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 macvlan_port *p,
2616 struct sk_buff *skb) __read_mostly;
2617 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2619 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2620 struct packet_type **pt_prev,
2622 struct net_device *orig_dev)
2624 struct macvlan_port *port;
2626 port = rcu_dereference(skb->dev->macvlan_port);
2631 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2634 return macvlan_handle_frame_hook(port, skb);
2637 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2640 #ifdef CONFIG_NET_CLS_ACT
2641 /* TODO: Maybe we should just force sch_ingress to be compiled in
2642 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2643 * a compare and 2 stores extra right now if we dont have it on
2644 * but have CONFIG_NET_CLS_ACT
2645 * NOTE: This doesnt stop any functionality; if you dont have
2646 * the ingress scheduler, you just cant add policies on ingress.
2649 static int ing_filter(struct sk_buff *skb)
2651 struct net_device *dev = skb->dev;
2652 u32 ttl = G_TC_RTTL(skb->tc_verd);
2653 struct netdev_queue *rxq;
2654 int result = TC_ACT_OK;
2657 if (MAX_RED_LOOP < ttl++) {
2659 "Redir loop detected Dropping packet (%d->%d)\n",
2660 skb->skb_iif, dev->ifindex);
2664 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2665 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2667 rxq = &dev->rx_queue;
2670 if (q != &noop_qdisc) {
2671 spin_lock(qdisc_lock(q));
2672 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2673 result = qdisc_enqueue_root(skb, q);
2674 spin_unlock(qdisc_lock(q));
2680 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2681 struct packet_type **pt_prev,
2682 int *ret, struct net_device *orig_dev)
2684 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2688 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2691 /* Huh? Why does turning on AF_PACKET affect this? */
2692 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2695 switch (ing_filter(skb)) {
2709 * netif_nit_deliver - deliver received packets to network taps
2712 * This function is used to deliver incoming packets to network
2713 * taps. It should be used when the normal netif_receive_skb path
2714 * is bypassed, for example because of VLAN acceleration.
2716 void netif_nit_deliver(struct sk_buff *skb)
2718 struct packet_type *ptype;
2720 if (list_empty(&ptype_all))
2723 skb_reset_network_header(skb);
2724 skb_reset_transport_header(skb);
2725 skb->mac_len = skb->network_header - skb->mac_header;
2728 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2729 if (!ptype->dev || ptype->dev == skb->dev)
2730 deliver_skb(skb, ptype, skb->dev);
2735 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2736 struct net_device *master)
2738 if (skb->pkt_type == PACKET_HOST) {
2739 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2741 memcpy(dest, master->dev_addr, ETH_ALEN);
2745 /* On bonding slaves other than the currently active slave, suppress
2746 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2747 * ARP on active-backup slaves with arp_validate enabled.
2749 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2751 struct net_device *dev = skb->dev;
2753 if (master->priv_flags & IFF_MASTER_ARPMON)
2754 dev->last_rx = jiffies;
2756 if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) {
2757 /* Do address unmangle. The local destination address
2758 * will be always the one master has. Provides the right
2759 * functionality in a bridge.
2761 skb_bond_set_mac_by_master(skb, master);
2764 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2765 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2766 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2769 if (master->priv_flags & IFF_MASTER_ALB) {
2770 if (skb->pkt_type != PACKET_BROADCAST &&
2771 skb->pkt_type != PACKET_MULTICAST)
2774 if (master->priv_flags & IFF_MASTER_8023AD &&
2775 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2782 EXPORT_SYMBOL(__skb_bond_should_drop);
2784 static int __netif_receive_skb(struct sk_buff *skb)
2786 struct packet_type *ptype, *pt_prev;
2787 struct net_device *orig_dev;
2788 struct net_device *master;
2789 struct net_device *null_or_orig;
2790 struct net_device *null_or_bond;
2791 int ret = NET_RX_DROP;
2794 if (!skb->tstamp.tv64)
2797 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2798 return NET_RX_SUCCESS;
2800 /* if we've gotten here through NAPI, check netpoll */
2801 if (netpoll_receive_skb(skb))
2805 skb->skb_iif = skb->dev->ifindex;
2807 null_or_orig = NULL;
2808 orig_dev = skb->dev;
2809 master = ACCESS_ONCE(orig_dev->master);
2811 if (skb_bond_should_drop(skb, master))
2812 null_or_orig = orig_dev; /* deliver only exact match */
2817 __get_cpu_var(softnet_data).processed++;
2819 skb_reset_network_header(skb);
2820 skb_reset_transport_header(skb);
2821 skb->mac_len = skb->network_header - skb->mac_header;
2827 #ifdef CONFIG_NET_CLS_ACT
2828 if (skb->tc_verd & TC_NCLS) {
2829 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2834 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2835 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2836 ptype->dev == orig_dev) {
2838 ret = deliver_skb(skb, pt_prev, orig_dev);
2843 #ifdef CONFIG_NET_CLS_ACT
2844 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2850 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2853 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2858 * Make sure frames received on VLAN interfaces stacked on
2859 * bonding interfaces still make their way to any base bonding
2860 * device that may have registered for a specific ptype. The
2861 * handler may have to adjust skb->dev and orig_dev.
2863 null_or_bond = NULL;
2864 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2865 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2866 null_or_bond = vlan_dev_real_dev(skb->dev);
2869 type = skb->protocol;
2870 list_for_each_entry_rcu(ptype,
2871 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2872 if (ptype->type == type && (ptype->dev == null_or_orig ||
2873 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2874 ptype->dev == null_or_bond)) {
2876 ret = deliver_skb(skb, pt_prev, orig_dev);
2882 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2885 /* Jamal, now you will not able to escape explaining
2886 * me how you were going to use this. :-)
2897 * netif_receive_skb - process receive buffer from network
2898 * @skb: buffer to process
2900 * netif_receive_skb() is the main receive data processing function.
2901 * It always succeeds. The buffer may be dropped during processing
2902 * for congestion control or by the protocol layers.
2904 * This function may only be called from softirq context and interrupts
2905 * should be enabled.
2907 * Return values (usually ignored):
2908 * NET_RX_SUCCESS: no congestion
2909 * NET_RX_DROP: packet was dropped
2911 int netif_receive_skb(struct sk_buff *skb)
2914 struct rps_dev_flow voidflow, *rflow = &voidflow;
2919 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2922 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2926 ret = __netif_receive_skb(skb);
2931 return __netif_receive_skb(skb);
2934 EXPORT_SYMBOL(netif_receive_skb);
2936 /* Network device is going away, flush any packets still pending
2937 * Called with irqs disabled.
2939 static void flush_backlog(void *arg)
2941 struct net_device *dev = arg;
2942 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2943 struct sk_buff *skb, *tmp;
2946 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
2947 if (skb->dev == dev) {
2948 __skb_unlink(skb, &sd->input_pkt_queue);
2950 input_queue_head_add(sd, 1);
2955 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
2956 if (skb->dev == dev) {
2957 __skb_unlink(skb, &sd->process_queue);
2963 static int napi_gro_complete(struct sk_buff *skb)
2965 struct packet_type *ptype;
2966 __be16 type = skb->protocol;
2967 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2970 if (NAPI_GRO_CB(skb)->count == 1) {
2971 skb_shinfo(skb)->gso_size = 0;
2976 list_for_each_entry_rcu(ptype, head, list) {
2977 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2980 err = ptype->gro_complete(skb);
2986 WARN_ON(&ptype->list == head);
2988 return NET_RX_SUCCESS;
2992 return netif_receive_skb(skb);
2995 static void napi_gro_flush(struct napi_struct *napi)
2997 struct sk_buff *skb, *next;
2999 for (skb = napi->gro_list; skb; skb = next) {
3002 napi_gro_complete(skb);
3005 napi->gro_count = 0;
3006 napi->gro_list = NULL;
3009 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3011 struct sk_buff **pp = NULL;
3012 struct packet_type *ptype;
3013 __be16 type = skb->protocol;
3014 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3017 enum gro_result ret;
3019 if (!(skb->dev->features & NETIF_F_GRO))
3022 if (skb_is_gso(skb) || skb_has_frags(skb))
3026 list_for_each_entry_rcu(ptype, head, list) {
3027 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3030 skb_set_network_header(skb, skb_gro_offset(skb));
3031 mac_len = skb->network_header - skb->mac_header;
3032 skb->mac_len = mac_len;
3033 NAPI_GRO_CB(skb)->same_flow = 0;
3034 NAPI_GRO_CB(skb)->flush = 0;
3035 NAPI_GRO_CB(skb)->free = 0;
3037 pp = ptype->gro_receive(&napi->gro_list, skb);
3042 if (&ptype->list == head)
3045 same_flow = NAPI_GRO_CB(skb)->same_flow;
3046 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3049 struct sk_buff *nskb = *pp;
3053 napi_gro_complete(nskb);
3060 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3064 NAPI_GRO_CB(skb)->count = 1;
3065 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3066 skb->next = napi->gro_list;
3067 napi->gro_list = skb;
3071 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3072 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3074 BUG_ON(skb->end - skb->tail < grow);
3076 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3079 skb->data_len -= grow;
3081 skb_shinfo(skb)->frags[0].page_offset += grow;
3082 skb_shinfo(skb)->frags[0].size -= grow;
3084 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3085 put_page(skb_shinfo(skb)->frags[0].page);
3086 memmove(skb_shinfo(skb)->frags,
3087 skb_shinfo(skb)->frags + 1,
3088 --skb_shinfo(skb)->nr_frags);
3099 EXPORT_SYMBOL(dev_gro_receive);
3102 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3106 if (netpoll_rx_on(skb))
3109 for (p = napi->gro_list; p; p = p->next) {
3110 NAPI_GRO_CB(p)->same_flow =
3111 (p->dev == skb->dev) &&
3112 !compare_ether_header(skb_mac_header(p),
3113 skb_gro_mac_header(skb));
3114 NAPI_GRO_CB(p)->flush = 0;
3117 return dev_gro_receive(napi, skb);
3120 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3124 if (netif_receive_skb(skb))
3129 case GRO_MERGED_FREE:
3140 EXPORT_SYMBOL(napi_skb_finish);
3142 void skb_gro_reset_offset(struct sk_buff *skb)
3144 NAPI_GRO_CB(skb)->data_offset = 0;
3145 NAPI_GRO_CB(skb)->frag0 = NULL;
3146 NAPI_GRO_CB(skb)->frag0_len = 0;
3148 if (skb->mac_header == skb->tail &&
3149 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3150 NAPI_GRO_CB(skb)->frag0 =
3151 page_address(skb_shinfo(skb)->frags[0].page) +
3152 skb_shinfo(skb)->frags[0].page_offset;
3153 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3156 EXPORT_SYMBOL(skb_gro_reset_offset);
3158 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3160 skb_gro_reset_offset(skb);
3162 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3164 EXPORT_SYMBOL(napi_gro_receive);
3166 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3168 __skb_pull(skb, skb_headlen(skb));
3169 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3173 EXPORT_SYMBOL(napi_reuse_skb);
3175 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3177 struct sk_buff *skb = napi->skb;
3180 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3186 EXPORT_SYMBOL(napi_get_frags);
3188 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3194 skb->protocol = eth_type_trans(skb, skb->dev);
3196 if (ret == GRO_HELD)
3197 skb_gro_pull(skb, -ETH_HLEN);
3198 else if (netif_receive_skb(skb))
3203 case GRO_MERGED_FREE:
3204 napi_reuse_skb(napi, skb);
3213 EXPORT_SYMBOL(napi_frags_finish);
3215 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3217 struct sk_buff *skb = napi->skb;
3224 skb_reset_mac_header(skb);
3225 skb_gro_reset_offset(skb);
3227 off = skb_gro_offset(skb);
3228 hlen = off + sizeof(*eth);
3229 eth = skb_gro_header_fast(skb, off);
3230 if (skb_gro_header_hard(skb, hlen)) {
3231 eth = skb_gro_header_slow(skb, hlen, off);
3232 if (unlikely(!eth)) {
3233 napi_reuse_skb(napi, skb);
3239 skb_gro_pull(skb, sizeof(*eth));
3242 * This works because the only protocols we care about don't require
3243 * special handling. We'll fix it up properly at the end.
3245 skb->protocol = eth->h_proto;
3250 EXPORT_SYMBOL(napi_frags_skb);
3252 gro_result_t napi_gro_frags(struct napi_struct *napi)
3254 struct sk_buff *skb = napi_frags_skb(napi);
3259 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3261 EXPORT_SYMBOL(napi_gro_frags);
3264 * net_rps_action sends any pending IPI's for rps.
3265 * Note: called with local irq disabled, but exits with local irq enabled.
3267 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3270 struct softnet_data *remsd = sd->rps_ipi_list;
3273 sd->rps_ipi_list = NULL;
3277 /* Send pending IPI's to kick RPS processing on remote cpus. */
3279 struct softnet_data *next = remsd->rps_ipi_next;
3281 if (cpu_online(remsd->cpu))
3282 __smp_call_function_single(remsd->cpu,
3291 static int process_backlog(struct napi_struct *napi, int quota)
3294 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3297 /* Check if we have pending ipi, its better to send them now,
3298 * not waiting net_rx_action() end.
3300 if (sd->rps_ipi_list) {
3301 local_irq_disable();
3302 net_rps_action_and_irq_enable(sd);
3305 napi->weight = weight_p;
3306 local_irq_disable();
3307 while (work < quota) {
3308 struct sk_buff *skb;
3311 while ((skb = __skb_dequeue(&sd->process_queue))) {
3313 __netif_receive_skb(skb);
3314 if (++work >= quota)
3316 local_irq_disable();
3320 qlen = skb_queue_len(&sd->input_pkt_queue);
3322 input_queue_head_add(sd, qlen);
3323 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3324 &sd->process_queue);
3326 if (qlen < quota - work) {
3328 * Inline a custom version of __napi_complete().
3329 * only current cpu owns and manipulates this napi,
3330 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3331 * we can use a plain write instead of clear_bit(),
3332 * and we dont need an smp_mb() memory barrier.
3334 list_del(&napi->poll_list);
3337 quota = work + qlen;
3347 * __napi_schedule - schedule for receive
3348 * @n: entry to schedule
3350 * The entry's receive function will be scheduled to run
3352 void __napi_schedule(struct napi_struct *n)
3354 unsigned long flags;
3356 local_irq_save(flags);
3357 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3358 local_irq_restore(flags);
3360 EXPORT_SYMBOL(__napi_schedule);
3362 void __napi_complete(struct napi_struct *n)
3364 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3365 BUG_ON(n->gro_list);
3367 list_del(&n->poll_list);
3368 smp_mb__before_clear_bit();
3369 clear_bit(NAPI_STATE_SCHED, &n->state);
3371 EXPORT_SYMBOL(__napi_complete);
3373 void napi_complete(struct napi_struct *n)
3375 unsigned long flags;
3378 * don't let napi dequeue from the cpu poll list
3379 * just in case its running on a different cpu
3381 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3385 local_irq_save(flags);
3387 local_irq_restore(flags);
3389 EXPORT_SYMBOL(napi_complete);
3391 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3392 int (*poll)(struct napi_struct *, int), int weight)
3394 INIT_LIST_HEAD(&napi->poll_list);
3395 napi->gro_count = 0;
3396 napi->gro_list = NULL;
3399 napi->weight = weight;
3400 list_add(&napi->dev_list, &dev->napi_list);
3402 #ifdef CONFIG_NETPOLL
3403 spin_lock_init(&napi->poll_lock);
3404 napi->poll_owner = -1;
3406 set_bit(NAPI_STATE_SCHED, &napi->state);
3408 EXPORT_SYMBOL(netif_napi_add);
3410 void netif_napi_del(struct napi_struct *napi)
3412 struct sk_buff *skb, *next;
3414 list_del_init(&napi->dev_list);
3415 napi_free_frags(napi);
3417 for (skb = napi->gro_list; skb; skb = next) {
3423 napi->gro_list = NULL;
3424 napi->gro_count = 0;
3426 EXPORT_SYMBOL(netif_napi_del);
3428 static void net_rx_action(struct softirq_action *h)
3430 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3431 unsigned long time_limit = jiffies + 2;
3432 int budget = netdev_budget;
3435 local_irq_disable();
3437 while (!list_empty(&sd->poll_list)) {
3438 struct napi_struct *n;
3441 /* If softirq window is exhuasted then punt.
3442 * Allow this to run for 2 jiffies since which will allow
3443 * an average latency of 1.5/HZ.
3445 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3450 /* Even though interrupts have been re-enabled, this
3451 * access is safe because interrupts can only add new
3452 * entries to the tail of this list, and only ->poll()
3453 * calls can remove this head entry from the list.
3455 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3457 have = netpoll_poll_lock(n);
3461 /* This NAPI_STATE_SCHED test is for avoiding a race
3462 * with netpoll's poll_napi(). Only the entity which
3463 * obtains the lock and sees NAPI_STATE_SCHED set will
3464 * actually make the ->poll() call. Therefore we avoid
3465 * accidently calling ->poll() when NAPI is not scheduled.
3468 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3469 work = n->poll(n, weight);
3473 WARN_ON_ONCE(work > weight);
3477 local_irq_disable();
3479 /* Drivers must not modify the NAPI state if they
3480 * consume the entire weight. In such cases this code
3481 * still "owns" the NAPI instance and therefore can
3482 * move the instance around on the list at-will.
3484 if (unlikely(work == weight)) {
3485 if (unlikely(napi_disable_pending(n))) {
3488 local_irq_disable();
3490 list_move_tail(&n->poll_list, &sd->poll_list);
3493 netpoll_poll_unlock(have);
3496 net_rps_action_and_irq_enable(sd);
3498 #ifdef CONFIG_NET_DMA
3500 * There may not be any more sk_buffs coming right now, so push
3501 * any pending DMA copies to hardware
3503 dma_issue_pending_all();
3510 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3514 static gifconf_func_t *gifconf_list[NPROTO];
3517 * register_gifconf - register a SIOCGIF handler
3518 * @family: Address family
3519 * @gifconf: Function handler
3521 * Register protocol dependent address dumping routines. The handler
3522 * that is passed must not be freed or reused until it has been replaced
3523 * by another handler.
3525 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3527 if (family >= NPROTO)
3529 gifconf_list[family] = gifconf;
3532 EXPORT_SYMBOL(register_gifconf);
3536 * Map an interface index to its name (SIOCGIFNAME)
3540 * We need this ioctl for efficient implementation of the
3541 * if_indextoname() function required by the IPv6 API. Without
3542 * it, we would have to search all the interfaces to find a
3546 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3548 struct net_device *dev;
3552 * Fetch the caller's info block.
3555 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3559 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3565 strcpy(ifr.ifr_name, dev->name);
3568 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3574 * Perform a SIOCGIFCONF call. This structure will change
3575 * size eventually, and there is nothing I can do about it.
3576 * Thus we will need a 'compatibility mode'.
3579 static int dev_ifconf(struct net *net, char __user *arg)
3582 struct net_device *dev;
3589 * Fetch the caller's info block.
3592 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3599 * Loop over the interfaces, and write an info block for each.
3603 for_each_netdev(net, dev) {
3604 for (i = 0; i < NPROTO; i++) {
3605 if (gifconf_list[i]) {
3608 done = gifconf_list[i](dev, NULL, 0);
3610 done = gifconf_list[i](dev, pos + total,
3620 * All done. Write the updated control block back to the caller.
3622 ifc.ifc_len = total;
3625 * Both BSD and Solaris return 0 here, so we do too.
3627 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3630 #ifdef CONFIG_PROC_FS
3632 * This is invoked by the /proc filesystem handler to display a device
3635 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3638 struct net *net = seq_file_net(seq);
3640 struct net_device *dev;
3644 return SEQ_START_TOKEN;
3647 for_each_netdev_rcu(net, dev)
3654 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3656 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3657 first_net_device(seq_file_net(seq)) :
3658 next_net_device((struct net_device *)v);
3661 return rcu_dereference(dev);
3664 void dev_seq_stop(struct seq_file *seq, void *v)
3670 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3672 const struct net_device_stats *stats = dev_get_stats(dev);
3674 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3675 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3676 dev->name, stats->rx_bytes, stats->rx_packets,
3678 stats->rx_dropped + stats->rx_missed_errors,
3679 stats->rx_fifo_errors,
3680 stats->rx_length_errors + stats->rx_over_errors +
3681 stats->rx_crc_errors + stats->rx_frame_errors,
3682 stats->rx_compressed, stats->multicast,
3683 stats->tx_bytes, stats->tx_packets,
3684 stats->tx_errors, stats->tx_dropped,
3685 stats->tx_fifo_errors, stats->collisions,
3686 stats->tx_carrier_errors +
3687 stats->tx_aborted_errors +
3688 stats->tx_window_errors +
3689 stats->tx_heartbeat_errors,
3690 stats->tx_compressed);
3694 * Called from the PROCfs module. This now uses the new arbitrary sized
3695 * /proc/net interface to create /proc/net/dev
3697 static int dev_seq_show(struct seq_file *seq, void *v)
3699 if (v == SEQ_START_TOKEN)
3700 seq_puts(seq, "Inter-| Receive "
3702 " face |bytes packets errs drop fifo frame "
3703 "compressed multicast|bytes packets errs "
3704 "drop fifo colls carrier compressed\n");
3706 dev_seq_printf_stats(seq, v);
3710 static struct softnet_data *softnet_get_online(loff_t *pos)
3712 struct softnet_data *sd = NULL;
3714 while (*pos < nr_cpu_ids)
3715 if (cpu_online(*pos)) {
3716 sd = &per_cpu(softnet_data, *pos);
3723 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3725 return softnet_get_online(pos);
3728 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3731 return softnet_get_online(pos);
3734 static void softnet_seq_stop(struct seq_file *seq, void *v)
3738 static int softnet_seq_show(struct seq_file *seq, void *v)
3740 struct softnet_data *sd = v;
3742 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3743 sd->processed, sd->dropped, sd->time_squeeze, 0,
3744 0, 0, 0, 0, /* was fastroute */
3745 sd->cpu_collision, sd->received_rps);
3749 static const struct seq_operations dev_seq_ops = {
3750 .start = dev_seq_start,
3751 .next = dev_seq_next,
3752 .stop = dev_seq_stop,
3753 .show = dev_seq_show,
3756 static int dev_seq_open(struct inode *inode, struct file *file)
3758 return seq_open_net(inode, file, &dev_seq_ops,
3759 sizeof(struct seq_net_private));
3762 static const struct file_operations dev_seq_fops = {
3763 .owner = THIS_MODULE,
3764 .open = dev_seq_open,
3766 .llseek = seq_lseek,
3767 .release = seq_release_net,
3770 static const struct seq_operations softnet_seq_ops = {
3771 .start = softnet_seq_start,
3772 .next = softnet_seq_next,
3773 .stop = softnet_seq_stop,
3774 .show = softnet_seq_show,
3777 static int softnet_seq_open(struct inode *inode, struct file *file)
3779 return seq_open(file, &softnet_seq_ops);
3782 static const struct file_operations softnet_seq_fops = {
3783 .owner = THIS_MODULE,
3784 .open = softnet_seq_open,
3786 .llseek = seq_lseek,
3787 .release = seq_release,
3790 static void *ptype_get_idx(loff_t pos)
3792 struct packet_type *pt = NULL;
3796 list_for_each_entry_rcu(pt, &ptype_all, list) {
3802 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3803 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3812 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3816 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3819 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3821 struct packet_type *pt;
3822 struct list_head *nxt;
3826 if (v == SEQ_START_TOKEN)
3827 return ptype_get_idx(0);
3830 nxt = pt->list.next;
3831 if (pt->type == htons(ETH_P_ALL)) {
3832 if (nxt != &ptype_all)
3835 nxt = ptype_base[0].next;
3837 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3839 while (nxt == &ptype_base[hash]) {
3840 if (++hash >= PTYPE_HASH_SIZE)
3842 nxt = ptype_base[hash].next;
3845 return list_entry(nxt, struct packet_type, list);
3848 static void ptype_seq_stop(struct seq_file *seq, void *v)
3854 static int ptype_seq_show(struct seq_file *seq, void *v)
3856 struct packet_type *pt = v;
3858 if (v == SEQ_START_TOKEN)
3859 seq_puts(seq, "Type Device Function\n");
3860 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3861 if (pt->type == htons(ETH_P_ALL))
3862 seq_puts(seq, "ALL ");
3864 seq_printf(seq, "%04x", ntohs(pt->type));
3866 seq_printf(seq, " %-8s %pF\n",
3867 pt->dev ? pt->dev->name : "", pt->func);
3873 static const struct seq_operations ptype_seq_ops = {
3874 .start = ptype_seq_start,
3875 .next = ptype_seq_next,
3876 .stop = ptype_seq_stop,
3877 .show = ptype_seq_show,
3880 static int ptype_seq_open(struct inode *inode, struct file *file)
3882 return seq_open_net(inode, file, &ptype_seq_ops,
3883 sizeof(struct seq_net_private));
3886 static const struct file_operations ptype_seq_fops = {
3887 .owner = THIS_MODULE,
3888 .open = ptype_seq_open,
3890 .llseek = seq_lseek,
3891 .release = seq_release_net,
3895 static int __net_init dev_proc_net_init(struct net *net)
3899 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3901 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3903 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3906 if (wext_proc_init(net))
3912 proc_net_remove(net, "ptype");
3914 proc_net_remove(net, "softnet_stat");
3916 proc_net_remove(net, "dev");
3920 static void __net_exit dev_proc_net_exit(struct net *net)
3922 wext_proc_exit(net);
3924 proc_net_remove(net, "ptype");
3925 proc_net_remove(net, "softnet_stat");
3926 proc_net_remove(net, "dev");
3929 static struct pernet_operations __net_initdata dev_proc_ops = {
3930 .init = dev_proc_net_init,
3931 .exit = dev_proc_net_exit,
3934 static int __init dev_proc_init(void)
3936 return register_pernet_subsys(&dev_proc_ops);
3939 #define dev_proc_init() 0
3940 #endif /* CONFIG_PROC_FS */
3944 * netdev_set_master - set up master/slave pair
3945 * @slave: slave device
3946 * @master: new master device
3948 * Changes the master device of the slave. Pass %NULL to break the
3949 * bonding. The caller must hold the RTNL semaphore. On a failure
3950 * a negative errno code is returned. On success the reference counts
3951 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3952 * function returns zero.
3954 int netdev_set_master(struct net_device *slave, struct net_device *master)
3956 struct net_device *old = slave->master;
3966 slave->master = master;
3973 slave->flags |= IFF_SLAVE;
3975 slave->flags &= ~IFF_SLAVE;
3977 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3980 EXPORT_SYMBOL(netdev_set_master);
3982 static void dev_change_rx_flags(struct net_device *dev, int flags)
3984 const struct net_device_ops *ops = dev->netdev_ops;
3986 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3987 ops->ndo_change_rx_flags(dev, flags);
3990 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3992 unsigned short old_flags = dev->flags;
3998 dev->flags |= IFF_PROMISC;
3999 dev->promiscuity += inc;
4000 if (dev->promiscuity == 0) {
4003 * If inc causes overflow, untouch promisc and return error.
4006 dev->flags &= ~IFF_PROMISC;
4008 dev->promiscuity -= inc;
4009 printk(KERN_WARNING "%s: promiscuity touches roof, "
4010 "set promiscuity failed, promiscuity feature "
4011 "of device might be broken.\n", dev->name);
4015 if (dev->flags != old_flags) {
4016 printk(KERN_INFO "device %s %s promiscuous mode\n",
4017 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4019 if (audit_enabled) {
4020 current_uid_gid(&uid, &gid);
4021 audit_log(current->audit_context, GFP_ATOMIC,
4022 AUDIT_ANOM_PROMISCUOUS,
4023 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4024 dev->name, (dev->flags & IFF_PROMISC),
4025 (old_flags & IFF_PROMISC),
4026 audit_get_loginuid(current),
4028 audit_get_sessionid(current));
4031 dev_change_rx_flags(dev, IFF_PROMISC);
4037 * dev_set_promiscuity - update promiscuity count on a device
4041 * Add or remove promiscuity from a device. While the count in the device
4042 * remains above zero the interface remains promiscuous. Once it hits zero
4043 * the device reverts back to normal filtering operation. A negative inc
4044 * value is used to drop promiscuity on the device.
4045 * Return 0 if successful or a negative errno code on error.
4047 int dev_set_promiscuity(struct net_device *dev, int inc)
4049 unsigned short old_flags = dev->flags;
4052 err = __dev_set_promiscuity(dev, inc);
4055 if (dev->flags != old_flags)
4056 dev_set_rx_mode(dev);
4059 EXPORT_SYMBOL(dev_set_promiscuity);
4062 * dev_set_allmulti - update allmulti count on a device
4066 * Add or remove reception of all multicast frames to a device. While the
4067 * count in the device remains above zero the interface remains listening
4068 * to all interfaces. Once it hits zero the device reverts back to normal
4069 * filtering operation. A negative @inc value is used to drop the counter
4070 * when releasing a resource needing all multicasts.
4071 * Return 0 if successful or a negative errno code on error.
4074 int dev_set_allmulti(struct net_device *dev, int inc)
4076 unsigned short old_flags = dev->flags;
4080 dev->flags |= IFF_ALLMULTI;
4081 dev->allmulti += inc;
4082 if (dev->allmulti == 0) {
4085 * If inc causes overflow, untouch allmulti and return error.
4088 dev->flags &= ~IFF_ALLMULTI;
4090 dev->allmulti -= inc;
4091 printk(KERN_WARNING "%s: allmulti touches roof, "
4092 "set allmulti failed, allmulti feature of "
4093 "device might be broken.\n", dev->name);
4097 if (dev->flags ^ old_flags) {
4098 dev_change_rx_flags(dev, IFF_ALLMULTI);
4099 dev_set_rx_mode(dev);
4103 EXPORT_SYMBOL(dev_set_allmulti);
4106 * Upload unicast and multicast address lists to device and
4107 * configure RX filtering. When the device doesn't support unicast
4108 * filtering it is put in promiscuous mode while unicast addresses
4111 void __dev_set_rx_mode(struct net_device *dev)
4113 const struct net_device_ops *ops = dev->netdev_ops;
4115 /* dev_open will call this function so the list will stay sane. */
4116 if (!(dev->flags&IFF_UP))
4119 if (!netif_device_present(dev))
4122 if (ops->ndo_set_rx_mode)
4123 ops->ndo_set_rx_mode(dev);
4125 /* Unicast addresses changes may only happen under the rtnl,
4126 * therefore calling __dev_set_promiscuity here is safe.
4128 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4129 __dev_set_promiscuity(dev, 1);
4130 dev->uc_promisc = 1;
4131 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4132 __dev_set_promiscuity(dev, -1);
4133 dev->uc_promisc = 0;
4136 if (ops->ndo_set_multicast_list)
4137 ops->ndo_set_multicast_list(dev);
4141 void dev_set_rx_mode(struct net_device *dev)
4143 netif_addr_lock_bh(dev);
4144 __dev_set_rx_mode(dev);
4145 netif_addr_unlock_bh(dev);
4149 * dev_get_flags - get flags reported to userspace
4152 * Get the combination of flag bits exported through APIs to userspace.
4154 unsigned dev_get_flags(const struct net_device *dev)
4158 flags = (dev->flags & ~(IFF_PROMISC |
4163 (dev->gflags & (IFF_PROMISC |
4166 if (netif_running(dev)) {
4167 if (netif_oper_up(dev))
4168 flags |= IFF_RUNNING;
4169 if (netif_carrier_ok(dev))
4170 flags |= IFF_LOWER_UP;
4171 if (netif_dormant(dev))
4172 flags |= IFF_DORMANT;
4177 EXPORT_SYMBOL(dev_get_flags);
4179 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4181 int old_flags = dev->flags;
4187 * Set the flags on our device.
4190 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4191 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4193 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4197 * Load in the correct multicast list now the flags have changed.
4200 if ((old_flags ^ flags) & IFF_MULTICAST)
4201 dev_change_rx_flags(dev, IFF_MULTICAST);
4203 dev_set_rx_mode(dev);
4206 * Have we downed the interface. We handle IFF_UP ourselves
4207 * according to user attempts to set it, rather than blindly
4212 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4213 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4216 dev_set_rx_mode(dev);
4219 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4220 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4222 dev->gflags ^= IFF_PROMISC;
4223 dev_set_promiscuity(dev, inc);
4226 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4227 is important. Some (broken) drivers set IFF_PROMISC, when
4228 IFF_ALLMULTI is requested not asking us and not reporting.
4230 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4231 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4233 dev->gflags ^= IFF_ALLMULTI;
4234 dev_set_allmulti(dev, inc);
4240 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4242 unsigned int changes = dev->flags ^ old_flags;
4244 if (changes & IFF_UP) {
4245 if (dev->flags & IFF_UP)
4246 call_netdevice_notifiers(NETDEV_UP, dev);
4248 call_netdevice_notifiers(NETDEV_DOWN, dev);
4251 if (dev->flags & IFF_UP &&
4252 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4253 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4257 * dev_change_flags - change device settings
4259 * @flags: device state flags
4261 * Change settings on device based state flags. The flags are
4262 * in the userspace exported format.
4264 int dev_change_flags(struct net_device *dev, unsigned flags)
4267 int old_flags = dev->flags;
4269 ret = __dev_change_flags(dev, flags);
4273 changes = old_flags ^ dev->flags;
4275 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4277 __dev_notify_flags(dev, old_flags);
4280 EXPORT_SYMBOL(dev_change_flags);
4283 * dev_set_mtu - Change maximum transfer unit
4285 * @new_mtu: new transfer unit
4287 * Change the maximum transfer size of the network device.
4289 int dev_set_mtu(struct net_device *dev, int new_mtu)
4291 const struct net_device_ops *ops = dev->netdev_ops;
4294 if (new_mtu == dev->mtu)
4297 /* MTU must be positive. */
4301 if (!netif_device_present(dev))
4305 if (ops->ndo_change_mtu)
4306 err = ops->ndo_change_mtu(dev, new_mtu);
4310 if (!err && dev->flags & IFF_UP)
4311 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4314 EXPORT_SYMBOL(dev_set_mtu);
4317 * dev_set_mac_address - Change Media Access Control Address
4321 * Change the hardware (MAC) address of the device
4323 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4325 const struct net_device_ops *ops = dev->netdev_ops;
4328 if (!ops->ndo_set_mac_address)
4330 if (sa->sa_family != dev->type)
4332 if (!netif_device_present(dev))
4334 err = ops->ndo_set_mac_address(dev, sa);
4336 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4339 EXPORT_SYMBOL(dev_set_mac_address);
4342 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4344 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4347 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4353 case SIOCGIFFLAGS: /* Get interface flags */
4354 ifr->ifr_flags = (short) dev_get_flags(dev);
4357 case SIOCGIFMETRIC: /* Get the metric on the interface
4358 (currently unused) */
4359 ifr->ifr_metric = 0;
4362 case SIOCGIFMTU: /* Get the MTU of a device */
4363 ifr->ifr_mtu = dev->mtu;
4368 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4370 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4371 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4372 ifr->ifr_hwaddr.sa_family = dev->type;
4380 ifr->ifr_map.mem_start = dev->mem_start;
4381 ifr->ifr_map.mem_end = dev->mem_end;
4382 ifr->ifr_map.base_addr = dev->base_addr;
4383 ifr->ifr_map.irq = dev->irq;
4384 ifr->ifr_map.dma = dev->dma;
4385 ifr->ifr_map.port = dev->if_port;
4389 ifr->ifr_ifindex = dev->ifindex;
4393 ifr->ifr_qlen = dev->tx_queue_len;
4397 /* dev_ioctl() should ensure this case
4409 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4411 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4414 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4415 const struct net_device_ops *ops;
4420 ops = dev->netdev_ops;
4423 case SIOCSIFFLAGS: /* Set interface flags */
4424 return dev_change_flags(dev, ifr->ifr_flags);
4426 case SIOCSIFMETRIC: /* Set the metric on the interface
4427 (currently unused) */
4430 case SIOCSIFMTU: /* Set the MTU of a device */
4431 return dev_set_mtu(dev, ifr->ifr_mtu);
4434 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4436 case SIOCSIFHWBROADCAST:
4437 if (ifr->ifr_hwaddr.sa_family != dev->type)
4439 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4440 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4441 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4445 if (ops->ndo_set_config) {
4446 if (!netif_device_present(dev))
4448 return ops->ndo_set_config(dev, &ifr->ifr_map);
4453 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4454 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4456 if (!netif_device_present(dev))
4458 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4461 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4462 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4464 if (!netif_device_present(dev))
4466 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4469 if (ifr->ifr_qlen < 0)
4471 dev->tx_queue_len = ifr->ifr_qlen;
4475 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4476 return dev_change_name(dev, ifr->ifr_newname);
4479 * Unknown or private ioctl
4482 if ((cmd >= SIOCDEVPRIVATE &&
4483 cmd <= SIOCDEVPRIVATE + 15) ||
4484 cmd == SIOCBONDENSLAVE ||
4485 cmd == SIOCBONDRELEASE ||
4486 cmd == SIOCBONDSETHWADDR ||
4487 cmd == SIOCBONDSLAVEINFOQUERY ||
4488 cmd == SIOCBONDINFOQUERY ||
4489 cmd == SIOCBONDCHANGEACTIVE ||
4490 cmd == SIOCGMIIPHY ||
4491 cmd == SIOCGMIIREG ||
4492 cmd == SIOCSMIIREG ||
4493 cmd == SIOCBRADDIF ||
4494 cmd == SIOCBRDELIF ||
4495 cmd == SIOCSHWTSTAMP ||
4496 cmd == SIOCWANDEV) {
4498 if (ops->ndo_do_ioctl) {
4499 if (netif_device_present(dev))
4500 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4512 * This function handles all "interface"-type I/O control requests. The actual
4513 * 'doing' part of this is dev_ifsioc above.
4517 * dev_ioctl - network device ioctl
4518 * @net: the applicable net namespace
4519 * @cmd: command to issue
4520 * @arg: pointer to a struct ifreq in user space
4522 * Issue ioctl functions to devices. This is normally called by the
4523 * user space syscall interfaces but can sometimes be useful for
4524 * other purposes. The return value is the return from the syscall if
4525 * positive or a negative errno code on error.
4528 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4534 /* One special case: SIOCGIFCONF takes ifconf argument
4535 and requires shared lock, because it sleeps writing
4539 if (cmd == SIOCGIFCONF) {
4541 ret = dev_ifconf(net, (char __user *) arg);
4545 if (cmd == SIOCGIFNAME)
4546 return dev_ifname(net, (struct ifreq __user *)arg);
4548 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4551 ifr.ifr_name[IFNAMSIZ-1] = 0;
4553 colon = strchr(ifr.ifr_name, ':');
4558 * See which interface the caller is talking about.
4563 * These ioctl calls:
4564 * - can be done by all.
4565 * - atomic and do not require locking.
4576 dev_load(net, ifr.ifr_name);
4578 ret = dev_ifsioc_locked(net, &ifr, cmd);
4583 if (copy_to_user(arg, &ifr,
4584 sizeof(struct ifreq)))
4590 dev_load(net, ifr.ifr_name);
4592 ret = dev_ethtool(net, &ifr);
4597 if (copy_to_user(arg, &ifr,
4598 sizeof(struct ifreq)))
4604 * These ioctl calls:
4605 * - require superuser power.
4606 * - require strict serialization.
4612 if (!capable(CAP_NET_ADMIN))
4614 dev_load(net, ifr.ifr_name);
4616 ret = dev_ifsioc(net, &ifr, cmd);
4621 if (copy_to_user(arg, &ifr,
4622 sizeof(struct ifreq)))
4628 * These ioctl calls:
4629 * - require superuser power.
4630 * - require strict serialization.
4631 * - do not return a value
4641 case SIOCSIFHWBROADCAST:
4644 case SIOCBONDENSLAVE:
4645 case SIOCBONDRELEASE:
4646 case SIOCBONDSETHWADDR:
4647 case SIOCBONDCHANGEACTIVE:
4651 if (!capable(CAP_NET_ADMIN))
4654 case SIOCBONDSLAVEINFOQUERY:
4655 case SIOCBONDINFOQUERY:
4656 dev_load(net, ifr.ifr_name);
4658 ret = dev_ifsioc(net, &ifr, cmd);
4663 /* Get the per device memory space. We can add this but
4664 * currently do not support it */
4666 /* Set the per device memory buffer space.
4667 * Not applicable in our case */
4672 * Unknown or private ioctl.
4675 if (cmd == SIOCWANDEV ||
4676 (cmd >= SIOCDEVPRIVATE &&
4677 cmd <= SIOCDEVPRIVATE + 15)) {
4678 dev_load(net, ifr.ifr_name);
4680 ret = dev_ifsioc(net, &ifr, cmd);
4682 if (!ret && copy_to_user(arg, &ifr,
4683 sizeof(struct ifreq)))
4687 /* Take care of Wireless Extensions */
4688 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4689 return wext_handle_ioctl(net, &ifr, cmd, arg);
4696 * dev_new_index - allocate an ifindex
4697 * @net: the applicable net namespace
4699 * Returns a suitable unique value for a new device interface
4700 * number. The caller must hold the rtnl semaphore or the
4701 * dev_base_lock to be sure it remains unique.
4703 static int dev_new_index(struct net *net)
4709 if (!__dev_get_by_index(net, ifindex))
4714 /* Delayed registration/unregisteration */
4715 static LIST_HEAD(net_todo_list);
4717 static void net_set_todo(struct net_device *dev)
4719 list_add_tail(&dev->todo_list, &net_todo_list);
4722 static void rollback_registered_many(struct list_head *head)
4724 struct net_device *dev, *tmp;
4726 BUG_ON(dev_boot_phase);
4729 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4730 /* Some devices call without registering
4731 * for initialization unwind. Remove those
4732 * devices and proceed with the remaining.
4734 if (dev->reg_state == NETREG_UNINITIALIZED) {
4735 pr_debug("unregister_netdevice: device %s/%p never "
4736 "was registered\n", dev->name, dev);
4739 list_del(&dev->unreg_list);
4743 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4745 /* If device is running, close it first. */
4748 /* And unlink it from device chain. */
4749 unlist_netdevice(dev);
4751 dev->reg_state = NETREG_UNREGISTERING;
4756 list_for_each_entry(dev, head, unreg_list) {
4757 /* Shutdown queueing discipline. */
4761 /* Notify protocols, that we are about to destroy
4762 this device. They should clean all the things.
4764 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4766 if (!dev->rtnl_link_ops ||
4767 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4768 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4771 * Flush the unicast and multicast chains
4776 if (dev->netdev_ops->ndo_uninit)
4777 dev->netdev_ops->ndo_uninit(dev);
4779 /* Notifier chain MUST detach us from master device. */
4780 WARN_ON(dev->master);
4782 /* Remove entries from kobject tree */
4783 netdev_unregister_kobject(dev);
4786 /* Process any work delayed until the end of the batch */
4787 dev = list_first_entry(head, struct net_device, unreg_list);
4788 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4792 list_for_each_entry(dev, head, unreg_list)
4796 static void rollback_registered(struct net_device *dev)
4800 list_add(&dev->unreg_list, &single);
4801 rollback_registered_many(&single);
4804 static void __netdev_init_queue_locks_one(struct net_device *dev,
4805 struct netdev_queue *dev_queue,
4808 spin_lock_init(&dev_queue->_xmit_lock);
4809 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4810 dev_queue->xmit_lock_owner = -1;
4813 static void netdev_init_queue_locks(struct net_device *dev)
4815 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4816 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4819 unsigned long netdev_fix_features(unsigned long features, const char *name)
4821 /* Fix illegal SG+CSUM combinations. */
4822 if ((features & NETIF_F_SG) &&
4823 !(features & NETIF_F_ALL_CSUM)) {
4825 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4826 "checksum feature.\n", name);
4827 features &= ~NETIF_F_SG;
4830 /* TSO requires that SG is present as well. */
4831 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4833 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4834 "SG feature.\n", name);
4835 features &= ~NETIF_F_TSO;
4838 if (features & NETIF_F_UFO) {
4839 if (!(features & NETIF_F_GEN_CSUM)) {
4841 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4842 "since no NETIF_F_HW_CSUM feature.\n",
4844 features &= ~NETIF_F_UFO;
4847 if (!(features & NETIF_F_SG)) {
4849 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4850 "since no NETIF_F_SG feature.\n", name);
4851 features &= ~NETIF_F_UFO;
4857 EXPORT_SYMBOL(netdev_fix_features);
4860 * netif_stacked_transfer_operstate - transfer operstate
4861 * @rootdev: the root or lower level device to transfer state from
4862 * @dev: the device to transfer operstate to
4864 * Transfer operational state from root to device. This is normally
4865 * called when a stacking relationship exists between the root
4866 * device and the device(a leaf device).
4868 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4869 struct net_device *dev)
4871 if (rootdev->operstate == IF_OPER_DORMANT)
4872 netif_dormant_on(dev);
4874 netif_dormant_off(dev);
4876 if (netif_carrier_ok(rootdev)) {
4877 if (!netif_carrier_ok(dev))
4878 netif_carrier_on(dev);
4880 if (netif_carrier_ok(dev))
4881 netif_carrier_off(dev);
4884 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4887 * register_netdevice - register a network device
4888 * @dev: device to register
4890 * Take a completed network device structure and add it to the kernel
4891 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4892 * chain. 0 is returned on success. A negative errno code is returned
4893 * on a failure to set up the device, or if the name is a duplicate.
4895 * Callers must hold the rtnl semaphore. You may want
4896 * register_netdev() instead of this.
4899 * The locking appears insufficient to guarantee two parallel registers
4900 * will not get the same name.
4903 int register_netdevice(struct net_device *dev)
4906 struct net *net = dev_net(dev);
4908 BUG_ON(dev_boot_phase);
4913 /* When net_device's are persistent, this will be fatal. */
4914 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4917 spin_lock_init(&dev->addr_list_lock);
4918 netdev_set_addr_lockdep_class(dev);
4919 netdev_init_queue_locks(dev);
4924 if (!dev->num_rx_queues) {
4926 * Allocate a single RX queue if driver never called
4930 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
4936 dev->_rx->first = dev->_rx;
4937 atomic_set(&dev->_rx->count, 1);
4938 dev->num_rx_queues = 1;
4941 /* Init, if this function is available */
4942 if (dev->netdev_ops->ndo_init) {
4943 ret = dev->netdev_ops->ndo_init(dev);
4951 ret = dev_get_valid_name(net, dev->name, dev->name, 0);
4955 dev->ifindex = dev_new_index(net);
4956 if (dev->iflink == -1)
4957 dev->iflink = dev->ifindex;
4959 /* Fix illegal checksum combinations */
4960 if ((dev->features & NETIF_F_HW_CSUM) &&
4961 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4962 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4964 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4967 if ((dev->features & NETIF_F_NO_CSUM) &&
4968 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4969 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4971 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4974 dev->features = netdev_fix_features(dev->features, dev->name);
4976 /* Enable software GSO if SG is supported. */
4977 if (dev->features & NETIF_F_SG)
4978 dev->features |= NETIF_F_GSO;
4980 netdev_initialize_kobject(dev);
4982 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4983 ret = notifier_to_errno(ret);
4987 ret = netdev_register_kobject(dev);
4990 dev->reg_state = NETREG_REGISTERED;
4993 * Default initial state at registry is that the
4994 * device is present.
4997 set_bit(__LINK_STATE_PRESENT, &dev->state);
4999 dev_init_scheduler(dev);
5001 list_netdevice(dev);
5003 /* Notify protocols, that a new device appeared. */
5004 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5005 ret = notifier_to_errno(ret);
5007 rollback_registered(dev);
5008 dev->reg_state = NETREG_UNREGISTERED;
5011 * Prevent userspace races by waiting until the network
5012 * device is fully setup before sending notifications.
5014 if (!dev->rtnl_link_ops ||
5015 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5016 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5022 if (dev->netdev_ops->ndo_uninit)
5023 dev->netdev_ops->ndo_uninit(dev);
5026 EXPORT_SYMBOL(register_netdevice);
5029 * init_dummy_netdev - init a dummy network device for NAPI
5030 * @dev: device to init
5032 * This takes a network device structure and initialize the minimum
5033 * amount of fields so it can be used to schedule NAPI polls without
5034 * registering a full blown interface. This is to be used by drivers
5035 * that need to tie several hardware interfaces to a single NAPI
5036 * poll scheduler due to HW limitations.
5038 int init_dummy_netdev(struct net_device *dev)
5040 /* Clear everything. Note we don't initialize spinlocks
5041 * are they aren't supposed to be taken by any of the
5042 * NAPI code and this dummy netdev is supposed to be
5043 * only ever used for NAPI polls
5045 memset(dev, 0, sizeof(struct net_device));
5047 /* make sure we BUG if trying to hit standard
5048 * register/unregister code path
5050 dev->reg_state = NETREG_DUMMY;
5052 /* initialize the ref count */
5053 atomic_set(&dev->refcnt, 1);
5055 /* NAPI wants this */
5056 INIT_LIST_HEAD(&dev->napi_list);
5058 /* a dummy interface is started by default */
5059 set_bit(__LINK_STATE_PRESENT, &dev->state);
5060 set_bit(__LINK_STATE_START, &dev->state);
5064 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5068 * register_netdev - register a network device
5069 * @dev: device to register
5071 * Take a completed network device structure and add it to the kernel
5072 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5073 * chain. 0 is returned on success. A negative errno code is returned
5074 * on a failure to set up the device, or if the name is a duplicate.
5076 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5077 * and expands the device name if you passed a format string to
5080 int register_netdev(struct net_device *dev)
5087 * If the name is a format string the caller wants us to do a
5090 if (strchr(dev->name, '%')) {
5091 err = dev_alloc_name(dev, dev->name);
5096 err = register_netdevice(dev);
5101 EXPORT_SYMBOL(register_netdev);
5104 * netdev_wait_allrefs - wait until all references are gone.
5106 * This is called when unregistering network devices.
5108 * Any protocol or device that holds a reference should register
5109 * for netdevice notification, and cleanup and put back the
5110 * reference if they receive an UNREGISTER event.
5111 * We can get stuck here if buggy protocols don't correctly
5114 static void netdev_wait_allrefs(struct net_device *dev)
5116 unsigned long rebroadcast_time, warning_time;
5118 linkwatch_forget_dev(dev);
5120 rebroadcast_time = warning_time = jiffies;
5121 while (atomic_read(&dev->refcnt) != 0) {
5122 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5125 /* Rebroadcast unregister notification */
5126 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5127 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5128 * should have already handle it the first time */
5130 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5132 /* We must not have linkwatch events
5133 * pending on unregister. If this
5134 * happens, we simply run the queue
5135 * unscheduled, resulting in a noop
5138 linkwatch_run_queue();
5143 rebroadcast_time = jiffies;
5148 if (time_after(jiffies, warning_time + 10 * HZ)) {
5149 printk(KERN_EMERG "unregister_netdevice: "
5150 "waiting for %s to become free. Usage "
5152 dev->name, atomic_read(&dev->refcnt));
5153 warning_time = jiffies;
5162 * register_netdevice(x1);
5163 * register_netdevice(x2);
5165 * unregister_netdevice(y1);
5166 * unregister_netdevice(y2);
5172 * We are invoked by rtnl_unlock().
5173 * This allows us to deal with problems:
5174 * 1) We can delete sysfs objects which invoke hotplug
5175 * without deadlocking with linkwatch via keventd.
5176 * 2) Since we run with the RTNL semaphore not held, we can sleep
5177 * safely in order to wait for the netdev refcnt to drop to zero.
5179 * We must not return until all unregister events added during
5180 * the interval the lock was held have been completed.
5182 void netdev_run_todo(void)
5184 struct list_head list;
5186 /* Snapshot list, allow later requests */
5187 list_replace_init(&net_todo_list, &list);
5191 while (!list_empty(&list)) {
5192 struct net_device *dev
5193 = list_first_entry(&list, struct net_device, todo_list);
5194 list_del(&dev->todo_list);
5196 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5197 printk(KERN_ERR "network todo '%s' but state %d\n",
5198 dev->name, dev->reg_state);
5203 dev->reg_state = NETREG_UNREGISTERED;
5205 on_each_cpu(flush_backlog, dev, 1);
5207 netdev_wait_allrefs(dev);
5210 BUG_ON(atomic_read(&dev->refcnt));
5211 WARN_ON(dev->ip_ptr);
5212 WARN_ON(dev->ip6_ptr);
5213 WARN_ON(dev->dn_ptr);
5215 if (dev->destructor)
5216 dev->destructor(dev);
5218 /* Free network device */
5219 kobject_put(&dev->dev.kobj);
5224 * dev_txq_stats_fold - fold tx_queues stats
5225 * @dev: device to get statistics from
5226 * @stats: struct net_device_stats to hold results
5228 void dev_txq_stats_fold(const struct net_device *dev,
5229 struct net_device_stats *stats)
5231 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5233 struct netdev_queue *txq;
5235 for (i = 0; i < dev->num_tx_queues; i++) {
5236 txq = netdev_get_tx_queue(dev, i);
5237 tx_bytes += txq->tx_bytes;
5238 tx_packets += txq->tx_packets;
5239 tx_dropped += txq->tx_dropped;
5241 if (tx_bytes || tx_packets || tx_dropped) {
5242 stats->tx_bytes = tx_bytes;
5243 stats->tx_packets = tx_packets;
5244 stats->tx_dropped = tx_dropped;
5247 EXPORT_SYMBOL(dev_txq_stats_fold);
5250 * dev_get_stats - get network device statistics
5251 * @dev: device to get statistics from
5253 * Get network statistics from device. The device driver may provide
5254 * its own method by setting dev->netdev_ops->get_stats; otherwise
5255 * the internal statistics structure is used.
5257 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5259 const struct net_device_ops *ops = dev->netdev_ops;
5261 if (ops->ndo_get_stats)
5262 return ops->ndo_get_stats(dev);
5264 dev_txq_stats_fold(dev, &dev->stats);
5267 EXPORT_SYMBOL(dev_get_stats);
5269 static void netdev_init_one_queue(struct net_device *dev,
5270 struct netdev_queue *queue,
5276 static void netdev_init_queues(struct net_device *dev)
5278 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5279 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5280 spin_lock_init(&dev->tx_global_lock);
5284 * alloc_netdev_mq - allocate network device
5285 * @sizeof_priv: size of private data to allocate space for
5286 * @name: device name format string
5287 * @setup: callback to initialize device
5288 * @queue_count: the number of subqueues to allocate
5290 * Allocates a struct net_device with private data area for driver use
5291 * and performs basic initialization. Also allocates subquue structs
5292 * for each queue on the device at the end of the netdevice.
5294 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5295 void (*setup)(struct net_device *), unsigned int queue_count)
5297 struct netdev_queue *tx;
5298 struct net_device *dev;
5300 struct net_device *p;
5302 struct netdev_rx_queue *rx;
5306 BUG_ON(strlen(name) >= sizeof(dev->name));
5308 alloc_size = sizeof(struct net_device);
5310 /* ensure 32-byte alignment of private area */
5311 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5312 alloc_size += sizeof_priv;
5314 /* ensure 32-byte alignment of whole construct */
5315 alloc_size += NETDEV_ALIGN - 1;
5317 p = kzalloc(alloc_size, GFP_KERNEL);
5319 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5323 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5325 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5331 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5333 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5338 atomic_set(&rx->count, queue_count);
5341 * Set a pointer to first element in the array which holds the
5344 for (i = 0; i < queue_count; i++)
5348 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5349 dev->padded = (char *)dev - (char *)p;
5351 if (dev_addr_init(dev))
5357 dev_net_set(dev, &init_net);
5360 dev->num_tx_queues = queue_count;
5361 dev->real_num_tx_queues = queue_count;
5365 dev->num_rx_queues = queue_count;
5368 dev->gso_max_size = GSO_MAX_SIZE;
5370 netdev_init_queues(dev);
5372 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5373 dev->ethtool_ntuple_list.count = 0;
5374 INIT_LIST_HEAD(&dev->napi_list);
5375 INIT_LIST_HEAD(&dev->unreg_list);
5376 INIT_LIST_HEAD(&dev->link_watch_list);
5377 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5379 strcpy(dev->name, name);
5392 EXPORT_SYMBOL(alloc_netdev_mq);
5395 * free_netdev - free network device
5398 * This function does the last stage of destroying an allocated device
5399 * interface. The reference to the device object is released.
5400 * If this is the last reference then it will be freed.
5402 void free_netdev(struct net_device *dev)
5404 struct napi_struct *p, *n;
5406 release_net(dev_net(dev));
5410 /* Flush device addresses */
5411 dev_addr_flush(dev);
5413 /* Clear ethtool n-tuple list */
5414 ethtool_ntuple_flush(dev);
5416 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5419 /* Compatibility with error handling in drivers */
5420 if (dev->reg_state == NETREG_UNINITIALIZED) {
5421 kfree((char *)dev - dev->padded);
5425 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5426 dev->reg_state = NETREG_RELEASED;
5428 /* will free via device release */
5429 put_device(&dev->dev);
5431 EXPORT_SYMBOL(free_netdev);
5434 * synchronize_net - Synchronize with packet receive processing
5436 * Wait for packets currently being received to be done.
5437 * Does not block later packets from starting.
5439 void synchronize_net(void)
5444 EXPORT_SYMBOL(synchronize_net);
5447 * unregister_netdevice_queue - remove device from the kernel
5451 * This function shuts down a device interface and removes it
5452 * from the kernel tables.
5453 * If head not NULL, device is queued to be unregistered later.
5455 * Callers must hold the rtnl semaphore. You may want
5456 * unregister_netdev() instead of this.
5459 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5464 list_move_tail(&dev->unreg_list, head);
5466 rollback_registered(dev);
5467 /* Finish processing unregister after unlock */
5471 EXPORT_SYMBOL(unregister_netdevice_queue);
5474 * unregister_netdevice_many - unregister many devices
5475 * @head: list of devices
5477 void unregister_netdevice_many(struct list_head *head)
5479 struct net_device *dev;
5481 if (!list_empty(head)) {
5482 rollback_registered_many(head);
5483 list_for_each_entry(dev, head, unreg_list)
5487 EXPORT_SYMBOL(unregister_netdevice_many);
5490 * unregister_netdev - remove device from the kernel
5493 * This function shuts down a device interface and removes it
5494 * from the kernel tables.
5496 * This is just a wrapper for unregister_netdevice that takes
5497 * the rtnl semaphore. In general you want to use this and not
5498 * unregister_netdevice.
5500 void unregister_netdev(struct net_device *dev)
5503 unregister_netdevice(dev);
5506 EXPORT_SYMBOL(unregister_netdev);
5509 * dev_change_net_namespace - move device to different nethost namespace
5511 * @net: network namespace
5512 * @pat: If not NULL name pattern to try if the current device name
5513 * is already taken in the destination network namespace.
5515 * This function shuts down a device interface and moves it
5516 * to a new network namespace. On success 0 is returned, on
5517 * a failure a netagive errno code is returned.
5519 * Callers must hold the rtnl semaphore.
5522 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5528 /* Don't allow namespace local devices to be moved. */
5530 if (dev->features & NETIF_F_NETNS_LOCAL)
5534 /* Don't allow real devices to be moved when sysfs
5538 if (dev->dev.parent)
5542 /* Ensure the device has been registrered */
5544 if (dev->reg_state != NETREG_REGISTERED)
5547 /* Get out if there is nothing todo */
5549 if (net_eq(dev_net(dev), net))
5552 /* Pick the destination device name, and ensure
5553 * we can use it in the destination network namespace.
5556 if (__dev_get_by_name(net, dev->name)) {
5557 /* We get here if we can't use the current device name */
5560 if (dev_get_valid_name(net, pat, dev->name, 1))
5565 * And now a mini version of register_netdevice unregister_netdevice.
5568 /* If device is running close it first. */
5571 /* And unlink it from device chain */
5573 unlist_netdevice(dev);
5577 /* Shutdown queueing discipline. */
5580 /* Notify protocols, that we are about to destroy
5581 this device. They should clean all the things.
5583 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5584 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5587 * Flush the unicast and multicast chains
5592 netdev_unregister_kobject(dev);
5594 /* Actually switch the network namespace */
5595 dev_net_set(dev, net);
5597 /* If there is an ifindex conflict assign a new one */
5598 if (__dev_get_by_index(net, dev->ifindex)) {
5599 int iflink = (dev->iflink == dev->ifindex);
5600 dev->ifindex = dev_new_index(net);
5602 dev->iflink = dev->ifindex;
5605 /* Fixup kobjects */
5606 err = netdev_register_kobject(dev);
5609 /* Add the device back in the hashes */
5610 list_netdevice(dev);
5612 /* Notify protocols, that a new device appeared. */
5613 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5616 * Prevent userspace races by waiting until the network
5617 * device is fully setup before sending notifications.
5619 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5626 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5628 static int dev_cpu_callback(struct notifier_block *nfb,
5629 unsigned long action,
5632 struct sk_buff **list_skb;
5633 struct sk_buff *skb;
5634 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5635 struct softnet_data *sd, *oldsd;
5637 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5640 local_irq_disable();
5641 cpu = smp_processor_id();
5642 sd = &per_cpu(softnet_data, cpu);
5643 oldsd = &per_cpu(softnet_data, oldcpu);
5645 /* Find end of our completion_queue. */
5646 list_skb = &sd->completion_queue;
5648 list_skb = &(*list_skb)->next;
5649 /* Append completion queue from offline CPU. */
5650 *list_skb = oldsd->completion_queue;
5651 oldsd->completion_queue = NULL;
5653 /* Append output queue from offline CPU. */
5654 if (oldsd->output_queue) {
5655 *sd->output_queue_tailp = oldsd->output_queue;
5656 sd->output_queue_tailp = oldsd->output_queue_tailp;
5657 oldsd->output_queue = NULL;
5658 oldsd->output_queue_tailp = &oldsd->output_queue;
5661 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5664 /* Process offline CPU's input_pkt_queue */
5665 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5667 input_queue_head_add(oldsd, 1);
5669 while ((skb = __skb_dequeue(&oldsd->process_queue)))
5677 * netdev_increment_features - increment feature set by one
5678 * @all: current feature set
5679 * @one: new feature set
5680 * @mask: mask feature set
5682 * Computes a new feature set after adding a device with feature set
5683 * @one to the master device with current feature set @all. Will not
5684 * enable anything that is off in @mask. Returns the new feature set.
5686 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5689 /* If device needs checksumming, downgrade to it. */
5690 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5691 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5692 else if (mask & NETIF_F_ALL_CSUM) {
5693 /* If one device supports v4/v6 checksumming, set for all. */
5694 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5695 !(all & NETIF_F_GEN_CSUM)) {
5696 all &= ~NETIF_F_ALL_CSUM;
5697 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5700 /* If one device supports hw checksumming, set for all. */
5701 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5702 all &= ~NETIF_F_ALL_CSUM;
5703 all |= NETIF_F_HW_CSUM;
5707 one |= NETIF_F_ALL_CSUM;
5709 one |= all & NETIF_F_ONE_FOR_ALL;
5710 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5711 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5715 EXPORT_SYMBOL(netdev_increment_features);
5717 static struct hlist_head *netdev_create_hash(void)
5720 struct hlist_head *hash;
5722 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5724 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5725 INIT_HLIST_HEAD(&hash[i]);
5730 /* Initialize per network namespace state */
5731 static int __net_init netdev_init(struct net *net)
5733 INIT_LIST_HEAD(&net->dev_base_head);
5735 net->dev_name_head = netdev_create_hash();
5736 if (net->dev_name_head == NULL)
5739 net->dev_index_head = netdev_create_hash();
5740 if (net->dev_index_head == NULL)
5746 kfree(net->dev_name_head);
5752 * netdev_drivername - network driver for the device
5753 * @dev: network device
5754 * @buffer: buffer for resulting name
5755 * @len: size of buffer
5757 * Determine network driver for device.
5759 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5761 const struct device_driver *driver;
5762 const struct device *parent;
5764 if (len <= 0 || !buffer)
5768 parent = dev->dev.parent;
5773 driver = parent->driver;
5774 if (driver && driver->name)
5775 strlcpy(buffer, driver->name, len);
5779 static void __net_exit netdev_exit(struct net *net)
5781 kfree(net->dev_name_head);
5782 kfree(net->dev_index_head);
5785 static struct pernet_operations __net_initdata netdev_net_ops = {
5786 .init = netdev_init,
5787 .exit = netdev_exit,
5790 static void __net_exit default_device_exit(struct net *net)
5792 struct net_device *dev, *aux;
5794 * Push all migratable network devices back to the
5795 * initial network namespace
5798 for_each_netdev_safe(net, dev, aux) {
5800 char fb_name[IFNAMSIZ];
5802 /* Ignore unmoveable devices (i.e. loopback) */
5803 if (dev->features & NETIF_F_NETNS_LOCAL)
5806 /* Leave virtual devices for the generic cleanup */
5807 if (dev->rtnl_link_ops)
5810 /* Push remaing network devices to init_net */
5811 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5812 err = dev_change_net_namespace(dev, &init_net, fb_name);
5814 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5815 __func__, dev->name, err);
5822 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5824 /* At exit all network devices most be removed from a network
5825 * namespace. Do this in the reverse order of registeration.
5826 * Do this across as many network namespaces as possible to
5827 * improve batching efficiency.
5829 struct net_device *dev;
5831 LIST_HEAD(dev_kill_list);
5834 list_for_each_entry(net, net_list, exit_list) {
5835 for_each_netdev_reverse(net, dev) {
5836 if (dev->rtnl_link_ops)
5837 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5839 unregister_netdevice_queue(dev, &dev_kill_list);
5842 unregister_netdevice_many(&dev_kill_list);
5846 static struct pernet_operations __net_initdata default_device_ops = {
5847 .exit = default_device_exit,
5848 .exit_batch = default_device_exit_batch,
5852 * Initialize the DEV module. At boot time this walks the device list and
5853 * unhooks any devices that fail to initialise (normally hardware not
5854 * present) and leaves us with a valid list of present and active devices.
5859 * This is called single threaded during boot, so no need
5860 * to take the rtnl semaphore.
5862 static int __init net_dev_init(void)
5864 int i, rc = -ENOMEM;
5866 BUG_ON(!dev_boot_phase);
5868 if (dev_proc_init())
5871 if (netdev_kobject_init())
5874 INIT_LIST_HEAD(&ptype_all);
5875 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5876 INIT_LIST_HEAD(&ptype_base[i]);
5878 if (register_pernet_subsys(&netdev_net_ops))
5882 * Initialise the packet receive queues.
5885 for_each_possible_cpu(i) {
5886 struct softnet_data *sd = &per_cpu(softnet_data, i);
5888 memset(sd, 0, sizeof(*sd));
5889 skb_queue_head_init(&sd->input_pkt_queue);
5890 skb_queue_head_init(&sd->process_queue);
5891 sd->completion_queue = NULL;
5892 INIT_LIST_HEAD(&sd->poll_list);
5893 sd->output_queue = NULL;
5894 sd->output_queue_tailp = &sd->output_queue;
5896 sd->csd.func = rps_trigger_softirq;
5902 sd->backlog.poll = process_backlog;
5903 sd->backlog.weight = weight_p;
5904 sd->backlog.gro_list = NULL;
5905 sd->backlog.gro_count = 0;
5910 /* The loopback device is special if any other network devices
5911 * is present in a network namespace the loopback device must
5912 * be present. Since we now dynamically allocate and free the
5913 * loopback device ensure this invariant is maintained by
5914 * keeping the loopback device as the first device on the
5915 * list of network devices. Ensuring the loopback devices
5916 * is the first device that appears and the last network device
5919 if (register_pernet_device(&loopback_net_ops))
5922 if (register_pernet_device(&default_device_ops))
5925 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5926 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5928 hotcpu_notifier(dev_cpu_callback, 0);
5936 subsys_initcall(net_dev_init);
5938 static int __init initialize_hashrnd(void)
5940 get_random_bytes(&hashrnd, sizeof(hashrnd));
5944 late_initcall_sync(initialize_hashrnd);