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 or the sk-reference
1906 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1908 static inline void skb_orphan_try(struct sk_buff *skb)
1910 struct sock *sk = skb->sk;
1912 if (sk && !skb_shinfo(skb)->tx_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_frag_list(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, poff;
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 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2290 ip_proto = ip->protocol;
2291 addr1 = (__force u32) ip->saddr;
2292 addr2 = (__force u32) ip->daddr;
2295 case __constant_htons(ETH_P_IPV6):
2296 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2299 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2300 ip_proto = ip6->nexthdr;
2301 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2302 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2310 poff = proto_ports_offset(ip_proto);
2312 nhoff += ihl * 4 + poff;
2313 if (pskb_may_pull(skb, nhoff + 4)) {
2314 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2315 if (ports.v16[1] < ports.v16[0])
2316 swap(ports.v16[0], ports.v16[1]);
2320 /* get a consistent hash (same value on both flow directions) */
2324 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2331 EXPORT_SYMBOL(__skb_get_rxhash);
2335 /* One global table that all flow-based protocols share. */
2336 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2337 EXPORT_SYMBOL(rps_sock_flow_table);
2340 * get_rps_cpu is called from netif_receive_skb and returns the target
2341 * CPU from the RPS map of the receiving queue for a given skb.
2342 * rcu_read_lock must be held on entry.
2344 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2345 struct rps_dev_flow **rflowp)
2347 struct netdev_rx_queue *rxqueue;
2348 struct rps_map *map;
2349 struct rps_dev_flow_table *flow_table;
2350 struct rps_sock_flow_table *sock_flow_table;
2354 if (skb_rx_queue_recorded(skb)) {
2355 u16 index = skb_get_rx_queue(skb);
2356 if (unlikely(index >= dev->num_rx_queues)) {
2357 WARN_ONCE(dev->num_rx_queues > 1, "%s received packet "
2358 "on queue %u, but number of RX queues is %u\n",
2359 dev->name, index, dev->num_rx_queues);
2362 rxqueue = dev->_rx + index;
2366 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2369 skb_reset_network_header(skb);
2370 if (!skb_get_rxhash(skb))
2373 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2374 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2375 if (flow_table && sock_flow_table) {
2377 struct rps_dev_flow *rflow;
2379 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2382 next_cpu = sock_flow_table->ents[skb->rxhash &
2383 sock_flow_table->mask];
2386 * If the desired CPU (where last recvmsg was done) is
2387 * different from current CPU (one in the rx-queue flow
2388 * table entry), switch if one of the following holds:
2389 * - Current CPU is unset (equal to RPS_NO_CPU).
2390 * - Current CPU is offline.
2391 * - The current CPU's queue tail has advanced beyond the
2392 * last packet that was enqueued using this table entry.
2393 * This guarantees that all previous packets for the flow
2394 * have been dequeued, thus preserving in order delivery.
2396 if (unlikely(tcpu != next_cpu) &&
2397 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2398 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2399 rflow->last_qtail)) >= 0)) {
2400 tcpu = rflow->cpu = next_cpu;
2401 if (tcpu != RPS_NO_CPU)
2402 rflow->last_qtail = per_cpu(softnet_data,
2403 tcpu).input_queue_head;
2405 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2412 map = rcu_dereference(rxqueue->rps_map);
2414 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2416 if (cpu_online(tcpu)) {
2426 /* Called from hardirq (IPI) context */
2427 static void rps_trigger_softirq(void *data)
2429 struct softnet_data *sd = data;
2431 ____napi_schedule(sd, &sd->backlog);
2435 #endif /* CONFIG_RPS */
2438 * Check if this softnet_data structure is another cpu one
2439 * If yes, queue it to our IPI list and return 1
2442 static int rps_ipi_queued(struct softnet_data *sd)
2445 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2448 sd->rps_ipi_next = mysd->rps_ipi_list;
2449 mysd->rps_ipi_list = sd;
2451 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2454 #endif /* CONFIG_RPS */
2459 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2460 * queue (may be a remote CPU queue).
2462 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2463 unsigned int *qtail)
2465 struct softnet_data *sd;
2466 unsigned long flags;
2468 sd = &per_cpu(softnet_data, cpu);
2470 local_irq_save(flags);
2473 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2474 if (skb_queue_len(&sd->input_pkt_queue)) {
2476 __skb_queue_tail(&sd->input_pkt_queue, skb);
2477 input_queue_tail_incr_save(sd, qtail);
2479 local_irq_restore(flags);
2480 return NET_RX_SUCCESS;
2483 /* Schedule NAPI for backlog device
2484 * We can use non atomic operation since we own the queue lock
2486 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2487 if (!rps_ipi_queued(sd))
2488 ____napi_schedule(sd, &sd->backlog);
2496 local_irq_restore(flags);
2503 * netif_rx - post buffer to the network code
2504 * @skb: buffer to post
2506 * This function receives a packet from a device driver and queues it for
2507 * the upper (protocol) levels to process. It always succeeds. The buffer
2508 * may be dropped during processing for congestion control or by the
2512 * NET_RX_SUCCESS (no congestion)
2513 * NET_RX_DROP (packet was dropped)
2517 int netif_rx(struct sk_buff *skb)
2521 /* if netpoll wants it, pretend we never saw it */
2522 if (netpoll_rx(skb))
2525 if (netdev_tstamp_prequeue)
2526 net_timestamp_check(skb);
2530 struct rps_dev_flow voidflow, *rflow = &voidflow;
2536 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2538 cpu = smp_processor_id();
2540 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2548 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2554 EXPORT_SYMBOL(netif_rx);
2556 int netif_rx_ni(struct sk_buff *skb)
2561 err = netif_rx(skb);
2562 if (local_softirq_pending())
2568 EXPORT_SYMBOL(netif_rx_ni);
2570 static void net_tx_action(struct softirq_action *h)
2572 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2574 if (sd->completion_queue) {
2575 struct sk_buff *clist;
2577 local_irq_disable();
2578 clist = sd->completion_queue;
2579 sd->completion_queue = NULL;
2583 struct sk_buff *skb = clist;
2584 clist = clist->next;
2586 WARN_ON(atomic_read(&skb->users));
2591 if (sd->output_queue) {
2594 local_irq_disable();
2595 head = sd->output_queue;
2596 sd->output_queue = NULL;
2597 sd->output_queue_tailp = &sd->output_queue;
2601 struct Qdisc *q = head;
2602 spinlock_t *root_lock;
2604 head = head->next_sched;
2606 root_lock = qdisc_lock(q);
2607 if (spin_trylock(root_lock)) {
2608 smp_mb__before_clear_bit();
2609 clear_bit(__QDISC_STATE_SCHED,
2612 spin_unlock(root_lock);
2614 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2616 __netif_reschedule(q);
2618 smp_mb__before_clear_bit();
2619 clear_bit(__QDISC_STATE_SCHED,
2627 static inline int deliver_skb(struct sk_buff *skb,
2628 struct packet_type *pt_prev,
2629 struct net_device *orig_dev)
2631 atomic_inc(&skb->users);
2632 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2635 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2636 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2637 /* This hook is defined here for ATM LANE */
2638 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2639 unsigned char *addr) __read_mostly;
2640 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2643 #ifdef CONFIG_NET_CLS_ACT
2644 /* TODO: Maybe we should just force sch_ingress to be compiled in
2645 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2646 * a compare and 2 stores extra right now if we dont have it on
2647 * but have CONFIG_NET_CLS_ACT
2648 * NOTE: This doesnt stop any functionality; if you dont have
2649 * the ingress scheduler, you just cant add policies on ingress.
2652 static int ing_filter(struct sk_buff *skb)
2654 struct net_device *dev = skb->dev;
2655 u32 ttl = G_TC_RTTL(skb->tc_verd);
2656 struct netdev_queue *rxq;
2657 int result = TC_ACT_OK;
2660 if (unlikely(MAX_RED_LOOP < ttl++)) {
2661 if (net_ratelimit())
2662 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2663 skb->skb_iif, dev->ifindex);
2667 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2668 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2670 rxq = &dev->rx_queue;
2673 if (q != &noop_qdisc) {
2674 spin_lock(qdisc_lock(q));
2675 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2676 result = qdisc_enqueue_root(skb, q);
2677 spin_unlock(qdisc_lock(q));
2683 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2684 struct packet_type **pt_prev,
2685 int *ret, struct net_device *orig_dev)
2687 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2691 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2695 switch (ing_filter(skb)) {
2709 * netif_nit_deliver - deliver received packets to network taps
2712 * This function is used to deliver incoming packets to network
2713 * taps. It should be used when the normal netif_receive_skb path
2714 * is bypassed, for example because of VLAN acceleration.
2716 void netif_nit_deliver(struct sk_buff *skb)
2718 struct packet_type *ptype;
2720 if (list_empty(&ptype_all))
2723 skb_reset_network_header(skb);
2724 skb_reset_transport_header(skb);
2725 skb->mac_len = skb->network_header - skb->mac_header;
2728 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2729 if (!ptype->dev || ptype->dev == skb->dev)
2730 deliver_skb(skb, ptype, skb->dev);
2736 * netdev_rx_handler_register - register receive handler
2737 * @dev: device to register a handler for
2738 * @rx_handler: receive handler to register
2739 * @rx_handler_data: data pointer that is used by rx handler
2741 * Register a receive hander for a device. This handler will then be
2742 * called from __netif_receive_skb. A negative errno code is returned
2745 * The caller must hold the rtnl_mutex.
2747 int netdev_rx_handler_register(struct net_device *dev,
2748 rx_handler_func_t *rx_handler,
2749 void *rx_handler_data)
2753 if (dev->rx_handler)
2756 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2757 rcu_assign_pointer(dev->rx_handler, rx_handler);
2761 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2764 * netdev_rx_handler_unregister - unregister receive handler
2765 * @dev: device to unregister a handler from
2767 * Unregister a receive hander from a device.
2769 * The caller must hold the rtnl_mutex.
2771 void netdev_rx_handler_unregister(struct net_device *dev)
2775 rcu_assign_pointer(dev->rx_handler, NULL);
2776 rcu_assign_pointer(dev->rx_handler_data, NULL);
2778 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2780 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2781 struct net_device *master)
2783 if (skb->pkt_type == PACKET_HOST) {
2784 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2786 memcpy(dest, master->dev_addr, ETH_ALEN);
2790 /* On bonding slaves other than the currently active slave, suppress
2791 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2792 * ARP on active-backup slaves with arp_validate enabled.
2794 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2796 struct net_device *dev = skb->dev;
2798 if (master->priv_flags & IFF_MASTER_ARPMON)
2799 dev->last_rx = jiffies;
2801 if ((master->priv_flags & IFF_MASTER_ALB) &&
2802 (master->priv_flags & IFF_BRIDGE_PORT)) {
2803 /* Do address unmangle. The local destination address
2804 * will be always the one master has. Provides the right
2805 * functionality in a bridge.
2807 skb_bond_set_mac_by_master(skb, master);
2810 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2811 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2812 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2815 if (master->priv_flags & IFF_MASTER_ALB) {
2816 if (skb->pkt_type != PACKET_BROADCAST &&
2817 skb->pkt_type != PACKET_MULTICAST)
2820 if (master->priv_flags & IFF_MASTER_8023AD &&
2821 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2828 EXPORT_SYMBOL(__skb_bond_should_drop);
2830 static int __netif_receive_skb(struct sk_buff *skb)
2832 struct packet_type *ptype, *pt_prev;
2833 rx_handler_func_t *rx_handler;
2834 struct net_device *orig_dev;
2835 struct net_device *master;
2836 struct net_device *null_or_orig;
2837 struct net_device *orig_or_bond;
2838 int ret = NET_RX_DROP;
2841 if (!netdev_tstamp_prequeue)
2842 net_timestamp_check(skb);
2844 if (vlan_tx_tag_present(skb))
2845 vlan_hwaccel_do_receive(skb);
2847 /* if we've gotten here through NAPI, check netpoll */
2848 if (netpoll_receive_skb(skb))
2852 skb->skb_iif = skb->dev->ifindex;
2855 * bonding note: skbs received on inactive slaves should only
2856 * be delivered to pkt handlers that are exact matches. Also
2857 * the deliver_no_wcard flag will be set. If packet handlers
2858 * are sensitive to duplicate packets these skbs will need to
2859 * be dropped at the handler. The vlan accel path may have
2860 * already set the deliver_no_wcard flag.
2862 null_or_orig = NULL;
2863 orig_dev = skb->dev;
2864 master = ACCESS_ONCE(orig_dev->master);
2865 if (skb->deliver_no_wcard)
2866 null_or_orig = orig_dev;
2868 if (skb_bond_should_drop(skb, master)) {
2869 skb->deliver_no_wcard = 1;
2870 null_or_orig = orig_dev; /* deliver only exact match */
2875 __this_cpu_inc(softnet_data.processed);
2876 skb_reset_network_header(skb);
2877 skb_reset_transport_header(skb);
2878 skb->mac_len = skb->network_header - skb->mac_header;
2884 #ifdef CONFIG_NET_CLS_ACT
2885 if (skb->tc_verd & TC_NCLS) {
2886 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2891 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2892 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2893 ptype->dev == orig_dev) {
2895 ret = deliver_skb(skb, pt_prev, orig_dev);
2900 #ifdef CONFIG_NET_CLS_ACT
2901 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2907 /* Handle special case of bridge or macvlan */
2908 rx_handler = rcu_dereference(skb->dev->rx_handler);
2911 ret = deliver_skb(skb, pt_prev, orig_dev);
2914 skb = rx_handler(skb);
2920 * Make sure frames received on VLAN interfaces stacked on
2921 * bonding interfaces still make their way to any base bonding
2922 * device that may have registered for a specific ptype. The
2923 * handler may have to adjust skb->dev and orig_dev.
2925 orig_or_bond = orig_dev;
2926 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2927 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2928 orig_or_bond = vlan_dev_real_dev(skb->dev);
2931 type = skb->protocol;
2932 list_for_each_entry_rcu(ptype,
2933 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2934 if (ptype->type == type && (ptype->dev == null_or_orig ||
2935 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2936 ptype->dev == orig_or_bond)) {
2938 ret = deliver_skb(skb, pt_prev, orig_dev);
2944 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2947 /* Jamal, now you will not able to escape explaining
2948 * me how you were going to use this. :-)
2959 * netif_receive_skb - process receive buffer from network
2960 * @skb: buffer to process
2962 * netif_receive_skb() is the main receive data processing function.
2963 * It always succeeds. The buffer may be dropped during processing
2964 * for congestion control or by the protocol layers.
2966 * This function may only be called from softirq context and interrupts
2967 * should be enabled.
2969 * Return values (usually ignored):
2970 * NET_RX_SUCCESS: no congestion
2971 * NET_RX_DROP: packet was dropped
2973 int netif_receive_skb(struct sk_buff *skb)
2975 if (netdev_tstamp_prequeue)
2976 net_timestamp_check(skb);
2978 if (skb_defer_rx_timestamp(skb))
2979 return NET_RX_SUCCESS;
2983 struct rps_dev_flow voidflow, *rflow = &voidflow;
2988 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2991 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2995 ret = __netif_receive_skb(skb);
3001 return __netif_receive_skb(skb);
3004 EXPORT_SYMBOL(netif_receive_skb);
3006 /* Network device is going away, flush any packets still pending
3007 * Called with irqs disabled.
3009 static void flush_backlog(void *arg)
3011 struct net_device *dev = arg;
3012 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3013 struct sk_buff *skb, *tmp;
3016 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3017 if (skb->dev == dev) {
3018 __skb_unlink(skb, &sd->input_pkt_queue);
3020 input_queue_head_incr(sd);
3025 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3026 if (skb->dev == dev) {
3027 __skb_unlink(skb, &sd->process_queue);
3029 input_queue_head_incr(sd);
3034 static int napi_gro_complete(struct sk_buff *skb)
3036 struct packet_type *ptype;
3037 __be16 type = skb->protocol;
3038 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3041 if (NAPI_GRO_CB(skb)->count == 1) {
3042 skb_shinfo(skb)->gso_size = 0;
3047 list_for_each_entry_rcu(ptype, head, list) {
3048 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3051 err = ptype->gro_complete(skb);
3057 WARN_ON(&ptype->list == head);
3059 return NET_RX_SUCCESS;
3063 return netif_receive_skb(skb);
3066 inline void napi_gro_flush(struct napi_struct *napi)
3068 struct sk_buff *skb, *next;
3070 for (skb = napi->gro_list; skb; skb = next) {
3073 napi_gro_complete(skb);
3076 napi->gro_count = 0;
3077 napi->gro_list = NULL;
3079 EXPORT_SYMBOL(napi_gro_flush);
3081 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3083 struct sk_buff **pp = NULL;
3084 struct packet_type *ptype;
3085 __be16 type = skb->protocol;
3086 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3089 enum gro_result ret;
3091 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3094 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3098 list_for_each_entry_rcu(ptype, head, list) {
3099 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3102 skb_set_network_header(skb, skb_gro_offset(skb));
3103 mac_len = skb->network_header - skb->mac_header;
3104 skb->mac_len = mac_len;
3105 NAPI_GRO_CB(skb)->same_flow = 0;
3106 NAPI_GRO_CB(skb)->flush = 0;
3107 NAPI_GRO_CB(skb)->free = 0;
3109 pp = ptype->gro_receive(&napi->gro_list, skb);
3114 if (&ptype->list == head)
3117 same_flow = NAPI_GRO_CB(skb)->same_flow;
3118 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3121 struct sk_buff *nskb = *pp;
3125 napi_gro_complete(nskb);
3132 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3136 NAPI_GRO_CB(skb)->count = 1;
3137 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3138 skb->next = napi->gro_list;
3139 napi->gro_list = skb;
3143 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3144 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3146 BUG_ON(skb->end - skb->tail < grow);
3148 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3151 skb->data_len -= grow;
3153 skb_shinfo(skb)->frags[0].page_offset += grow;
3154 skb_shinfo(skb)->frags[0].size -= grow;
3156 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3157 put_page(skb_shinfo(skb)->frags[0].page);
3158 memmove(skb_shinfo(skb)->frags,
3159 skb_shinfo(skb)->frags + 1,
3160 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3171 EXPORT_SYMBOL(dev_gro_receive);
3173 static inline gro_result_t
3174 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3178 for (p = napi->gro_list; p; p = p->next) {
3179 unsigned long diffs;
3181 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3182 diffs |= compare_ether_header(skb_mac_header(p),
3183 skb_gro_mac_header(skb));
3184 NAPI_GRO_CB(p)->same_flow = !diffs;
3185 NAPI_GRO_CB(p)->flush = 0;
3188 return dev_gro_receive(napi, skb);
3191 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3195 if (netif_receive_skb(skb))
3200 case GRO_MERGED_FREE:
3211 EXPORT_SYMBOL(napi_skb_finish);
3213 void skb_gro_reset_offset(struct sk_buff *skb)
3215 NAPI_GRO_CB(skb)->data_offset = 0;
3216 NAPI_GRO_CB(skb)->frag0 = NULL;
3217 NAPI_GRO_CB(skb)->frag0_len = 0;
3219 if (skb->mac_header == skb->tail &&
3220 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3221 NAPI_GRO_CB(skb)->frag0 =
3222 page_address(skb_shinfo(skb)->frags[0].page) +
3223 skb_shinfo(skb)->frags[0].page_offset;
3224 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3227 EXPORT_SYMBOL(skb_gro_reset_offset);
3229 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3231 skb_gro_reset_offset(skb);
3233 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3235 EXPORT_SYMBOL(napi_gro_receive);
3237 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3239 __skb_pull(skb, skb_headlen(skb));
3240 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3244 EXPORT_SYMBOL(napi_reuse_skb);
3246 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3248 struct sk_buff *skb = napi->skb;
3251 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3257 EXPORT_SYMBOL(napi_get_frags);
3259 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3265 skb->protocol = eth_type_trans(skb, skb->dev);
3267 if (ret == GRO_HELD)
3268 skb_gro_pull(skb, -ETH_HLEN);
3269 else if (netif_receive_skb(skb))
3274 case GRO_MERGED_FREE:
3275 napi_reuse_skb(napi, skb);
3284 EXPORT_SYMBOL(napi_frags_finish);
3286 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3288 struct sk_buff *skb = napi->skb;
3295 skb_reset_mac_header(skb);
3296 skb_gro_reset_offset(skb);
3298 off = skb_gro_offset(skb);
3299 hlen = off + sizeof(*eth);
3300 eth = skb_gro_header_fast(skb, off);
3301 if (skb_gro_header_hard(skb, hlen)) {
3302 eth = skb_gro_header_slow(skb, hlen, off);
3303 if (unlikely(!eth)) {
3304 napi_reuse_skb(napi, skb);
3310 skb_gro_pull(skb, sizeof(*eth));
3313 * This works because the only protocols we care about don't require
3314 * special handling. We'll fix it up properly at the end.
3316 skb->protocol = eth->h_proto;
3321 EXPORT_SYMBOL(napi_frags_skb);
3323 gro_result_t napi_gro_frags(struct napi_struct *napi)
3325 struct sk_buff *skb = napi_frags_skb(napi);
3330 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3332 EXPORT_SYMBOL(napi_gro_frags);
3335 * net_rps_action sends any pending IPI's for rps.
3336 * Note: called with local irq disabled, but exits with local irq enabled.
3338 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3341 struct softnet_data *remsd = sd->rps_ipi_list;
3344 sd->rps_ipi_list = NULL;
3348 /* Send pending IPI's to kick RPS processing on remote cpus. */
3350 struct softnet_data *next = remsd->rps_ipi_next;
3352 if (cpu_online(remsd->cpu))
3353 __smp_call_function_single(remsd->cpu,
3362 static int process_backlog(struct napi_struct *napi, int quota)
3365 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3368 /* Check if we have pending ipi, its better to send them now,
3369 * not waiting net_rx_action() end.
3371 if (sd->rps_ipi_list) {
3372 local_irq_disable();
3373 net_rps_action_and_irq_enable(sd);
3376 napi->weight = weight_p;
3377 local_irq_disable();
3378 while (work < quota) {
3379 struct sk_buff *skb;
3382 while ((skb = __skb_dequeue(&sd->process_queue))) {
3384 __netif_receive_skb(skb);
3385 local_irq_disable();
3386 input_queue_head_incr(sd);
3387 if (++work >= quota) {
3394 qlen = skb_queue_len(&sd->input_pkt_queue);
3396 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3397 &sd->process_queue);
3399 if (qlen < quota - work) {
3401 * Inline a custom version of __napi_complete().
3402 * only current cpu owns and manipulates this napi,
3403 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3404 * we can use a plain write instead of clear_bit(),
3405 * and we dont need an smp_mb() memory barrier.
3407 list_del(&napi->poll_list);
3410 quota = work + qlen;
3420 * __napi_schedule - schedule for receive
3421 * @n: entry to schedule
3423 * The entry's receive function will be scheduled to run
3425 void __napi_schedule(struct napi_struct *n)
3427 unsigned long flags;
3429 local_irq_save(flags);
3430 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3431 local_irq_restore(flags);
3433 EXPORT_SYMBOL(__napi_schedule);
3435 void __napi_complete(struct napi_struct *n)
3437 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3438 BUG_ON(n->gro_list);
3440 list_del(&n->poll_list);
3441 smp_mb__before_clear_bit();
3442 clear_bit(NAPI_STATE_SCHED, &n->state);
3444 EXPORT_SYMBOL(__napi_complete);
3446 void napi_complete(struct napi_struct *n)
3448 unsigned long flags;
3451 * don't let napi dequeue from the cpu poll list
3452 * just in case its running on a different cpu
3454 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3458 local_irq_save(flags);
3460 local_irq_restore(flags);
3462 EXPORT_SYMBOL(napi_complete);
3464 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3465 int (*poll)(struct napi_struct *, int), int weight)
3467 INIT_LIST_HEAD(&napi->poll_list);
3468 napi->gro_count = 0;
3469 napi->gro_list = NULL;
3472 napi->weight = weight;
3473 list_add(&napi->dev_list, &dev->napi_list);
3475 #ifdef CONFIG_NETPOLL
3476 spin_lock_init(&napi->poll_lock);
3477 napi->poll_owner = -1;
3479 set_bit(NAPI_STATE_SCHED, &napi->state);
3481 EXPORT_SYMBOL(netif_napi_add);
3483 void netif_napi_del(struct napi_struct *napi)
3485 struct sk_buff *skb, *next;
3487 list_del_init(&napi->dev_list);
3488 napi_free_frags(napi);
3490 for (skb = napi->gro_list; skb; skb = next) {
3496 napi->gro_list = NULL;
3497 napi->gro_count = 0;
3499 EXPORT_SYMBOL(netif_napi_del);
3501 static void net_rx_action(struct softirq_action *h)
3503 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3504 unsigned long time_limit = jiffies + 2;
3505 int budget = netdev_budget;
3508 local_irq_disable();
3510 while (!list_empty(&sd->poll_list)) {
3511 struct napi_struct *n;
3514 /* If softirq window is exhuasted then punt.
3515 * Allow this to run for 2 jiffies since which will allow
3516 * an average latency of 1.5/HZ.
3518 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3523 /* Even though interrupts have been re-enabled, this
3524 * access is safe because interrupts can only add new
3525 * entries to the tail of this list, and only ->poll()
3526 * calls can remove this head entry from the list.
3528 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3530 have = netpoll_poll_lock(n);
3534 /* This NAPI_STATE_SCHED test is for avoiding a race
3535 * with netpoll's poll_napi(). Only the entity which
3536 * obtains the lock and sees NAPI_STATE_SCHED set will
3537 * actually make the ->poll() call. Therefore we avoid
3538 * accidently calling ->poll() when NAPI is not scheduled.
3541 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3542 work = n->poll(n, weight);
3546 WARN_ON_ONCE(work > weight);
3550 local_irq_disable();
3552 /* Drivers must not modify the NAPI state if they
3553 * consume the entire weight. In such cases this code
3554 * still "owns" the NAPI instance and therefore can
3555 * move the instance around on the list at-will.
3557 if (unlikely(work == weight)) {
3558 if (unlikely(napi_disable_pending(n))) {
3561 local_irq_disable();
3563 list_move_tail(&n->poll_list, &sd->poll_list);
3566 netpoll_poll_unlock(have);
3569 net_rps_action_and_irq_enable(sd);
3571 #ifdef CONFIG_NET_DMA
3573 * There may not be any more sk_buffs coming right now, so push
3574 * any pending DMA copies to hardware
3576 dma_issue_pending_all();
3583 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3587 static gifconf_func_t *gifconf_list[NPROTO];
3590 * register_gifconf - register a SIOCGIF handler
3591 * @family: Address family
3592 * @gifconf: Function handler
3594 * Register protocol dependent address dumping routines. The handler
3595 * that is passed must not be freed or reused until it has been replaced
3596 * by another handler.
3598 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3600 if (family >= NPROTO)
3602 gifconf_list[family] = gifconf;
3605 EXPORT_SYMBOL(register_gifconf);
3609 * Map an interface index to its name (SIOCGIFNAME)
3613 * We need this ioctl for efficient implementation of the
3614 * if_indextoname() function required by the IPv6 API. Without
3615 * it, we would have to search all the interfaces to find a
3619 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3621 struct net_device *dev;
3625 * Fetch the caller's info block.
3628 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3632 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3638 strcpy(ifr.ifr_name, dev->name);
3641 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3647 * Perform a SIOCGIFCONF call. This structure will change
3648 * size eventually, and there is nothing I can do about it.
3649 * Thus we will need a 'compatibility mode'.
3652 static int dev_ifconf(struct net *net, char __user *arg)
3655 struct net_device *dev;
3662 * Fetch the caller's info block.
3665 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3672 * Loop over the interfaces, and write an info block for each.
3676 for_each_netdev(net, dev) {
3677 for (i = 0; i < NPROTO; i++) {
3678 if (gifconf_list[i]) {
3681 done = gifconf_list[i](dev, NULL, 0);
3683 done = gifconf_list[i](dev, pos + total,
3693 * All done. Write the updated control block back to the caller.
3695 ifc.ifc_len = total;
3698 * Both BSD and Solaris return 0 here, so we do too.
3700 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3703 #ifdef CONFIG_PROC_FS
3705 * This is invoked by the /proc filesystem handler to display a device
3708 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3711 struct net *net = seq_file_net(seq);
3713 struct net_device *dev;
3717 return SEQ_START_TOKEN;
3720 for_each_netdev_rcu(net, dev)
3727 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3729 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3730 first_net_device(seq_file_net(seq)) :
3731 next_net_device((struct net_device *)v);
3734 return rcu_dereference(dev);
3737 void dev_seq_stop(struct seq_file *seq, void *v)
3743 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3745 struct rtnl_link_stats64 temp;
3746 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3748 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3749 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3750 dev->name, stats->rx_bytes, stats->rx_packets,
3752 stats->rx_dropped + stats->rx_missed_errors,
3753 stats->rx_fifo_errors,
3754 stats->rx_length_errors + stats->rx_over_errors +
3755 stats->rx_crc_errors + stats->rx_frame_errors,
3756 stats->rx_compressed, stats->multicast,
3757 stats->tx_bytes, stats->tx_packets,
3758 stats->tx_errors, stats->tx_dropped,
3759 stats->tx_fifo_errors, stats->collisions,
3760 stats->tx_carrier_errors +
3761 stats->tx_aborted_errors +
3762 stats->tx_window_errors +
3763 stats->tx_heartbeat_errors,
3764 stats->tx_compressed);
3768 * Called from the PROCfs module. This now uses the new arbitrary sized
3769 * /proc/net interface to create /proc/net/dev
3771 static int dev_seq_show(struct seq_file *seq, void *v)
3773 if (v == SEQ_START_TOKEN)
3774 seq_puts(seq, "Inter-| Receive "
3776 " face |bytes packets errs drop fifo frame "
3777 "compressed multicast|bytes packets errs "
3778 "drop fifo colls carrier compressed\n");
3780 dev_seq_printf_stats(seq, v);
3784 static struct softnet_data *softnet_get_online(loff_t *pos)
3786 struct softnet_data *sd = NULL;
3788 while (*pos < nr_cpu_ids)
3789 if (cpu_online(*pos)) {
3790 sd = &per_cpu(softnet_data, *pos);
3797 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3799 return softnet_get_online(pos);
3802 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3805 return softnet_get_online(pos);
3808 static void softnet_seq_stop(struct seq_file *seq, void *v)
3812 static int softnet_seq_show(struct seq_file *seq, void *v)
3814 struct softnet_data *sd = v;
3816 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3817 sd->processed, sd->dropped, sd->time_squeeze, 0,
3818 0, 0, 0, 0, /* was fastroute */
3819 sd->cpu_collision, sd->received_rps);
3823 static const struct seq_operations dev_seq_ops = {
3824 .start = dev_seq_start,
3825 .next = dev_seq_next,
3826 .stop = dev_seq_stop,
3827 .show = dev_seq_show,
3830 static int dev_seq_open(struct inode *inode, struct file *file)
3832 return seq_open_net(inode, file, &dev_seq_ops,
3833 sizeof(struct seq_net_private));
3836 static const struct file_operations dev_seq_fops = {
3837 .owner = THIS_MODULE,
3838 .open = dev_seq_open,
3840 .llseek = seq_lseek,
3841 .release = seq_release_net,
3844 static const struct seq_operations softnet_seq_ops = {
3845 .start = softnet_seq_start,
3846 .next = softnet_seq_next,
3847 .stop = softnet_seq_stop,
3848 .show = softnet_seq_show,
3851 static int softnet_seq_open(struct inode *inode, struct file *file)
3853 return seq_open(file, &softnet_seq_ops);
3856 static const struct file_operations softnet_seq_fops = {
3857 .owner = THIS_MODULE,
3858 .open = softnet_seq_open,
3860 .llseek = seq_lseek,
3861 .release = seq_release,
3864 static void *ptype_get_idx(loff_t pos)
3866 struct packet_type *pt = NULL;
3870 list_for_each_entry_rcu(pt, &ptype_all, list) {
3876 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3877 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3886 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3890 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3893 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3895 struct packet_type *pt;
3896 struct list_head *nxt;
3900 if (v == SEQ_START_TOKEN)
3901 return ptype_get_idx(0);
3904 nxt = pt->list.next;
3905 if (pt->type == htons(ETH_P_ALL)) {
3906 if (nxt != &ptype_all)
3909 nxt = ptype_base[0].next;
3911 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3913 while (nxt == &ptype_base[hash]) {
3914 if (++hash >= PTYPE_HASH_SIZE)
3916 nxt = ptype_base[hash].next;
3919 return list_entry(nxt, struct packet_type, list);
3922 static void ptype_seq_stop(struct seq_file *seq, void *v)
3928 static int ptype_seq_show(struct seq_file *seq, void *v)
3930 struct packet_type *pt = v;
3932 if (v == SEQ_START_TOKEN)
3933 seq_puts(seq, "Type Device Function\n");
3934 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3935 if (pt->type == htons(ETH_P_ALL))
3936 seq_puts(seq, "ALL ");
3938 seq_printf(seq, "%04x", ntohs(pt->type));
3940 seq_printf(seq, " %-8s %pF\n",
3941 pt->dev ? pt->dev->name : "", pt->func);
3947 static const struct seq_operations ptype_seq_ops = {
3948 .start = ptype_seq_start,
3949 .next = ptype_seq_next,
3950 .stop = ptype_seq_stop,
3951 .show = ptype_seq_show,
3954 static int ptype_seq_open(struct inode *inode, struct file *file)
3956 return seq_open_net(inode, file, &ptype_seq_ops,
3957 sizeof(struct seq_net_private));
3960 static const struct file_operations ptype_seq_fops = {
3961 .owner = THIS_MODULE,
3962 .open = ptype_seq_open,
3964 .llseek = seq_lseek,
3965 .release = seq_release_net,
3969 static int __net_init dev_proc_net_init(struct net *net)
3973 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3975 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3977 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3980 if (wext_proc_init(net))
3986 proc_net_remove(net, "ptype");
3988 proc_net_remove(net, "softnet_stat");
3990 proc_net_remove(net, "dev");
3994 static void __net_exit dev_proc_net_exit(struct net *net)
3996 wext_proc_exit(net);
3998 proc_net_remove(net, "ptype");
3999 proc_net_remove(net, "softnet_stat");
4000 proc_net_remove(net, "dev");
4003 static struct pernet_operations __net_initdata dev_proc_ops = {
4004 .init = dev_proc_net_init,
4005 .exit = dev_proc_net_exit,
4008 static int __init dev_proc_init(void)
4010 return register_pernet_subsys(&dev_proc_ops);
4013 #define dev_proc_init() 0
4014 #endif /* CONFIG_PROC_FS */
4018 * netdev_set_master - set up master/slave pair
4019 * @slave: slave device
4020 * @master: new master device
4022 * Changes the master device of the slave. Pass %NULL to break the
4023 * bonding. The caller must hold the RTNL semaphore. On a failure
4024 * a negative errno code is returned. On success the reference counts
4025 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4026 * function returns zero.
4028 int netdev_set_master(struct net_device *slave, struct net_device *master)
4030 struct net_device *old = slave->master;
4040 slave->master = master;
4047 slave->flags |= IFF_SLAVE;
4049 slave->flags &= ~IFF_SLAVE;
4051 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4054 EXPORT_SYMBOL(netdev_set_master);
4056 static void dev_change_rx_flags(struct net_device *dev, int flags)
4058 const struct net_device_ops *ops = dev->netdev_ops;
4060 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4061 ops->ndo_change_rx_flags(dev, flags);
4064 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4066 unsigned short old_flags = dev->flags;
4072 dev->flags |= IFF_PROMISC;
4073 dev->promiscuity += inc;
4074 if (dev->promiscuity == 0) {
4077 * If inc causes overflow, untouch promisc and return error.
4080 dev->flags &= ~IFF_PROMISC;
4082 dev->promiscuity -= inc;
4083 printk(KERN_WARNING "%s: promiscuity touches roof, "
4084 "set promiscuity failed, promiscuity feature "
4085 "of device might be broken.\n", dev->name);
4089 if (dev->flags != old_flags) {
4090 printk(KERN_INFO "device %s %s promiscuous mode\n",
4091 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4093 if (audit_enabled) {
4094 current_uid_gid(&uid, &gid);
4095 audit_log(current->audit_context, GFP_ATOMIC,
4096 AUDIT_ANOM_PROMISCUOUS,
4097 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4098 dev->name, (dev->flags & IFF_PROMISC),
4099 (old_flags & IFF_PROMISC),
4100 audit_get_loginuid(current),
4102 audit_get_sessionid(current));
4105 dev_change_rx_flags(dev, IFF_PROMISC);
4111 * dev_set_promiscuity - update promiscuity count on a device
4115 * Add or remove promiscuity from a device. While the count in the device
4116 * remains above zero the interface remains promiscuous. Once it hits zero
4117 * the device reverts back to normal filtering operation. A negative inc
4118 * value is used to drop promiscuity on the device.
4119 * Return 0 if successful or a negative errno code on error.
4121 int dev_set_promiscuity(struct net_device *dev, int inc)
4123 unsigned short old_flags = dev->flags;
4126 err = __dev_set_promiscuity(dev, inc);
4129 if (dev->flags != old_flags)
4130 dev_set_rx_mode(dev);
4133 EXPORT_SYMBOL(dev_set_promiscuity);
4136 * dev_set_allmulti - update allmulti count on a device
4140 * Add or remove reception of all multicast frames to a device. While the
4141 * count in the device remains above zero the interface remains listening
4142 * to all interfaces. Once it hits zero the device reverts back to normal
4143 * filtering operation. A negative @inc value is used to drop the counter
4144 * when releasing a resource needing all multicasts.
4145 * Return 0 if successful or a negative errno code on error.
4148 int dev_set_allmulti(struct net_device *dev, int inc)
4150 unsigned short old_flags = dev->flags;
4154 dev->flags |= IFF_ALLMULTI;
4155 dev->allmulti += inc;
4156 if (dev->allmulti == 0) {
4159 * If inc causes overflow, untouch allmulti and return error.
4162 dev->flags &= ~IFF_ALLMULTI;
4164 dev->allmulti -= inc;
4165 printk(KERN_WARNING "%s: allmulti touches roof, "
4166 "set allmulti failed, allmulti feature of "
4167 "device might be broken.\n", dev->name);
4171 if (dev->flags ^ old_flags) {
4172 dev_change_rx_flags(dev, IFF_ALLMULTI);
4173 dev_set_rx_mode(dev);
4177 EXPORT_SYMBOL(dev_set_allmulti);
4180 * Upload unicast and multicast address lists to device and
4181 * configure RX filtering. When the device doesn't support unicast
4182 * filtering it is put in promiscuous mode while unicast addresses
4185 void __dev_set_rx_mode(struct net_device *dev)
4187 const struct net_device_ops *ops = dev->netdev_ops;
4189 /* dev_open will call this function so the list will stay sane. */
4190 if (!(dev->flags&IFF_UP))
4193 if (!netif_device_present(dev))
4196 if (ops->ndo_set_rx_mode)
4197 ops->ndo_set_rx_mode(dev);
4199 /* Unicast addresses changes may only happen under the rtnl,
4200 * therefore calling __dev_set_promiscuity here is safe.
4202 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4203 __dev_set_promiscuity(dev, 1);
4204 dev->uc_promisc = 1;
4205 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4206 __dev_set_promiscuity(dev, -1);
4207 dev->uc_promisc = 0;
4210 if (ops->ndo_set_multicast_list)
4211 ops->ndo_set_multicast_list(dev);
4215 void dev_set_rx_mode(struct net_device *dev)
4217 netif_addr_lock_bh(dev);
4218 __dev_set_rx_mode(dev);
4219 netif_addr_unlock_bh(dev);
4223 * dev_get_flags - get flags reported to userspace
4226 * Get the combination of flag bits exported through APIs to userspace.
4228 unsigned dev_get_flags(const struct net_device *dev)
4232 flags = (dev->flags & ~(IFF_PROMISC |
4237 (dev->gflags & (IFF_PROMISC |
4240 if (netif_running(dev)) {
4241 if (netif_oper_up(dev))
4242 flags |= IFF_RUNNING;
4243 if (netif_carrier_ok(dev))
4244 flags |= IFF_LOWER_UP;
4245 if (netif_dormant(dev))
4246 flags |= IFF_DORMANT;
4251 EXPORT_SYMBOL(dev_get_flags);
4253 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4255 int old_flags = dev->flags;
4261 * Set the flags on our device.
4264 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4265 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4267 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4271 * Load in the correct multicast list now the flags have changed.
4274 if ((old_flags ^ flags) & IFF_MULTICAST)
4275 dev_change_rx_flags(dev, IFF_MULTICAST);
4277 dev_set_rx_mode(dev);
4280 * Have we downed the interface. We handle IFF_UP ourselves
4281 * according to user attempts to set it, rather than blindly
4286 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4287 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4290 dev_set_rx_mode(dev);
4293 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4294 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4296 dev->gflags ^= IFF_PROMISC;
4297 dev_set_promiscuity(dev, inc);
4300 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4301 is important. Some (broken) drivers set IFF_PROMISC, when
4302 IFF_ALLMULTI is requested not asking us and not reporting.
4304 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4305 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4307 dev->gflags ^= IFF_ALLMULTI;
4308 dev_set_allmulti(dev, inc);
4314 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4316 unsigned int changes = dev->flags ^ old_flags;
4318 if (changes & IFF_UP) {
4319 if (dev->flags & IFF_UP)
4320 call_netdevice_notifiers(NETDEV_UP, dev);
4322 call_netdevice_notifiers(NETDEV_DOWN, dev);
4325 if (dev->flags & IFF_UP &&
4326 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4327 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4331 * dev_change_flags - change device settings
4333 * @flags: device state flags
4335 * Change settings on device based state flags. The flags are
4336 * in the userspace exported format.
4338 int dev_change_flags(struct net_device *dev, unsigned flags)
4341 int old_flags = dev->flags;
4343 ret = __dev_change_flags(dev, flags);
4347 changes = old_flags ^ dev->flags;
4349 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4351 __dev_notify_flags(dev, old_flags);
4354 EXPORT_SYMBOL(dev_change_flags);
4357 * dev_set_mtu - Change maximum transfer unit
4359 * @new_mtu: new transfer unit
4361 * Change the maximum transfer size of the network device.
4363 int dev_set_mtu(struct net_device *dev, int new_mtu)
4365 const struct net_device_ops *ops = dev->netdev_ops;
4368 if (new_mtu == dev->mtu)
4371 /* MTU must be positive. */
4375 if (!netif_device_present(dev))
4379 if (ops->ndo_change_mtu)
4380 err = ops->ndo_change_mtu(dev, new_mtu);
4384 if (!err && dev->flags & IFF_UP)
4385 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4388 EXPORT_SYMBOL(dev_set_mtu);
4391 * dev_set_mac_address - Change Media Access Control Address
4395 * Change the hardware (MAC) address of the device
4397 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4399 const struct net_device_ops *ops = dev->netdev_ops;
4402 if (!ops->ndo_set_mac_address)
4404 if (sa->sa_family != dev->type)
4406 if (!netif_device_present(dev))
4408 err = ops->ndo_set_mac_address(dev, sa);
4410 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4413 EXPORT_SYMBOL(dev_set_mac_address);
4416 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4418 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4421 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4427 case SIOCGIFFLAGS: /* Get interface flags */
4428 ifr->ifr_flags = (short) dev_get_flags(dev);
4431 case SIOCGIFMETRIC: /* Get the metric on the interface
4432 (currently unused) */
4433 ifr->ifr_metric = 0;
4436 case SIOCGIFMTU: /* Get the MTU of a device */
4437 ifr->ifr_mtu = dev->mtu;
4442 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4444 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4445 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4446 ifr->ifr_hwaddr.sa_family = dev->type;
4454 ifr->ifr_map.mem_start = dev->mem_start;
4455 ifr->ifr_map.mem_end = dev->mem_end;
4456 ifr->ifr_map.base_addr = dev->base_addr;
4457 ifr->ifr_map.irq = dev->irq;
4458 ifr->ifr_map.dma = dev->dma;
4459 ifr->ifr_map.port = dev->if_port;
4463 ifr->ifr_ifindex = dev->ifindex;
4467 ifr->ifr_qlen = dev->tx_queue_len;
4471 /* dev_ioctl() should ensure this case
4483 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4485 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4488 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4489 const struct net_device_ops *ops;
4494 ops = dev->netdev_ops;
4497 case SIOCSIFFLAGS: /* Set interface flags */
4498 return dev_change_flags(dev, ifr->ifr_flags);
4500 case SIOCSIFMETRIC: /* Set the metric on the interface
4501 (currently unused) */
4504 case SIOCSIFMTU: /* Set the MTU of a device */
4505 return dev_set_mtu(dev, ifr->ifr_mtu);
4508 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4510 case SIOCSIFHWBROADCAST:
4511 if (ifr->ifr_hwaddr.sa_family != dev->type)
4513 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4514 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4515 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4519 if (ops->ndo_set_config) {
4520 if (!netif_device_present(dev))
4522 return ops->ndo_set_config(dev, &ifr->ifr_map);
4527 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4528 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4530 if (!netif_device_present(dev))
4532 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4535 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4536 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4538 if (!netif_device_present(dev))
4540 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4543 if (ifr->ifr_qlen < 0)
4545 dev->tx_queue_len = ifr->ifr_qlen;
4549 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4550 return dev_change_name(dev, ifr->ifr_newname);
4553 * Unknown or private ioctl
4556 if ((cmd >= SIOCDEVPRIVATE &&
4557 cmd <= SIOCDEVPRIVATE + 15) ||
4558 cmd == SIOCBONDENSLAVE ||
4559 cmd == SIOCBONDRELEASE ||
4560 cmd == SIOCBONDSETHWADDR ||
4561 cmd == SIOCBONDSLAVEINFOQUERY ||
4562 cmd == SIOCBONDINFOQUERY ||
4563 cmd == SIOCBONDCHANGEACTIVE ||
4564 cmd == SIOCGMIIPHY ||
4565 cmd == SIOCGMIIREG ||
4566 cmd == SIOCSMIIREG ||
4567 cmd == SIOCBRADDIF ||
4568 cmd == SIOCBRDELIF ||
4569 cmd == SIOCSHWTSTAMP ||
4570 cmd == SIOCWANDEV) {
4572 if (ops->ndo_do_ioctl) {
4573 if (netif_device_present(dev))
4574 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4586 * This function handles all "interface"-type I/O control requests. The actual
4587 * 'doing' part of this is dev_ifsioc above.
4591 * dev_ioctl - network device ioctl
4592 * @net: the applicable net namespace
4593 * @cmd: command to issue
4594 * @arg: pointer to a struct ifreq in user space
4596 * Issue ioctl functions to devices. This is normally called by the
4597 * user space syscall interfaces but can sometimes be useful for
4598 * other purposes. The return value is the return from the syscall if
4599 * positive or a negative errno code on error.
4602 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4608 /* One special case: SIOCGIFCONF takes ifconf argument
4609 and requires shared lock, because it sleeps writing
4613 if (cmd == SIOCGIFCONF) {
4615 ret = dev_ifconf(net, (char __user *) arg);
4619 if (cmd == SIOCGIFNAME)
4620 return dev_ifname(net, (struct ifreq __user *)arg);
4622 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4625 ifr.ifr_name[IFNAMSIZ-1] = 0;
4627 colon = strchr(ifr.ifr_name, ':');
4632 * See which interface the caller is talking about.
4637 * These ioctl calls:
4638 * - can be done by all.
4639 * - atomic and do not require locking.
4650 dev_load(net, ifr.ifr_name);
4652 ret = dev_ifsioc_locked(net, &ifr, cmd);
4657 if (copy_to_user(arg, &ifr,
4658 sizeof(struct ifreq)))
4664 dev_load(net, ifr.ifr_name);
4666 ret = dev_ethtool(net, &ifr);
4671 if (copy_to_user(arg, &ifr,
4672 sizeof(struct ifreq)))
4678 * These ioctl calls:
4679 * - require superuser power.
4680 * - require strict serialization.
4686 if (!capable(CAP_NET_ADMIN))
4688 dev_load(net, ifr.ifr_name);
4690 ret = dev_ifsioc(net, &ifr, cmd);
4695 if (copy_to_user(arg, &ifr,
4696 sizeof(struct ifreq)))
4702 * These ioctl calls:
4703 * - require superuser power.
4704 * - require strict serialization.
4705 * - do not return a value
4715 case SIOCSIFHWBROADCAST:
4718 case SIOCBONDENSLAVE:
4719 case SIOCBONDRELEASE:
4720 case SIOCBONDSETHWADDR:
4721 case SIOCBONDCHANGEACTIVE:
4725 if (!capable(CAP_NET_ADMIN))
4728 case SIOCBONDSLAVEINFOQUERY:
4729 case SIOCBONDINFOQUERY:
4730 dev_load(net, ifr.ifr_name);
4732 ret = dev_ifsioc(net, &ifr, cmd);
4737 /* Get the per device memory space. We can add this but
4738 * currently do not support it */
4740 /* Set the per device memory buffer space.
4741 * Not applicable in our case */
4746 * Unknown or private ioctl.
4749 if (cmd == SIOCWANDEV ||
4750 (cmd >= SIOCDEVPRIVATE &&
4751 cmd <= SIOCDEVPRIVATE + 15)) {
4752 dev_load(net, ifr.ifr_name);
4754 ret = dev_ifsioc(net, &ifr, cmd);
4756 if (!ret && copy_to_user(arg, &ifr,
4757 sizeof(struct ifreq)))
4761 /* Take care of Wireless Extensions */
4762 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4763 return wext_handle_ioctl(net, &ifr, cmd, arg);
4770 * dev_new_index - allocate an ifindex
4771 * @net: the applicable net namespace
4773 * Returns a suitable unique value for a new device interface
4774 * number. The caller must hold the rtnl semaphore or the
4775 * dev_base_lock to be sure it remains unique.
4777 static int dev_new_index(struct net *net)
4783 if (!__dev_get_by_index(net, ifindex))
4788 /* Delayed registration/unregisteration */
4789 static LIST_HEAD(net_todo_list);
4791 static void net_set_todo(struct net_device *dev)
4793 list_add_tail(&dev->todo_list, &net_todo_list);
4796 static void rollback_registered_many(struct list_head *head)
4798 struct net_device *dev, *tmp;
4800 BUG_ON(dev_boot_phase);
4803 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4804 /* Some devices call without registering
4805 * for initialization unwind. Remove those
4806 * devices and proceed with the remaining.
4808 if (dev->reg_state == NETREG_UNINITIALIZED) {
4809 pr_debug("unregister_netdevice: device %s/%p never "
4810 "was registered\n", dev->name, dev);
4813 list_del(&dev->unreg_list);
4817 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4819 /* If device is running, close it first. */
4822 /* And unlink it from device chain. */
4823 unlist_netdevice(dev);
4825 dev->reg_state = NETREG_UNREGISTERING;
4830 list_for_each_entry(dev, head, unreg_list) {
4831 /* Shutdown queueing discipline. */
4835 /* Notify protocols, that we are about to destroy
4836 this device. They should clean all the things.
4838 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4840 if (!dev->rtnl_link_ops ||
4841 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4842 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4845 * Flush the unicast and multicast chains
4850 if (dev->netdev_ops->ndo_uninit)
4851 dev->netdev_ops->ndo_uninit(dev);
4853 /* Notifier chain MUST detach us from master device. */
4854 WARN_ON(dev->master);
4856 /* Remove entries from kobject tree */
4857 netdev_unregister_kobject(dev);
4860 /* Process any work delayed until the end of the batch */
4861 dev = list_first_entry(head, struct net_device, unreg_list);
4862 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4866 list_for_each_entry(dev, head, unreg_list)
4870 static void rollback_registered(struct net_device *dev)
4874 list_add(&dev->unreg_list, &single);
4875 rollback_registered_many(&single);
4878 static void __netdev_init_queue_locks_one(struct net_device *dev,
4879 struct netdev_queue *dev_queue,
4882 spin_lock_init(&dev_queue->_xmit_lock);
4883 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4884 dev_queue->xmit_lock_owner = -1;
4887 static void netdev_init_queue_locks(struct net_device *dev)
4889 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4890 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4893 unsigned long netdev_fix_features(unsigned long features, const char *name)
4895 /* Fix illegal SG+CSUM combinations. */
4896 if ((features & NETIF_F_SG) &&
4897 !(features & NETIF_F_ALL_CSUM)) {
4899 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4900 "checksum feature.\n", name);
4901 features &= ~NETIF_F_SG;
4904 /* TSO requires that SG is present as well. */
4905 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4907 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4908 "SG feature.\n", name);
4909 features &= ~NETIF_F_TSO;
4912 if (features & NETIF_F_UFO) {
4913 if (!(features & NETIF_F_GEN_CSUM)) {
4915 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4916 "since no NETIF_F_HW_CSUM feature.\n",
4918 features &= ~NETIF_F_UFO;
4921 if (!(features & NETIF_F_SG)) {
4923 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4924 "since no NETIF_F_SG feature.\n", name);
4925 features &= ~NETIF_F_UFO;
4931 EXPORT_SYMBOL(netdev_fix_features);
4934 * netif_stacked_transfer_operstate - transfer operstate
4935 * @rootdev: the root or lower level device to transfer state from
4936 * @dev: the device to transfer operstate to
4938 * Transfer operational state from root to device. This is normally
4939 * called when a stacking relationship exists between the root
4940 * device and the device(a leaf device).
4942 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4943 struct net_device *dev)
4945 if (rootdev->operstate == IF_OPER_DORMANT)
4946 netif_dormant_on(dev);
4948 netif_dormant_off(dev);
4950 if (netif_carrier_ok(rootdev)) {
4951 if (!netif_carrier_ok(dev))
4952 netif_carrier_on(dev);
4954 if (netif_carrier_ok(dev))
4955 netif_carrier_off(dev);
4958 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4961 * register_netdevice - register a network device
4962 * @dev: device to register
4964 * Take a completed network device structure and add it to the kernel
4965 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4966 * chain. 0 is returned on success. A negative errno code is returned
4967 * on a failure to set up the device, or if the name is a duplicate.
4969 * Callers must hold the rtnl semaphore. You may want
4970 * register_netdev() instead of this.
4973 * The locking appears insufficient to guarantee two parallel registers
4974 * will not get the same name.
4977 int register_netdevice(struct net_device *dev)
4980 struct net *net = dev_net(dev);
4982 BUG_ON(dev_boot_phase);
4987 /* When net_device's are persistent, this will be fatal. */
4988 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4991 spin_lock_init(&dev->addr_list_lock);
4992 netdev_set_addr_lockdep_class(dev);
4993 netdev_init_queue_locks(dev);
4998 if (!dev->num_rx_queues) {
5000 * Allocate a single RX queue if driver never called
5004 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
5010 dev->_rx->first = dev->_rx;
5011 atomic_set(&dev->_rx->count, 1);
5012 dev->num_rx_queues = 1;
5015 /* Init, if this function is available */
5016 if (dev->netdev_ops->ndo_init) {
5017 ret = dev->netdev_ops->ndo_init(dev);
5025 ret = dev_get_valid_name(dev, dev->name, 0);
5029 dev->ifindex = dev_new_index(net);
5030 if (dev->iflink == -1)
5031 dev->iflink = dev->ifindex;
5033 /* Fix illegal checksum combinations */
5034 if ((dev->features & NETIF_F_HW_CSUM) &&
5035 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5036 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5038 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5041 if ((dev->features & NETIF_F_NO_CSUM) &&
5042 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5043 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5045 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5048 dev->features = netdev_fix_features(dev->features, dev->name);
5050 /* Enable software GSO if SG is supported. */
5051 if (dev->features & NETIF_F_SG)
5052 dev->features |= NETIF_F_GSO;
5054 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5055 ret = notifier_to_errno(ret);
5059 ret = netdev_register_kobject(dev);
5062 dev->reg_state = NETREG_REGISTERED;
5065 * Default initial state at registry is that the
5066 * device is present.
5069 set_bit(__LINK_STATE_PRESENT, &dev->state);
5071 dev_init_scheduler(dev);
5073 list_netdevice(dev);
5075 /* Notify protocols, that a new device appeared. */
5076 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5077 ret = notifier_to_errno(ret);
5079 rollback_registered(dev);
5080 dev->reg_state = NETREG_UNREGISTERED;
5083 * Prevent userspace races by waiting until the network
5084 * device is fully setup before sending notifications.
5086 if (!dev->rtnl_link_ops ||
5087 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5088 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5094 if (dev->netdev_ops->ndo_uninit)
5095 dev->netdev_ops->ndo_uninit(dev);
5098 EXPORT_SYMBOL(register_netdevice);
5101 * init_dummy_netdev - init a dummy network device for NAPI
5102 * @dev: device to init
5104 * This takes a network device structure and initialize the minimum
5105 * amount of fields so it can be used to schedule NAPI polls without
5106 * registering a full blown interface. This is to be used by drivers
5107 * that need to tie several hardware interfaces to a single NAPI
5108 * poll scheduler due to HW limitations.
5110 int init_dummy_netdev(struct net_device *dev)
5112 /* Clear everything. Note we don't initialize spinlocks
5113 * are they aren't supposed to be taken by any of the
5114 * NAPI code and this dummy netdev is supposed to be
5115 * only ever used for NAPI polls
5117 memset(dev, 0, sizeof(struct net_device));
5119 /* make sure we BUG if trying to hit standard
5120 * register/unregister code path
5122 dev->reg_state = NETREG_DUMMY;
5124 /* initialize the ref count */
5125 atomic_set(&dev->refcnt, 1);
5127 /* NAPI wants this */
5128 INIT_LIST_HEAD(&dev->napi_list);
5130 /* a dummy interface is started by default */
5131 set_bit(__LINK_STATE_PRESENT, &dev->state);
5132 set_bit(__LINK_STATE_START, &dev->state);
5136 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5140 * register_netdev - register a network device
5141 * @dev: device to register
5143 * Take a completed network device structure and add it to the kernel
5144 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5145 * chain. 0 is returned on success. A negative errno code is returned
5146 * on a failure to set up the device, or if the name is a duplicate.
5148 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5149 * and expands the device name if you passed a format string to
5152 int register_netdev(struct net_device *dev)
5159 * If the name is a format string the caller wants us to do a
5162 if (strchr(dev->name, '%')) {
5163 err = dev_alloc_name(dev, dev->name);
5168 err = register_netdevice(dev);
5173 EXPORT_SYMBOL(register_netdev);
5176 * netdev_wait_allrefs - wait until all references are gone.
5178 * This is called when unregistering network devices.
5180 * Any protocol or device that holds a reference should register
5181 * for netdevice notification, and cleanup and put back the
5182 * reference if they receive an UNREGISTER event.
5183 * We can get stuck here if buggy protocols don't correctly
5186 static void netdev_wait_allrefs(struct net_device *dev)
5188 unsigned long rebroadcast_time, warning_time;
5190 linkwatch_forget_dev(dev);
5192 rebroadcast_time = warning_time = jiffies;
5193 while (atomic_read(&dev->refcnt) != 0) {
5194 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5197 /* Rebroadcast unregister notification */
5198 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5199 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5200 * should have already handle it the first time */
5202 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5204 /* We must not have linkwatch events
5205 * pending on unregister. If this
5206 * happens, we simply run the queue
5207 * unscheduled, resulting in a noop
5210 linkwatch_run_queue();
5215 rebroadcast_time = jiffies;
5220 if (time_after(jiffies, warning_time + 10 * HZ)) {
5221 printk(KERN_EMERG "unregister_netdevice: "
5222 "waiting for %s to become free. Usage "
5224 dev->name, atomic_read(&dev->refcnt));
5225 warning_time = jiffies;
5234 * register_netdevice(x1);
5235 * register_netdevice(x2);
5237 * unregister_netdevice(y1);
5238 * unregister_netdevice(y2);
5244 * We are invoked by rtnl_unlock().
5245 * This allows us to deal with problems:
5246 * 1) We can delete sysfs objects which invoke hotplug
5247 * without deadlocking with linkwatch via keventd.
5248 * 2) Since we run with the RTNL semaphore not held, we can sleep
5249 * safely in order to wait for the netdev refcnt to drop to zero.
5251 * We must not return until all unregister events added during
5252 * the interval the lock was held have been completed.
5254 void netdev_run_todo(void)
5256 struct list_head list;
5258 /* Snapshot list, allow later requests */
5259 list_replace_init(&net_todo_list, &list);
5263 while (!list_empty(&list)) {
5264 struct net_device *dev
5265 = list_first_entry(&list, struct net_device, todo_list);
5266 list_del(&dev->todo_list);
5268 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5269 printk(KERN_ERR "network todo '%s' but state %d\n",
5270 dev->name, dev->reg_state);
5275 dev->reg_state = NETREG_UNREGISTERED;
5277 on_each_cpu(flush_backlog, dev, 1);
5279 netdev_wait_allrefs(dev);
5282 BUG_ON(atomic_read(&dev->refcnt));
5283 WARN_ON(dev->ip_ptr);
5284 WARN_ON(dev->ip6_ptr);
5285 WARN_ON(dev->dn_ptr);
5287 if (dev->destructor)
5288 dev->destructor(dev);
5290 /* Free network device */
5291 kobject_put(&dev->dev.kobj);
5296 * dev_txq_stats_fold - fold tx_queues stats
5297 * @dev: device to get statistics from
5298 * @stats: struct rtnl_link_stats64 to hold results
5300 void dev_txq_stats_fold(const struct net_device *dev,
5301 struct rtnl_link_stats64 *stats)
5303 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5305 struct netdev_queue *txq;
5307 for (i = 0; i < dev->num_tx_queues; i++) {
5308 txq = netdev_get_tx_queue(dev, i);
5309 spin_lock_bh(&txq->_xmit_lock);
5310 tx_bytes += txq->tx_bytes;
5311 tx_packets += txq->tx_packets;
5312 tx_dropped += txq->tx_dropped;
5313 spin_unlock_bh(&txq->_xmit_lock);
5315 if (tx_bytes || tx_packets || tx_dropped) {
5316 stats->tx_bytes = tx_bytes;
5317 stats->tx_packets = tx_packets;
5318 stats->tx_dropped = tx_dropped;
5321 EXPORT_SYMBOL(dev_txq_stats_fold);
5323 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5324 * fields in the same order, with only the type differing.
5326 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5327 const struct net_device_stats *netdev_stats)
5329 #if BITS_PER_LONG == 64
5330 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5331 memcpy(stats64, netdev_stats, sizeof(*stats64));
5333 size_t i, n = sizeof(*stats64) / sizeof(u64);
5334 const unsigned long *src = (const unsigned long *)netdev_stats;
5335 u64 *dst = (u64 *)stats64;
5337 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5338 sizeof(*stats64) / sizeof(u64));
5339 for (i = 0; i < n; i++)
5345 * dev_get_stats - get network device statistics
5346 * @dev: device to get statistics from
5347 * @storage: place to store stats
5349 * Get network statistics from device. Return @storage.
5350 * The device driver may provide its own method by setting
5351 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5352 * otherwise the internal statistics structure is used.
5354 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5355 struct rtnl_link_stats64 *storage)
5357 const struct net_device_ops *ops = dev->netdev_ops;
5359 if (ops->ndo_get_stats64) {
5360 memset(storage, 0, sizeof(*storage));
5361 return ops->ndo_get_stats64(dev, storage);
5363 if (ops->ndo_get_stats) {
5364 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5367 netdev_stats_to_stats64(storage, &dev->stats);
5368 dev_txq_stats_fold(dev, storage);
5371 EXPORT_SYMBOL(dev_get_stats);
5373 static void netdev_init_one_queue(struct net_device *dev,
5374 struct netdev_queue *queue,
5380 static void netdev_init_queues(struct net_device *dev)
5382 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5383 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5384 spin_lock_init(&dev->tx_global_lock);
5388 * alloc_netdev_mq - allocate network device
5389 * @sizeof_priv: size of private data to allocate space for
5390 * @name: device name format string
5391 * @setup: callback to initialize device
5392 * @queue_count: the number of subqueues to allocate
5394 * Allocates a struct net_device with private data area for driver use
5395 * and performs basic initialization. Also allocates subquue structs
5396 * for each queue on the device at the end of the netdevice.
5398 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5399 void (*setup)(struct net_device *), unsigned int queue_count)
5401 struct netdev_queue *tx;
5402 struct net_device *dev;
5404 struct net_device *p;
5406 struct netdev_rx_queue *rx;
5410 BUG_ON(strlen(name) >= sizeof(dev->name));
5412 alloc_size = sizeof(struct net_device);
5414 /* ensure 32-byte alignment of private area */
5415 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5416 alloc_size += sizeof_priv;
5418 /* ensure 32-byte alignment of whole construct */
5419 alloc_size += NETDEV_ALIGN - 1;
5421 p = kzalloc(alloc_size, GFP_KERNEL);
5423 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5427 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5429 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5435 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5437 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5442 atomic_set(&rx->count, queue_count);
5445 * Set a pointer to first element in the array which holds the
5448 for (i = 0; i < queue_count; i++)
5452 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5453 dev->padded = (char *)dev - (char *)p;
5455 if (dev_addr_init(dev))
5461 dev_net_set(dev, &init_net);
5464 dev->num_tx_queues = queue_count;
5465 dev->real_num_tx_queues = queue_count;
5469 dev->num_rx_queues = queue_count;
5472 dev->gso_max_size = GSO_MAX_SIZE;
5474 netdev_init_queues(dev);
5476 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5477 dev->ethtool_ntuple_list.count = 0;
5478 INIT_LIST_HEAD(&dev->napi_list);
5479 INIT_LIST_HEAD(&dev->unreg_list);
5480 INIT_LIST_HEAD(&dev->link_watch_list);
5481 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5483 strcpy(dev->name, name);
5496 EXPORT_SYMBOL(alloc_netdev_mq);
5499 * free_netdev - free network device
5502 * This function does the last stage of destroying an allocated device
5503 * interface. The reference to the device object is released.
5504 * If this is the last reference then it will be freed.
5506 void free_netdev(struct net_device *dev)
5508 struct napi_struct *p, *n;
5510 release_net(dev_net(dev));
5514 /* Flush device addresses */
5515 dev_addr_flush(dev);
5517 /* Clear ethtool n-tuple list */
5518 ethtool_ntuple_flush(dev);
5520 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5523 /* Compatibility with error handling in drivers */
5524 if (dev->reg_state == NETREG_UNINITIALIZED) {
5525 kfree((char *)dev - dev->padded);
5529 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5530 dev->reg_state = NETREG_RELEASED;
5532 /* will free via device release */
5533 put_device(&dev->dev);
5535 EXPORT_SYMBOL(free_netdev);
5538 * synchronize_net - Synchronize with packet receive processing
5540 * Wait for packets currently being received to be done.
5541 * Does not block later packets from starting.
5543 void synchronize_net(void)
5548 EXPORT_SYMBOL(synchronize_net);
5551 * unregister_netdevice_queue - remove device from the kernel
5555 * This function shuts down a device interface and removes it
5556 * from the kernel tables.
5557 * If head not NULL, device is queued to be unregistered later.
5559 * Callers must hold the rtnl semaphore. You may want
5560 * unregister_netdev() instead of this.
5563 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5568 list_move_tail(&dev->unreg_list, head);
5570 rollback_registered(dev);
5571 /* Finish processing unregister after unlock */
5575 EXPORT_SYMBOL(unregister_netdevice_queue);
5578 * unregister_netdevice_many - unregister many devices
5579 * @head: list of devices
5581 void unregister_netdevice_many(struct list_head *head)
5583 struct net_device *dev;
5585 if (!list_empty(head)) {
5586 rollback_registered_many(head);
5587 list_for_each_entry(dev, head, unreg_list)
5591 EXPORT_SYMBOL(unregister_netdevice_many);
5594 * unregister_netdev - remove device from the kernel
5597 * This function shuts down a device interface and removes it
5598 * from the kernel tables.
5600 * This is just a wrapper for unregister_netdevice that takes
5601 * the rtnl semaphore. In general you want to use this and not
5602 * unregister_netdevice.
5604 void unregister_netdev(struct net_device *dev)
5607 unregister_netdevice(dev);
5610 EXPORT_SYMBOL(unregister_netdev);
5613 * dev_change_net_namespace - move device to different nethost namespace
5615 * @net: network namespace
5616 * @pat: If not NULL name pattern to try if the current device name
5617 * is already taken in the destination network namespace.
5619 * This function shuts down a device interface and moves it
5620 * to a new network namespace. On success 0 is returned, on
5621 * a failure a netagive errno code is returned.
5623 * Callers must hold the rtnl semaphore.
5626 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5632 /* Don't allow namespace local devices to be moved. */
5634 if (dev->features & NETIF_F_NETNS_LOCAL)
5637 /* Ensure the device has been registrered */
5639 if (dev->reg_state != NETREG_REGISTERED)
5642 /* Get out if there is nothing todo */
5644 if (net_eq(dev_net(dev), net))
5647 /* Pick the destination device name, and ensure
5648 * we can use it in the destination network namespace.
5651 if (__dev_get_by_name(net, dev->name)) {
5652 /* We get here if we can't use the current device name */
5655 if (dev_get_valid_name(dev, pat, 1))
5660 * And now a mini version of register_netdevice unregister_netdevice.
5663 /* If device is running close it first. */
5666 /* And unlink it from device chain */
5668 unlist_netdevice(dev);
5672 /* Shutdown queueing discipline. */
5675 /* Notify protocols, that we are about to destroy
5676 this device. They should clean all the things.
5678 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5679 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5682 * Flush the unicast and multicast chains
5687 /* Actually switch the network namespace */
5688 dev_net_set(dev, net);
5690 /* If there is an ifindex conflict assign a new one */
5691 if (__dev_get_by_index(net, dev->ifindex)) {
5692 int iflink = (dev->iflink == dev->ifindex);
5693 dev->ifindex = dev_new_index(net);
5695 dev->iflink = dev->ifindex;
5698 /* Fixup kobjects */
5699 err = device_rename(&dev->dev, dev->name);
5702 /* Add the device back in the hashes */
5703 list_netdevice(dev);
5705 /* Notify protocols, that a new device appeared. */
5706 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5709 * Prevent userspace races by waiting until the network
5710 * device is fully setup before sending notifications.
5712 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5719 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5721 static int dev_cpu_callback(struct notifier_block *nfb,
5722 unsigned long action,
5725 struct sk_buff **list_skb;
5726 struct sk_buff *skb;
5727 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5728 struct softnet_data *sd, *oldsd;
5730 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5733 local_irq_disable();
5734 cpu = smp_processor_id();
5735 sd = &per_cpu(softnet_data, cpu);
5736 oldsd = &per_cpu(softnet_data, oldcpu);
5738 /* Find end of our completion_queue. */
5739 list_skb = &sd->completion_queue;
5741 list_skb = &(*list_skb)->next;
5742 /* Append completion queue from offline CPU. */
5743 *list_skb = oldsd->completion_queue;
5744 oldsd->completion_queue = NULL;
5746 /* Append output queue from offline CPU. */
5747 if (oldsd->output_queue) {
5748 *sd->output_queue_tailp = oldsd->output_queue;
5749 sd->output_queue_tailp = oldsd->output_queue_tailp;
5750 oldsd->output_queue = NULL;
5751 oldsd->output_queue_tailp = &oldsd->output_queue;
5754 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5757 /* Process offline CPU's input_pkt_queue */
5758 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5760 input_queue_head_incr(oldsd);
5762 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5764 input_queue_head_incr(oldsd);
5772 * netdev_increment_features - increment feature set by one
5773 * @all: current feature set
5774 * @one: new feature set
5775 * @mask: mask feature set
5777 * Computes a new feature set after adding a device with feature set
5778 * @one to the master device with current feature set @all. Will not
5779 * enable anything that is off in @mask. Returns the new feature set.
5781 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5784 /* If device needs checksumming, downgrade to it. */
5785 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5786 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5787 else if (mask & NETIF_F_ALL_CSUM) {
5788 /* If one device supports v4/v6 checksumming, set for all. */
5789 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5790 !(all & NETIF_F_GEN_CSUM)) {
5791 all &= ~NETIF_F_ALL_CSUM;
5792 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5795 /* If one device supports hw checksumming, set for all. */
5796 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5797 all &= ~NETIF_F_ALL_CSUM;
5798 all |= NETIF_F_HW_CSUM;
5802 one |= NETIF_F_ALL_CSUM;
5804 one |= all & NETIF_F_ONE_FOR_ALL;
5805 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5806 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5810 EXPORT_SYMBOL(netdev_increment_features);
5812 static struct hlist_head *netdev_create_hash(void)
5815 struct hlist_head *hash;
5817 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5819 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5820 INIT_HLIST_HEAD(&hash[i]);
5825 /* Initialize per network namespace state */
5826 static int __net_init netdev_init(struct net *net)
5828 INIT_LIST_HEAD(&net->dev_base_head);
5830 net->dev_name_head = netdev_create_hash();
5831 if (net->dev_name_head == NULL)
5834 net->dev_index_head = netdev_create_hash();
5835 if (net->dev_index_head == NULL)
5841 kfree(net->dev_name_head);
5847 * netdev_drivername - network driver for the device
5848 * @dev: network device
5849 * @buffer: buffer for resulting name
5850 * @len: size of buffer
5852 * Determine network driver for device.
5854 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5856 const struct device_driver *driver;
5857 const struct device *parent;
5859 if (len <= 0 || !buffer)
5863 parent = dev->dev.parent;
5868 driver = parent->driver;
5869 if (driver && driver->name)
5870 strlcpy(buffer, driver->name, len);
5874 static int __netdev_printk(const char *level, const struct net_device *dev,
5875 struct va_format *vaf)
5879 if (dev && dev->dev.parent)
5880 r = dev_printk(level, dev->dev.parent, "%s: %pV",
5881 netdev_name(dev), vaf);
5883 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
5885 r = printk("%s(NULL net_device): %pV", level, vaf);
5890 int netdev_printk(const char *level, const struct net_device *dev,
5891 const char *format, ...)
5893 struct va_format vaf;
5897 va_start(args, format);
5902 r = __netdev_printk(level, dev, &vaf);
5907 EXPORT_SYMBOL(netdev_printk);
5909 #define define_netdev_printk_level(func, level) \
5910 int func(const struct net_device *dev, const char *fmt, ...) \
5913 struct va_format vaf; \
5916 va_start(args, fmt); \
5921 r = __netdev_printk(level, dev, &vaf); \
5926 EXPORT_SYMBOL(func);
5928 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
5929 define_netdev_printk_level(netdev_alert, KERN_ALERT);
5930 define_netdev_printk_level(netdev_crit, KERN_CRIT);
5931 define_netdev_printk_level(netdev_err, KERN_ERR);
5932 define_netdev_printk_level(netdev_warn, KERN_WARNING);
5933 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
5934 define_netdev_printk_level(netdev_info, KERN_INFO);
5936 static void __net_exit netdev_exit(struct net *net)
5938 kfree(net->dev_name_head);
5939 kfree(net->dev_index_head);
5942 static struct pernet_operations __net_initdata netdev_net_ops = {
5943 .init = netdev_init,
5944 .exit = netdev_exit,
5947 static void __net_exit default_device_exit(struct net *net)
5949 struct net_device *dev, *aux;
5951 * Push all migratable network devices back to the
5952 * initial network namespace
5955 for_each_netdev_safe(net, dev, aux) {
5957 char fb_name[IFNAMSIZ];
5959 /* Ignore unmoveable devices (i.e. loopback) */
5960 if (dev->features & NETIF_F_NETNS_LOCAL)
5963 /* Leave virtual devices for the generic cleanup */
5964 if (dev->rtnl_link_ops)
5967 /* Push remaing network devices to init_net */
5968 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5969 err = dev_change_net_namespace(dev, &init_net, fb_name);
5971 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5972 __func__, dev->name, err);
5979 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5981 /* At exit all network devices most be removed from a network
5982 * namespace. Do this in the reverse order of registeration.
5983 * Do this across as many network namespaces as possible to
5984 * improve batching efficiency.
5986 struct net_device *dev;
5988 LIST_HEAD(dev_kill_list);
5991 list_for_each_entry(net, net_list, exit_list) {
5992 for_each_netdev_reverse(net, dev) {
5993 if (dev->rtnl_link_ops)
5994 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5996 unregister_netdevice_queue(dev, &dev_kill_list);
5999 unregister_netdevice_many(&dev_kill_list);
6003 static struct pernet_operations __net_initdata default_device_ops = {
6004 .exit = default_device_exit,
6005 .exit_batch = default_device_exit_batch,
6009 * Initialize the DEV module. At boot time this walks the device list and
6010 * unhooks any devices that fail to initialise (normally hardware not
6011 * present) and leaves us with a valid list of present and active devices.
6016 * This is called single threaded during boot, so no need
6017 * to take the rtnl semaphore.
6019 static int __init net_dev_init(void)
6021 int i, rc = -ENOMEM;
6023 BUG_ON(!dev_boot_phase);
6025 if (dev_proc_init())
6028 if (netdev_kobject_init())
6031 INIT_LIST_HEAD(&ptype_all);
6032 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6033 INIT_LIST_HEAD(&ptype_base[i]);
6035 if (register_pernet_subsys(&netdev_net_ops))
6039 * Initialise the packet receive queues.
6042 for_each_possible_cpu(i) {
6043 struct softnet_data *sd = &per_cpu(softnet_data, i);
6045 memset(sd, 0, sizeof(*sd));
6046 skb_queue_head_init(&sd->input_pkt_queue);
6047 skb_queue_head_init(&sd->process_queue);
6048 sd->completion_queue = NULL;
6049 INIT_LIST_HEAD(&sd->poll_list);
6050 sd->output_queue = NULL;
6051 sd->output_queue_tailp = &sd->output_queue;
6053 sd->csd.func = rps_trigger_softirq;
6059 sd->backlog.poll = process_backlog;
6060 sd->backlog.weight = weight_p;
6061 sd->backlog.gro_list = NULL;
6062 sd->backlog.gro_count = 0;
6067 /* The loopback device is special if any other network devices
6068 * is present in a network namespace the loopback device must
6069 * be present. Since we now dynamically allocate and free the
6070 * loopback device ensure this invariant is maintained by
6071 * keeping the loopback device as the first device on the
6072 * list of network devices. Ensuring the loopback devices
6073 * is the first device that appears and the last network device
6076 if (register_pernet_device(&loopback_net_ops))
6079 if (register_pernet_device(&default_device_ops))
6082 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6083 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6085 hotcpu_notifier(dev_cpu_callback, 0);
6093 subsys_initcall(net_dev_init);
6095 static int __init initialize_hashrnd(void)
6097 get_random_bytes(&hashrnd, sizeof(hashrnd));
6101 late_initcall_sync(initialize_hashrnd);