_bit(__LINK_STATE_PRESENT, &dev->state);
4995 dev_init_scheduler(dev);
4997 list_netdevice(dev);
4999 /* Notify protocols, that a new device appeared. */
5000 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5001 ret = notifier_to_errno(ret);
5003 rollback_registered(dev);
5004 dev->reg_state = NETREG_UNREGISTERED;
5007 * Prevent userspace races by waiting until the network
5008 * device is fully setup before sending notifications.
5010 if (!dev->rtnl_link_ops ||
5011 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5012 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5018 if (dev->netdev_ops->ndo_uninit)
5019 dev->netdev_ops->ndo_uninit(dev);
5022 EXPORT_SYMBOL(register_netdevice);
5025 * init_dummy_netdev - init a dummy network device for NAPI
5026 * @dev: device to init
5028 * This takes a network device structure and initialize the minimum
5029 * amount of fields so it can be used to schedule NAPI polls without
5030 * registering a full blown interface. This is to be used by drivers
5031 * that need to tie several hardware interfaces to a single NAPI
5032 * poll scheduler due to HW limitations.
5034 int init_dummy_netdev(struct net_device *dev)
5036 /* Clear everything. Note we don't initialize spinlocks
5037 * are they aren't supposed to be taken by any of the
5038 * NAPI code and this dummy netdev is supposed to be
5039 * only ever used for NAPI polls
5041 memset(dev, 0, sizeof(struct net_device));
5043 /* make sure we BUG if trying to hit standard
5044 * register/unregister code path
5046 dev->reg_state = NETREG_DUMMY;
5048 /* initialize the ref count */
5049 atomic_set(&dev->refcnt, 1);
5051 /* NAPI wants this */
5052 INIT_LIST_HEAD(&dev->napi_list);
5054 /* a dummy interface is started by default */
5055 set_bit(__LINK_STATE_PRESENT, &dev->state);
5056 set_bit(__LINK_STATE_START, &dev->state);
5060 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5064 * register_netdev - register a network device
5065 * @dev: device to register
5067 * Take a completed network device structure and add it to the kernel
5068 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5069 * chain. 0 is returned on success. A negative errno code is returned
5070 * on a failure to set up the device, or if the name is a duplicate.
5072 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5073 * and expands the device name if you passed a format string to
5076 int register_netdev(struct net_device *dev)
5083 * If the name is a format string the caller wants us to do a
5086 if (strchr(dev->name, '%')) {
5087 err = dev_alloc_name(dev, dev->name);
5092 err = register_netdevice(dev);
5097 EXPORT_SYMBOL(register_netdev);
5100 * netdev_wait_allrefs - wait until all references are gone.
5102 * This is called when unregistering network devices.
5104 * Any protocol or device that holds a reference should register
5105 * for netdevice notification, and cleanup and put back the
5106 * reference if they receive an UNREGISTER event.
5107 * We can get stuck here if buggy protocols don't correctly
5110 static void netdev_wait_allrefs(struct net_device *dev)
5112 unsigned long rebroadcast_time, warning_time;
5114 linkwatch_forget_dev(dev);
5116 rebroadcast_time = warning_time = jiffies;
5117 while (atomic_read(&dev->refcnt) != 0) {
5118 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5121 /* Rebroadcast unregister notification */
5122 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5123 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5124 * should have already handle it the first time */
5126 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5128 /* We must not have linkwatch events
5129 * pending on unregister. If this
5130 * happens, we simply run the queue
5131 * unscheduled, resulting in a noop
5134 linkwatch_run_queue();
5139 rebroadcast_time = jiffies;
5144 if (time_after(jiffies, warning_time + 10 * HZ)) {
5145 printk(KERN_EMERG "unregister_netdevice: "
5146 "waiting for %s to become free. Usage "
5148 dev->name, atomic_read(&dev->refcnt));
5149 warning_time = jiffies;
5158 * register_netdevice(x1);
5159 * register_netdevice(x2);
5161 * unregister_netdevice(y1);
5162 * unregister_netdevice(y2);
5168 * We are invoked by rtnl_unlock().
5169 * This allows us to deal with problems:
5170 * 1) We can delete sysfs objects which invoke hotplug
5171 * without deadlocking with linkwatch via keventd.
5172 * 2) Since we run with the RTNL semaphore not held, we can sleep
5173 * safely in order to wait for the netdev refcnt to drop to zero.
5175 * We must not return until all unregister events added during
5176 * the interval the lock was held have been completed.
5178 void netdev_run_todo(void)
5180 struct list_head list;
5182 /* Snapshot list, allow later requests */
5183 list_replace_init(&net_todo_list, &list);
5187 while (!list_empty(&list)) {
5188 struct net_device *dev
5189 = list_first_entry(&list, struct net_device, todo_list);
5190 list_del(&dev->todo_list);
5192 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5193 printk(KERN_ERR "network todo '%s' but state %d\n",
5194 dev->name, dev->reg_state);
5199 dev->reg_state = NETREG_UNREGISTERED;
5201 on_each_cpu(flush_backlog, dev, 1);
5203 netdev_wait_allrefs(dev);
5206 BUG_ON(atomic_read(&dev->refcnt));
5207 WARN_ON(dev->ip_ptr);
5208 WARN_ON(dev->ip6_ptr);
5209 WARN_ON(dev->dn_ptr);
5211 if (dev->destructor)
5212 dev->destructor(dev);
5214 /* Free network device */
5215 kobject_put(&dev->dev.kobj);
5220 * dev_txq_stats_fold - fold tx_queues stats
5221 * @dev: device to get statistics from
5222 * @stats: struct net_device_stats to hold results
5224 void dev_txq_stats_fold(const struct net_device *dev,
5225 struct net_device_stats *stats)
5227 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5229 struct netdev_queue *txq;
5231 for (i = 0; i < dev->num_tx_queues; i++) {
5232 txq = netdev_get_tx_queue(dev, i);
5233 tx_bytes += txq->tx_bytes;
5234 tx_packets += txq->tx_packets;
5235 tx_dropped += txq->tx_dropped;
5237 if (tx_bytes || tx_packets || tx_dropped) {
5238 stats->tx_bytes = tx_bytes;
5239 stats->tx_packets = tx_packets;
5240 stats->tx_dropped = tx_dropped;
5243 EXPORT_SYMBOL(dev_txq_stats_fold);
5246 * dev_get_stats - get network device statistics
5247 * @dev: device to get statistics from
5249 * Get network statistics from device. The device driver may provide
5250 * its own method by setting dev->netdev_ops->get_stats; otherwise
5251 * the internal statistics structure is used.
5253 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5255 const struct net_device_ops *ops = dev->netdev_ops;
5257 if (ops->ndo_get_stats)
5258 return ops->ndo_get_stats(dev);
5260 dev_txq_stats_fold(dev, &dev->stats);
5263 EXPORT_SYMBOL(dev_get_stats);
5265 static void netdev_init_one_queue(struct net_device *dev,
5266 struct netdev_queue *queue,
5272 static void netdev_init_queues(struct net_device *dev)
5274 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5275 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5276 spin_lock_init(&dev->tx_global_lock);
5280 * alloc_netdev_mq - allocate network device
5281 * @sizeof_priv: size of private data to allocate space for
5282 * @name: device name format string
5283 * @setup: callback to initialize device
5284 * @queue_count: the number of subqueues to allocate
5286 * Allocates a struct net_device with private data area for driver use
5287 * and performs basic initialization. Also allocates subquue structs
5288 * for each queue on the device at the end of the netdevice.
5290 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5291 void (*setup)(struct net_device *), unsigned int queue_count)
5293 struct netdev_queue *tx;
5294 struct net_device *dev;
5296 struct net_device *p;
5298 struct netdev_rx_queue *rx;
5302 BUG_ON(strlen(name) >= sizeof(dev->name));
5304 alloc_size = sizeof(struct net_device);
5306 /* ensure 32-byte alignment of private area */
5307 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5308 alloc_size += sizeof_priv;
5310 /* ensure 32-byte alignment of whole construct */
5311 alloc_size += NETDEV_ALIGN - 1;
5313 p = kzalloc(alloc_size, GFP_KERNEL);
5315 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5319 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5321 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5327 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5329 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5334 atomic_set(&rx->count, queue_count);
5337 * Set a pointer to first element in the array which holds the
5340 for (i = 0; i < queue_count; i++)
5344 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5345 dev->padded = (char *)dev - (char *)p;
5347 if (dev_addr_init(dev))
5353 dev_net_set(dev, &init_net);
5356 dev->num_tx_queues = queue_count;
5357 dev->real_num_tx_queues = queue_count;
5361 dev->num_rx_queues = queue_count;
5364 dev->gso_max_size = GSO_MAX_SIZE;
5366 netdev_init_queues(dev);
5368 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5369 dev->ethtool_ntuple_list.count = 0;
5370 INIT_LIST_HEAD(&dev->napi_list);
5371 INIT_LIST_HEAD(&dev->unreg_list);
5372 INIT_LIST_HEAD(&dev->link_watch_list);
5373 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5375 strcpy(dev->name, name);
5388 EXPORT_SYMBOL(alloc_netdev_mq);
5391 * free_netdev - free network device
5394 * This function does the last stage of destroying an allocated device
5395 * interface. The reference to the device object is released.
5396 * If this is the last reference then it will be freed.
5398 void free_netdev(struct net_device *dev)
5400 struct napi_struct *p, *n;
5402 release_net(dev_net(dev));
5406 /* Flush device addresses */
5407 dev_addr_flush(dev);
5409 /* Clear ethtool n-tuple list */
5410 ethtool_ntuple_flush(dev);
5412 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5415 /* Compatibility with error handling in drivers */
5416 if (dev->reg_state == NETREG_UNINITIALIZED) {
5417 kfree((char *)dev - dev->padded);
5421 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5422 dev->reg_state = NETREG_RELEASED;
5424 /* will free via device release */
5425 put_device(&dev->dev);
5427 EXPORT_SYMBOL(free_netdev);
5430 * synchronize_net - Synchronize with packet receive processing
5432 * Wait for packets currently being received to be done.
5433 * Does not block later packets from starting.
5435 void synchronize_net(void)
5440 EXPORT_SYMBOL(synchronize_net);
5443 * unregister_netdevice_queue - remove device from the kernel
5447 * This function shuts down a device interface and removes it
5448 * from the kernel tables.
5449 * If head not NULL, device is queued to be unregistered later.
5451 * Callers must hold the rtnl semaphore. You may want
5452 * unregister_netdev() instead of this.
5455 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5460 list_move_tail(&dev->unreg_list, head);
5462 rollback_registered(dev);
5463 /* Finish processing unregister after unlock */
5467 EXPORT_SYMBOL(unregister_netdevice_queue);
5470 * unregister_netdevice_many - unregister many devices
5471 * @head: list of devices
5473 void unregister_netdevice_many(struct list_head *head)
5475 struct net_device *dev;
5477 if (!list_empty(head)) {
5478 rollback_registered_many(head);
5479 list_for_each_entry(dev, head, unreg_list)
5483 EXPORT_SYMBOL(unregister_netdevice_many);
5486 * unregister_netdev - remove device from the kernel
5489 * This function shuts down a device interface and removes it
5490 * from the kernel tables.
5492 * This is just a wrapper for unregister_netdevice that takes
5493 * the rtnl semaphore. In general you want to use this and not
5494 * unregister_netdevice.
5496 void unregister_netdev(struct net_device *dev)
5499 unregister_netdevice(dev);
5502 EXPORT_SYMBOL(unregister_netdev);
5505 * dev_change_net_namespace - move device to different nethost namespace
5507 * @net: network namespace
5508 * @pat: If not NULL name pattern to try if the current device name
5509 * is already taken in the destination network namespace.
5511 * This function shuts down a device interface and moves it
5512 * to a new network namespace. On success 0 is returned, on
5513 * a failure a netagive errno code is returned.
5515 * Callers must hold the rtnl semaphore.
5518 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5524 /* Don't allow namespace local devices to be moved. */
5526 if (dev->features & NETIF_F_NETNS_LOCAL)
5530 /* Don't allow real devices to be moved when sysfs
5534 if (dev->dev.parent)
5538 /* Ensure the device has been registrered */
5540 if (dev->reg_state != NETREG_REGISTERED)
5543 /* Get out if there is nothing todo */
5545 if (net_eq(dev_net(dev), net))
5548 /* Pick the destination device name, and ensure
5549 * we can use it in the destination network namespace.
5552 if (__dev_get_by_name(net, dev->name)) {
5553 /* We get here if we can't use the current device name */
5556 if (dev_get_valid_name(net, pat, dev->name, 1))
5561 * And now a mini version of register_netdevice unregister_netdevice.
5564 /* If device is running close it first. */
5567 /* And unlink it from device chain */
5569 unlist_netdevice(dev);
5573 /* Shutdown queueing discipline. */
5576 /* Notify protocols, that we are about to destroy
5577 this device. They should clean all the things.
5579 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5580 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5583 * Flush the unicast and multicast chains
5588 netdev_unregister_kobject(dev);
5590 /* Actually switch the network namespace */
5591 dev_net_set(dev, net);
5593 /* If there is an ifindex conflict assign a new one */
5594 if (__dev_get_by_index(net, dev->ifindex)) {
5595 int iflink = (dev->iflink == dev->ifindex);
5596 dev->ifindex = dev_new_index(net);
5598 dev->iflink = dev->ifindex;
5601 /* Fixup kobjects */
5602 err = netdev_register_kobject(dev);
5605 /* Add the device back in the hashes */
5606 list_netdevice(dev);
5608 /* Notify protocols, that a new device appeared. */
5609 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5612 * Prevent userspace races by waiting until the network
5613 * device is fully setup before sending notifications.
5615 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5622 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5624 static int dev_cpu_callback(struct notifier_block *nfb,
5625 unsigned long action,
5628 struct sk_buff **list_skb;
5629 struct sk_buff *skb;
5630 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5631 struct softnet_data *sd, *oldsd;
5633 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5636 local_irq_disable();
5637 cpu = smp_processor_id();
5638 sd = &per_cpu(softnet_data, cpu);
5639 oldsd = &per_cpu(softnet_data, oldcpu);
5641 /* Find end of our completion_queue. */
5642 list_skb = &sd->completion_queue;
5644 list_skb = &(*list_skb)->next;
5645 /* Append completion queue from offline CPU. */
5646 *list_skb = oldsd->completion_queue;
5647 oldsd->completion_queue = NULL;
5649 /* Append output queue from offline CPU. */
5650 if (oldsd->output_queue) {
5651 *sd->output_queue_tailp = oldsd->output_queue;
5652 sd->output_queue_tailp = oldsd->output_queue_tailp;
5653 oldsd->output_queue = NULL;
5654 oldsd->output_queue_tailp = &oldsd->output_queue;
5657 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5660 /* Process offline CPU's input_pkt_queue */
5661 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5663 input_queue_head_add(oldsd, 1);
5665 while ((skb = __skb_dequeue(&oldsd->process_queue)))
5673 * netdev_increment_features - increment feature set by one
5674 * @all: current feature set
5675 * @one: new feature set
5676 * @mask: mask feature set
5678 * Computes a new feature set after adding a device with feature set
5679 * @one to the master device with current feature set @all. Will not
5680 * enable anything that is off in @mask. Returns the new feature set.
5682 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5685 /* If device needs checksumming, downgrade to it. */
5686 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5687 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5688 else if (mask & NETIF_F_ALL_CSUM) {
5689 /* If one device supports v4/v6 checksumming, set for all. */
5690 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5691 !(all & NETIF_F_GEN_CSUM)) {
5692 all &= ~NETIF_F_ALL_CSUM;
5693 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5696 /* If one device supports hw checksumming, set for all. */
5697 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5698 all &= ~NETIF_F_ALL_CSUM;
5699 all |= NETIF_F_HW_CSUM;
5703 one |= NETIF_F_ALL_CSUM;
5705 one |= all & NETIF_F_ONE_FOR_ALL;
5706 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5707 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5711 EXPORT_SYMBOL(netdev_increment_features);
5713 static struct hlist_head *netdev_create_hash(void)
5716 struct hlist_head *hash;
5718 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5720 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5721 INIT_HLIST_HEAD(&hash[i]);
5726 /* Initialize per network namespace state */
5727 static int __net_init netdev_init(struct net *net)
5729 INIT_LIST_HEAD(&net->dev_base_head);
5731 net->dev_name_head = netdev_create_hash();
5732 if (net->dev_name_head == NULL)
5735 net->dev_index_head = netdev_create_hash();
5736 if (net->dev_index_head == NULL)
5742 kfree(net->dev_name_head);
5748 * netdev_drivername - network driver for the device
5749 * @dev: network device
5750 * @buffer: buffer for resulting name
5751 * @len: size of buffer
5753 * Determine network driver for device.
5755 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5757 const struct device_driver *driver;
5758 const struct device *parent;
5760 if (len <= 0 || !buffer)
5764 parent = dev->dev.parent;
5769 driver = parent->driver;
5770 if (driver && driver->name)
5771 strlcpy(buffer, driver->name, len);
5775 static void __net_exit netdev_exit(struct net *net)
5777 kfree(net->dev_name_head);
5778 kfree(net->dev_index_head);
5781 static struct pernet_operations __net_initdata netdev_net_ops = {
5782 .init = netdev_init,
5783 .exit = netdev_exit,
5786 static void __net_exit default_device_exit(struct net *net)
5788 struct net_device *dev, *aux;
5790 * Push all migratable network devices back to the
5791 * initial network namespace
5794 for_each_netdev_safe(net, dev, aux) {
5796 char fb_name[IFNAMSIZ];
5798 /* Ignore unmoveable devices (i.e. loopback) */
5799 if (dev->features & NETIF_F_NETNS_LOCAL)
5802 /* Leave virtual devices for the generic cleanup */
5803 if (dev->rtnl_link_ops)
5806 /* Push remaing network devices to init_net */
5807 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5808 err = dev_change_net_namespace(dev, &init_net, fb_name);
5810 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5811 __func__, dev->name, err);
5818 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5820 /* At exit all network devices most be removed from a network
5821 * namespace. Do this in the reverse order of registeration.
5822 * Do this across as many network namespaces as possible to
5823 * improve batching efficiency.
5825 struct net_device *dev;
5827 LIST_HEAD(dev_kill_list);
5830 list_for_each_entry(net, net_list, exit_list) {
5831 for_each_netdev_reverse(net, dev) {
5832 if (dev->rtnl_link_ops)
5833 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5835 unregister_netdevice_queue(dev, &dev_kill_list);
5838 unregister_netdevice_many(&dev_kill_list);
5842 static struct pernet_operations __net_initdata default_device_ops = {
5843 .exit = default_device_exit,
5844 .exit_batch = default_device_exit_batch,
5848 * Initialize the DEV module. At boot time this walks the device list and
5849 * unhooks any devices that fail to initialise (normally hardware not
5850 * present) and leaves us with a valid list of present and active devices.
5855 * This is called single threaded during boot, so no need
5856 * to take the rtnl semaphore.
5858 static int __init net_dev_init(void)
5860 int i, rc = -ENOMEM;
5862 BUG_ON(!dev_boot_phase);
5864 if (dev_proc_init())
5867 if (netdev_kobject_init())
5870 INIT_LIST_HEAD(&ptype_all);
5871 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5872 INIT_LIST_HEAD(&ptype_base[i]);
5874 if (register_pernet_subsys(&netdev_net_ops))
5878 * Initialise the packet receive queues.
5881 for_each_possible_cpu(i) {
5882 struct softnet_data *sd = &per_cpu(softnet_data, i);
5884 memset(sd, 0, sizeof(*sd));
5885 skb_queue_head_init(&sd->input_pkt_queue);
5886 skb_queue_head_init(&sd->process_queue);
5887 sd->completion_queue = NULL;
5888 INIT_LIST_HEAD(&sd->poll_list);
5889 sd->output_queue = NULL;
5890 sd->output_queue_tailp = &sd->output_queue;
5892 sd->csd.func = rps_trigger_softirq;
5898 sd->backlog.poll = process_backlog;
5899 sd->backlog.weight = weight_p;
5900 sd->backlog.gro_list = NULL;
5901 sd->backlog.gro_count = 0;
5906 /* The loopback device is special if any other network devices
5907 * is present in a network namespace the loopback device must
5908 * be present. Since we now dynamically allocate and free the
5909 * loopback device ensure this invariant is maintained by
5910 * keeping the loopback device as the first device on the
5911 * list of network devices. Ensuring the loopback devices
5912 * is the first device that appears and the last network device
5915 if (register_pernet_device(&loopback_net_ops))
5918 if (register_pernet_device(&default_device_ops))
5921 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5922 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5924 hotcpu_notifier(dev_cpu_callback, 0);
5932 subsys_initcall(net_dev_init);
5934 static int __init initialize_hashrnd(void)
5936 get_random_bytes(&hashrnd, sizeof(hashrnd));
5940 late_initcall_sync(initialize_hashrnd);