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>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <linux/pci.h>
133 #include "net-sysfs.h"
135 /* Instead of increasing this, you should create a hash table. */
136 #define MAX_GRO_SKBS 8
138 /* This should be increased if a protocol with a bigger head is added. */
139 #define GRO_MAX_HEAD (MAX_HEADER + 128)
142 * The list of packet types we will receive (as opposed to discard)
143 * and the routines to invoke.
145 * Why 16. Because with 16 the only overlap we get on a hash of the
146 * low nibble of the protocol value is RARP/SNAP/X.25.
148 * NOTE: That is no longer true with the addition of VLAN tags. Not
149 * sure which should go first, but I bet it won't make much
150 * difference if we are running VLANs. The good news is that
151 * this protocol won't be in the list unless compiled in, so
152 * the average user (w/out VLANs) will not be adversely affected.
169 #define PTYPE_HASH_SIZE (16)
170 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
172 static DEFINE_SPINLOCK(ptype_lock);
173 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
174 static struct list_head ptype_all __read_mostly; /* Taps */
177 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
180 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
182 * Writers must hold the rtnl semaphore while they loop through the
183 * dev_base_head list, and hold dev_base_lock for writing when they do the
184 * actual updates. This allows pure readers to access the list even
185 * while a writer is preparing to update it.
187 * To put it another way, dev_base_lock is held for writing only to
188 * protect against pure readers; the rtnl semaphore provides the
189 * protection against other writers.
191 * See, for example usages, register_netdevice() and
192 * unregister_netdevice(), which must be called with the rtnl
195 DEFINE_RWLOCK(dev_base_lock);
196 EXPORT_SYMBOL(dev_base_lock);
198 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
200 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
201 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
204 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
206 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
209 static inline void rps_lock(struct softnet_data *sd)
212 spin_lock(&sd->input_pkt_queue.lock);
216 static inline void rps_unlock(struct softnet_data *sd)
219 spin_unlock(&sd->input_pkt_queue.lock);
223 /* Device list insertion */
224 static int list_netdevice(struct net_device *dev)
226 struct net *net = dev_net(dev);
230 write_lock_bh(&dev_base_lock);
231 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
232 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
233 hlist_add_head_rcu(&dev->index_hlist,
234 dev_index_hash(net, dev->ifindex));
235 write_unlock_bh(&dev_base_lock);
239 /* Device list removal
240 * caller must respect a RCU grace period before freeing/reusing dev
242 static void unlist_netdevice(struct net_device *dev)
246 /* Unlink dev from the device chain */
247 write_lock_bh(&dev_base_lock);
248 list_del_rcu(&dev->dev_list);
249 hlist_del_rcu(&dev->name_hlist);
250 hlist_del_rcu(&dev->index_hlist);
251 write_unlock_bh(&dev_base_lock);
258 static RAW_NOTIFIER_HEAD(netdev_chain);
261 * Device drivers call our routines to queue packets here. We empty the
262 * queue in the local softnet handler.
265 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
266 EXPORT_PER_CPU_SYMBOL(softnet_data);
268 #ifdef CONFIG_LOCKDEP
270 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
271 * according to dev->type
273 static const unsigned short netdev_lock_type[] =
274 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
275 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
276 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
277 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
278 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
279 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
280 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
281 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
282 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
283 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
284 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
285 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
286 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
287 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
288 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
289 ARPHRD_VOID, ARPHRD_NONE};
291 static const char *const netdev_lock_name[] =
292 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
293 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
294 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
295 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
296 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
297 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
298 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
299 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
300 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
301 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
302 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
303 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
304 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
305 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
306 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
307 "_xmit_VOID", "_xmit_NONE"};
309 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
310 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
316 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
317 if (netdev_lock_type[i] == dev_type)
319 /* the last key is used by default */
320 return ARRAY_SIZE(netdev_lock_type) - 1;
323 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
324 unsigned short dev_type)
328 i = netdev_lock_pos(dev_type);
329 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
330 netdev_lock_name[i]);
333 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
337 i = netdev_lock_pos(dev->type);
338 lockdep_set_class_and_name(&dev->addr_list_lock,
339 &netdev_addr_lock_key[i],
340 netdev_lock_name[i]);
343 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
344 unsigned short dev_type)
347 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
352 /*******************************************************************************
354 Protocol management and registration routines
356 *******************************************************************************/
359 * Add a protocol ID to the list. Now that the input handler is
360 * smarter we can dispense with all the messy stuff that used to be
363 * BEWARE!!! Protocol handlers, mangling input packets,
364 * MUST BE last in hash buckets and checking protocol handlers
365 * MUST start from promiscuous ptype_all chain in net_bh.
366 * It is true now, do not change it.
367 * Explanation follows: if protocol handler, mangling packet, will
368 * be the first on list, it is not able to sense, that packet
369 * is cloned and should be copied-on-write, so that it will
370 * change it and subsequent readers will get broken packet.
375 * dev_add_pack - add packet handler
376 * @pt: packet type declaration
378 * Add a protocol handler to the networking stack. The passed &packet_type
379 * is linked into kernel lists and may not be freed until it has been
380 * removed from the kernel lists.
382 * This call does not sleep therefore it can not
383 * guarantee all CPU's that are in middle of receiving packets
384 * will see the new packet type (until the next received packet).
387 void dev_add_pack(struct packet_type *pt)
391 spin_lock_bh(&ptype_lock);
392 if (pt->type == htons(ETH_P_ALL))
393 list_add_rcu(&pt->list, &ptype_all);
395 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
396 list_add_rcu(&pt->list, &ptype_base[hash]);
398 spin_unlock_bh(&ptype_lock);
400 EXPORT_SYMBOL(dev_add_pack);
403 * __dev_remove_pack - remove packet handler
404 * @pt: packet type declaration
406 * Remove a protocol handler that was previously added to the kernel
407 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
408 * from the kernel lists and can be freed or reused once this function
411 * The packet type might still be in use by receivers
412 * and must not be freed until after all the CPU's have gone
413 * through a quiescent state.
415 void __dev_remove_pack(struct packet_type *pt)
417 struct list_head *head;
418 struct packet_type *pt1;
420 spin_lock_bh(&ptype_lock);
422 if (pt->type == htons(ETH_P_ALL))
425 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
427 list_for_each_entry(pt1, head, list) {
429 list_del_rcu(&pt->list);
434 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
436 spin_unlock_bh(&ptype_lock);
438 EXPORT_SYMBOL(__dev_remove_pack);
441 * dev_remove_pack - remove packet handler
442 * @pt: packet type declaration
444 * Remove a protocol handler that was previously added to the kernel
445 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
446 * from the kernel lists and can be freed or reused once this function
449 * This call sleeps to guarantee that no CPU is looking at the packet
452 void dev_remove_pack(struct packet_type *pt)
454 __dev_remove_pack(pt);
458 EXPORT_SYMBOL(dev_remove_pack);
460 /******************************************************************************
462 Device Boot-time Settings Routines
464 *******************************************************************************/
466 /* Boot time configuration table */
467 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
470 * netdev_boot_setup_add - add new setup entry
471 * @name: name of the device
472 * @map: configured settings for the device
474 * Adds new setup entry to the dev_boot_setup list. The function
475 * returns 0 on error and 1 on success. This is a generic routine to
478 static int netdev_boot_setup_add(char *name, struct ifmap *map)
480 struct netdev_boot_setup *s;
484 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
485 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
486 memset(s[i].name, 0, sizeof(s[i].name));
487 strlcpy(s[i].name, name, IFNAMSIZ);
488 memcpy(&s[i].map, map, sizeof(s[i].map));
493 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
497 * netdev_boot_setup_check - check boot time settings
498 * @dev: the netdevice
500 * Check boot time settings for the device.
501 * The found settings are set for the device to be used
502 * later in the device probing.
503 * Returns 0 if no settings found, 1 if they are.
505 int netdev_boot_setup_check(struct net_device *dev)
507 struct netdev_boot_setup *s = dev_boot_setup;
510 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
511 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
512 !strcmp(dev->name, s[i].name)) {
513 dev->irq = s[i].map.irq;
514 dev->base_addr = s[i].map.base_addr;
515 dev->mem_start = s[i].map.mem_start;
516 dev->mem_end = s[i].map.mem_end;
522 EXPORT_SYMBOL(netdev_boot_setup_check);
526 * netdev_boot_base - get address from boot time settings
527 * @prefix: prefix for network device
528 * @unit: id for network device
530 * Check boot time settings for the base address of device.
531 * The found settings are set for the device to be used
532 * later in the device probing.
533 * Returns 0 if no settings found.
535 unsigned long netdev_boot_base(const char *prefix, int unit)
537 const struct netdev_boot_setup *s = dev_boot_setup;
541 sprintf(name, "%s%d", prefix, unit);
544 * If device already registered then return base of 1
545 * to indicate not to probe for this interface
547 if (__dev_get_by_name(&init_net, name))
550 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
551 if (!strcmp(name, s[i].name))
552 return s[i].map.base_addr;
557 * Saves at boot time configured settings for any netdevice.
559 int __init netdev_boot_setup(char *str)
564 str = get_options(str, ARRAY_SIZE(ints), ints);
569 memset(&map, 0, sizeof(map));
573 map.base_addr = ints[2];
575 map.mem_start = ints[3];
577 map.mem_end = ints[4];
579 /* Add new entry to the list */
580 return netdev_boot_setup_add(str, &map);
583 __setup("netdev=", netdev_boot_setup);
585 /*******************************************************************************
587 Device Interface Subroutines
589 *******************************************************************************/
592 * __dev_get_by_name - find a device by its name
593 * @net: the applicable net namespace
594 * @name: name to find
596 * Find an interface by name. Must be called under RTNL semaphore
597 * or @dev_base_lock. If the name is found a pointer to the device
598 * is returned. If the name is not found then %NULL is returned. The
599 * reference counters are not incremented so the caller must be
600 * careful with locks.
603 struct net_device *__dev_get_by_name(struct net *net, const char *name)
605 struct hlist_node *p;
606 struct net_device *dev;
607 struct hlist_head *head = dev_name_hash(net, name);
609 hlist_for_each_entry(dev, p, head, name_hlist)
610 if (!strncmp(dev->name, name, IFNAMSIZ))
615 EXPORT_SYMBOL(__dev_get_by_name);
618 * dev_get_by_name_rcu - find a device by its name
619 * @net: the applicable net namespace
620 * @name: name to find
622 * Find an interface by name.
623 * If the name is found a pointer to the device is returned.
624 * If the name is not found then %NULL is returned.
625 * The reference counters are not incremented so the caller must be
626 * careful with locks. The caller must hold RCU lock.
629 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
631 struct hlist_node *p;
632 struct net_device *dev;
633 struct hlist_head *head = dev_name_hash(net, name);
635 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
636 if (!strncmp(dev->name, name, IFNAMSIZ))
641 EXPORT_SYMBOL(dev_get_by_name_rcu);
644 * dev_get_by_name - find a device by its name
645 * @net: the applicable net namespace
646 * @name: name to find
648 * Find an interface by name. This can be called from any
649 * context and does its own locking. The returned handle has
650 * the usage count incremented and the caller must use dev_put() to
651 * release it when it is no longer needed. %NULL is returned if no
652 * matching device is found.
655 struct net_device *dev_get_by_name(struct net *net, const char *name)
657 struct net_device *dev;
660 dev = dev_get_by_name_rcu(net, name);
666 EXPORT_SYMBOL(dev_get_by_name);
669 * __dev_get_by_index - find a device by its ifindex
670 * @net: the applicable net namespace
671 * @ifindex: index of device
673 * Search for an interface by index. Returns %NULL if the device
674 * is not found or a pointer to the device. The device has not
675 * had its reference counter increased so the caller must be careful
676 * about locking. The caller must hold either the RTNL semaphore
680 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
682 struct hlist_node *p;
683 struct net_device *dev;
684 struct hlist_head *head = dev_index_hash(net, ifindex);
686 hlist_for_each_entry(dev, p, head, index_hlist)
687 if (dev->ifindex == ifindex)
692 EXPORT_SYMBOL(__dev_get_by_index);
695 * dev_get_by_index_rcu - find a device by its ifindex
696 * @net: the applicable net namespace
697 * @ifindex: index of device
699 * Search for an interface by index. Returns %NULL if the device
700 * is not found or a pointer to the device. The device has not
701 * had its reference counter increased so the caller must be careful
702 * about locking. The caller must hold RCU lock.
705 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
707 struct hlist_node *p;
708 struct net_device *dev;
709 struct hlist_head *head = dev_index_hash(net, ifindex);
711 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
712 if (dev->ifindex == ifindex)
717 EXPORT_SYMBOL(dev_get_by_index_rcu);
721 * dev_get_by_index - find a device by its ifindex
722 * @net: the applicable net namespace
723 * @ifindex: index of device
725 * Search for an interface by index. Returns NULL if the device
726 * is not found or a pointer to the device. The device returned has
727 * had a reference added and the pointer is safe until the user calls
728 * dev_put to indicate they have finished with it.
731 struct net_device *dev_get_by_index(struct net *net, int ifindex)
733 struct net_device *dev;
736 dev = dev_get_by_index_rcu(net, ifindex);
742 EXPORT_SYMBOL(dev_get_by_index);
745 * dev_getbyhwaddr - find a device by its hardware address
746 * @net: the applicable net namespace
747 * @type: media type of device
748 * @ha: hardware address
750 * Search for an interface by MAC address. Returns NULL if the device
751 * is not found or a pointer to the device. The caller must hold the
752 * rtnl semaphore. The returned device has not had its ref count increased
753 * and the caller must therefore be careful about locking
756 * If the API was consistent this would be __dev_get_by_hwaddr
759 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
761 struct net_device *dev;
765 for_each_netdev(net, dev)
766 if (dev->type == type &&
767 !memcmp(dev->dev_addr, ha, dev->addr_len))
772 EXPORT_SYMBOL(dev_getbyhwaddr);
774 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
776 struct net_device *dev;
779 for_each_netdev(net, dev)
780 if (dev->type == type)
785 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
787 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
789 struct net_device *dev, *ret = NULL;
792 for_each_netdev_rcu(net, dev)
793 if (dev->type == type) {
801 EXPORT_SYMBOL(dev_getfirstbyhwtype);
804 * dev_get_by_flags_rcu - find any device with given flags
805 * @net: the applicable net namespace
806 * @if_flags: IFF_* values
807 * @mask: bitmask of bits in if_flags to check
809 * Search for any interface with the given flags. Returns NULL if a device
810 * is not found or a pointer to the device. Must be called inside
811 * rcu_read_lock(), and result refcount is unchanged.
814 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
817 struct net_device *dev, *ret;
820 for_each_netdev_rcu(net, dev) {
821 if (((dev->flags ^ if_flags) & mask) == 0) {
828 EXPORT_SYMBOL(dev_get_by_flags_rcu);
831 * dev_valid_name - check if name is okay for network device
834 * Network device names need to be valid file names to
835 * to allow sysfs to work. We also disallow any kind of
838 int dev_valid_name(const char *name)
842 if (strlen(name) >= IFNAMSIZ)
844 if (!strcmp(name, ".") || !strcmp(name, ".."))
848 if (*name == '/' || isspace(*name))
854 EXPORT_SYMBOL(dev_valid_name);
857 * __dev_alloc_name - allocate a name for a device
858 * @net: network namespace to allocate the device name in
859 * @name: name format string
860 * @buf: scratch buffer and result name string
862 * Passed a format string - eg "lt%d" it will try and find a suitable
863 * id. It scans list of devices to build up a free map, then chooses
864 * the first empty slot. The caller must hold the dev_base or rtnl lock
865 * while allocating the name and adding the device in order to avoid
867 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
868 * Returns the number of the unit assigned or a negative errno code.
871 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
875 const int max_netdevices = 8*PAGE_SIZE;
876 unsigned long *inuse;
877 struct net_device *d;
879 p = strnchr(name, IFNAMSIZ-1, '%');
882 * Verify the string as this thing may have come from
883 * the user. There must be either one "%d" and no other "%"
886 if (p[1] != 'd' || strchr(p + 2, '%'))
889 /* Use one page as a bit array of possible slots */
890 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
894 for_each_netdev(net, d) {
895 if (!sscanf(d->name, name, &i))
897 if (i < 0 || i >= max_netdevices)
900 /* avoid cases where sscanf is not exact inverse of printf */
901 snprintf(buf, IFNAMSIZ, name, i);
902 if (!strncmp(buf, d->name, IFNAMSIZ))
906 i = find_first_zero_bit(inuse, max_netdevices);
907 free_page((unsigned long) inuse);
911 snprintf(buf, IFNAMSIZ, name, i);
912 if (!__dev_get_by_name(net, buf))
915 /* It is possible to run out of possible slots
916 * when the name is long and there isn't enough space left
917 * for the digits, or if all bits are used.
923 * dev_alloc_name - allocate a name for a device
925 * @name: name format string
927 * Passed a format string - eg "lt%d" it will try and find a suitable
928 * id. It scans list of devices to build up a free map, then chooses
929 * the first empty slot. The caller must hold the dev_base or rtnl lock
930 * while allocating the name and adding the device in order to avoid
932 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
933 * Returns the number of the unit assigned or a negative errno code.
936 int dev_alloc_name(struct net_device *dev, const char *name)
942 BUG_ON(!dev_net(dev));
944 ret = __dev_alloc_name(net, name, buf);
946 strlcpy(dev->name, buf, IFNAMSIZ);
949 EXPORT_SYMBOL(dev_alloc_name);
951 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
955 BUG_ON(!dev_net(dev));
958 if (!dev_valid_name(name))
961 if (fmt && strchr(name, '%'))
962 return dev_alloc_name(dev, name);
963 else if (__dev_get_by_name(net, name))
965 else if (dev->name != name)
966 strlcpy(dev->name, name, IFNAMSIZ);
972 * dev_change_name - change name of a device
974 * @newname: name (or format string) must be at least IFNAMSIZ
976 * Change name of a device, can pass format strings "eth%d".
979 int dev_change_name(struct net_device *dev, const char *newname)
981 char oldname[IFNAMSIZ];
987 BUG_ON(!dev_net(dev));
990 if (dev->flags & IFF_UP)
993 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
996 memcpy(oldname, dev->name, IFNAMSIZ);
998 err = dev_get_valid_name(dev, newname, 1);
1003 ret = device_rename(&dev->dev, dev->name);
1005 memcpy(dev->name, oldname, IFNAMSIZ);
1009 write_lock_bh(&dev_base_lock);
1010 hlist_del(&dev->name_hlist);
1011 write_unlock_bh(&dev_base_lock);
1015 write_lock_bh(&dev_base_lock);
1016 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1017 write_unlock_bh(&dev_base_lock);
1019 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1020 ret = notifier_to_errno(ret);
1023 /* err >= 0 after dev_alloc_name() or stores the first errno */
1026 memcpy(dev->name, oldname, IFNAMSIZ);
1030 "%s: name change rollback failed: %d.\n",
1039 * dev_set_alias - change ifalias of a device
1041 * @alias: name up to IFALIASZ
1042 * @len: limit of bytes to copy from info
1044 * Set ifalias for a device,
1046 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1050 if (len >= IFALIASZ)
1055 kfree(dev->ifalias);
1056 dev->ifalias = NULL;
1061 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1065 strlcpy(dev->ifalias, alias, len+1);
1071 * netdev_features_change - device changes features
1072 * @dev: device to cause notification
1074 * Called to indicate a device has changed features.
1076 void netdev_features_change(struct net_device *dev)
1078 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1080 EXPORT_SYMBOL(netdev_features_change);
1083 * netdev_state_change - device changes state
1084 * @dev: device to cause notification
1086 * Called to indicate a device has changed state. This function calls
1087 * the notifier chains for netdev_chain and sends a NEWLINK message
1088 * to the routing socket.
1090 void netdev_state_change(struct net_device *dev)
1092 if (dev->flags & IFF_UP) {
1093 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1094 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1097 EXPORT_SYMBOL(netdev_state_change);
1099 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1101 return call_netdevice_notifiers(event, dev);
1103 EXPORT_SYMBOL(netdev_bonding_change);
1106 * dev_load - load a network module
1107 * @net: the applicable net namespace
1108 * @name: name of interface
1110 * If a network interface is not present and the process has suitable
1111 * privileges this function loads the module. If module loading is not
1112 * available in this kernel then it becomes a nop.
1115 void dev_load(struct net *net, const char *name)
1117 struct net_device *dev;
1120 dev = dev_get_by_name_rcu(net, name);
1123 if (!dev && capable(CAP_NET_ADMIN))
1124 request_module("%s", name);
1126 EXPORT_SYMBOL(dev_load);
1128 static int __dev_open(struct net_device *dev)
1130 const struct net_device_ops *ops = dev->netdev_ops;
1136 * Is it even present?
1138 if (!netif_device_present(dev))
1141 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1142 ret = notifier_to_errno(ret);
1147 * Call device private open method
1149 set_bit(__LINK_STATE_START, &dev->state);
1151 if (ops->ndo_validate_addr)
1152 ret = ops->ndo_validate_addr(dev);
1154 if (!ret && ops->ndo_open)
1155 ret = ops->ndo_open(dev);
1158 * If it went open OK then:
1162 clear_bit(__LINK_STATE_START, &dev->state);
1167 dev->flags |= IFF_UP;
1172 net_dmaengine_get();
1175 * Initialize multicasting status
1177 dev_set_rx_mode(dev);
1180 * Wakeup transmit queue engine
1189 * dev_open - prepare an interface for use.
1190 * @dev: device to open
1192 * Takes a device from down to up state. The device's private open
1193 * function is invoked and then the multicast lists are loaded. Finally
1194 * the device is moved into the up state and a %NETDEV_UP message is
1195 * sent to the netdev notifier chain.
1197 * Calling this function on an active interface is a nop. On a failure
1198 * a negative errno code is returned.
1200 int dev_open(struct net_device *dev)
1207 if (dev->flags & IFF_UP)
1213 ret = __dev_open(dev);
1218 * ... and announce new interface.
1220 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1221 call_netdevice_notifiers(NETDEV_UP, dev);
1225 EXPORT_SYMBOL(dev_open);
1227 static int __dev_close(struct net_device *dev)
1229 const struct net_device_ops *ops = dev->netdev_ops;
1235 * Tell people we are going down, so that they can
1236 * prepare to death, when device is still operating.
1238 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1240 clear_bit(__LINK_STATE_START, &dev->state);
1242 /* Synchronize to scheduled poll. We cannot touch poll list,
1243 * it can be even on different cpu. So just clear netif_running().
1245 * dev->stop() will invoke napi_disable() on all of it's
1246 * napi_struct instances on this device.
1248 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1250 dev_deactivate(dev);
1253 * Call the device specific close. This cannot fail.
1254 * Only if device is UP
1256 * We allow it to be called even after a DETACH hot-plug
1263 * Device is now down.
1266 dev->flags &= ~IFF_UP;
1271 net_dmaengine_put();
1277 * dev_close - shutdown an interface.
1278 * @dev: device to shutdown
1280 * This function moves an active device into down state. A
1281 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1282 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1285 int dev_close(struct net_device *dev)
1287 if (!(dev->flags & IFF_UP))
1293 * Tell people we are down
1295 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1296 call_netdevice_notifiers(NETDEV_DOWN, dev);
1300 EXPORT_SYMBOL(dev_close);
1304 * dev_disable_lro - disable Large Receive Offload on a device
1307 * Disable Large Receive Offload (LRO) on a net device. Must be
1308 * called under RTNL. This is needed if received packets may be
1309 * forwarded to another interface.
1311 void dev_disable_lro(struct net_device *dev)
1313 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1314 dev->ethtool_ops->set_flags) {
1315 u32 flags = dev->ethtool_ops->get_flags(dev);
1316 if (flags & ETH_FLAG_LRO) {
1317 flags &= ~ETH_FLAG_LRO;
1318 dev->ethtool_ops->set_flags(dev, flags);
1321 WARN_ON(dev->features & NETIF_F_LRO);
1323 EXPORT_SYMBOL(dev_disable_lro);
1326 static int dev_boot_phase = 1;
1329 * Device change register/unregister. These are not inline or static
1330 * as we export them to the world.
1334 * register_netdevice_notifier - register a network notifier block
1337 * Register a notifier to be called when network device events occur.
1338 * The notifier passed is linked into the kernel structures and must
1339 * not be reused until it has been unregistered. A negative errno code
1340 * is returned on a failure.
1342 * When registered all registration and up events are replayed
1343 * to the new notifier to allow device to have a race free
1344 * view of the network device list.
1347 int register_netdevice_notifier(struct notifier_block *nb)
1349 struct net_device *dev;
1350 struct net_device *last;
1355 err = raw_notifier_chain_register(&netdev_chain, nb);
1361 for_each_netdev(net, dev) {
1362 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1363 err = notifier_to_errno(err);
1367 if (!(dev->flags & IFF_UP))
1370 nb->notifier_call(nb, NETDEV_UP, dev);
1381 for_each_netdev(net, dev) {
1385 if (dev->flags & IFF_UP) {
1386 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1387 nb->notifier_call(nb, NETDEV_DOWN, dev);
1389 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1390 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1394 raw_notifier_chain_unregister(&netdev_chain, nb);
1397 EXPORT_SYMBOL(register_netdevice_notifier);
1400 * unregister_netdevice_notifier - unregister a network notifier block
1403 * Unregister a notifier previously registered by
1404 * register_netdevice_notifier(). The notifier is unlinked into the
1405 * kernel structures and may then be reused. A negative errno code
1406 * is returned on a failure.
1409 int unregister_netdevice_notifier(struct notifier_block *nb)
1414 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1418 EXPORT_SYMBOL(unregister_netdevice_notifier);
1421 * call_netdevice_notifiers - call all network notifier blocks
1422 * @val: value passed unmodified to notifier function
1423 * @dev: net_device pointer passed unmodified to notifier function
1425 * Call all network notifier blocks. Parameters and return value
1426 * are as for raw_notifier_call_chain().
1429 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1432 return raw_notifier_call_chain(&netdev_chain, val, dev);
1435 /* When > 0 there are consumers of rx skb time stamps */
1436 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1438 void net_enable_timestamp(void)
1440 atomic_inc(&netstamp_needed);
1442 EXPORT_SYMBOL(net_enable_timestamp);
1444 void net_disable_timestamp(void)
1446 atomic_dec(&netstamp_needed);
1448 EXPORT_SYMBOL(net_disable_timestamp);
1450 static inline void net_timestamp_set(struct sk_buff *skb)
1452 if (atomic_read(&netstamp_needed))
1453 __net_timestamp(skb);
1455 skb->tstamp.tv64 = 0;
1458 static inline void net_timestamp_check(struct sk_buff *skb)
1460 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1461 __net_timestamp(skb);
1465 * dev_forward_skb - loopback an skb to another netif
1467 * @dev: destination network device
1468 * @skb: buffer to forward
1471 * NET_RX_SUCCESS (no congestion)
1472 * NET_RX_DROP (packet was dropped, but freed)
1474 * dev_forward_skb can be used for injecting an skb from the
1475 * start_xmit function of one device into the receive queue
1476 * of another device.
1478 * The receiving device may be in another namespace, so
1479 * we have to clear all information in the skb that could
1480 * impact namespace isolation.
1482 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
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)))
1511 net_timestamp_set(skb);
1513 net_timestamp_set(skb);
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 ntohs(skb2->protocol),
1541 skb_reset_network_header(skb2);
1544 skb2->transport_header = skb2->network_header;
1545 skb2->pkt_type = PACKET_OUTGOING;
1546 ptype->func(skb2, skb->dev, ptype, skb->dev);
1553 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1554 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1556 void netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1558 unsigned int real_num = dev->real_num_tx_queues;
1560 if (unlikely(txq > dev->num_tx_queues))
1562 else if (txq > real_num)
1563 dev->real_num_tx_queues = txq;
1564 else if (txq < real_num) {
1565 dev->real_num_tx_queues = txq;
1566 qdisc_reset_all_tx_gt(dev, txq);
1569 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1571 static inline void __netif_reschedule(struct Qdisc *q)
1573 struct softnet_data *sd;
1574 unsigned long flags;
1576 local_irq_save(flags);
1577 sd = &__get_cpu_var(softnet_data);
1578 q->next_sched = NULL;
1579 *sd->output_queue_tailp = q;
1580 sd->output_queue_tailp = &q->next_sched;
1581 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1582 local_irq_restore(flags);
1585 void __netif_schedule(struct Qdisc *q)
1587 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1588 __netif_reschedule(q);
1590 EXPORT_SYMBOL(__netif_schedule);
1592 void dev_kfree_skb_irq(struct sk_buff *skb)
1594 if (atomic_dec_and_test(&skb->users)) {
1595 struct softnet_data *sd;
1596 unsigned long flags;
1598 local_irq_save(flags);
1599 sd = &__get_cpu_var(softnet_data);
1600 skb->next = sd->completion_queue;
1601 sd->completion_queue = skb;
1602 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1603 local_irq_restore(flags);
1606 EXPORT_SYMBOL(dev_kfree_skb_irq);
1608 void dev_kfree_skb_any(struct sk_buff *skb)
1610 if (in_irq() || irqs_disabled())
1611 dev_kfree_skb_irq(skb);
1615 EXPORT_SYMBOL(dev_kfree_skb_any);
1619 * netif_device_detach - mark device as removed
1620 * @dev: network device
1622 * Mark device as removed from system and therefore no longer available.
1624 void netif_device_detach(struct net_device *dev)
1626 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1627 netif_running(dev)) {
1628 netif_tx_stop_all_queues(dev);
1631 EXPORT_SYMBOL(netif_device_detach);
1634 * netif_device_attach - mark device as attached
1635 * @dev: network device
1637 * Mark device as attached from system and restart if needed.
1639 void netif_device_attach(struct net_device *dev)
1641 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1642 netif_running(dev)) {
1643 netif_tx_wake_all_queues(dev);
1644 __netdev_watchdog_up(dev);
1647 EXPORT_SYMBOL(netif_device_attach);
1649 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1651 return ((features & NETIF_F_GEN_CSUM) ||
1652 ((features & NETIF_F_IP_CSUM) &&
1653 protocol == htons(ETH_P_IP)) ||
1654 ((features & NETIF_F_IPV6_CSUM) &&
1655 protocol == htons(ETH_P_IPV6)) ||
1656 ((features & NETIF_F_FCOE_CRC) &&
1657 protocol == htons(ETH_P_FCOE)));
1660 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1662 if (can_checksum_protocol(dev->features, skb->protocol))
1665 if (skb->protocol == htons(ETH_P_8021Q)) {
1666 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1667 if (can_checksum_protocol(dev->features & dev->vlan_features,
1668 veh->h_vlan_encapsulated_proto))
1676 * skb_dev_set -- assign a new device to a buffer
1677 * @skb: buffer for the new device
1678 * @dev: network device
1680 * If an skb is owned by a device already, we have to reset
1681 * all data private to the namespace a device belongs to
1682 * before assigning it a new device.
1684 #ifdef CONFIG_NET_NS
1685 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1688 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1691 skb_init_secmark(skb);
1695 skb->ipvs_property = 0;
1696 #ifdef CONFIG_NET_SCHED
1702 EXPORT_SYMBOL(skb_set_dev);
1703 #endif /* CONFIG_NET_NS */
1706 * Invalidate hardware checksum when packet is to be mangled, and
1707 * complete checksum manually on outgoing path.
1709 int skb_checksum_help(struct sk_buff *skb)
1712 int ret = 0, offset;
1714 if (skb->ip_summed == CHECKSUM_COMPLETE)
1715 goto out_set_summed;
1717 if (unlikely(skb_shinfo(skb)->gso_size)) {
1718 /* Let GSO fix up the checksum. */
1719 goto out_set_summed;
1722 offset = skb->csum_start - skb_headroom(skb);
1723 BUG_ON(offset >= skb_headlen(skb));
1724 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1726 offset += skb->csum_offset;
1727 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1729 if (skb_cloned(skb) &&
1730 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1731 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1736 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1738 skb->ip_summed = CHECKSUM_NONE;
1742 EXPORT_SYMBOL(skb_checksum_help);
1745 * skb_gso_segment - Perform segmentation on skb.
1746 * @skb: buffer to segment
1747 * @features: features for the output path (see dev->features)
1749 * This function segments the given skb and returns a list of segments.
1751 * It may return NULL if the skb requires no segmentation. This is
1752 * only possible when GSO is used for verifying header integrity.
1754 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1756 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1757 struct packet_type *ptype;
1758 __be16 type = skb->protocol;
1761 skb_reset_mac_header(skb);
1762 skb->mac_len = skb->network_header - skb->mac_header;
1763 __skb_pull(skb, skb->mac_len);
1765 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1766 struct net_device *dev = skb->dev;
1767 struct ethtool_drvinfo info = {};
1769 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1770 dev->ethtool_ops->get_drvinfo(dev, &info);
1772 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1774 info.driver, dev ? dev->features : 0L,
1775 skb->sk ? skb->sk->sk_route_caps : 0L,
1776 skb->len, skb->data_len, skb->ip_summed);
1778 if (skb_header_cloned(skb) &&
1779 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1780 return ERR_PTR(err);
1784 list_for_each_entry_rcu(ptype,
1785 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1786 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1787 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1788 err = ptype->gso_send_check(skb);
1789 segs = ERR_PTR(err);
1790 if (err || skb_gso_ok(skb, features))
1792 __skb_push(skb, (skb->data -
1793 skb_network_header(skb)));
1795 segs = ptype->gso_segment(skb, features);
1801 __skb_push(skb, skb->data - skb_mac_header(skb));
1805 EXPORT_SYMBOL(skb_gso_segment);
1807 /* Take action when hardware reception checksum errors are detected. */
1809 void netdev_rx_csum_fault(struct net_device *dev)
1811 if (net_ratelimit()) {
1812 printk(KERN_ERR "%s: hw csum failure.\n",
1813 dev ? dev->name : "<unknown>");
1817 EXPORT_SYMBOL(netdev_rx_csum_fault);
1820 /* Actually, we should eliminate this check as soon as we know, that:
1821 * 1. IOMMU is present and allows to map all the memory.
1822 * 2. No high memory really exists on this machine.
1825 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1827 #ifdef CONFIG_HIGHMEM
1829 if (!(dev->features & NETIF_F_HIGHDMA)) {
1830 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1831 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1835 if (PCI_DMA_BUS_IS_PHYS) {
1836 struct device *pdev = dev->dev.parent;
1840 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1841 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1842 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1851 void (*destructor)(struct sk_buff *skb);
1854 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1856 static void dev_gso_skb_destructor(struct sk_buff *skb)
1858 struct dev_gso_cb *cb;
1861 struct sk_buff *nskb = skb->next;
1863 skb->next = nskb->next;
1866 } while (skb->next);
1868 cb = DEV_GSO_CB(skb);
1870 cb->destructor(skb);
1874 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1875 * @skb: buffer to segment
1877 * This function segments the given skb and stores the list of segments
1880 static int dev_gso_segment(struct sk_buff *skb)
1882 struct net_device *dev = skb->dev;
1883 struct sk_buff *segs;
1884 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1887 segs = skb_gso_segment(skb, features);
1889 /* Verifying header integrity only. */
1894 return PTR_ERR(segs);
1897 DEV_GSO_CB(skb)->destructor = skb->destructor;
1898 skb->destructor = dev_gso_skb_destructor;
1904 * Try to orphan skb early, right before transmission by the device.
1905 * We cannot orphan skb if tx timestamp is requested, since
1906 * drivers need to call skb_tstamp_tx() to send the timestamp.
1908 static inline void skb_orphan_try(struct sk_buff *skb)
1910 struct sock *sk = skb->sk;
1912 if (sk && !skb_tx(skb)->flags) {
1913 /* skb_tx_hash() wont be able to get sk.
1914 * We copy sk_hash into skb->rxhash
1917 skb->rxhash = sk->sk_hash;
1923 * Returns true if either:
1924 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1925 * 2. skb is fragmented and the device does not support SG, or if
1926 * at least one of fragments is in highmem and device does not
1927 * support DMA from it.
1929 static inline int skb_needs_linearize(struct sk_buff *skb,
1930 struct net_device *dev)
1932 return skb_is_nonlinear(skb) &&
1933 ((skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
1934 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
1935 illegal_highdma(dev, skb))));
1938 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1939 struct netdev_queue *txq)
1941 const struct net_device_ops *ops = dev->netdev_ops;
1942 int rc = NETDEV_TX_OK;
1944 if (likely(!skb->next)) {
1945 if (!list_empty(&ptype_all))
1946 dev_queue_xmit_nit(skb, dev);
1949 * If device doesnt need skb->dst, release it right now while
1950 * its hot in this cpu cache
1952 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1955 skb_orphan_try(skb);
1957 if (netif_needs_gso(dev, skb)) {
1958 if (unlikely(dev_gso_segment(skb)))
1963 if (skb_needs_linearize(skb, dev) &&
1964 __skb_linearize(skb))
1967 /* If packet is not checksummed and device does not
1968 * support checksumming for this protocol, complete
1969 * checksumming here.
1971 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1972 skb_set_transport_header(skb, skb->csum_start -
1974 if (!dev_can_checksum(dev, skb) &&
1975 skb_checksum_help(skb))
1980 rc = ops->ndo_start_xmit(skb, dev);
1981 if (rc == NETDEV_TX_OK)
1982 txq_trans_update(txq);
1988 struct sk_buff *nskb = skb->next;
1990 skb->next = nskb->next;
1994 * If device doesnt need nskb->dst, release it right now while
1995 * its hot in this cpu cache
1997 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2000 rc = ops->ndo_start_xmit(nskb, dev);
2001 if (unlikely(rc != NETDEV_TX_OK)) {
2002 if (rc & ~NETDEV_TX_MASK)
2003 goto out_kfree_gso_skb;
2004 nskb->next = skb->next;
2008 txq_trans_update(txq);
2009 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2010 return NETDEV_TX_BUSY;
2011 } while (skb->next);
2014 if (likely(skb->next == NULL))
2015 skb->destructor = DEV_GSO_CB(skb)->destructor;
2021 static u32 hashrnd __read_mostly;
2023 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
2027 if (skb_rx_queue_recorded(skb)) {
2028 hash = skb_get_rx_queue(skb);
2029 while (unlikely(hash >= dev->real_num_tx_queues))
2030 hash -= dev->real_num_tx_queues;
2034 if (skb->sk && skb->sk->sk_hash)
2035 hash = skb->sk->sk_hash;
2037 hash = (__force u16) skb->protocol ^ skb->rxhash;
2038 hash = jhash_1word(hash, hashrnd);
2040 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2042 EXPORT_SYMBOL(skb_tx_hash);
2044 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2046 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2047 if (net_ratelimit()) {
2048 pr_warning("%s selects TX queue %d, but "
2049 "real number of TX queues is %d\n",
2050 dev->name, queue_index, dev->real_num_tx_queues);
2057 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2058 struct sk_buff *skb)
2061 struct sock *sk = skb->sk;
2063 queue_index = sk_tx_queue_get(sk);
2064 if (queue_index < 0) {
2065 const struct net_device_ops *ops = dev->netdev_ops;
2067 if (ops->ndo_select_queue) {
2068 queue_index = ops->ndo_select_queue(dev, skb);
2069 queue_index = dev_cap_txqueue(dev, queue_index);
2072 if (dev->real_num_tx_queues > 1)
2073 queue_index = skb_tx_hash(dev, skb);
2076 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2078 if (dst && skb_dst(skb) == dst)
2079 sk_tx_queue_set(sk, queue_index);
2084 skb_set_queue_mapping(skb, queue_index);
2085 return netdev_get_tx_queue(dev, queue_index);
2088 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2089 struct net_device *dev,
2090 struct netdev_queue *txq)
2092 spinlock_t *root_lock = qdisc_lock(q);
2093 bool contended = qdisc_is_running(q);
2097 * Heuristic to force contended enqueues to serialize on a
2098 * separate lock before trying to get qdisc main lock.
2099 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2100 * and dequeue packets faster.
2102 if (unlikely(contended))
2103 spin_lock(&q->busylock);
2105 spin_lock(root_lock);
2106 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2109 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2110 qdisc_run_begin(q)) {
2112 * This is a work-conserving queue; there are no old skbs
2113 * waiting to be sent out; and the qdisc is not running -
2114 * xmit the skb directly.
2116 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2118 __qdisc_update_bstats(q, skb->len);
2119 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2120 if (unlikely(contended)) {
2121 spin_unlock(&q->busylock);
2128 rc = NET_XMIT_SUCCESS;
2131 rc = qdisc_enqueue_root(skb, q);
2132 if (qdisc_run_begin(q)) {
2133 if (unlikely(contended)) {
2134 spin_unlock(&q->busylock);
2140 spin_unlock(root_lock);
2141 if (unlikely(contended))
2142 spin_unlock(&q->busylock);
2147 * dev_queue_xmit - transmit a buffer
2148 * @skb: buffer to transmit
2150 * Queue a buffer for transmission to a network device. The caller must
2151 * have set the device and priority and built the buffer before calling
2152 * this function. The function can be called from an interrupt.
2154 * A negative errno code is returned on a failure. A success does not
2155 * guarantee the frame will be transmitted as it may be dropped due
2156 * to congestion or traffic shaping.
2158 * -----------------------------------------------------------------------------------
2159 * I notice this method can also return errors from the queue disciplines,
2160 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2163 * Regardless of the return value, the skb is consumed, so it is currently
2164 * difficult to retry a send to this method. (You can bump the ref count
2165 * before sending to hold a reference for retry if you are careful.)
2167 * When calling this method, interrupts MUST be enabled. This is because
2168 * the BH enable code must have IRQs enabled so that it will not deadlock.
2171 int dev_queue_xmit(struct sk_buff *skb)
2173 struct net_device *dev = skb->dev;
2174 struct netdev_queue *txq;
2178 /* Disable soft irqs for various locks below. Also
2179 * stops preemption for RCU.
2183 txq = dev_pick_tx(dev, skb);
2184 q = rcu_dereference_bh(txq->qdisc);
2186 #ifdef CONFIG_NET_CLS_ACT
2187 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2190 rc = __dev_xmit_skb(skb, q, dev, txq);
2194 /* The device has no queue. Common case for software devices:
2195 loopback, all the sorts of tunnels...
2197 Really, it is unlikely that netif_tx_lock protection is necessary
2198 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2200 However, it is possible, that they rely on protection
2203 Check this and shot the lock. It is not prone from deadlocks.
2204 Either shot noqueue qdisc, it is even simpler 8)
2206 if (dev->flags & IFF_UP) {
2207 int cpu = smp_processor_id(); /* ok because BHs are off */
2209 if (txq->xmit_lock_owner != cpu) {
2211 HARD_TX_LOCK(dev, txq, cpu);
2213 if (!netif_tx_queue_stopped(txq)) {
2214 rc = dev_hard_start_xmit(skb, dev, txq);
2215 if (dev_xmit_complete(rc)) {
2216 HARD_TX_UNLOCK(dev, txq);
2220 HARD_TX_UNLOCK(dev, txq);
2221 if (net_ratelimit())
2222 printk(KERN_CRIT "Virtual device %s asks to "
2223 "queue packet!\n", dev->name);
2225 /* Recursion is detected! It is possible,
2227 if (net_ratelimit())
2228 printk(KERN_CRIT "Dead loop on virtual device "
2229 "%s, fix it urgently!\n", dev->name);
2234 rcu_read_unlock_bh();
2239 rcu_read_unlock_bh();
2242 EXPORT_SYMBOL(dev_queue_xmit);
2245 /*=======================================================================
2247 =======================================================================*/
2249 int netdev_max_backlog __read_mostly = 1000;
2250 int netdev_tstamp_prequeue __read_mostly = 1;
2251 int netdev_budget __read_mostly = 300;
2252 int weight_p __read_mostly = 64; /* old backlog weight */
2254 /* Called with irq disabled */
2255 static inline void ____napi_schedule(struct softnet_data *sd,
2256 struct napi_struct *napi)
2258 list_add_tail(&napi->poll_list, &sd->poll_list);
2259 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2263 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2264 * and src/dst port numbers. Returns a non-zero hash number on success
2267 __u32 __skb_get_rxhash(struct sk_buff *skb)
2269 int nhoff, hash = 0;
2270 struct ipv6hdr *ip6;
2273 u32 addr1, addr2, ihl;
2279 nhoff = skb_network_offset(skb);
2281 switch (skb->protocol) {
2282 case __constant_htons(ETH_P_IP):
2283 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2286 ip = (struct iphdr *) skb->data + nhoff;
2287 ip_proto = ip->protocol;
2288 addr1 = (__force u32) ip->saddr;
2289 addr2 = (__force u32) ip->daddr;
2292 case __constant_htons(ETH_P_IPV6):
2293 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2296 ip6 = (struct ipv6hdr *) skb->data + nhoff;
2297 ip_proto = ip6->nexthdr;
2298 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2299 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2313 case IPPROTO_UDPLITE:
2314 if (pskb_may_pull(skb, (ihl * 4) + 4 + nhoff)) {
2315 ports.v32 = * (__force u32 *) (skb->data + nhoff +
2317 if (ports.v16[1] < ports.v16[0])
2318 swap(ports.v16[0], ports.v16[1]);
2326 /* get a consistent hash (same value on both flow directions) */
2330 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2337 EXPORT_SYMBOL(__skb_get_rxhash);
2341 /* One global table that all flow-based protocols share. */
2342 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2343 EXPORT_SYMBOL(rps_sock_flow_table);
2346 * get_rps_cpu is called from netif_receive_skb and returns the target
2347 * CPU from the RPS map of the receiving queue for a given skb.
2348 * rcu_read_lock must be held on entry.
2350 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2351 struct rps_dev_flow **rflowp)
2353 struct netdev_rx_queue *rxqueue;
2354 struct rps_map *map;
2355 struct rps_dev_flow_table *flow_table;
2356 struct rps_sock_flow_table *sock_flow_table;
2360 if (skb_rx_queue_recorded(skb)) {
2361 u16 index = skb_get_rx_queue(skb);
2362 if (unlikely(index >= dev->num_rx_queues)) {
2363 WARN_ONCE(dev->num_rx_queues > 1, "%s received packet "
2364 "on queue %u, but number of RX queues is %u\n",
2365 dev->name, index, dev->num_rx_queues);
2368 rxqueue = dev->_rx + index;
2372 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2375 if (!skb_get_rxhash(skb))
2378 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2379 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2380 if (flow_table && sock_flow_table) {
2382 struct rps_dev_flow *rflow;
2384 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2387 next_cpu = sock_flow_table->ents[skb->rxhash &
2388 sock_flow_table->mask];
2391 * If the desired CPU (where last recvmsg was done) is
2392 * different from current CPU (one in the rx-queue flow
2393 * table entry), switch if one of the following holds:
2394 * - Current CPU is unset (equal to RPS_NO_CPU).
2395 * - Current CPU is offline.
2396 * - The current CPU's queue tail has advanced beyond the
2397 * last packet that was enqueued using this table entry.
2398 * This guarantees that all previous packets for the flow
2399 * have been dequeued, thus preserving in order delivery.
2401 if (unlikely(tcpu != next_cpu) &&
2402 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2403 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2404 rflow->last_qtail)) >= 0)) {
2405 tcpu = rflow->cpu = next_cpu;
2406 if (tcpu != RPS_NO_CPU)
2407 rflow->last_qtail = per_cpu(softnet_data,
2408 tcpu).input_queue_head;
2410 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2417 map = rcu_dereference(rxqueue->rps_map);
2419 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2421 if (cpu_online(tcpu)) {
2431 /* Called from hardirq (IPI) context */
2432 static void rps_trigger_softirq(void *data)
2434 struct softnet_data *sd = data;
2436 ____napi_schedule(sd, &sd->backlog);
2440 #endif /* CONFIG_RPS */
2443 * Check if this softnet_data structure is another cpu one
2444 * If yes, queue it to our IPI list and return 1
2447 static int rps_ipi_queued(struct softnet_data *sd)
2450 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2453 sd->rps_ipi_next = mysd->rps_ipi_list;
2454 mysd->rps_ipi_list = sd;
2456 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2459 #endif /* CONFIG_RPS */
2464 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2465 * queue (may be a remote CPU queue).
2467 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2468 unsigned int *qtail)
2470 struct softnet_data *sd;
2471 unsigned long flags;
2473 sd = &per_cpu(softnet_data, cpu);
2475 local_irq_save(flags);
2478 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2479 if (skb_queue_len(&sd->input_pkt_queue)) {
2481 __skb_queue_tail(&sd->input_pkt_queue, skb);
2482 input_queue_tail_incr_save(sd, qtail);
2484 local_irq_restore(flags);
2485 return NET_RX_SUCCESS;
2488 /* Schedule NAPI for backlog device
2489 * We can use non atomic operation since we own the queue lock
2491 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2492 if (!rps_ipi_queued(sd))
2493 ____napi_schedule(sd, &sd->backlog);
2501 local_irq_restore(flags);
2508 * netif_rx - post buffer to the network code
2509 * @skb: buffer to post
2511 * This function receives a packet from a device driver and queues it for
2512 * the upper (protocol) levels to process. It always succeeds. The buffer
2513 * may be dropped during processing for congestion control or by the
2517 * NET_RX_SUCCESS (no congestion)
2518 * NET_RX_DROP (packet was dropped)
2522 int netif_rx(struct sk_buff *skb)
2526 /* if netpoll wants it, pretend we never saw it */
2527 if (netpoll_rx(skb))
2530 if (netdev_tstamp_prequeue)
2531 net_timestamp_check(skb);
2535 struct rps_dev_flow voidflow, *rflow = &voidflow;
2541 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2543 cpu = smp_processor_id();
2545 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2553 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2559 EXPORT_SYMBOL(netif_rx);
2561 int netif_rx_ni(struct sk_buff *skb)
2566 err = netif_rx(skb);
2567 if (local_softirq_pending())
2573 EXPORT_SYMBOL(netif_rx_ni);
2575 static void net_tx_action(struct softirq_action *h)
2577 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2579 if (sd->completion_queue) {
2580 struct sk_buff *clist;
2582 local_irq_disable();
2583 clist = sd->completion_queue;
2584 sd->completion_queue = NULL;
2588 struct sk_buff *skb = clist;
2589 clist = clist->next;
2591 WARN_ON(atomic_read(&skb->users));
2596 if (sd->output_queue) {
2599 local_irq_disable();
2600 head = sd->output_queue;
2601 sd->output_queue = NULL;
2602 sd->output_queue_tailp = &sd->output_queue;
2606 struct Qdisc *q = head;
2607 spinlock_t *root_lock;
2609 head = head->next_sched;
2611 root_lock = qdisc_lock(q);
2612 if (spin_trylock(root_lock)) {
2613 smp_mb__before_clear_bit();
2614 clear_bit(__QDISC_STATE_SCHED,
2617 spin_unlock(root_lock);
2619 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2621 __netif_reschedule(q);
2623 smp_mb__before_clear_bit();
2624 clear_bit(__QDISC_STATE_SCHED,
2632 static inline int deliver_skb(struct sk_buff *skb,
2633 struct packet_type *pt_prev,
2634 struct net_device *orig_dev)
2636 atomic_inc(&skb->users);
2637 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2640 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2641 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2642 /* This hook is defined here for ATM LANE */
2643 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2644 unsigned char *addr) __read_mostly;
2645 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2648 #ifdef CONFIG_NET_CLS_ACT
2649 /* TODO: Maybe we should just force sch_ingress to be compiled in
2650 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2651 * a compare and 2 stores extra right now if we dont have it on
2652 * but have CONFIG_NET_CLS_ACT
2653 * NOTE: This doesnt stop any functionality; if you dont have
2654 * the ingress scheduler, you just cant add policies on ingress.
2657 static int ing_filter(struct sk_buff *skb)
2659 struct net_device *dev = skb->dev;
2660 u32 ttl = G_TC_RTTL(skb->tc_verd);
2661 struct netdev_queue *rxq;
2662 int result = TC_ACT_OK;
2665 if (unlikely(MAX_RED_LOOP < ttl++)) {
2666 if (net_ratelimit())
2667 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2668 skb->skb_iif, dev->ifindex);
2672 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2673 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2675 rxq = &dev->rx_queue;
2678 if (q != &noop_qdisc) {
2679 spin_lock(qdisc_lock(q));
2680 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2681 result = qdisc_enqueue_root(skb, q);
2682 spin_unlock(qdisc_lock(q));
2688 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2689 struct packet_type **pt_prev,
2690 int *ret, struct net_device *orig_dev)
2692 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2696 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2700 switch (ing_filter(skb)) {
2714 * netif_nit_deliver - deliver received packets to network taps
2717 * This function is used to deliver incoming packets to network
2718 * taps. It should be used when the normal netif_receive_skb path
2719 * is bypassed, for example because of VLAN acceleration.
2721 void netif_nit_deliver(struct sk_buff *skb)
2723 struct packet_type *ptype;
2725 if (list_empty(&ptype_all))
2728 skb_reset_network_header(skb);
2729 skb_reset_transport_header(skb);
2730 skb->mac_len = skb->network_header - skb->mac_header;
2733 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2734 if (!ptype->dev || ptype->dev == skb->dev)
2735 deliver_skb(skb, ptype, skb->dev);
2741 * netdev_rx_handler_register - register receive handler
2742 * @dev: device to register a handler for
2743 * @rx_handler: receive handler to register
2744 * @rx_handler_data: data pointer that is used by rx handler
2746 * Register a receive hander for a device. This handler will then be
2747 * called from __netif_receive_skb. A negative errno code is returned
2750 * The caller must hold the rtnl_mutex.
2752 int netdev_rx_handler_register(struct net_device *dev,
2753 rx_handler_func_t *rx_handler,
2754 void *rx_handler_data)
2758 if (dev->rx_handler)
2761 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2762 rcu_assign_pointer(dev->rx_handler, rx_handler);
2766 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2769 * netdev_rx_handler_unregister - unregister receive handler
2770 * @dev: device to unregister a handler from
2772 * Unregister a receive hander from a device.
2774 * The caller must hold the rtnl_mutex.
2776 void netdev_rx_handler_unregister(struct net_device *dev)
2780 rcu_assign_pointer(dev->rx_handler, NULL);
2781 rcu_assign_pointer(dev->rx_handler_data, NULL);
2783 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2785 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2786 struct net_device *master)
2788 if (skb->pkt_type == PACKET_HOST) {
2789 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2791 memcpy(dest, master->dev_addr, ETH_ALEN);
2795 /* On bonding slaves other than the currently active slave, suppress
2796 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2797 * ARP on active-backup slaves with arp_validate enabled.
2799 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2801 struct net_device *dev = skb->dev;
2803 if (master->priv_flags & IFF_MASTER_ARPMON)
2804 dev->last_rx = jiffies;
2806 if ((master->priv_flags & IFF_MASTER_ALB) &&
2807 (master->priv_flags & IFF_BRIDGE_PORT)) {
2808 /* Do address unmangle. The local destination address
2809 * will be always the one master has. Provides the right
2810 * functionality in a bridge.
2812 skb_bond_set_mac_by_master(skb, master);
2815 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2816 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2817 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2820 if (master->priv_flags & IFF_MASTER_ALB) {
2821 if (skb->pkt_type != PACKET_BROADCAST &&
2822 skb->pkt_type != PACKET_MULTICAST)
2825 if (master->priv_flags & IFF_MASTER_8023AD &&
2826 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2833 EXPORT_SYMBOL(__skb_bond_should_drop);
2835 static int __netif_receive_skb(struct sk_buff *skb)
2837 struct packet_type *ptype, *pt_prev;
2838 rx_handler_func_t *rx_handler;
2839 struct net_device *orig_dev;
2840 struct net_device *master;
2841 struct net_device *null_or_orig;
2842 struct net_device *orig_or_bond;
2843 int ret = NET_RX_DROP;
2846 if (!netdev_tstamp_prequeue)
2847 net_timestamp_check(skb);
2849 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2850 return NET_RX_SUCCESS;
2852 /* if we've gotten here through NAPI, check netpoll */
2853 if (netpoll_receive_skb(skb))
2857 skb->skb_iif = skb->dev->ifindex;
2860 * bonding note: skbs received on inactive slaves should only
2861 * be delivered to pkt handlers that are exact matches. Also
2862 * the deliver_no_wcard flag will be set. If packet handlers
2863 * are sensitive to duplicate packets these skbs will need to
2864 * be dropped at the handler. The vlan accel path may have
2865 * already set the deliver_no_wcard flag.
2867 null_or_orig = NULL;
2868 orig_dev = skb->dev;
2869 master = ACCESS_ONCE(orig_dev->master);
2870 if (skb->deliver_no_wcard)
2871 null_or_orig = orig_dev;
2873 if (skb_bond_should_drop(skb, master)) {
2874 skb->deliver_no_wcard = 1;
2875 null_or_orig = orig_dev; /* deliver only exact match */
2880 __this_cpu_inc(softnet_data.processed);
2881 skb_reset_network_header(skb);
2882 skb_reset_transport_header(skb);
2883 skb->mac_len = skb->network_header - skb->mac_header;
2889 #ifdef CONFIG_NET_CLS_ACT
2890 if (skb->tc_verd & TC_NCLS) {
2891 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2896 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2897 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2898 ptype->dev == orig_dev) {
2900 ret = deliver_skb(skb, pt_prev, orig_dev);
2905 #ifdef CONFIG_NET_CLS_ACT
2906 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2912 /* Handle special case of bridge or macvlan */
2913 rx_handler = rcu_dereference(skb->dev->rx_handler);
2916 ret = deliver_skb(skb, pt_prev, orig_dev);
2919 skb = rx_handler(skb);
2925 * Make sure frames received on VLAN interfaces stacked on
2926 * bonding interfaces still make their way to any base bonding
2927 * device that may have registered for a specific ptype. The
2928 * handler may have to adjust skb->dev and orig_dev.
2930 orig_or_bond = orig_dev;
2931 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2932 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2933 orig_or_bond = vlan_dev_real_dev(skb->dev);
2936 type = skb->protocol;
2937 list_for_each_entry_rcu(ptype,
2938 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2939 if (ptype->type == type && (ptype->dev == null_or_orig ||
2940 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2941 ptype->dev == orig_or_bond)) {
2943 ret = deliver_skb(skb, pt_prev, orig_dev);
2949 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2952 /* Jamal, now you will not able to escape explaining
2953 * me how you were going to use this. :-)
2964 * netif_receive_skb - process receive buffer from network
2965 * @skb: buffer to process
2967 * netif_receive_skb() is the main receive data processing function.
2968 * It always succeeds. The buffer may be dropped during processing
2969 * for congestion control or by the protocol layers.
2971 * This function may only be called from softirq context and interrupts
2972 * should be enabled.
2974 * Return values (usually ignored):
2975 * NET_RX_SUCCESS: no congestion
2976 * NET_RX_DROP: packet was dropped
2978 int netif_receive_skb(struct sk_buff *skb)
2980 if (netdev_tstamp_prequeue)
2981 net_timestamp_check(skb);
2983 if (skb_defer_rx_timestamp(skb))
2984 return NET_RX_SUCCESS;
2988 struct rps_dev_flow voidflow, *rflow = &voidflow;
2993 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2996 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3000 ret = __netif_receive_skb(skb);
3006 return __netif_receive_skb(skb);
3009 EXPORT_SYMBOL(netif_receive_skb);
3011 /* Network device is going away, flush any packets still pending
3012 * Called with irqs disabled.
3014 static void flush_backlog(void *arg)
3016 struct net_device *dev = arg;
3017 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3018 struct sk_buff *skb, *tmp;
3021 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3022 if (skb->dev == dev) {
3023 __skb_unlink(skb, &sd->input_pkt_queue);
3025 input_queue_head_incr(sd);
3030 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3031 if (skb->dev == dev) {
3032 __skb_unlink(skb, &sd->process_queue);
3034 input_queue_head_incr(sd);
3039 static int napi_gro_complete(struct sk_buff *skb)
3041 struct packet_type *ptype;
3042 __be16 type = skb->protocol;
3043 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3046 if (NAPI_GRO_CB(skb)->count == 1) {
3047 skb_shinfo(skb)->gso_size = 0;
3052 list_for_each_entry_rcu(ptype, head, list) {
3053 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3056 err = ptype->gro_complete(skb);
3062 WARN_ON(&ptype->list == head);
3064 return NET_RX_SUCCESS;
3068 return netif_receive_skb(skb);
3071 static void napi_gro_flush(struct napi_struct *napi)
3073 struct sk_buff *skb, *next;
3075 for (skb = napi->gro_list; skb; skb = next) {
3078 napi_gro_complete(skb);
3081 napi->gro_count = 0;
3082 napi->gro_list = NULL;
3085 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3087 struct sk_buff **pp = NULL;
3088 struct packet_type *ptype;
3089 __be16 type = skb->protocol;
3090 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3093 enum gro_result ret;
3095 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3098 if (skb_is_gso(skb) || skb_has_frags(skb))
3102 list_for_each_entry_rcu(ptype, head, list) {
3103 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3106 skb_set_network_header(skb, skb_gro_offset(skb));
3107 mac_len = skb->network_header - skb->mac_header;
3108 skb->mac_len = mac_len;
3109 NAPI_GRO_CB(skb)->same_flow = 0;
3110 NAPI_GRO_CB(skb)->flush = 0;
3111 NAPI_GRO_CB(skb)->free = 0;
3113 pp = ptype->gro_receive(&napi->gro_list, skb);
3118 if (&ptype->list == head)
3121 same_flow = NAPI_GRO_CB(skb)->same_flow;
3122 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3125 struct sk_buff *nskb = *pp;
3129 napi_gro_complete(nskb);
3136 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3140 NAPI_GRO_CB(skb)->count = 1;
3141 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3142 skb->next = napi->gro_list;
3143 napi->gro_list = skb;
3147 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3148 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3150 BUG_ON(skb->end - skb->tail < grow);
3152 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3155 skb->data_len -= grow;
3157 skb_shinfo(skb)->frags[0].page_offset += grow;
3158 skb_shinfo(skb)->frags[0].size -= grow;
3160 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3161 put_page(skb_shinfo(skb)->frags[0].page);
3162 memmove(skb_shinfo(skb)->frags,
3163 skb_shinfo(skb)->frags + 1,
3164 --skb_shinfo(skb)->nr_frags);
3175 EXPORT_SYMBOL(dev_gro_receive);
3178 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3182 for (p = napi->gro_list; p; p = p->next) {
3183 NAPI_GRO_CB(p)->same_flow =
3184 (p->dev == skb->dev) &&
3185 !compare_ether_header(skb_mac_header(p),
3186 skb_gro_mac_header(skb));
3187 NAPI_GRO_CB(p)->flush = 0;
3190 return dev_gro_receive(napi, skb);
3193 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3197 if (netif_receive_skb(skb))
3202 case GRO_MERGED_FREE:
3213 EXPORT_SYMBOL(napi_skb_finish);
3215 void skb_gro_reset_offset(struct sk_buff *skb)
3217 NAPI_GRO_CB(skb)->data_offset = 0;
3218 NAPI_GRO_CB(skb)->frag0 = NULL;
3219 NAPI_GRO_CB(skb)->frag0_len = 0;
3221 if (skb->mac_header == skb->tail &&
3222 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3223 NAPI_GRO_CB(skb)->frag0 =
3224 page_address(skb_shinfo(skb)->frags[0].page) +
3225 skb_shinfo(skb)->frags[0].page_offset;
3226 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3229 EXPORT_SYMBOL(skb_gro_reset_offset);
3231 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3233 skb_gro_reset_offset(skb);
3235 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3237 EXPORT_SYMBOL(napi_gro_receive);
3239 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3241 __skb_pull(skb, skb_headlen(skb));
3242 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3246 EXPORT_SYMBOL(napi_reuse_skb);
3248 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3250 struct sk_buff *skb = napi->skb;
3253 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3259 EXPORT_SYMBOL(napi_get_frags);
3261 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3267 skb->protocol = eth_type_trans(skb, skb->dev);
3269 if (ret == GRO_HELD)
3270 skb_gro_pull(skb, -ETH_HLEN);
3271 else if (netif_receive_skb(skb))
3276 case GRO_MERGED_FREE:
3277 napi_reuse_skb(napi, skb);
3286 EXPORT_SYMBOL(napi_frags_finish);
3288 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3290 struct sk_buff *skb = napi->skb;
3297 skb_reset_mac_header(skb);
3298 skb_gro_reset_offset(skb);
3300 off = skb_gro_offset(skb);
3301 hlen = off + sizeof(*eth);
3302 eth = skb_gro_header_fast(skb, off);
3303 if (skb_gro_header_hard(skb, hlen)) {
3304 eth = skb_gro_header_slow(skb, hlen, off);
3305 if (unlikely(!eth)) {
3306 napi_reuse_skb(napi, skb);
3312 skb_gro_pull(skb, sizeof(*eth));
3315 * This works because the only protocols we care about don't require
3316 * special handling. We'll fix it up properly at the end.
3318 skb->protocol = eth->h_proto;
3323 EXPORT_SYMBOL(napi_frags_skb);
3325 gro_result_t napi_gro_frags(struct napi_struct *napi)
3327 struct sk_buff *skb = napi_frags_skb(napi);
3332 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3334 EXPORT_SYMBOL(napi_gro_frags);
3337 * net_rps_action sends any pending IPI's for rps.
3338 * Note: called with local irq disabled, but exits with local irq enabled.
3340 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3343 struct softnet_data *remsd = sd->rps_ipi_list;
3346 sd->rps_ipi_list = NULL;
3350 /* Send pending IPI's to kick RPS processing on remote cpus. */
3352 struct softnet_data *next = remsd->rps_ipi_next;
3354 if (cpu_online(remsd->cpu))
3355 __smp_call_function_single(remsd->cpu,
3364 static int process_backlog(struct napi_struct *napi, int quota)
3367 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3370 /* Check if we have pending ipi, its better to send them now,
3371 * not waiting net_rx_action() end.
3373 if (sd->rps_ipi_list) {
3374 local_irq_disable();
3375 net_rps_action_and_irq_enable(sd);
3378 napi->weight = weight_p;
3379 local_irq_disable();
3380 while (work < quota) {
3381 struct sk_buff *skb;
3384 while ((skb = __skb_dequeue(&sd->process_queue))) {
3386 __netif_receive_skb(skb);
3387 local_irq_disable();
3388 input_queue_head_incr(sd);
3389 if (++work >= quota) {
3396 qlen = skb_queue_len(&sd->input_pkt_queue);
3398 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3399 &sd->process_queue);
3401 if (qlen < quota - work) {
3403 * Inline a custom version of __napi_complete().
3404 * only current cpu owns and manipulates this napi,
3405 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3406 * we can use a plain write instead of clear_bit(),
3407 * and we dont need an smp_mb() memory barrier.
3409 list_del(&napi->poll_list);
3412 quota = work + qlen;
3422 * __napi_schedule - schedule for receive
3423 * @n: entry to schedule
3425 * The entry's receive function will be scheduled to run
3427 void __napi_schedule(struct napi_struct *n)
3429 unsigned long flags;
3431 local_irq_save(flags);
3432 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3433 local_irq_restore(flags);
3435 EXPORT_SYMBOL(__napi_schedule);
3437 void __napi_complete(struct napi_struct *n)
3439 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3440 BUG_ON(n->gro_list);
3442 list_del(&n->poll_list);
3443 smp_mb__before_clear_bit();
3444 clear_bit(NAPI_STATE_SCHED, &n->state);
3446 EXPORT_SYMBOL(__napi_complete);
3448 void napi_complete(struct napi_struct *n)
3450 unsigned long flags;
3453 * don't let napi dequeue from the cpu poll list
3454 * just in case its running on a different cpu
3456 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3460 local_irq_save(flags);
3462 local_irq_restore(flags);
3464 EXPORT_SYMBOL(napi_complete);
3466 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3467 int (*poll)(struct napi_struct *, int), int weight)
3469 INIT_LIST_HEAD(&napi->poll_list);
3470 napi->gro_count = 0;
3471 napi->gro_list = NULL;
3474 napi->weight = weight;
3475 list_add(&napi->dev_list, &dev->napi_list);
3477 #ifdef CONFIG_NETPOLL
3478 spin_lock_init(&napi->poll_lock);
3479 napi->poll_owner = -1;
3481 set_bit(NAPI_STATE_SCHED, &napi->state);
3483 EXPORT_SYMBOL(netif_napi_add);
3485 void netif_napi_del(struct napi_struct *napi)
3487 struct sk_buff *skb, *next;
3489 list_del_init(&napi->dev_list);
3490 napi_free_frags(napi);
3492 for (skb = napi->gro_list; skb; skb = next) {
3498 napi->gro_list = NULL;
3499 napi->gro_count = 0;
3501 EXPORT_SYMBOL(netif_napi_del);
3503 static void net_rx_action(struct softirq_action *h)
3505 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3506 unsigned long time_limit = jiffies + 2;
3507 int budget = netdev_budget;
3510 local_irq_disable();
3512 while (!list_empty(&sd->poll_list)) {
3513 struct napi_struct *n;
3516 /* If softirq window is exhuasted then punt.
3517 * Allow this to run for 2 jiffies since which will allow
3518 * an average latency of 1.5/HZ.
3520 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3525 /* Even though interrupts have been re-enabled, this
3526 * access is safe because interrupts can only add new
3527 * entries to the tail of this list, and only ->poll()
3528 * calls can remove this head entry from the list.
3530 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3532 have = netpoll_poll_lock(n);
3536 /* This NAPI_STATE_SCHED test is for avoiding a race
3537 * with netpoll's poll_napi(). Only the entity which
3538 * obtains the lock and sees NAPI_STATE_SCHED set will
3539 * actually make the ->poll() call. Therefore we avoid
3540 * accidently calling ->poll() when NAPI is not scheduled.
3543 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3544 work = n->poll(n, weight);
3548 WARN_ON_ONCE(work > weight);
3552 local_irq_disable();
3554 /* Drivers must not modify the NAPI state if they
3555 * consume the entire weight. In such cases this code
3556 * still "owns" the NAPI instance and therefore can
3557 * move the instance around on the list at-will.
3559 if (unlikely(work == weight)) {
3560 if (unlikely(napi_disable_pending(n))) {
3563 local_irq_disable();
3565 list_move_tail(&n->poll_list, &sd->poll_list);
3568 netpoll_poll_unlock(have);
3571 net_rps_action_and_irq_enable(sd);
3573 #ifdef CONFIG_NET_DMA
3575 * There may not be any more sk_buffs coming right now, so push
3576 * any pending DMA copies to hardware
3578 dma_issue_pending_all();
3585 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3589 static gifconf_func_t *gifconf_list[NPROTO];
3592 * register_gifconf - register a SIOCGIF handler
3593 * @family: Address family
3594 * @gifconf: Function handler
3596 * Register protocol dependent address dumping routines. The handler
3597 * that is passed must not be freed or reused until it has been replaced
3598 * by another handler.
3600 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3602 if (family >= NPROTO)
3604 gifconf_list[family] = gifconf;
3607 EXPORT_SYMBOL(register_gifconf);
3611 * Map an interface index to its name (SIOCGIFNAME)
3615 * We need this ioctl for efficient implementation of the
3616 * if_indextoname() function required by the IPv6 API. Without
3617 * it, we would have to search all the interfaces to find a
3621 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3623 struct net_device *dev;
3627 * Fetch the caller's info block.
3630 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3634 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3640 strcpy(ifr.ifr_name, dev->name);
3643 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3649 * Perform a SIOCGIFCONF call. This structure will change
3650 * size eventually, and there is nothing I can do about it.
3651 * Thus we will need a 'compatibility mode'.
3654 static int dev_ifconf(struct net *net, char __user *arg)
3657 struct net_device *dev;
3664 * Fetch the caller's info block.
3667 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3674 * Loop over the interfaces, and write an info block for each.
3678 for_each_netdev(net, dev) {
3679 for (i = 0; i < NPROTO; i++) {
3680 if (gifconf_list[i]) {
3683 done = gifconf_list[i](dev, NULL, 0);
3685 done = gifconf_list[i](dev, pos + total,
3695 * All done. Write the updated control block back to the caller.
3697 ifc.ifc_len = total;
3700 * Both BSD and Solaris return 0 here, so we do too.
3702 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3705 #ifdef CONFIG_PROC_FS
3707 * This is invoked by the /proc filesystem handler to display a device
3710 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3713 struct net *net = seq_file_net(seq);
3715 struct net_device *dev;
3719 return SEQ_START_TOKEN;
3722 for_each_netdev_rcu(net, dev)
3729 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3731 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3732 first_net_device(seq_file_net(seq)) :
3733 next_net_device((struct net_device *)v);
3736 return rcu_dereference(dev);
3739 void dev_seq_stop(struct seq_file *seq, void *v)
3745 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3747 struct rtnl_link_stats64 temp;
3748 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3750 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3751 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3752 dev->name, stats->rx_bytes, stats->rx_packets,
3754 stats->rx_dropped + stats->rx_missed_errors,
3755 stats->rx_fifo_errors,
3756 stats->rx_length_errors + stats->rx_over_errors +
3757 stats->rx_crc_errors + stats->rx_frame_errors,
3758 stats->rx_compressed, stats->multicast,
3759 stats->tx_bytes, stats->tx_packets,
3760 stats->tx_errors, stats->tx_dropped,
3761 stats->tx_fifo_errors, stats->collisions,
3762 stats->tx_carrier_errors +
3763 stats->tx_aborted_errors +
3764 stats->tx_window_errors +
3765 stats->tx_heartbeat_errors,
3766 stats->tx_compressed);
3770 * Called from the PROCfs module. This now uses the new arbitrary sized
3771 * /proc/net interface to create /proc/net/dev
3773 static int dev_seq_show(struct seq_file *seq, void *v)
3775 if (v == SEQ_START_TOKEN)
3776 seq_puts(seq, "Inter-| Receive "
3778 " face |bytes packets errs drop fifo frame "
3779 "compressed multicast|bytes packets errs "
3780 "drop fifo colls carrier compressed\n");
3782 dev_seq_printf_stats(seq, v);
3786 static struct softnet_data *softnet_get_online(loff_t *pos)
3788 struct softnet_data *sd = NULL;
3790 while (*pos < nr_cpu_ids)
3791 if (cpu_online(*pos)) {
3792 sd = &per_cpu(softnet_data, *pos);
3799 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3801 return softnet_get_online(pos);
3804 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3807 return softnet_get_online(pos);
3810 static void softnet_seq_stop(struct seq_file *seq, void *v)
3814 static int softnet_seq_show(struct seq_file *seq, void *v)
3816 struct softnet_data *sd = v;
3818 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3819 sd->processed, sd->dropped, sd->time_squeeze, 0,
3820 0, 0, 0, 0, /* was fastroute */
3821 sd->cpu_collision, sd->received_rps);
3825 static const struct seq_operations dev_seq_ops = {
3826 .start = dev_seq_start,
3827 .next = dev_seq_next,
3828 .stop = dev_seq_stop,
3829 .show = dev_seq_show,
3832 static int dev_seq_open(struct inode *inode, struct file *file)
3834 return seq_open_net(inode, file, &dev_seq_ops,
3835 sizeof(struct seq_net_private));
3838 static const struct file_operations dev_seq_fops = {
3839 .owner = THIS_MODULE,
3840 .open = dev_seq_open,
3842 .llseek = seq_lseek,
3843 .release = seq_release_net,
3846 static const struct seq_operations softnet_seq_ops = {
3847 .start = softnet_seq_start,
3848 .next = softnet_seq_next,
3849 .stop = softnet_seq_stop,
3850 .show = softnet_seq_show,
3853 static int softnet_seq_open(struct inode *inode, struct file *file)
3855 return seq_open(file, &softnet_seq_ops);
3858 static const struct file_operations softnet_seq_fops = {
3859 .owner = THIS_MODULE,
3860 .open = softnet_seq_open,
3862 .llseek = seq_lseek,
3863 .release = seq_release,
3866 static void *ptype_get_idx(loff_t pos)
3868 struct packet_type *pt = NULL;
3872 list_for_each_entry_rcu(pt, &ptype_all, list) {
3878 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3879 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3888 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3892 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3895 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3897 struct packet_type *pt;
3898 struct list_head *nxt;
3902 if (v == SEQ_START_TOKEN)
3903 return ptype_get_idx(0);
3906 nxt = pt->list.next;
3907 if (pt->type == htons(ETH_P_ALL)) {
3908 if (nxt != &ptype_all)
3911 nxt = ptype_base[0].next;
3913 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3915 while (nxt == &ptype_base[hash]) {
3916 if (++hash >= PTYPE_HASH_SIZE)
3918 nxt = ptype_base[hash].next;
3921 return list_entry(nxt, struct packet_type, list);
3924 static void ptype_seq_stop(struct seq_file *seq, void *v)
3930 static int ptype_seq_show(struct seq_file *seq, void *v)
3932 struct packet_type *pt = v;
3934 if (v == SEQ_START_TOKEN)
3935 seq_puts(seq, "Type Device Function\n");
3936 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3937 if (pt->type == htons(ETH_P_ALL))
3938 seq_puts(seq, "ALL ");
3940 seq_printf(seq, "%04x", ntohs(pt->type));
3942 seq_printf(seq, " %-8s %pF\n",
3943 pt->dev ? pt->dev->name : "", pt->func);
3949 static const struct seq_operations ptype_seq_ops = {
3950 .start = ptype_seq_start,
3951 .next = ptype_seq_next,
3952 .stop = ptype_seq_stop,
3953 .show = ptype_seq_show,
3956 static int ptype_seq_open(struct inode *inode, struct file *file)
3958 return seq_open_net(inode, file, &ptype_seq_ops,
3959 sizeof(struct seq_net_private));
3962 static const struct file_operations ptype_seq_fops = {
3963 .owner = THIS_MODULE,
3964 .open = ptype_seq_open,
3966 .llseek = seq_lseek,
3967 .release = seq_release_net,
3971 static int __net_init dev_proc_net_init(struct net *net)
3975 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3977 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3979 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3982 if (wext_proc_init(net))
3988 proc_net_remove(net, "ptype");
3990 proc_net_remove(net, "softnet_stat");
3992 proc_net_remove(net, "dev");
3996 static void __net_exit dev_proc_net_exit(struct net *net)
3998 wext_proc_exit(net);
4000 proc_net_remove(net, "ptype");
4001 proc_net_remove(net, "softnet_stat");
4002 proc_net_remove(net, "dev");
4005 static struct pernet_operations __net_initdata dev_proc_ops = {
4006 .init = dev_proc_net_init,
4007 .exit = dev_proc_net_exit,
4010 static int __init dev_proc_init(void)
4012 return register_pernet_subsys(&dev_proc_ops);
4015 #define dev_proc_init() 0
4016 #endif /* CONFIG_PROC_FS */
4020 * netdev_set_master - set up master/slave pair
4021 * @slave: slave device
4022 * @master: new master device
4024 * Changes the master device of the slave. Pass %NULL to break the
4025 * bonding. The caller must hold the RTNL semaphore. On a failure
4026 * a negative errno code is returned. On success the reference counts
4027 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4028 * function returns zero.
4030 int netdev_set_master(struct net_device *slave, struct net_device *master)
4032 struct net_device *old = slave->master;
4042 slave->master = master;
4049 slave->flags |= IFF_SLAVE;
4051 slave->flags &= ~IFF_SLAVE;
4053 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4056 EXPORT_SYMBOL(netdev_set_master);
4058 static void dev_change_rx_flags(struct net_device *dev, int flags)
4060 const struct net_device_ops *ops = dev->netdev_ops;
4062 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4063 ops->ndo_change_rx_flags(dev, flags);
4066 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4068 unsigned short old_flags = dev->flags;
4074 dev->flags |= IFF_PROMISC;
4075 dev->promiscuity += inc;
4076 if (dev->promiscuity == 0) {
4079 * If inc causes overflow, untouch promisc and return error.
4082 dev->flags &= ~IFF_PROMISC;
4084 dev->promiscuity -= inc;
4085 printk(KERN_WARNING "%s: promiscuity touches roof, "
4086 "set promiscuity failed, promiscuity feature "
4087 "of device might be broken.\n", dev->name);
4091 if (dev->flags != old_flags) {
4092 printk(KERN_INFO "device %s %s promiscuous mode\n",
4093 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4095 if (audit_enabled) {
4096 current_uid_gid(&uid, &gid);
4097 audit_log(current->audit_context, GFP_ATOMIC,
4098 AUDIT_ANOM_PROMISCUOUS,
4099 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4100 dev->name, (dev->flags & IFF_PROMISC),
4101 (old_flags & IFF_PROMISC),
4102 audit_get_loginuid(current),
4104 audit_get_sessionid(current));
4107 dev_change_rx_flags(dev, IFF_PROMISC);
4113 * dev_set_promiscuity - update promiscuity count on a device
4117 * Add or remove promiscuity from a device. While the count in the device
4118 * remains above zero the interface remains promiscuous. Once it hits zero
4119 * the device reverts back to normal filtering operation. A negative inc
4120 * value is used to drop promiscuity on the device.
4121 * Return 0 if successful or a negative errno code on error.
4123 int dev_set_promiscuity(struct net_device *dev, int inc)
4125 unsigned short old_flags = dev->flags;
4128 err = __dev_set_promiscuity(dev, inc);
4131 if (dev->flags != old_flags)
4132 dev_set_rx_mode(dev);
4135 EXPORT_SYMBOL(dev_set_promiscuity);
4138 * dev_set_allmulti - update allmulti count on a device
4142 * Add or remove reception of all multicast frames to a device. While the
4143 * count in the device remains above zero the interface remains listening
4144 * to all interfaces. Once it hits zero the device reverts back to normal
4145 * filtering operation. A negative @inc value is used to drop the counter
4146 * when releasing a resource needing all multicasts.
4147 * Return 0 if successful or a negative errno code on error.
4150 int dev_set_allmulti(struct net_device *dev, int inc)
4152 unsigned short old_flags = dev->flags;
4156 dev->flags |= IFF_ALLMULTI;
4157 dev->allmulti += inc;
4158 if (dev->allmulti == 0) {
4161 * If inc causes overflow, untouch allmulti and return error.
4164 dev->flags &= ~IFF_ALLMULTI;
4166 dev->allmulti -= inc;
4167 printk(KERN_WARNING "%s: allmulti touches roof, "
4168 "set allmulti failed, allmulti feature of "
4169 "device might be broken.\n", dev->name);
4173 if (dev->flags ^ old_flags) {
4174 dev_change_rx_flags(dev, IFF_ALLMULTI);
4175 dev_set_rx_mode(dev);
4179 EXPORT_SYMBOL(dev_set_allmulti);
4182 * Upload unicast and multicast address lists to device and
4183 * configure RX filtering. When the device doesn't support unicast
4184 * filtering it is put in promiscuous mode while unicast addresses
4187 void __dev_set_rx_mode(struct net_device *dev)
4189 const struct net_device_ops *ops = dev->netdev_ops;
4191 /* dev_open will call this function so the list will stay sane. */
4192 if (!(dev->flags&IFF_UP))
4195 if (!netif_device_present(dev))
4198 if (ops->ndo_set_rx_mode)
4199 ops->ndo_set_rx_mode(dev);
4201 /* Unicast addresses changes may only happen under the rtnl,
4202 * therefore calling __dev_set_promiscuity here is safe.
4204 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4205 __dev_set_promiscuity(dev, 1);
4206 dev->uc_promisc = 1;
4207 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4208 __dev_set_promiscuity(dev, -1);
4209 dev->uc_promisc = 0;
4212 if (ops->ndo_set_multicast_list)
4213 ops->ndo_set_multicast_list(dev);
4217 void dev_set_rx_mode(struct net_device *dev)
4219 netif_addr_lock_bh(dev);
4220 __dev_set_rx_mode(dev);
4221 netif_addr_unlock_bh(dev);
4225 * dev_get_flags - get flags reported to userspace
4228 * Get the combination of flag bits exported through APIs to userspace.
4230 unsigned dev_get_flags(const struct net_device *dev)
4234 flags = (dev->flags & ~(IFF_PROMISC |
4239 (dev->gflags & (IFF_PROMISC |
4242 if (netif_running(dev)) {
4243 if (netif_oper_up(dev))
4244 flags |= IFF_RUNNING;
4245 if (netif_carrier_ok(dev))
4246 flags |= IFF_LOWER_UP;
4247 if (netif_dormant(dev))
4248 flags |= IFF_DORMANT;
4253 EXPORT_SYMBOL(dev_get_flags);
4255 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4257 int old_flags = dev->flags;
4263 * Set the flags on our device.
4266 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4267 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4269 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4273 * Load in the correct multicast list now the flags have changed.
4276 if ((old_flags ^ flags) & IFF_MULTICAST)
4277 dev_change_rx_flags(dev, IFF_MULTICAST);
4279 dev_set_rx_mode(dev);
4282 * Have we downed the interface. We handle IFF_UP ourselves
4283 * according to user attempts to set it, rather than blindly
4288 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4289 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4292 dev_set_rx_mode(dev);
4295 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4296 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4298 dev->gflags ^= IFF_PROMISC;
4299 dev_set_promiscuity(dev, inc);
4302 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4303 is important. Some (broken) drivers set IFF_PROMISC, when
4304 IFF_ALLMULTI is requested not asking us and not reporting.
4306 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4307 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4309 dev->gflags ^= IFF_ALLMULTI;
4310 dev_set_allmulti(dev, inc);
4316 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4318 unsigned int changes = dev->flags ^ old_flags;
4320 if (changes & IFF_UP) {
4321 if (dev->flags & IFF_UP)
4322 call_netdevice_notifiers(NETDEV_UP, dev);
4324 call_netdevice_notifiers(NETDEV_DOWN, dev);
4327 if (dev->flags & IFF_UP &&
4328 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4329 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4333 * dev_change_flags - change device settings
4335 * @flags: device state flags
4337 * Change settings on device based state flags. The flags are
4338 * in the userspace exported format.
4340 int dev_change_flags(struct net_device *dev, unsigned flags)
4343 int old_flags = dev->flags;
4345 ret = __dev_change_flags(dev, flags);
4349 changes = old_flags ^ dev->flags;
4351 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4353 __dev_notify_flags(dev, old_flags);
4356 EXPORT_SYMBOL(dev_change_flags);
4359 * dev_set_mtu - Change maximum transfer unit
4361 * @new_mtu: new transfer unit
4363 * Change the maximum transfer size of the network device.
4365 int dev_set_mtu(struct net_device *dev, int new_mtu)
4367 const struct net_device_ops *ops = dev->netdev_ops;
4370 if (new_mtu == dev->mtu)
4373 /* MTU must be positive. */
4377 if (!netif_device_present(dev))
4381 if (ops->ndo_change_mtu)
4382 err = ops->ndo_change_mtu(dev, new_mtu);
4386 if (!err && dev->flags & IFF_UP)
4387 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4390 EXPORT_SYMBOL(dev_set_mtu);
4393 * dev_set_mac_address - Change Media Access Control Address
4397 * Change the hardware (MAC) address of the device
4399 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4401 const struct net_device_ops *ops = dev->netdev_ops;
4404 if (!ops->ndo_set_mac_address)
4406 if (sa->sa_family != dev->type)
4408 if (!netif_device_present(dev))
4410 err = ops->ndo_set_mac_address(dev, sa);
4412 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4415 EXPORT_SYMBOL(dev_set_mac_address);
4418 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4420 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4423 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4429 case SIOCGIFFLAGS: /* Get interface flags */
4430 ifr->ifr_flags = (short) dev_get_flags(dev);
4433 case SIOCGIFMETRIC: /* Get the metric on the interface
4434 (currently unused) */
4435 ifr->ifr_metric = 0;
4438 case SIOCGIFMTU: /* Get the MTU of a device */
4439 ifr->ifr_mtu = dev->mtu;
4444 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4446 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4447 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4448 ifr->ifr_hwaddr.sa_family = dev->type;
4456 ifr->ifr_map.mem_start = dev->mem_start;
4457 ifr->ifr_map.mem_end = dev->mem_end;
4458 ifr->ifr_map.base_addr = dev->base_addr;
4459 ifr->ifr_map.irq = dev->irq;
4460 ifr->ifr_map.dma = dev->dma;
4461 ifr->ifr_map.port = dev->if_port;
4465 ifr->ifr_ifindex = dev->ifindex;
4469 ifr->ifr_qlen = dev->tx_queue_len;
4473 /* dev_ioctl() should ensure this case
4485 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4487 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4490 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4491 const struct net_device_ops *ops;
4496 ops = dev->netdev_ops;
4499 case SIOCSIFFLAGS: /* Set interface flags */
4500 return dev_change_flags(dev, ifr->ifr_flags);
4502 case SIOCSIFMETRIC: /* Set the metric on the interface
4503 (currently unused) */
4506 case SIOCSIFMTU: /* Set the MTU of a device */
4507 return dev_set_mtu(dev, ifr->ifr_mtu);
4510 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4512 case SIOCSIFHWBROADCAST:
4513 if (ifr->ifr_hwaddr.sa_family != dev->type)
4515 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4516 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4517 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4521 if (ops->ndo_set_config) {
4522 if (!netif_device_present(dev))
4524 return ops->ndo_set_config(dev, &ifr->ifr_map);
4529 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4530 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4532 if (!netif_device_present(dev))
4534 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4537 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4538 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4540 if (!netif_device_present(dev))
4542 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4545 if (ifr->ifr_qlen < 0)
4547 dev->tx_queue_len = ifr->ifr_qlen;
4551 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4552 return dev_change_name(dev, ifr->ifr_newname);
4555 * Unknown or private ioctl
4558 if ((cmd >= SIOCDEVPRIVATE &&
4559 cmd <= SIOCDEVPRIVATE + 15) ||
4560 cmd == SIOCBONDENSLAVE ||
4561 cmd == SIOCBONDRELEASE ||
4562 cmd == SIOCBONDSETHWADDR ||
4563 cmd == SIOCBONDSLAVEINFOQUERY ||
4564 cmd == SIOCBONDINFOQUERY ||
4565 cmd == SIOCBONDCHANGEACTIVE ||
4566 cmd == SIOCGMIIPHY ||
4567 cmd == SIOCGMIIREG ||
4568 cmd == SIOCSMIIREG ||
4569 cmd == SIOCBRADDIF ||
4570 cmd == SIOCBRDELIF ||
4571 cmd == SIOCSHWTSTAMP ||
4572 cmd == SIOCWANDEV) {
4574 if (ops->ndo_do_ioctl) {
4575 if (netif_device_present(dev))
4576 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4588 * This function handles all "interface"-type I/O control requests. The actual
4589 * 'doing' part of this is dev_ifsioc above.
4593 * dev_ioctl - network device ioctl
4594 * @net: the applicable net namespace
4595 * @cmd: command to issue
4596 * @arg: pointer to a struct ifreq in user space
4598 * Issue ioctl functions to devices. This is normally called by the
4599 * user space syscall interfaces but can sometimes be useful for
4600 * other purposes. The return value is the return from the syscall if
4601 * positive or a negative errno code on error.
4604 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4610 /* One special case: SIOCGIFCONF takes ifconf argument
4611 and requires shared lock, because it sleeps writing
4615 if (cmd == SIOCGIFCONF) {
4617 ret = dev_ifconf(net, (char __user *) arg);
4621 if (cmd == SIOCGIFNAME)
4622 return dev_ifname(net, (struct ifreq __user *)arg);
4624 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4627 ifr.ifr_name[IFNAMSIZ-1] = 0;
4629 colon = strchr(ifr.ifr_name, ':');
4634 * See which interface the caller is talking about.
4639 * These ioctl calls:
4640 * - can be done by all.
4641 * - atomic and do not require locking.
4652 dev_load(net, ifr.ifr_name);
4654 ret = dev_ifsioc_locked(net, &ifr, cmd);
4659 if (copy_to_user(arg, &ifr,
4660 sizeof(struct ifreq)))
4666 dev_load(net, ifr.ifr_name);
4668 ret = dev_ethtool(net, &ifr);
4673 if (copy_to_user(arg, &ifr,
4674 sizeof(struct ifreq)))
4680 * These ioctl calls:
4681 * - require superuser power.
4682 * - require strict serialization.
4688 if (!capable(CAP_NET_ADMIN))
4690 dev_load(net, ifr.ifr_name);
4692 ret = dev_ifsioc(net, &ifr, cmd);
4697 if (copy_to_user(arg, &ifr,
4698 sizeof(struct ifreq)))
4704 * These ioctl calls:
4705 * - require superuser power.
4706 * - require strict serialization.
4707 * - do not return a value
4717 case SIOCSIFHWBROADCAST:
4720 case SIOCBONDENSLAVE:
4721 case SIOCBONDRELEASE:
4722 case SIOCBONDSETHWADDR:
4723 case SIOCBONDCHANGEACTIVE:
4727 if (!capable(CAP_NET_ADMIN))
4730 case SIOCBONDSLAVEINFOQUERY:
4731 case SIOCBONDINFOQUERY:
4732 dev_load(net, ifr.ifr_name);
4734 ret = dev_ifsioc(net, &ifr, cmd);
4739 /* Get the per device memory space. We can add this but
4740 * currently do not support it */
4742 /* Set the per device memory buffer space.
4743 * Not applicable in our case */
4748 * Unknown or private ioctl.
4751 if (cmd == SIOCWANDEV ||
4752 (cmd >= SIOCDEVPRIVATE &&
4753 cmd <= SIOCDEVPRIVATE + 15)) {
4754 dev_load(net, ifr.ifr_name);
4756 ret = dev_ifsioc(net, &ifr, cmd);
4758 if (!ret && copy_to_user(arg, &ifr,
4759 sizeof(struct ifreq)))
4763 /* Take care of Wireless Extensions */
4764 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4765 return wext_handle_ioctl(net, &ifr, cmd, arg);
4772 * dev_new_index - allocate an ifindex
4773 * @net: the applicable net namespace
4775 * Returns a suitable unique value for a new device interface
4776 * number. The caller must hold the rtnl semaphore or the
4777 * dev_base_lock to be sure it remains unique.
4779 static int dev_new_index(struct net *net)
4785 if (!__dev_get_by_index(net, ifindex))
4790 /* Delayed registration/unregisteration */
4791 static LIST_HEAD(net_todo_list);
4793 static void net_set_todo(struct net_device *dev)
4795 list_add_tail(&dev->todo_list, &net_todo_list);
4798 static void rollback_registered_many(struct list_head *head)
4800 struct net_device *dev, *tmp;
4802 BUG_ON(dev_boot_phase);
4805 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4806 /* Some devices call without registering
4807 * for initialization unwind. Remove those
4808 * devices and proceed with the remaining.
4810 if (dev->reg_state == NETREG_UNINITIALIZED) {
4811 pr_debug("unregister_netdevice: device %s/%p never "
4812 "was registered\n", dev->name, dev);
4815 list_del(&dev->unreg_list);
4819 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4821 /* If device is running, close it first. */
4824 /* And unlink it from device chain. */
4825 unlist_netdevice(dev);
4827 dev->reg_state = NETREG_UNREGISTERING;
4832 list_for_each_entry(dev, head, unreg_list) {
4833 /* Shutdown queueing discipline. */
4837 /* Notify protocols, that we are about to destroy
4838 this device. They should clean all the things.
4840 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4842 if (!dev->rtnl_link_ops ||
4843 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4844 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4847 * Flush the unicast and multicast chains
4852 if (dev->netdev_ops->ndo_uninit)
4853 dev->netdev_ops->ndo_uninit(dev);
4855 /* Notifier chain MUST detach us from master device. */
4856 WARN_ON(dev->master);
4858 /* Remove entries from kobject tree */
4859 netdev_unregister_kobject(dev);
4862 /* Process any work delayed until the end of the batch */
4863 dev = list_first_entry(head, struct net_device, unreg_list);
4864 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4868 list_for_each_entry(dev, head, unreg_list)
4872 static void rollback_registered(struct net_device *dev)
4876 list_add(&dev->unreg_list, &single);
4877 rollback_registered_many(&single);
4880 static void __netdev_init_queue_locks_one(struct net_device *dev,
4881 struct netdev_queue *dev_queue,
4884 spin_lock_init(&dev_queue->_xmit_lock);
4885 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4886 dev_queue->xmit_lock_owner = -1;
4889 static void netdev_init_queue_locks(struct net_device *dev)
4891 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4892 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4895 unsigned long netdev_fix_features(unsigned long features, const char *name)
4897 /* Fix illegal SG+CSUM combinations. */
4898 if ((features & NETIF_F_SG) &&
4899 !(features & NETIF_F_ALL_CSUM)) {
4901 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4902 "checksum feature.\n", name);
4903 features &= ~NETIF_F_SG;
4906 /* TSO requires that SG is present as well. */
4907 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4909 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4910 "SG feature.\n", name);
4911 features &= ~NETIF_F_TSO;
4914 if (features & NETIF_F_UFO) {
4915 if (!(features & NETIF_F_GEN_CSUM)) {
4917 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4918 "since no NETIF_F_HW_CSUM feature.\n",
4920 features &= ~NETIF_F_UFO;
4923 if (!(features & NETIF_F_SG)) {
4925 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4926 "since no NETIF_F_SG feature.\n", name);
4927 features &= ~NETIF_F_UFO;
4933 EXPORT_SYMBOL(netdev_fix_features);
4936 * netif_stacked_transfer_operstate - transfer operstate
4937 * @rootdev: the root or lower level device to transfer state from
4938 * @dev: the device to transfer operstate to
4940 * Transfer operational state from root to device. This is normally
4941 * called when a stacking relationship exists between the root
4942 * device and the device(a leaf device).
4944 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4945 struct net_device *dev)
4947 if (rootdev->operstate == IF_OPER_DORMANT)
4948 netif_dormant_on(dev);
4950 netif_dormant_off(dev);
4952 if (netif_carrier_ok(rootdev)) {
4953 if (!netif_carrier_ok(dev))
4954 netif_carrier_on(dev);
4956 if (netif_carrier_ok(dev))
4957 netif_carrier_off(dev);
4960 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4963 * register_netdevice - register a network device
4964 * @dev: device to register
4966 * Take a completed network device structure and add it to the kernel
4967 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4968 * chain. 0 is returned on success. A negative errno code is returned
4969 * on a failure to set up the device, or if the name is a duplicate.
4971 * Callers must hold the rtnl semaphore. You may want
4972 * register_netdev() instead of this.
4975 * The locking appears insufficient to guarantee two parallel registers
4976 * will not get the same name.
4979 int register_netdevice(struct net_device *dev)
4982 struct net *net = dev_net(dev);
4984 BUG_ON(dev_boot_phase);
4989 /* When net_device's are persistent, this will be fatal. */
4990 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4993 spin_lock_init(&dev->addr_list_lock);
4994 netdev_set_addr_lockdep_class(dev);
4995 netdev_init_queue_locks(dev);
5000 if (!dev->num_rx_queues) {
5002 * Allocate a single RX queue if driver never called
5006 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
5012 dev->_rx->first = dev->_rx;
5013 atomic_set(&dev->_rx->count, 1);
5014 dev->num_rx_queues = 1;
5017 /* Init, if this function is available */
5018 if (dev->netdev_ops->ndo_init) {
5019 ret = dev->netdev_ops->ndo_init(dev);
5027 ret = dev_get_valid_name(dev, dev->name, 0);
5031 dev->ifindex = dev_new_index(net);
5032 if (dev->iflink == -1)
5033 dev->iflink = dev->ifindex;
5035 /* Fix illegal checksum combinations */
5036 if ((dev->features & NETIF_F_HW_CSUM) &&
5037 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5038 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5040 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5043 if ((dev->features & NETIF_F_NO_CSUM) &&
5044 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5045 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5047 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5050 dev->features = netdev_fix_features(dev->features, dev->name);
5052 /* Enable software GSO if SG is supported. */
5053 if (dev->features & NETIF_F_SG)
5054 dev->features |= NETIF_F_GSO;
5056 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5057 ret = notifier_to_errno(ret);
5061 ret = netdev_register_kobject(dev);
5064 dev->reg_state = NETREG_REGISTERED;
5067 * Default initial state at registry is that the
5068 * device is present.
5071 set_bit(__LINK_STATE_PRESENT, &dev->state);
5073 dev_init_scheduler(dev);
5075 list_netdevice(dev);
5077 /* Notify protocols, that a new device appeared. */
5078 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5079 ret = notifier_to_errno(ret);
5081 rollback_registered(dev);
5082 dev->reg_state = NETREG_UNREGISTERED;
5085 * Prevent userspace races by waiting until the network
5086 * device is fully setup before sending notifications.
5088 if (!dev->rtnl_link_ops ||
5089 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5090 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5096 if (dev->netdev_ops->ndo_uninit)
5097 dev->netdev_ops->ndo_uninit(dev);
5100 EXPORT_SYMBOL(register_netdevice);
5103 * init_dummy_netdev - init a dummy network device for NAPI
5104 * @dev: device to init
5106 * This takes a network device structure and initialize the minimum
5107 * amount of fields so it can be used to schedule NAPI polls without
5108 * registering a full blown interface. This is to be used by drivers
5109 * that need to tie several hardware interfaces to a single NAPI
5110 * poll scheduler due to HW limitations.
5112 int init_dummy_netdev(struct net_device *dev)
5114 /* Clear everything. Note we don't initialize spinlocks
5115 * are they aren't supposed to be taken by any of the
5116 * NAPI code and this dummy netdev is supposed to be
5117 * only ever used for NAPI polls
5119 memset(dev, 0, sizeof(struct net_device));
5121 /* make sure we BUG if trying to hit standard
5122 * register/unregister code path
5124 dev->reg_state = NETREG_DUMMY;
5126 /* initialize the ref count */
5127 atomic_set(&dev->refcnt, 1);
5129 /* NAPI wants this */
5130 INIT_LIST_HEAD(&dev->napi_list);
5132 /* a dummy interface is started by default */
5133 set_bit(__LINK_STATE_PRESENT, &dev->state);
5134 set_bit(__LINK_STATE_START, &dev->state);
5138 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5142 * register_netdev - register a network device
5143 * @dev: device to register
5145 * Take a completed network device structure and add it to the kernel
5146 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5147 * chain. 0 is returned on success. A negative errno code is returned
5148 * on a failure to set up the device, or if the name is a duplicate.
5150 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5151 * and expands the device name if you passed a format string to
5154 int register_netdev(struct net_device *dev)
5161 * If the name is a format string the caller wants us to do a
5164 if (strchr(dev->name, '%')) {
5165 err = dev_alloc_name(dev, dev->name);
5170 err = register_netdevice(dev);
5175 EXPORT_SYMBOL(register_netdev);
5178 * netdev_wait_allrefs - wait until all references are gone.
5180 * This is called when unregistering network devices.
5182 * Any protocol or device that holds a reference should register
5183 * for netdevice notification, and cleanup and put back the
5184 * reference if they receive an UNREGISTER event.
5185 * We can get stuck here if buggy protocols don't correctly
5188 static void netdev_wait_allrefs(struct net_device *dev)
5190 unsigned long rebroadcast_time, warning_time;
5192 linkwatch_forget_dev(dev);
5194 rebroadcast_time = warning_time = jiffies;
5195 while (atomic_read(&dev->refcnt) != 0) {
5196 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5199 /* Rebroadcast unregister notification */
5200 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5201 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5202 * should have already handle it the first time */
5204 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5206 /* We must not have linkwatch events
5207 * pending on unregister. If this
5208 * happens, we simply run the queue
5209 * unscheduled, resulting in a noop
5212 linkwatch_run_queue();
5217 rebroadcast_time = jiffies;
5222 if (time_after(jiffies, warning_time + 10 * HZ)) {
5223 printk(KERN_EMERG "unregister_netdevice: "
5224 "waiting for %s to become free. Usage "
5226 dev->name, atomic_read(&dev->refcnt));
5227 warning_time = jiffies;
5236 * register_netdevice(x1);
5237 * register_netdevice(x2);
5239 * unregister_netdevice(y1);
5240 * unregister_netdevice(y2);
5246 * We are invoked by rtnl_unlock().
5247 * This allows us to deal with problems:
5248 * 1) We can delete sysfs objects which invoke hotplug
5249 * without deadlocking with linkwatch via keventd.
5250 * 2) Since we run with the RTNL semaphore not held, we can sleep
5251 * safely in order to wait for the netdev refcnt to drop to zero.
5253 * We must not return until all unregister events added during
5254 * the interval the lock was held have been completed.
5256 void netdev_run_todo(void)
5258 struct list_head list;
5260 /* Snapshot list, allow later requests */
5261 list_replace_init(&net_todo_list, &list);
5265 while (!list_empty(&list)) {
5266 struct net_device *dev
5267 = list_first_entry(&list, struct net_device, todo_list);
5268 list_del(&dev->todo_list);
5270 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5271 printk(KERN_ERR "network todo '%s' but state %d\n",
5272 dev->name, dev->reg_state);
5277 dev->reg_state = NETREG_UNREGISTERED;
5279 on_each_cpu(flush_backlog, dev, 1);
5281 netdev_wait_allrefs(dev);
5284 BUG_ON(atomic_read(&dev->refcnt));
5285 WARN_ON(dev->ip_ptr);
5286 WARN_ON(dev->ip6_ptr);
5287 WARN_ON(dev->dn_ptr);
5289 if (dev->destructor)
5290 dev->destructor(dev);
5292 /* Free network device */
5293 kobject_put(&dev->dev.kobj);
5298 * dev_txq_stats_fold - fold tx_queues stats
5299 * @dev: device to get statistics from
5300 * @stats: struct rtnl_link_stats64 to hold results
5302 void dev_txq_stats_fold(const struct net_device *dev,
5303 struct rtnl_link_stats64 *stats)
5305 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5307 struct netdev_queue *txq;
5309 for (i = 0; i < dev->num_tx_queues; i++) {
5310 txq = netdev_get_tx_queue(dev, i);
5311 spin_lock_bh(&txq->_xmit_lock);
5312 tx_bytes += txq->tx_bytes;
5313 tx_packets += txq->tx_packets;
5314 tx_dropped += txq->tx_dropped;
5315 spin_unlock_bh(&txq->_xmit_lock);
5317 if (tx_bytes || tx_packets || tx_dropped) {
5318 stats->tx_bytes = tx_bytes;
5319 stats->tx_packets = tx_packets;
5320 stats->tx_dropped = tx_dropped;
5323 EXPORT_SYMBOL(dev_txq_stats_fold);
5325 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5326 * fields in the same order, with only the type differing.
5328 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5329 const struct net_device_stats *netdev_stats)
5331 #if BITS_PER_LONG == 64
5332 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5333 memcpy(stats64, netdev_stats, sizeof(*stats64));
5335 size_t i, n = sizeof(*stats64) / sizeof(u64);
5336 const unsigned long *src = (const unsigned long *)netdev_stats;
5337 u64 *dst = (u64 *)stats64;
5339 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5340 sizeof(*stats64) / sizeof(u64));
5341 for (i = 0; i < n; i++)
5347 * dev_get_stats - get network device statistics
5348 * @dev: device to get statistics from
5349 * @storage: place to store stats
5351 * Get network statistics from device. Return @storage.
5352 * The device driver may provide its own method by setting
5353 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5354 * otherwise the internal statistics structure is used.
5356 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5357 struct rtnl_link_stats64 *storage)
5359 const struct net_device_ops *ops = dev->netdev_ops;
5361 if (ops->ndo_get_stats64) {
5362 memset(storage, 0, sizeof(*storage));
5363 return ops->ndo_get_stats64(dev, storage);
5365 if (ops->ndo_get_stats) {
5366 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5369 netdev_stats_to_stats64(storage, &dev->stats);
5370 dev_txq_stats_fold(dev, storage);
5373 EXPORT_SYMBOL(dev_get_stats);
5375 static void netdev_init_one_queue(struct net_device *dev,
5376 struct netdev_queue *queue,
5382 static void netdev_init_queues(struct net_device *dev)
5384 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5385 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5386 spin_lock_init(&dev->tx_global_lock);
5390 * alloc_netdev_mq - allocate network device
5391 * @sizeof_priv: size of private data to allocate space for
5392 * @name: device name format string
5393 * @setup: callback to initialize device
5394 * @queue_count: the number of subqueues to allocate
5396 * Allocates a struct net_device with private data area for driver use
5397 * and performs basic initialization. Also allocates subquue structs
5398 * for each queue on the device at the end of the netdevice.
5400 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5401 void (*setup)(struct net_device *), unsigned int queue_count)
5403 struct netdev_queue *tx;
5404 struct net_device *dev;
5406 struct net_device *p;
5408 struct netdev_rx_queue *rx;
5412 BUG_ON(strlen(name) >= sizeof(dev->name));
5414 alloc_size = sizeof(struct net_device);
5416 /* ensure 32-byte alignment of private area */
5417 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5418 alloc_size += sizeof_priv;
5420 /* ensure 32-byte alignment of whole construct */
5421 alloc_size += NETDEV_ALIGN - 1;
5423 p = kzalloc(alloc_size, GFP_KERNEL);
5425 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5429 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5431 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5437 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5439 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5444 atomic_set(&rx->count, queue_count);
5447 * Set a pointer to first element in the array which holds the
5450 for (i = 0; i < queue_count; i++)
5454 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5455 dev->padded = (char *)dev - (char *)p;
5457 if (dev_addr_init(dev))
5463 dev_net_set(dev, &init_net);
5466 dev->num_tx_queues = queue_count;
5467 dev->real_num_tx_queues = queue_count;
5471 dev->num_rx_queues = queue_count;
5474 dev->gso_max_size = GSO_MAX_SIZE;
5476 netdev_init_queues(dev);
5478 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5479 dev->ethtool_ntuple_list.count = 0;
5480 INIT_LIST_HEAD(&dev->napi_list);
5481 INIT_LIST_HEAD(&dev->unreg_list);
5482 INIT_LIST_HEAD(&dev->link_watch_list);
5483 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5485 strcpy(dev->name, name);
5498 EXPORT_SYMBOL(alloc_netdev_mq);
5501 * free_netdev - free network device
5504 * This function does the last stage of destroying an allocated device
5505 * interface. The reference to the device object is released.
5506 * If this is the last reference then it will be freed.
5508 void free_netdev(struct net_device *dev)
5510 struct napi_struct *p, *n;
5512 release_net(dev_net(dev));
5516 /* Flush device addresses */
5517 dev_addr_flush(dev);
5519 /* Clear ethtool n-tuple list */
5520 ethtool_ntuple_flush(dev);
5522 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5525 /* Compatibility with error handling in drivers */
5526 if (dev->reg_state == NETREG_UNINITIALIZED) {
5527 kfree((char *)dev - dev->padded);
5531 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5532 dev->reg_state = NETREG_RELEASED;
5534 /* will free via device release */
5535 put_device(&dev->dev);
5537 EXPORT_SYMBOL(free_netdev);
5540 * synchronize_net - Synchronize with packet receive processing
5542 * Wait for packets currently being received to be done.
5543 * Does not block later packets from starting.
5545 void synchronize_net(void)
5550 EXPORT_SYMBOL(synchronize_net);
5553 * unregister_netdevice_queue - remove device from the kernel
5557 * This function shuts down a device interface and removes it
5558 * from the kernel tables.
5559 * If head not NULL, device is queued to be unregistered later.
5561 * Callers must hold the rtnl semaphore. You may want
5562 * unregister_netdev() instead of this.
5565 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5570 list_move_tail(&dev->unreg_list, head);
5572 rollback_registered(dev);
5573 /* Finish processing unregister after unlock */
5577 EXPORT_SYMBOL(unregister_netdevice_queue);
5580 * unregister_netdevice_many - unregister many devices
5581 * @head: list of devices
5583 void unregister_netdevice_many(struct list_head *head)
5585 struct net_device *dev;
5587 if (!list_empty(head)) {
5588 rollback_registered_many(head);
5589 list_for_each_entry(dev, head, unreg_list)
5593 EXPORT_SYMBOL(unregister_netdevice_many);
5596 * unregister_netdev - remove device from the kernel
5599 * This function shuts down a device interface and removes it
5600 * from the kernel tables.
5602 * This is just a wrapper for unregister_netdevice that takes
5603 * the rtnl semaphore. In general you want to use this and not
5604 * unregister_netdevice.
5606 void unregister_netdev(struct net_device *dev)
5609 unregister_netdevice(dev);
5612 EXPORT_SYMBOL(unregister_netdev);
5615 * dev_change_net_namespace - move device to different nethost namespace
5617 * @net: network namespace
5618 * @pat: If not NULL name pattern to try if the current device name
5619 * is already taken in the destination network namespace.
5621 * This function shuts down a device interface and moves it
5622 * to a new network namespace. On success 0 is returned, on
5623 * a failure a netagive errno code is returned.
5625 * Callers must hold the rtnl semaphore.
5628 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5634 /* Don't allow namespace local devices to be moved. */
5636 if (dev->features & NETIF_F_NETNS_LOCAL)
5639 /* Ensure the device has been registrered */
5641 if (dev->reg_state != NETREG_REGISTERED)
5644 /* Get out if there is nothing todo */
5646 if (net_eq(dev_net(dev), net))
5649 /* Pick the destination device name, and ensure
5650 * we can use it in the destination network namespace.
5653 if (__dev_get_by_name(net, dev->name)) {
5654 /* We get here if we can't use the current device name */
5657 if (dev_get_valid_name(dev, pat, 1))
5662 * And now a mini version of register_netdevice unregister_netdevice.
5665 /* If device is running close it first. */
5668 /* And unlink it from device chain */
5670 unlist_netdevice(dev);
5674 /* Shutdown queueing discipline. */
5677 /* Notify protocols, that we are about to destroy
5678 this device. They should clean all the things.
5680 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5681 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5684 * Flush the unicast and multicast chains
5689 /* Actually switch the network namespace */
5690 dev_net_set(dev, net);
5692 /* If there is an ifindex conflict assign a new one */
5693 if (__dev_get_by_index(net, dev->ifindex)) {
5694 int iflink = (dev->iflink == dev->ifindex);
5695 dev->ifindex = dev_new_index(net);
5697 dev->iflink = dev->ifindex;
5700 /* Fixup kobjects */
5701 err = device_rename(&dev->dev, dev->name);
5704 /* Add the device back in the hashes */
5705 list_netdevice(dev);
5707 /* Notify protocols, that a new device appeared. */
5708 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5711 * Prevent userspace races by waiting until the network
5712 * device is fully setup before sending notifications.
5714 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5721 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5723 static int dev_cpu_callback(struct notifier_block *nfb,
5724 unsigned long action,
5727 struct sk_buff **list_skb;
5728 struct sk_buff *skb;
5729 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5730 struct softnet_data *sd, *oldsd;
5732 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5735 local_irq_disable();
5736 cpu = smp_processor_id();
5737 sd = &per_cpu(softnet_data, cpu);
5738 oldsd = &per_cpu(softnet_data, oldcpu);
5740 /* Find end of our completion_queue. */
5741 list_skb = &sd->completion_queue;
5743 list_skb = &(*list_skb)->next;
5744 /* Append completion queue from offline CPU. */
5745 *list_skb = oldsd->completion_queue;
5746 oldsd->completion_queue = NULL;
5748 /* Append output queue from offline CPU. */
5749 if (oldsd->output_queue) {
5750 *sd->output_queue_tailp = oldsd->output_queue;
5751 sd->output_queue_tailp = oldsd->output_queue_tailp;
5752 oldsd->output_queue = NULL;
5753 oldsd->output_queue_tailp = &oldsd->output_queue;
5756 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5759 /* Process offline CPU's input_pkt_queue */
5760 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5762 input_queue_head_incr(oldsd);
5764 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5766 input_queue_head_incr(oldsd);
5774 * netdev_increment_features - increment feature set by one
5775 * @all: current feature set
5776 * @one: new feature set
5777 * @mask: mask feature set
5779 * Computes a new feature set after adding a device with feature set
5780 * @one to the master device with current feature set @all. Will not
5781 * enable anything that is off in @mask. Returns the new feature set.
5783 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5786 /* If device needs checksumming, downgrade to it. */
5787 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5788 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5789 else if (mask & NETIF_F_ALL_CSUM) {
5790 /* If one device supports v4/v6 checksumming, set for all. */
5791 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5792 !(all & NETIF_F_GEN_CSUM)) {
5793 all &= ~NETIF_F_ALL_CSUM;
5794 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5797 /* If one device supports hw checksumming, set for all. */
5798 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5799 all &= ~NETIF_F_ALL_CSUM;
5800 all |= NETIF_F_HW_CSUM;
5804 one |= NETIF_F_ALL_CSUM;
5806 one |= all & NETIF_F_ONE_FOR_ALL;
5807 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5808 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5812 EXPORT_SYMBOL(netdev_increment_features);
5814 static struct hlist_head *netdev_create_hash(void)
5817 struct hlist_head *hash;
5819 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5821 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5822 INIT_HLIST_HEAD(&hash[i]);
5827 /* Initialize per network namespace state */
5828 static int __net_init netdev_init(struct net *net)
5830 INIT_LIST_HEAD(&net->dev_base_head);
5832 net->dev_name_head = netdev_create_hash();
5833 if (net->dev_name_head == NULL)
5836 net->dev_index_head = netdev_create_hash();
5837 if (net->dev_index_head == NULL)
5843 kfree(net->dev_name_head);
5849 * netdev_drivername - network driver for the device
5850 * @dev: network device
5851 * @buffer: buffer for resulting name
5852 * @len: size of buffer
5854 * Determine network driver for device.
5856 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5858 const struct device_driver *driver;
5859 const struct device *parent;
5861 if (len <= 0 || !buffer)
5865 parent = dev->dev.parent;
5870 driver = parent->driver;
5871 if (driver && driver->name)
5872 strlcpy(buffer, driver->name, len);
5876 static int __netdev_printk(const char *level, const struct net_device *dev,
5877 struct va_format *vaf)
5881 if (dev && dev->dev.parent)
5882 r = dev_printk(level, dev->dev.parent, "%s: %pV",
5883 netdev_name(dev), vaf);
5885 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
5887 r = printk("%s(NULL net_device): %pV", level, vaf);
5892 int netdev_printk(const char *level, const struct net_device *dev,
5893 const char *format, ...)
5895 struct va_format vaf;
5899 va_start(args, format);
5904 r = __netdev_printk(level, dev, &vaf);
5909 EXPORT_SYMBOL(netdev_printk);
5911 #define define_netdev_printk_level(func, level) \
5912 int func(const struct net_device *dev, const char *fmt, ...) \
5915 struct va_format vaf; \
5918 va_start(args, fmt); \
5923 r = __netdev_printk(level, dev, &vaf); \
5928 EXPORT_SYMBOL(func);
5930 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
5931 define_netdev_printk_level(netdev_alert, KERN_ALERT);
5932 define_netdev_printk_level(netdev_crit, KERN_CRIT);
5933 define_netdev_printk_level(netdev_err, KERN_ERR);
5934 define_netdev_printk_level(netdev_warn, KERN_WARNING);
5935 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
5936 define_netdev_printk_level(netdev_info, KERN_INFO);
5938 static void __net_exit netdev_exit(struct net *net)
5940 kfree(net->dev_name_head);
5941 kfree(net->dev_index_head);
5944 static struct pernet_operations __net_initdata netdev_net_ops = {
5945 .init = netdev_init,
5946 .exit = netdev_exit,
5949 static void __net_exit default_device_exit(struct net *net)
5951 struct net_device *dev, *aux;
5953 * Push all migratable network devices back to the
5954 * initial network namespace
5957 for_each_netdev_safe(net, dev, aux) {
5959 char fb_name[IFNAMSIZ];
5961 /* Ignore unmoveable devices (i.e. loopback) */
5962 if (dev->features & NETIF_F_NETNS_LOCAL)
5965 /* Leave virtual devices for the generic cleanup */
5966 if (dev->rtnl_link_ops)
5969 /* Push remaing network devices to init_net */
5970 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5971 err = dev_change_net_namespace(dev, &init_net, fb_name);
5973 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5974 __func__, dev->name, err);
5981 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5983 /* At exit all network devices most be removed from a network
5984 * namespace. Do this in the reverse order of registeration.
5985 * Do this across as many network namespaces as possible to
5986 * improve batching efficiency.
5988 struct net_device *dev;
5990 LIST_HEAD(dev_kill_list);
5993 list_for_each_entry(net, net_list, exit_list) {
5994 for_each_netdev_reverse(net, dev) {
5995 if (dev->rtnl_link_ops)
5996 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5998 unregister_netdevice_queue(dev, &dev_kill_list);
6001 unregister_netdevice_many(&dev_kill_list);
6005 static struct pernet_operations __net_initdata default_device_ops = {
6006 .exit = default_device_exit,
6007 .exit_batch = default_device_exit_batch,
6011 * Initialize the DEV module. At boot time this walks the device list and
6012 * unhooks any devices that fail to initialise (normally hardware not
6013 * present) and leaves us with a valid list of present and active devices.
6018 * This is called single threaded during boot, so no need
6019 * to take the rtnl semaphore.
6021 static int __init net_dev_init(void)
6023 int i, rc = -ENOMEM;
6025 BUG_ON(!dev_boot_phase);
6027 if (dev_proc_init())
6030 if (netdev_kobject_init())
6033 INIT_LIST_HEAD(&ptype_all);
6034 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6035 INIT_LIST_HEAD(&ptype_base[i]);
6037 if (register_pernet_subsys(&netdev_net_ops))
6041 * Initialise the packet receive queues.
6044 for_each_possible_cpu(i) {
6045 struct softnet_data *sd = &per_cpu(softnet_data, i);
6047 memset(sd, 0, sizeof(*sd));
6048 skb_queue_head_init(&sd->input_pkt_queue);
6049 skb_queue_head_init(&sd->process_queue);
6050 sd->completion_queue = NULL;
6051 INIT_LIST_HEAD(&sd->poll_list);
6052 sd->output_queue = NULL;
6053 sd->output_queue_tailp = &sd->output_queue;
6055 sd->csd.func = rps_trigger_softirq;
6061 sd->backlog.poll = process_backlog;
6062 sd->backlog.weight = weight_p;
6063 sd->backlog.gro_list = NULL;
6064 sd->backlog.gro_count = 0;
6069 /* The loopback device is special if any other network devices
6070 * is present in a network namespace the loopback device must
6071 * be present. Since we now dynamically allocate and free the
6072 * loopback device ensure this invariant is maintained by
6073 * keeping the loopback device as the first device on the
6074 * list of network devices. Ensuring the loopback devices
6075 * is the first device that appears and the last network device
6078 if (register_pernet_device(&loopback_net_ops))
6081 if (register_pernet_device(&default_device_ops))
6084 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6085 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6087 hotcpu_notifier(dev_cpu_callback, 0);
6095 subsys_initcall(net_dev_init);
6097 static int __init initialize_hashrnd(void)
6099 get_random_bytes(&hashrnd, sizeof(hashrnd));
6103 late_initcall_sync(initialize_hashrnd);