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>
132 #include <linux/inetdevice.h>
134 #include "net-sysfs.h"
136 /* Instead of increasing this, you should create a hash table. */
137 #define MAX_GRO_SKBS 8
139 /* This should be increased if a protocol with a bigger head is added. */
140 #define GRO_MAX_HEAD (MAX_HEADER + 128)
143 * The list of packet types we will receive (as opposed to discard)
144 * and the routines to invoke.
146 * Why 16. Because with 16 the only overlap we get on a hash of the
147 * low nibble of the protocol value is RARP/SNAP/X.25.
149 * NOTE: That is no longer true with the addition of VLAN tags. Not
150 * sure which should go first, but I bet it won't make much
151 * difference if we are running VLANs. The good news is that
152 * this protocol won't be in the list unless compiled in, so
153 * the average user (w/out VLANs) will not be adversely affected.
170 #define PTYPE_HASH_SIZE (16)
171 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
173 static DEFINE_SPINLOCK(ptype_lock);
174 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
175 static struct list_head ptype_all __read_mostly; /* Taps */
178 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
181 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
183 * Writers must hold the rtnl semaphore while they loop through the
184 * dev_base_head list, and hold dev_base_lock for writing when they do the
185 * actual updates. This allows pure readers to access the list even
186 * while a writer is preparing to update it.
188 * To put it another way, dev_base_lock is held for writing only to
189 * protect against pure readers; the rtnl semaphore provides the
190 * protection against other writers.
192 * See, for example usages, register_netdevice() and
193 * unregister_netdevice(), which must be called with the rtnl
196 DEFINE_RWLOCK(dev_base_lock);
197 EXPORT_SYMBOL(dev_base_lock);
199 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
201 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
202 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
205 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
207 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
210 static inline void rps_lock(struct softnet_data *sd)
213 spin_lock(&sd->input_pkt_queue.lock);
217 static inline void rps_unlock(struct softnet_data *sd)
220 spin_unlock(&sd->input_pkt_queue.lock);
224 /* Device list insertion */
225 static int list_netdevice(struct net_device *dev)
227 struct net *net = dev_net(dev);
231 write_lock_bh(&dev_base_lock);
232 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
233 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
234 hlist_add_head_rcu(&dev->index_hlist,
235 dev_index_hash(net, dev->ifindex));
236 write_unlock_bh(&dev_base_lock);
240 /* Device list removal
241 * caller must respect a RCU grace period before freeing/reusing dev
243 static void unlist_netdevice(struct net_device *dev)
247 /* Unlink dev from the device chain */
248 write_lock_bh(&dev_base_lock);
249 list_del_rcu(&dev->dev_list);
250 hlist_del_rcu(&dev->name_hlist);
251 hlist_del_rcu(&dev->index_hlist);
252 write_unlock_bh(&dev_base_lock);
259 static RAW_NOTIFIER_HEAD(netdev_chain);
262 * Device drivers call our routines to queue packets here. We empty the
263 * queue in the local softnet handler.
266 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
267 EXPORT_PER_CPU_SYMBOL(softnet_data);
269 #ifdef CONFIG_LOCKDEP
271 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
272 * according to dev->type
274 static const unsigned short netdev_lock_type[] =
275 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
276 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
277 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
278 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
279 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
280 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
281 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
282 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
283 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
284 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
285 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
286 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
287 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
288 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
289 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
290 ARPHRD_VOID, ARPHRD_NONE};
292 static const char *const netdev_lock_name[] =
293 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
294 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
295 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
296 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
297 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
298 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
299 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
300 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
301 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
302 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
303 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
304 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
305 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
306 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
307 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
308 "_xmit_VOID", "_xmit_NONE"};
310 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
311 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
313 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
317 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
318 if (netdev_lock_type[i] == dev_type)
320 /* the last key is used by default */
321 return ARRAY_SIZE(netdev_lock_type) - 1;
324 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
325 unsigned short dev_type)
329 i = netdev_lock_pos(dev_type);
330 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
331 netdev_lock_name[i]);
334 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
338 i = netdev_lock_pos(dev->type);
339 lockdep_set_class_and_name(&dev->addr_list_lock,
340 &netdev_addr_lock_key[i],
341 netdev_lock_name[i]);
344 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
345 unsigned short dev_type)
348 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
353 /*******************************************************************************
355 Protocol management and registration routines
357 *******************************************************************************/
360 * Add a protocol ID to the list. Now that the input handler is
361 * smarter we can dispense with all the messy stuff that used to be
364 * BEWARE!!! Protocol handlers, mangling input packets,
365 * MUST BE last in hash buckets and checking protocol handlers
366 * MUST start from promiscuous ptype_all chain in net_bh.
367 * It is true now, do not change it.
368 * Explanation follows: if protocol handler, mangling packet, will
369 * be the first on list, it is not able to sense, that packet
370 * is cloned and should be copied-on-write, so that it will
371 * change it and subsequent readers will get broken packet.
375 static inline struct list_head *ptype_head(const struct packet_type *pt)
377 if (pt->type == htons(ETH_P_ALL))
380 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
384 * dev_add_pack - add packet handler
385 * @pt: packet type declaration
387 * Add a protocol handler to the networking stack. The passed &packet_type
388 * is linked into kernel lists and may not be freed until it has been
389 * removed from the kernel lists.
391 * This call does not sleep therefore it can not
392 * guarantee all CPU's that are in middle of receiving packets
393 * will see the new packet type (until the next received packet).
396 void dev_add_pack(struct packet_type *pt)
398 struct list_head *head = ptype_head(pt);
400 spin_lock(&ptype_lock);
401 list_add_rcu(&pt->list, head);
402 spin_unlock(&ptype_lock);
404 EXPORT_SYMBOL(dev_add_pack);
407 * __dev_remove_pack - remove packet handler
408 * @pt: packet type declaration
410 * Remove a protocol handler that was previously added to the kernel
411 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
412 * from the kernel lists and can be freed or reused once this function
415 * The packet type might still be in use by receivers
416 * and must not be freed until after all the CPU's have gone
417 * through a quiescent state.
419 void __dev_remove_pack(struct packet_type *pt)
421 struct list_head *head = ptype_head(pt);
422 struct packet_type *pt1;
424 spin_lock(&ptype_lock);
426 list_for_each_entry(pt1, head, list) {
428 list_del_rcu(&pt->list);
433 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
435 spin_unlock(&ptype_lock);
437 EXPORT_SYMBOL(__dev_remove_pack);
440 * dev_remove_pack - remove packet handler
441 * @pt: packet type declaration
443 * Remove a protocol handler that was previously added to the kernel
444 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
445 * from the kernel lists and can be freed or reused once this function
448 * This call sleeps to guarantee that no CPU is looking at the packet
451 void dev_remove_pack(struct packet_type *pt)
453 __dev_remove_pack(pt);
457 EXPORT_SYMBOL(dev_remove_pack);
459 /******************************************************************************
461 Device Boot-time Settings Routines
463 *******************************************************************************/
465 /* Boot time configuration table */
466 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
469 * netdev_boot_setup_add - add new setup entry
470 * @name: name of the device
471 * @map: configured settings for the device
473 * Adds new setup entry to the dev_boot_setup list. The function
474 * returns 0 on error and 1 on success. This is a generic routine to
477 static int netdev_boot_setup_add(char *name, struct ifmap *map)
479 struct netdev_boot_setup *s;
483 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
484 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
485 memset(s[i].name, 0, sizeof(s[i].name));
486 strlcpy(s[i].name, name, IFNAMSIZ);
487 memcpy(&s[i].map, map, sizeof(s[i].map));
492 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
496 * netdev_boot_setup_check - check boot time settings
497 * @dev: the netdevice
499 * Check boot time settings for the device.
500 * The found settings are set for the device to be used
501 * later in the device probing.
502 * Returns 0 if no settings found, 1 if they are.
504 int netdev_boot_setup_check(struct net_device *dev)
506 struct netdev_boot_setup *s = dev_boot_setup;
509 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
510 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
511 !strcmp(dev->name, s[i].name)) {
512 dev->irq = s[i].map.irq;
513 dev->base_addr = s[i].map.base_addr;
514 dev->mem_start = s[i].map.mem_start;
515 dev->mem_end = s[i].map.mem_end;
521 EXPORT_SYMBOL(netdev_boot_setup_check);
525 * netdev_boot_base - get address from boot time settings
526 * @prefix: prefix for network device
527 * @unit: id for network device
529 * Check boot time settings for the base address of device.
530 * The found settings are set for the device to be used
531 * later in the device probing.
532 * Returns 0 if no settings found.
534 unsigned long netdev_boot_base(const char *prefix, int unit)
536 const struct netdev_boot_setup *s = dev_boot_setup;
540 sprintf(name, "%s%d", prefix, unit);
543 * If device already registered then return base of 1
544 * to indicate not to probe for this interface
546 if (__dev_get_by_name(&init_net, name))
549 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
550 if (!strcmp(name, s[i].name))
551 return s[i].map.base_addr;
556 * Saves at boot time configured settings for any netdevice.
558 int __init netdev_boot_setup(char *str)
563 str = get_options(str, ARRAY_SIZE(ints), ints);
568 memset(&map, 0, sizeof(map));
572 map.base_addr = ints[2];
574 map.mem_start = ints[3];
576 map.mem_end = ints[4];
578 /* Add new entry to the list */
579 return netdev_boot_setup_add(str, &map);
582 __setup("netdev=", netdev_boot_setup);
584 /*******************************************************************************
586 Device Interface Subroutines
588 *******************************************************************************/
591 * __dev_get_by_name - find a device by its name
592 * @net: the applicable net namespace
593 * @name: name to find
595 * Find an interface by name. Must be called under RTNL semaphore
596 * or @dev_base_lock. If the name is found a pointer to the device
597 * is returned. If the name is not found then %NULL is returned. The
598 * reference counters are not incremented so the caller must be
599 * careful with locks.
602 struct net_device *__dev_get_by_name(struct net *net, const char *name)
604 struct hlist_node *p;
605 struct net_device *dev;
606 struct hlist_head *head = dev_name_hash(net, name);
608 hlist_for_each_entry(dev, p, head, name_hlist)
609 if (!strncmp(dev->name, name, IFNAMSIZ))
614 EXPORT_SYMBOL(__dev_get_by_name);
617 * dev_get_by_name_rcu - find a device by its name
618 * @net: the applicable net namespace
619 * @name: name to find
621 * Find an interface by name.
622 * If the name is found a pointer to the device is returned.
623 * If the name is not found then %NULL is returned.
624 * The reference counters are not incremented so the caller must be
625 * careful with locks. The caller must hold RCU lock.
628 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
630 struct hlist_node *p;
631 struct net_device *dev;
632 struct hlist_head *head = dev_name_hash(net, name);
634 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
635 if (!strncmp(dev->name, name, IFNAMSIZ))
640 EXPORT_SYMBOL(dev_get_by_name_rcu);
643 * dev_get_by_name - find a device by its name
644 * @net: the applicable net namespace
645 * @name: name to find
647 * Find an interface by name. This can be called from any
648 * context and does its own locking. The returned handle has
649 * the usage count incremented and the caller must use dev_put() to
650 * release it when it is no longer needed. %NULL is returned if no
651 * matching device is found.
654 struct net_device *dev_get_by_name(struct net *net, const char *name)
656 struct net_device *dev;
659 dev = dev_get_by_name_rcu(net, name);
665 EXPORT_SYMBOL(dev_get_by_name);
668 * __dev_get_by_index - find a device by its ifindex
669 * @net: the applicable net namespace
670 * @ifindex: index of device
672 * Search for an interface by index. Returns %NULL if the device
673 * is not found or a pointer to the device. The device has not
674 * had its reference counter increased so the caller must be careful
675 * about locking. The caller must hold either the RTNL semaphore
679 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
681 struct hlist_node *p;
682 struct net_device *dev;
683 struct hlist_head *head = dev_index_hash(net, ifindex);
685 hlist_for_each_entry(dev, p, head, index_hlist)
686 if (dev->ifindex == ifindex)
691 EXPORT_SYMBOL(__dev_get_by_index);
694 * dev_get_by_index_rcu - find a device by its ifindex
695 * @net: the applicable net namespace
696 * @ifindex: index of device
698 * Search for an interface by index. Returns %NULL if the device
699 * is not found or a pointer to the device. The device has not
700 * had its reference counter increased so the caller must be careful
701 * about locking. The caller must hold RCU lock.
704 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
706 struct hlist_node *p;
707 struct net_device *dev;
708 struct hlist_head *head = dev_index_hash(net, ifindex);
710 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
711 if (dev->ifindex == ifindex)
716 EXPORT_SYMBOL(dev_get_by_index_rcu);
720 * dev_get_by_index - find a device by its ifindex
721 * @net: the applicable net namespace
722 * @ifindex: index of device
724 * Search for an interface by index. Returns NULL if the device
725 * is not found or a pointer to the device. The device returned has
726 * had a reference added and the pointer is safe until the user calls
727 * dev_put to indicate they have finished with it.
730 struct net_device *dev_get_by_index(struct net *net, int ifindex)
732 struct net_device *dev;
735 dev = dev_get_by_index_rcu(net, ifindex);
741 EXPORT_SYMBOL(dev_get_by_index);
744 * dev_getbyhwaddr - find a device by its hardware address
745 * @net: the applicable net namespace
746 * @type: media type of device
747 * @ha: hardware address
749 * Search for an interface by MAC address. Returns NULL if the device
750 * is not found or a pointer to the device. The caller must hold the
751 * rtnl semaphore. The returned device has not had its ref count increased
752 * and the caller must therefore be careful about locking
755 * If the API was consistent this would be __dev_get_by_hwaddr
758 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
760 struct net_device *dev;
764 for_each_netdev(net, dev)
765 if (dev->type == type &&
766 !memcmp(dev->dev_addr, ha, dev->addr_len))
771 EXPORT_SYMBOL(dev_getbyhwaddr);
773 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
775 struct net_device *dev;
778 for_each_netdev(net, dev)
779 if (dev->type == type)
784 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
786 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
788 struct net_device *dev, *ret = NULL;
791 for_each_netdev_rcu(net, dev)
792 if (dev->type == type) {
800 EXPORT_SYMBOL(dev_getfirstbyhwtype);
803 * dev_get_by_flags_rcu - find any device with given flags
804 * @net: the applicable net namespace
805 * @if_flags: IFF_* values
806 * @mask: bitmask of bits in if_flags to check
808 * Search for any interface with the given flags. Returns NULL if a device
809 * is not found or a pointer to the device. Must be called inside
810 * rcu_read_lock(), and result refcount is unchanged.
813 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
816 struct net_device *dev, *ret;
819 for_each_netdev_rcu(net, dev) {
820 if (((dev->flags ^ if_flags) & mask) == 0) {
827 EXPORT_SYMBOL(dev_get_by_flags_rcu);
830 * dev_valid_name - check if name is okay for network device
833 * Network device names need to be valid file names to
834 * to allow sysfs to work. We also disallow any kind of
837 int dev_valid_name(const char *name)
841 if (strlen(name) >= IFNAMSIZ)
843 if (!strcmp(name, ".") || !strcmp(name, ".."))
847 if (*name == '/' || isspace(*name))
853 EXPORT_SYMBOL(dev_valid_name);
856 * __dev_alloc_name - allocate a name for a device
857 * @net: network namespace to allocate the device name in
858 * @name: name format string
859 * @buf: scratch buffer and result name string
861 * Passed a format string - eg "lt%d" it will try and find a suitable
862 * id. It scans list of devices to build up a free map, then chooses
863 * the first empty slot. The caller must hold the dev_base or rtnl lock
864 * while allocating the name and adding the device in order to avoid
866 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
867 * Returns the number of the unit assigned or a negative errno code.
870 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
874 const int max_netdevices = 8*PAGE_SIZE;
875 unsigned long *inuse;
876 struct net_device *d;
878 p = strnchr(name, IFNAMSIZ-1, '%');
881 * Verify the string as this thing may have come from
882 * the user. There must be either one "%d" and no other "%"
885 if (p[1] != 'd' || strchr(p + 2, '%'))
888 /* Use one page as a bit array of possible slots */
889 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
893 for_each_netdev(net, d) {
894 if (!sscanf(d->name, name, &i))
896 if (i < 0 || i >= max_netdevices)
899 /* avoid cases where sscanf is not exact inverse of printf */
900 snprintf(buf, IFNAMSIZ, name, i);
901 if (!strncmp(buf, d->name, IFNAMSIZ))
905 i = find_first_zero_bit(inuse, max_netdevices);
906 free_page((unsigned long) inuse);
910 snprintf(buf, IFNAMSIZ, name, i);
911 if (!__dev_get_by_name(net, buf))
914 /* It is possible to run out of possible slots
915 * when the name is long and there isn't enough space left
916 * for the digits, or if all bits are used.
922 * dev_alloc_name - allocate a name for a device
924 * @name: name format string
926 * Passed a format string - eg "lt%d" it will try and find a suitable
927 * id. It scans list of devices to build up a free map, then chooses
928 * the first empty slot. The caller must hold the dev_base or rtnl lock
929 * while allocating the name and adding the device in order to avoid
931 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
932 * Returns the number of the unit assigned or a negative errno code.
935 int dev_alloc_name(struct net_device *dev, const char *name)
941 BUG_ON(!dev_net(dev));
943 ret = __dev_alloc_name(net, name, buf);
945 strlcpy(dev->name, buf, IFNAMSIZ);
948 EXPORT_SYMBOL(dev_alloc_name);
950 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
954 BUG_ON(!dev_net(dev));
957 if (!dev_valid_name(name))
960 if (fmt && strchr(name, '%'))
961 return dev_alloc_name(dev, name);
962 else if (__dev_get_by_name(net, name))
964 else if (dev->name != name)
965 strlcpy(dev->name, name, IFNAMSIZ);
971 * dev_change_name - change name of a device
973 * @newname: name (or format string) must be at least IFNAMSIZ
975 * Change name of a device, can pass format strings "eth%d".
978 int dev_change_name(struct net_device *dev, const char *newname)
980 char oldname[IFNAMSIZ];
986 BUG_ON(!dev_net(dev));
989 if (dev->flags & IFF_UP)
992 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
995 memcpy(oldname, dev->name, IFNAMSIZ);
997 err = dev_get_valid_name(dev, newname, 1);
1002 ret = device_rename(&dev->dev, dev->name);
1004 memcpy(dev->name, oldname, IFNAMSIZ);
1008 write_lock_bh(&dev_base_lock);
1009 hlist_del(&dev->name_hlist);
1010 write_unlock_bh(&dev_base_lock);
1014 write_lock_bh(&dev_base_lock);
1015 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1016 write_unlock_bh(&dev_base_lock);
1018 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1019 ret = notifier_to_errno(ret);
1022 /* err >= 0 after dev_alloc_name() or stores the first errno */
1025 memcpy(dev->name, oldname, IFNAMSIZ);
1029 "%s: name change rollback failed: %d.\n",
1038 * dev_set_alias - change ifalias of a device
1040 * @alias: name up to IFALIASZ
1041 * @len: limit of bytes to copy from info
1043 * Set ifalias for a device,
1045 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1049 if (len >= IFALIASZ)
1054 kfree(dev->ifalias);
1055 dev->ifalias = NULL;
1060 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1064 strlcpy(dev->ifalias, alias, len+1);
1070 * netdev_features_change - device changes features
1071 * @dev: device to cause notification
1073 * Called to indicate a device has changed features.
1075 void netdev_features_change(struct net_device *dev)
1077 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1079 EXPORT_SYMBOL(netdev_features_change);
1082 * netdev_state_change - device changes state
1083 * @dev: device to cause notification
1085 * Called to indicate a device has changed state. This function calls
1086 * the notifier chains for netdev_chain and sends a NEWLINK message
1087 * to the routing socket.
1089 void netdev_state_change(struct net_device *dev)
1091 if (dev->flags & IFF_UP) {
1092 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1093 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1096 EXPORT_SYMBOL(netdev_state_change);
1098 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1100 return call_netdevice_notifiers(event, dev);
1102 EXPORT_SYMBOL(netdev_bonding_change);
1105 * dev_load - load a network module
1106 * @net: the applicable net namespace
1107 * @name: name of interface
1109 * If a network interface is not present and the process has suitable
1110 * privileges this function loads the module. If module loading is not
1111 * available in this kernel then it becomes a nop.
1114 void dev_load(struct net *net, const char *name)
1116 struct net_device *dev;
1119 dev = dev_get_by_name_rcu(net, name);
1122 if (!dev && capable(CAP_NET_ADMIN))
1123 request_module("%s", name);
1125 EXPORT_SYMBOL(dev_load);
1127 static int __dev_open(struct net_device *dev)
1129 const struct net_device_ops *ops = dev->netdev_ops;
1135 * Is it even present?
1137 if (!netif_device_present(dev))
1140 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1141 ret = notifier_to_errno(ret);
1146 * Call device private open method
1148 set_bit(__LINK_STATE_START, &dev->state);
1150 if (ops->ndo_validate_addr)
1151 ret = ops->ndo_validate_addr(dev);
1153 if (!ret && ops->ndo_open)
1154 ret = ops->ndo_open(dev);
1157 * If it went open OK then:
1161 clear_bit(__LINK_STATE_START, &dev->state);
1166 dev->flags |= IFF_UP;
1171 net_dmaengine_get();
1174 * Initialize multicasting status
1176 dev_set_rx_mode(dev);
1179 * Wakeup transmit queue engine
1188 * dev_open - prepare an interface for use.
1189 * @dev: device to open
1191 * Takes a device from down to up state. The device's private open
1192 * function is invoked and then the multicast lists are loaded. Finally
1193 * the device is moved into the up state and a %NETDEV_UP message is
1194 * sent to the netdev notifier chain.
1196 * Calling this function on an active interface is a nop. On a failure
1197 * a negative errno code is returned.
1199 int dev_open(struct net_device *dev)
1206 if (dev->flags & IFF_UP)
1212 ret = __dev_open(dev);
1217 * ... and announce new interface.
1219 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1220 call_netdevice_notifiers(NETDEV_UP, dev);
1224 EXPORT_SYMBOL(dev_open);
1226 static int __dev_close(struct net_device *dev)
1228 const struct net_device_ops *ops = dev->netdev_ops;
1234 * Tell people we are going down, so that they can
1235 * prepare to death, when device is still operating.
1237 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1239 clear_bit(__LINK_STATE_START, &dev->state);
1241 /* Synchronize to scheduled poll. We cannot touch poll list,
1242 * it can be even on different cpu. So just clear netif_running().
1244 * dev->stop() will invoke napi_disable() on all of it's
1245 * napi_struct instances on this device.
1247 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1249 dev_deactivate(dev);
1252 * Call the device specific close. This cannot fail.
1253 * Only if device is UP
1255 * We allow it to be called even after a DETACH hot-plug
1262 * Device is now down.
1265 dev->flags &= ~IFF_UP;
1270 net_dmaengine_put();
1276 * dev_close - shutdown an interface.
1277 * @dev: device to shutdown
1279 * This function moves an active device into down state. A
1280 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1281 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1284 int dev_close(struct net_device *dev)
1286 if (!(dev->flags & IFF_UP))
1292 * Tell people we are down
1294 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1295 call_netdevice_notifiers(NETDEV_DOWN, dev);
1299 EXPORT_SYMBOL(dev_close);
1303 * dev_disable_lro - disable Large Receive Offload on a device
1306 * Disable Large Receive Offload (LRO) on a net device. Must be
1307 * called under RTNL. This is needed if received packets may be
1308 * forwarded to another interface.
1310 void dev_disable_lro(struct net_device *dev)
1312 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1313 dev->ethtool_ops->set_flags) {
1314 u32 flags = dev->ethtool_ops->get_flags(dev);
1315 if (flags & ETH_FLAG_LRO) {
1316 flags &= ~ETH_FLAG_LRO;
1317 dev->ethtool_ops->set_flags(dev, flags);
1320 WARN_ON(dev->features & NETIF_F_LRO);
1322 EXPORT_SYMBOL(dev_disable_lro);
1325 static int dev_boot_phase = 1;
1328 * Device change register/unregister. These are not inline or static
1329 * as we export them to the world.
1333 * register_netdevice_notifier - register a network notifier block
1336 * Register a notifier to be called when network device events occur.
1337 * The notifier passed is linked into the kernel structures and must
1338 * not be reused until it has been unregistered. A negative errno code
1339 * is returned on a failure.
1341 * When registered all registration and up events are replayed
1342 * to the new notifier to allow device to have a race free
1343 * view of the network device list.
1346 int register_netdevice_notifier(struct notifier_block *nb)
1348 struct net_device *dev;
1349 struct net_device *last;
1354 err = raw_notifier_chain_register(&netdev_chain, nb);
1360 for_each_netdev(net, dev) {
1361 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1362 err = notifier_to_errno(err);
1366 if (!(dev->flags & IFF_UP))
1369 nb->notifier_call(nb, NETDEV_UP, dev);
1380 for_each_netdev(net, dev) {
1384 if (dev->flags & IFF_UP) {
1385 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1386 nb->notifier_call(nb, NETDEV_DOWN, dev);
1388 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1389 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1393 raw_notifier_chain_unregister(&netdev_chain, nb);
1396 EXPORT_SYMBOL(register_netdevice_notifier);
1399 * unregister_netdevice_notifier - unregister a network notifier block
1402 * Unregister a notifier previously registered by
1403 * register_netdevice_notifier(). The notifier is unlinked into the
1404 * kernel structures and may then be reused. A negative errno code
1405 * is returned on a failure.
1408 int unregister_netdevice_notifier(struct notifier_block *nb)
1413 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1417 EXPORT_SYMBOL(unregister_netdevice_notifier);
1420 * call_netdevice_notifiers - call all network notifier blocks
1421 * @val: value passed unmodified to notifier function
1422 * @dev: net_device pointer passed unmodified to notifier function
1424 * Call all network notifier blocks. Parameters and return value
1425 * are as for raw_notifier_call_chain().
1428 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1431 return raw_notifier_call_chain(&netdev_chain, val, dev);
1434 /* When > 0 there are consumers of rx skb time stamps */
1435 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1437 void net_enable_timestamp(void)
1439 atomic_inc(&netstamp_needed);
1441 EXPORT_SYMBOL(net_enable_timestamp);
1443 void net_disable_timestamp(void)
1445 atomic_dec(&netstamp_needed);
1447 EXPORT_SYMBOL(net_disable_timestamp);
1449 static inline void net_timestamp_set(struct sk_buff *skb)
1451 if (atomic_read(&netstamp_needed))
1452 __net_timestamp(skb);
1454 skb->tstamp.tv64 = 0;
1457 static inline void net_timestamp_check(struct sk_buff *skb)
1459 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1460 __net_timestamp(skb);
1464 * dev_forward_skb - loopback an skb to another netif
1466 * @dev: destination network device
1467 * @skb: buffer to forward
1470 * NET_RX_SUCCESS (no congestion)
1471 * NET_RX_DROP (packet was dropped, but freed)
1473 * dev_forward_skb can be used for injecting an skb from the
1474 * start_xmit function of one device into the receive queue
1475 * of another device.
1477 * The receiving device may be in another namespace, so
1478 * we have to clear all information in the skb that could
1479 * impact namespace isolation.
1481 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1486 if (unlikely(!(dev->flags & IFF_UP) ||
1487 (skb->len > (dev->mtu + dev->hard_header_len)))) {
1488 atomic_long_inc(&dev->rx_dropped);
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 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1558 if (txq < 1 || txq > dev->num_tx_queues)
1561 if (dev->reg_state == NETREG_REGISTERED) {
1564 if (txq < dev->real_num_tx_queues)
1565 qdisc_reset_all_tx_gt(dev, txq);
1568 dev->real_num_tx_queues = txq;
1571 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1575 * netif_set_real_num_rx_queues - set actual number of RX queues used
1576 * @dev: Network device
1577 * @rxq: Actual number of RX queues
1579 * This must be called either with the rtnl_lock held or before
1580 * registration of the net device. Returns 0 on success, or a
1581 * negative error code. If called before registration, it always
1584 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1588 if (rxq < 1 || rxq > dev->num_rx_queues)
1591 if (dev->reg_state == NETREG_REGISTERED) {
1594 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1600 dev->real_num_rx_queues = rxq;
1603 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1606 static inline void __netif_reschedule(struct Qdisc *q)
1608 struct softnet_data *sd;
1609 unsigned long flags;
1611 local_irq_save(flags);
1612 sd = &__get_cpu_var(softnet_data);
1613 q->next_sched = NULL;
1614 *sd->output_queue_tailp = q;
1615 sd->output_queue_tailp = &q->next_sched;
1616 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1617 local_irq_restore(flags);
1620 void __netif_schedule(struct Qdisc *q)
1622 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1623 __netif_reschedule(q);
1625 EXPORT_SYMBOL(__netif_schedule);
1627 void dev_kfree_skb_irq(struct sk_buff *skb)
1629 if (atomic_dec_and_test(&skb->users)) {
1630 struct softnet_data *sd;
1631 unsigned long flags;
1633 local_irq_save(flags);
1634 sd = &__get_cpu_var(softnet_data);
1635 skb->next = sd->completion_queue;
1636 sd->completion_queue = skb;
1637 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1638 local_irq_restore(flags);
1641 EXPORT_SYMBOL(dev_kfree_skb_irq);
1643 void dev_kfree_skb_any(struct sk_buff *skb)
1645 if (in_irq() || irqs_disabled())
1646 dev_kfree_skb_irq(skb);
1650 EXPORT_SYMBOL(dev_kfree_skb_any);
1654 * netif_device_detach - mark device as removed
1655 * @dev: network device
1657 * Mark device as removed from system and therefore no longer available.
1659 void netif_device_detach(struct net_device *dev)
1661 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1662 netif_running(dev)) {
1663 netif_tx_stop_all_queues(dev);
1666 EXPORT_SYMBOL(netif_device_detach);
1669 * netif_device_attach - mark device as attached
1670 * @dev: network device
1672 * Mark device as attached from system and restart if needed.
1674 void netif_device_attach(struct net_device *dev)
1676 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1677 netif_running(dev)) {
1678 netif_tx_wake_all_queues(dev);
1679 __netdev_watchdog_up(dev);
1682 EXPORT_SYMBOL(netif_device_attach);
1684 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1686 return ((features & NETIF_F_GEN_CSUM) ||
1687 ((features & NETIF_F_IP_CSUM) &&
1688 protocol == htons(ETH_P_IP)) ||
1689 ((features & NETIF_F_IPV6_CSUM) &&
1690 protocol == htons(ETH_P_IPV6)) ||
1691 ((features & NETIF_F_FCOE_CRC) &&
1692 protocol == htons(ETH_P_FCOE)));
1695 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1697 if (can_checksum_protocol(dev->features, skb->protocol))
1700 if (skb->protocol == htons(ETH_P_8021Q)) {
1701 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1702 if (can_checksum_protocol(dev->features & dev->vlan_features,
1703 veh->h_vlan_encapsulated_proto))
1711 * skb_dev_set -- assign a new device to a buffer
1712 * @skb: buffer for the new device
1713 * @dev: network device
1715 * If an skb is owned by a device already, we have to reset
1716 * all data private to the namespace a device belongs to
1717 * before assigning it a new device.
1719 #ifdef CONFIG_NET_NS
1720 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1723 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1726 skb_init_secmark(skb);
1730 skb->ipvs_property = 0;
1731 #ifdef CONFIG_NET_SCHED
1737 EXPORT_SYMBOL(skb_set_dev);
1738 #endif /* CONFIG_NET_NS */
1741 * Invalidate hardware checksum when packet is to be mangled, and
1742 * complete checksum manually on outgoing path.
1744 int skb_checksum_help(struct sk_buff *skb)
1747 int ret = 0, offset;
1749 if (skb->ip_summed == CHECKSUM_COMPLETE)
1750 goto out_set_summed;
1752 if (unlikely(skb_shinfo(skb)->gso_size)) {
1753 /* Let GSO fix up the checksum. */
1754 goto out_set_summed;
1757 offset = skb->csum_start - skb_headroom(skb);
1758 BUG_ON(offset >= skb_headlen(skb));
1759 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1761 offset += skb->csum_offset;
1762 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1764 if (skb_cloned(skb) &&
1765 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1766 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1771 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1773 skb->ip_summed = CHECKSUM_NONE;
1777 EXPORT_SYMBOL(skb_checksum_help);
1780 * skb_gso_segment - Perform segmentation on skb.
1781 * @skb: buffer to segment
1782 * @features: features for the output path (see dev->features)
1784 * This function segments the given skb and returns a list of segments.
1786 * It may return NULL if the skb requires no segmentation. This is
1787 * only possible when GSO is used for verifying header integrity.
1789 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1791 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1792 struct packet_type *ptype;
1793 __be16 type = skb->protocol;
1796 skb_reset_mac_header(skb);
1797 skb->mac_len = skb->network_header - skb->mac_header;
1798 __skb_pull(skb, skb->mac_len);
1800 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1801 struct net_device *dev = skb->dev;
1802 struct ethtool_drvinfo info = {};
1804 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1805 dev->ethtool_ops->get_drvinfo(dev, &info);
1807 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1809 info.driver, dev ? dev->features : 0L,
1810 skb->sk ? skb->sk->sk_route_caps : 0L,
1811 skb->len, skb->data_len, skb->ip_summed);
1813 if (skb_header_cloned(skb) &&
1814 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1815 return ERR_PTR(err);
1819 list_for_each_entry_rcu(ptype,
1820 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1821 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1822 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1823 err = ptype->gso_send_check(skb);
1824 segs = ERR_PTR(err);
1825 if (err || skb_gso_ok(skb, features))
1827 __skb_push(skb, (skb->data -
1828 skb_network_header(skb)));
1830 segs = ptype->gso_segment(skb, features);
1836 __skb_push(skb, skb->data - skb_mac_header(skb));
1840 EXPORT_SYMBOL(skb_gso_segment);
1842 /* Take action when hardware reception checksum errors are detected. */
1844 void netdev_rx_csum_fault(struct net_device *dev)
1846 if (net_ratelimit()) {
1847 printk(KERN_ERR "%s: hw csum failure.\n",
1848 dev ? dev->name : "<unknown>");
1852 EXPORT_SYMBOL(netdev_rx_csum_fault);
1855 /* Actually, we should eliminate this check as soon as we know, that:
1856 * 1. IOMMU is present and allows to map all the memory.
1857 * 2. No high memory really exists on this machine.
1860 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1862 #ifdef CONFIG_HIGHMEM
1864 if (!(dev->features & NETIF_F_HIGHDMA)) {
1865 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1866 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1870 if (PCI_DMA_BUS_IS_PHYS) {
1871 struct device *pdev = dev->dev.parent;
1875 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1876 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1877 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1886 void (*destructor)(struct sk_buff *skb);
1889 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1891 static void dev_gso_skb_destructor(struct sk_buff *skb)
1893 struct dev_gso_cb *cb;
1896 struct sk_buff *nskb = skb->next;
1898 skb->next = nskb->next;
1901 } while (skb->next);
1903 cb = DEV_GSO_CB(skb);
1905 cb->destructor(skb);
1909 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1910 * @skb: buffer to segment
1912 * This function segments the given skb and stores the list of segments
1915 static int dev_gso_segment(struct sk_buff *skb)
1917 struct net_device *dev = skb->dev;
1918 struct sk_buff *segs;
1919 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1922 segs = skb_gso_segment(skb, features);
1924 /* Verifying header integrity only. */
1929 return PTR_ERR(segs);
1932 DEV_GSO_CB(skb)->destructor = skb->destructor;
1933 skb->destructor = dev_gso_skb_destructor;
1939 * Try to orphan skb early, right before transmission by the device.
1940 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1941 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1943 static inline void skb_orphan_try(struct sk_buff *skb)
1945 struct sock *sk = skb->sk;
1947 if (sk && !skb_shinfo(skb)->tx_flags) {
1948 /* skb_tx_hash() wont be able to get sk.
1949 * We copy sk_hash into skb->rxhash
1952 skb->rxhash = sk->sk_hash;
1958 * Returns true if either:
1959 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1960 * 2. skb is fragmented and the device does not support SG, or if
1961 * at least one of fragments is in highmem and device does not
1962 * support DMA from it.
1964 static inline int skb_needs_linearize(struct sk_buff *skb,
1965 struct net_device *dev)
1967 return skb_is_nonlinear(skb) &&
1968 ((skb_has_frag_list(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
1969 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
1970 illegal_highdma(dev, skb))));
1973 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1974 struct netdev_queue *txq)
1976 const struct net_device_ops *ops = dev->netdev_ops;
1977 int rc = NETDEV_TX_OK;
1979 if (likely(!skb->next)) {
1980 if (!list_empty(&ptype_all))
1981 dev_queue_xmit_nit(skb, dev);
1984 * If device doesnt need skb->dst, release it right now while
1985 * its hot in this cpu cache
1987 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1990 skb_orphan_try(skb);
1992 if (netif_needs_gso(dev, skb)) {
1993 if (unlikely(dev_gso_segment(skb)))
1998 if (skb_needs_linearize(skb, dev) &&
1999 __skb_linearize(skb))
2002 /* If packet is not checksummed and device does not
2003 * support checksumming for this protocol, complete
2004 * checksumming here.
2006 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2007 skb_set_transport_header(skb, skb->csum_start -
2009 if (!dev_can_checksum(dev, skb) &&
2010 skb_checksum_help(skb))
2015 rc = ops->ndo_start_xmit(skb, dev);
2016 if (rc == NETDEV_TX_OK)
2017 txq_trans_update(txq);
2023 struct sk_buff *nskb = skb->next;
2025 skb->next = nskb->next;
2029 * If device doesnt need nskb->dst, release it right now while
2030 * its hot in this cpu cache
2032 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2035 rc = ops->ndo_start_xmit(nskb, dev);
2036 if (unlikely(rc != NETDEV_TX_OK)) {
2037 if (rc & ~NETDEV_TX_MASK)
2038 goto out_kfree_gso_skb;
2039 nskb->next = skb->next;
2043 txq_trans_update(txq);
2044 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2045 return NETDEV_TX_BUSY;
2046 } while (skb->next);
2049 if (likely(skb->next == NULL))
2050 skb->destructor = DEV_GSO_CB(skb)->destructor;
2056 static u32 hashrnd __read_mostly;
2058 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
2062 if (skb_rx_queue_recorded(skb)) {
2063 hash = skb_get_rx_queue(skb);
2064 while (unlikely(hash >= dev->real_num_tx_queues))
2065 hash -= dev->real_num_tx_queues;
2069 if (skb->sk && skb->sk->sk_hash)
2070 hash = skb->sk->sk_hash;
2072 hash = (__force u16) skb->protocol ^ skb->rxhash;
2073 hash = jhash_1word(hash, hashrnd);
2075 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2077 EXPORT_SYMBOL(skb_tx_hash);
2079 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2081 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2082 if (net_ratelimit()) {
2083 pr_warning("%s selects TX queue %d, but "
2084 "real number of TX queues is %d\n",
2085 dev->name, queue_index, dev->real_num_tx_queues);
2092 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2093 struct sk_buff *skb)
2096 const struct net_device_ops *ops = dev->netdev_ops;
2098 if (ops->ndo_select_queue) {
2099 queue_index = ops->ndo_select_queue(dev, skb);
2100 queue_index = dev_cap_txqueue(dev, queue_index);
2102 struct sock *sk = skb->sk;
2103 queue_index = sk_tx_queue_get(sk);
2104 if (queue_index < 0) {
2107 if (dev->real_num_tx_queues > 1)
2108 queue_index = skb_tx_hash(dev, skb);
2111 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2113 if (dst && skb_dst(skb) == dst)
2114 sk_tx_queue_set(sk, queue_index);
2119 skb_set_queue_mapping(skb, queue_index);
2120 return netdev_get_tx_queue(dev, queue_index);
2123 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2124 struct net_device *dev,
2125 struct netdev_queue *txq)
2127 spinlock_t *root_lock = qdisc_lock(q);
2128 bool contended = qdisc_is_running(q);
2132 * Heuristic to force contended enqueues to serialize on a
2133 * separate lock before trying to get qdisc main lock.
2134 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2135 * and dequeue packets faster.
2137 if (unlikely(contended))
2138 spin_lock(&q->busylock);
2140 spin_lock(root_lock);
2141 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2144 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2145 qdisc_run_begin(q)) {
2147 * This is a work-conserving queue; there are no old skbs
2148 * waiting to be sent out; and the qdisc is not running -
2149 * xmit the skb directly.
2151 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2153 __qdisc_update_bstats(q, skb->len);
2154 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2155 if (unlikely(contended)) {
2156 spin_unlock(&q->busylock);
2163 rc = NET_XMIT_SUCCESS;
2166 rc = qdisc_enqueue_root(skb, q);
2167 if (qdisc_run_begin(q)) {
2168 if (unlikely(contended)) {
2169 spin_unlock(&q->busylock);
2175 spin_unlock(root_lock);
2176 if (unlikely(contended))
2177 spin_unlock(&q->busylock);
2181 static DEFINE_PER_CPU(int, xmit_recursion);
2182 #define RECURSION_LIMIT 3
2185 * dev_queue_xmit - transmit a buffer
2186 * @skb: buffer to transmit
2188 * Queue a buffer for transmission to a network device. The caller must
2189 * have set the device and priority and built the buffer before calling
2190 * this function. The function can be called from an interrupt.
2192 * A negative errno code is returned on a failure. A success does not
2193 * guarantee the frame will be transmitted as it may be dropped due
2194 * to congestion or traffic shaping.
2196 * -----------------------------------------------------------------------------------
2197 * I notice this method can also return errors from the queue disciplines,
2198 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2201 * Regardless of the return value, the skb is consumed, so it is currently
2202 * difficult to retry a send to this method. (You can bump the ref count
2203 * before sending to hold a reference for retry if you are careful.)
2205 * When calling this method, interrupts MUST be enabled. This is because
2206 * the BH enable code must have IRQs enabled so that it will not deadlock.
2209 int dev_queue_xmit(struct sk_buff *skb)
2211 struct net_device *dev = skb->dev;
2212 struct netdev_queue *txq;
2216 /* Disable soft irqs for various locks below. Also
2217 * stops preemption for RCU.
2221 txq = dev_pick_tx(dev, skb);
2222 q = rcu_dereference_bh(txq->qdisc);
2224 #ifdef CONFIG_NET_CLS_ACT
2225 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2228 rc = __dev_xmit_skb(skb, q, dev, txq);
2232 /* The device has no queue. Common case for software devices:
2233 loopback, all the sorts of tunnels...
2235 Really, it is unlikely that netif_tx_lock protection is necessary
2236 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2238 However, it is possible, that they rely on protection
2241 Check this and shot the lock. It is not prone from deadlocks.
2242 Either shot noqueue qdisc, it is even simpler 8)
2244 if (dev->flags & IFF_UP) {
2245 int cpu = smp_processor_id(); /* ok because BHs are off */
2247 if (txq->xmit_lock_owner != cpu) {
2249 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2250 goto recursion_alert;
2252 HARD_TX_LOCK(dev, txq, cpu);
2254 if (!netif_tx_queue_stopped(txq)) {
2255 __this_cpu_inc(xmit_recursion);
2256 rc = dev_hard_start_xmit(skb, dev, txq);
2257 __this_cpu_dec(xmit_recursion);
2258 if (dev_xmit_complete(rc)) {
2259 HARD_TX_UNLOCK(dev, txq);
2263 HARD_TX_UNLOCK(dev, txq);
2264 if (net_ratelimit())
2265 printk(KERN_CRIT "Virtual device %s asks to "
2266 "queue packet!\n", dev->name);
2268 /* Recursion is detected! It is possible,
2272 if (net_ratelimit())
2273 printk(KERN_CRIT "Dead loop on virtual device "
2274 "%s, fix it urgently!\n", dev->name);
2279 rcu_read_unlock_bh();
2284 rcu_read_unlock_bh();
2287 EXPORT_SYMBOL(dev_queue_xmit);
2290 /*=======================================================================
2292 =======================================================================*/
2294 int netdev_max_backlog __read_mostly = 1000;
2295 int netdev_tstamp_prequeue __read_mostly = 1;
2296 int netdev_budget __read_mostly = 300;
2297 int weight_p __read_mostly = 64; /* old backlog weight */
2299 /* Called with irq disabled */
2300 static inline void ____napi_schedule(struct softnet_data *sd,
2301 struct napi_struct *napi)
2303 list_add_tail(&napi->poll_list, &sd->poll_list);
2304 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2308 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2309 * and src/dst port numbers. Returns a non-zero hash number on success
2312 __u32 __skb_get_rxhash(struct sk_buff *skb)
2314 int nhoff, hash = 0, poff;
2315 struct ipv6hdr *ip6;
2318 u32 addr1, addr2, ihl;
2324 nhoff = skb_network_offset(skb);
2326 switch (skb->protocol) {
2327 case __constant_htons(ETH_P_IP):
2328 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2331 ip = (struct iphdr *) (skb->data + nhoff);
2332 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2335 ip_proto = ip->protocol;
2336 addr1 = (__force u32) ip->saddr;
2337 addr2 = (__force u32) ip->daddr;
2340 case __constant_htons(ETH_P_IPV6):
2341 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2344 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2345 ip_proto = ip6->nexthdr;
2346 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2347 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2355 poff = proto_ports_offset(ip_proto);
2357 nhoff += ihl * 4 + poff;
2358 if (pskb_may_pull(skb, nhoff + 4)) {
2359 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2360 if (ports.v16[1] < ports.v16[0])
2361 swap(ports.v16[0], ports.v16[1]);
2365 /* get a consistent hash (same value on both flow directions) */
2369 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2376 EXPORT_SYMBOL(__skb_get_rxhash);
2380 /* One global table that all flow-based protocols share. */
2381 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2382 EXPORT_SYMBOL(rps_sock_flow_table);
2385 * get_rps_cpu is called from netif_receive_skb and returns the target
2386 * CPU from the RPS map of the receiving queue for a given skb.
2387 * rcu_read_lock must be held on entry.
2389 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2390 struct rps_dev_flow **rflowp)
2392 struct netdev_rx_queue *rxqueue;
2393 struct rps_map *map = NULL;
2394 struct rps_dev_flow_table *flow_table;
2395 struct rps_sock_flow_table *sock_flow_table;
2399 if (skb_rx_queue_recorded(skb)) {
2400 u16 index = skb_get_rx_queue(skb);
2401 if (unlikely(index >= dev->real_num_rx_queues)) {
2402 WARN_ONCE(dev->real_num_rx_queues > 1,
2403 "%s received packet on queue %u, but number "
2404 "of RX queues is %u\n",
2405 dev->name, index, dev->real_num_rx_queues);
2408 rxqueue = dev->_rx + index;
2412 if (rxqueue->rps_map) {
2413 map = rcu_dereference(rxqueue->rps_map);
2414 if (map && map->len == 1) {
2415 tcpu = map->cpus[0];
2416 if (cpu_online(tcpu))
2420 } else if (!rxqueue->rps_flow_table) {
2424 skb_reset_network_header(skb);
2425 if (!skb_get_rxhash(skb))
2428 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2429 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2430 if (flow_table && sock_flow_table) {
2432 struct rps_dev_flow *rflow;
2434 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2437 next_cpu = sock_flow_table->ents[skb->rxhash &
2438 sock_flow_table->mask];
2441 * If the desired CPU (where last recvmsg was done) is
2442 * different from current CPU (one in the rx-queue flow
2443 * table entry), switch if one of the following holds:
2444 * - Current CPU is unset (equal to RPS_NO_CPU).
2445 * - Current CPU is offline.
2446 * - The current CPU's queue tail has advanced beyond the
2447 * last packet that was enqueued using this table entry.
2448 * This guarantees that all previous packets for the flow
2449 * have been dequeued, thus preserving in order delivery.
2451 if (unlikely(tcpu != next_cpu) &&
2452 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2453 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2454 rflow->last_qtail)) >= 0)) {
2455 tcpu = rflow->cpu = next_cpu;
2456 if (tcpu != RPS_NO_CPU)
2457 rflow->last_qtail = per_cpu(softnet_data,
2458 tcpu).input_queue_head;
2460 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2468 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2470 if (cpu_online(tcpu)) {
2480 /* Called from hardirq (IPI) context */
2481 static void rps_trigger_softirq(void *data)
2483 struct softnet_data *sd = data;
2485 ____napi_schedule(sd, &sd->backlog);
2489 #endif /* CONFIG_RPS */
2492 * Check if this softnet_data structure is another cpu one
2493 * If yes, queue it to our IPI list and return 1
2496 static int rps_ipi_queued(struct softnet_data *sd)
2499 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2502 sd->rps_ipi_next = mysd->rps_ipi_list;
2503 mysd->rps_ipi_list = sd;
2505 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2508 #endif /* CONFIG_RPS */
2513 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2514 * queue (may be a remote CPU queue).
2516 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2517 unsigned int *qtail)
2519 struct softnet_data *sd;
2520 unsigned long flags;
2522 sd = &per_cpu(softnet_data, cpu);
2524 local_irq_save(flags);
2527 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2528 if (skb_queue_len(&sd->input_pkt_queue)) {
2530 __skb_queue_tail(&sd->input_pkt_queue, skb);
2531 input_queue_tail_incr_save(sd, qtail);
2533 local_irq_restore(flags);
2534 return NET_RX_SUCCESS;
2537 /* Schedule NAPI for backlog device
2538 * We can use non atomic operation since we own the queue lock
2540 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2541 if (!rps_ipi_queued(sd))
2542 ____napi_schedule(sd, &sd->backlog);
2550 local_irq_restore(flags);
2552 atomic_long_inc(&skb->dev->rx_dropped);
2558 * netif_rx - post buffer to the network code
2559 * @skb: buffer to post
2561 * This function receives a packet from a device driver and queues it for
2562 * the upper (protocol) levels to process. It always succeeds. The buffer
2563 * may be dropped during processing for congestion control or by the
2567 * NET_RX_SUCCESS (no congestion)
2568 * NET_RX_DROP (packet was dropped)
2572 int netif_rx(struct sk_buff *skb)
2576 /* if netpoll wants it, pretend we never saw it */
2577 if (netpoll_rx(skb))
2580 if (netdev_tstamp_prequeue)
2581 net_timestamp_check(skb);
2585 struct rps_dev_flow voidflow, *rflow = &voidflow;
2591 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2593 cpu = smp_processor_id();
2595 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2603 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2609 EXPORT_SYMBOL(netif_rx);
2611 int netif_rx_ni(struct sk_buff *skb)
2616 err = netif_rx(skb);
2617 if (local_softirq_pending())
2623 EXPORT_SYMBOL(netif_rx_ni);
2625 static void net_tx_action(struct softirq_action *h)
2627 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2629 if (sd->completion_queue) {
2630 struct sk_buff *clist;
2632 local_irq_disable();
2633 clist = sd->completion_queue;
2634 sd->completion_queue = NULL;
2638 struct sk_buff *skb = clist;
2639 clist = clist->next;
2641 WARN_ON(atomic_read(&skb->users));
2646 if (sd->output_queue) {
2649 local_irq_disable();
2650 head = sd->output_queue;
2651 sd->output_queue = NULL;
2652 sd->output_queue_tailp = &sd->output_queue;
2656 struct Qdisc *q = head;
2657 spinlock_t *root_lock;
2659 head = head->next_sched;
2661 root_lock = qdisc_lock(q);
2662 if (spin_trylock(root_lock)) {
2663 smp_mb__before_clear_bit();
2664 clear_bit(__QDISC_STATE_SCHED,
2667 spin_unlock(root_lock);
2669 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2671 __netif_reschedule(q);
2673 smp_mb__before_clear_bit();
2674 clear_bit(__QDISC_STATE_SCHED,
2682 static inline int deliver_skb(struct sk_buff *skb,
2683 struct packet_type *pt_prev,
2684 struct net_device *orig_dev)
2686 atomic_inc(&skb->users);
2687 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2690 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2691 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2692 /* This hook is defined here for ATM LANE */
2693 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2694 unsigned char *addr) __read_mostly;
2695 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2698 #ifdef CONFIG_NET_CLS_ACT
2699 /* TODO: Maybe we should just force sch_ingress to be compiled in
2700 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2701 * a compare and 2 stores extra right now if we dont have it on
2702 * but have CONFIG_NET_CLS_ACT
2703 * NOTE: This doesnt stop any functionality; if you dont have
2704 * the ingress scheduler, you just cant add policies on ingress.
2707 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2709 struct net_device *dev = skb->dev;
2710 u32 ttl = G_TC_RTTL(skb->tc_verd);
2711 int result = TC_ACT_OK;
2714 if (unlikely(MAX_RED_LOOP < ttl++)) {
2715 if (net_ratelimit())
2716 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2717 skb->skb_iif, dev->ifindex);
2721 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2722 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2725 if (q != &noop_qdisc) {
2726 spin_lock(qdisc_lock(q));
2727 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2728 result = qdisc_enqueue_root(skb, q);
2729 spin_unlock(qdisc_lock(q));
2735 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2736 struct packet_type **pt_prev,
2737 int *ret, struct net_device *orig_dev)
2739 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
2741 if (!rxq || rxq->qdisc == &noop_qdisc)
2745 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2749 switch (ing_filter(skb, rxq)) {
2763 * netif_nit_deliver - deliver received packets to network taps
2766 * This function is used to deliver incoming packets to network
2767 * taps. It should be used when the normal netif_receive_skb path
2768 * is bypassed, for example because of VLAN acceleration.
2770 void netif_nit_deliver(struct sk_buff *skb)
2772 struct packet_type *ptype;
2774 if (list_empty(&ptype_all))
2777 skb_reset_network_header(skb);
2778 skb_reset_transport_header(skb);
2779 skb->mac_len = skb->network_header - skb->mac_header;
2782 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2783 if (!ptype->dev || ptype->dev == skb->dev)
2784 deliver_skb(skb, ptype, skb->dev);
2790 * netdev_rx_handler_register - register receive handler
2791 * @dev: device to register a handler for
2792 * @rx_handler: receive handler to register
2793 * @rx_handler_data: data pointer that is used by rx handler
2795 * Register a receive hander for a device. This handler will then be
2796 * called from __netif_receive_skb. A negative errno code is returned
2799 * The caller must hold the rtnl_mutex.
2801 int netdev_rx_handler_register(struct net_device *dev,
2802 rx_handler_func_t *rx_handler,
2803 void *rx_handler_data)
2807 if (dev->rx_handler)
2810 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2811 rcu_assign_pointer(dev->rx_handler, rx_handler);
2815 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2818 * netdev_rx_handler_unregister - unregister receive handler
2819 * @dev: device to unregister a handler from
2821 * Unregister a receive hander from a device.
2823 * The caller must hold the rtnl_mutex.
2825 void netdev_rx_handler_unregister(struct net_device *dev)
2829 rcu_assign_pointer(dev->rx_handler, NULL);
2830 rcu_assign_pointer(dev->rx_handler_data, NULL);
2832 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2834 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2835 struct net_device *master)
2837 if (skb->pkt_type == PACKET_HOST) {
2838 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2840 memcpy(dest, master->dev_addr, ETH_ALEN);
2844 /* On bonding slaves other than the currently active slave, suppress
2845 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2846 * ARP on active-backup slaves with arp_validate enabled.
2848 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2850 struct net_device *dev = skb->dev;
2852 if (master->priv_flags & IFF_MASTER_ARPMON)
2853 dev->last_rx = jiffies;
2855 if ((master->priv_flags & IFF_MASTER_ALB) &&
2856 (master->priv_flags & IFF_BRIDGE_PORT)) {
2857 /* Do address unmangle. The local destination address
2858 * will be always the one master has. Provides the right
2859 * functionality in a bridge.
2861 skb_bond_set_mac_by_master(skb, master);
2864 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2865 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2866 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2869 if (master->priv_flags & IFF_MASTER_ALB) {
2870 if (skb->pkt_type != PACKET_BROADCAST &&
2871 skb->pkt_type != PACKET_MULTICAST)
2874 if (master->priv_flags & IFF_MASTER_8023AD &&
2875 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2882 EXPORT_SYMBOL(__skb_bond_should_drop);
2884 static int __netif_receive_skb(struct sk_buff *skb)
2886 struct packet_type *ptype, *pt_prev;
2887 rx_handler_func_t *rx_handler;
2888 struct net_device *orig_dev;
2889 struct net_device *master;
2890 struct net_device *null_or_orig;
2891 struct net_device *orig_or_bond;
2892 int ret = NET_RX_DROP;
2895 if (!netdev_tstamp_prequeue)
2896 net_timestamp_check(skb);
2898 if (vlan_tx_tag_present(skb))
2899 vlan_hwaccel_do_receive(skb);
2901 /* if we've gotten here through NAPI, check netpoll */
2902 if (netpoll_receive_skb(skb))
2906 skb->skb_iif = skb->dev->ifindex;
2909 * bonding note: skbs received on inactive slaves should only
2910 * be delivered to pkt handlers that are exact matches. Also
2911 * the deliver_no_wcard flag will be set. If packet handlers
2912 * are sensitive to duplicate packets these skbs will need to
2913 * be dropped at the handler. The vlan accel path may have
2914 * already set the deliver_no_wcard flag.
2916 null_or_orig = NULL;
2917 orig_dev = skb->dev;
2918 master = ACCESS_ONCE(orig_dev->master);
2919 if (skb->deliver_no_wcard)
2920 null_or_orig = orig_dev;
2922 if (skb_bond_should_drop(skb, master)) {
2923 skb->deliver_no_wcard = 1;
2924 null_or_orig = orig_dev; /* deliver only exact match */
2929 __this_cpu_inc(softnet_data.processed);
2930 skb_reset_network_header(skb);
2931 skb_reset_transport_header(skb);
2932 skb->mac_len = skb->network_header - skb->mac_header;
2938 #ifdef CONFIG_NET_CLS_ACT
2939 if (skb->tc_verd & TC_NCLS) {
2940 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2945 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2946 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2947 ptype->dev == orig_dev) {
2949 ret = deliver_skb(skb, pt_prev, orig_dev);
2954 #ifdef CONFIG_NET_CLS_ACT
2955 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2961 /* Handle special case of bridge or macvlan */
2962 rx_handler = rcu_dereference(skb->dev->rx_handler);
2965 ret = deliver_skb(skb, pt_prev, orig_dev);
2968 skb = rx_handler(skb);
2974 * Make sure frames received on VLAN interfaces stacked on
2975 * bonding interfaces still make their way to any base bonding
2976 * device that may have registered for a specific ptype. The
2977 * handler may have to adjust skb->dev and orig_dev.
2979 orig_or_bond = orig_dev;
2980 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2981 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2982 orig_or_bond = vlan_dev_real_dev(skb->dev);
2985 type = skb->protocol;
2986 list_for_each_entry_rcu(ptype,
2987 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2988 if (ptype->type == type && (ptype->dev == null_or_orig ||
2989 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2990 ptype->dev == orig_or_bond)) {
2992 ret = deliver_skb(skb, pt_prev, orig_dev);
2998 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3000 atomic_long_inc(&skb->dev->rx_dropped);
3002 /* Jamal, now you will not able to escape explaining
3003 * me how you were going to use this. :-)
3014 * netif_receive_skb - process receive buffer from network
3015 * @skb: buffer to process
3017 * netif_receive_skb() is the main receive data processing function.
3018 * It always succeeds. The buffer may be dropped during processing
3019 * for congestion control or by the protocol layers.
3021 * This function may only be called from softirq context and interrupts
3022 * should be enabled.
3024 * Return values (usually ignored):
3025 * NET_RX_SUCCESS: no congestion
3026 * NET_RX_DROP: packet was dropped
3028 int netif_receive_skb(struct sk_buff *skb)
3030 if (netdev_tstamp_prequeue)
3031 net_timestamp_check(skb);
3033 if (skb_defer_rx_timestamp(skb))
3034 return NET_RX_SUCCESS;
3038 struct rps_dev_flow voidflow, *rflow = &voidflow;
3043 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3046 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3050 ret = __netif_receive_skb(skb);
3056 return __netif_receive_skb(skb);
3059 EXPORT_SYMBOL(netif_receive_skb);
3061 /* Network device is going away, flush any packets still pending
3062 * Called with irqs disabled.
3064 static void flush_backlog(void *arg)
3066 struct net_device *dev = arg;
3067 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3068 struct sk_buff *skb, *tmp;
3071 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3072 if (skb->dev == dev) {
3073 __skb_unlink(skb, &sd->input_pkt_queue);
3075 input_queue_head_incr(sd);
3080 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3081 if (skb->dev == dev) {
3082 __skb_unlink(skb, &sd->process_queue);
3084 input_queue_head_incr(sd);
3089 static int napi_gro_complete(struct sk_buff *skb)
3091 struct packet_type *ptype;
3092 __be16 type = skb->protocol;
3093 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3096 if (NAPI_GRO_CB(skb)->count == 1) {
3097 skb_shinfo(skb)->gso_size = 0;
3102 list_for_each_entry_rcu(ptype, head, list) {
3103 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3106 err = ptype->gro_complete(skb);
3112 WARN_ON(&ptype->list == head);
3114 return NET_RX_SUCCESS;
3118 return netif_receive_skb(skb);
3121 inline void napi_gro_flush(struct napi_struct *napi)
3123 struct sk_buff *skb, *next;
3125 for (skb = napi->gro_list; skb; skb = next) {
3128 napi_gro_complete(skb);
3131 napi->gro_count = 0;
3132 napi->gro_list = NULL;
3134 EXPORT_SYMBOL(napi_gro_flush);
3136 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3138 struct sk_buff **pp = NULL;
3139 struct packet_type *ptype;
3140 __be16 type = skb->protocol;
3141 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3144 enum gro_result ret;
3146 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3149 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3153 list_for_each_entry_rcu(ptype, head, list) {
3154 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3157 skb_set_network_header(skb, skb_gro_offset(skb));
3158 mac_len = skb->network_header - skb->mac_header;
3159 skb->mac_len = mac_len;
3160 NAPI_GRO_CB(skb)->same_flow = 0;
3161 NAPI_GRO_CB(skb)->flush = 0;
3162 NAPI_GRO_CB(skb)->free = 0;
3164 pp = ptype->gro_receive(&napi->gro_list, skb);
3169 if (&ptype->list == head)
3172 same_flow = NAPI_GRO_CB(skb)->same_flow;
3173 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3176 struct sk_buff *nskb = *pp;
3180 napi_gro_complete(nskb);
3187 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3191 NAPI_GRO_CB(skb)->count = 1;
3192 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3193 skb->next = napi->gro_list;
3194 napi->gro_list = skb;
3198 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3199 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3201 BUG_ON(skb->end - skb->tail < grow);
3203 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3206 skb->data_len -= grow;
3208 skb_shinfo(skb)->frags[0].page_offset += grow;
3209 skb_shinfo(skb)->frags[0].size -= grow;
3211 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3212 put_page(skb_shinfo(skb)->frags[0].page);
3213 memmove(skb_shinfo(skb)->frags,
3214 skb_shinfo(skb)->frags + 1,
3215 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3226 EXPORT_SYMBOL(dev_gro_receive);
3228 static inline gro_result_t
3229 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3233 for (p = napi->gro_list; p; p = p->next) {
3234 unsigned long diffs;
3236 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3237 diffs |= compare_ether_header(skb_mac_header(p),
3238 skb_gro_mac_header(skb));
3239 NAPI_GRO_CB(p)->same_flow = !diffs;
3240 NAPI_GRO_CB(p)->flush = 0;
3243 return dev_gro_receive(napi, skb);
3246 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3250 if (netif_receive_skb(skb))
3255 case GRO_MERGED_FREE:
3266 EXPORT_SYMBOL(napi_skb_finish);
3268 void skb_gro_reset_offset(struct sk_buff *skb)
3270 NAPI_GRO_CB(skb)->data_offset = 0;
3271 NAPI_GRO_CB(skb)->frag0 = NULL;
3272 NAPI_GRO_CB(skb)->frag0_len = 0;
3274 if (skb->mac_header == skb->tail &&
3275 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3276 NAPI_GRO_CB(skb)->frag0 =
3277 page_address(skb_shinfo(skb)->frags[0].page) +
3278 skb_shinfo(skb)->frags[0].page_offset;
3279 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3282 EXPORT_SYMBOL(skb_gro_reset_offset);
3284 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3286 skb_gro_reset_offset(skb);
3288 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3290 EXPORT_SYMBOL(napi_gro_receive);
3292 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3294 __skb_pull(skb, skb_headlen(skb));
3295 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3299 EXPORT_SYMBOL(napi_reuse_skb);
3301 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3303 struct sk_buff *skb = napi->skb;
3306 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3312 EXPORT_SYMBOL(napi_get_frags);
3314 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3320 skb->protocol = eth_type_trans(skb, skb->dev);
3322 if (ret == GRO_HELD)
3323 skb_gro_pull(skb, -ETH_HLEN);
3324 else if (netif_receive_skb(skb))
3329 case GRO_MERGED_FREE:
3330 napi_reuse_skb(napi, skb);
3339 EXPORT_SYMBOL(napi_frags_finish);
3341 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3343 struct sk_buff *skb = napi->skb;
3350 skb_reset_mac_header(skb);
3351 skb_gro_reset_offset(skb);
3353 off = skb_gro_offset(skb);
3354 hlen = off + sizeof(*eth);
3355 eth = skb_gro_header_fast(skb, off);
3356 if (skb_gro_header_hard(skb, hlen)) {
3357 eth = skb_gro_header_slow(skb, hlen, off);
3358 if (unlikely(!eth)) {
3359 napi_reuse_skb(napi, skb);
3365 skb_gro_pull(skb, sizeof(*eth));
3368 * This works because the only protocols we care about don't require
3369 * special handling. We'll fix it up properly at the end.
3371 skb->protocol = eth->h_proto;
3376 EXPORT_SYMBOL(napi_frags_skb);
3378 gro_result_t napi_gro_frags(struct napi_struct *napi)
3380 struct sk_buff *skb = napi_frags_skb(napi);
3385 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3387 EXPORT_SYMBOL(napi_gro_frags);
3390 * net_rps_action sends any pending IPI's for rps.
3391 * Note: called with local irq disabled, but exits with local irq enabled.
3393 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3396 struct softnet_data *remsd = sd->rps_ipi_list;
3399 sd->rps_ipi_list = NULL;
3403 /* Send pending IPI's to kick RPS processing on remote cpus. */
3405 struct softnet_data *next = remsd->rps_ipi_next;
3407 if (cpu_online(remsd->cpu))
3408 __smp_call_function_single(remsd->cpu,
3417 static int process_backlog(struct napi_struct *napi, int quota)
3420 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3423 /* Check if we have pending ipi, its better to send them now,
3424 * not waiting net_rx_action() end.
3426 if (sd->rps_ipi_list) {
3427 local_irq_disable();
3428 net_rps_action_and_irq_enable(sd);
3431 napi->weight = weight_p;
3432 local_irq_disable();
3433 while (work < quota) {
3434 struct sk_buff *skb;
3437 while ((skb = __skb_dequeue(&sd->process_queue))) {
3439 __netif_receive_skb(skb);
3440 local_irq_disable();
3441 input_queue_head_incr(sd);
3442 if (++work >= quota) {
3449 qlen = skb_queue_len(&sd->input_pkt_queue);
3451 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3452 &sd->process_queue);
3454 if (qlen < quota - work) {
3456 * Inline a custom version of __napi_complete().
3457 * only current cpu owns and manipulates this napi,
3458 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3459 * we can use a plain write instead of clear_bit(),
3460 * and we dont need an smp_mb() memory barrier.
3462 list_del(&napi->poll_list);
3465 quota = work + qlen;
3475 * __napi_schedule - schedule for receive
3476 * @n: entry to schedule
3478 * The entry's receive function will be scheduled to run
3480 void __napi_schedule(struct napi_struct *n)
3482 unsigned long flags;
3484 local_irq_save(flags);
3485 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3486 local_irq_restore(flags);
3488 EXPORT_SYMBOL(__napi_schedule);
3490 void __napi_complete(struct napi_struct *n)
3492 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3493 BUG_ON(n->gro_list);
3495 list_del(&n->poll_list);
3496 smp_mb__before_clear_bit();
3497 clear_bit(NAPI_STATE_SCHED, &n->state);
3499 EXPORT_SYMBOL(__napi_complete);
3501 void napi_complete(struct napi_struct *n)
3503 unsigned long flags;
3506 * don't let napi dequeue from the cpu poll list
3507 * just in case its running on a different cpu
3509 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3513 local_irq_save(flags);
3515 local_irq_restore(flags);
3517 EXPORT_SYMBOL(napi_complete);
3519 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3520 int (*poll)(struct napi_struct *, int), int weight)
3522 INIT_LIST_HEAD(&napi->poll_list);
3523 napi->gro_count = 0;
3524 napi->gro_list = NULL;
3527 napi->weight = weight;
3528 list_add(&napi->dev_list, &dev->napi_list);
3530 #ifdef CONFIG_NETPOLL
3531 spin_lock_init(&napi->poll_lock);
3532 napi->poll_owner = -1;
3534 set_bit(NAPI_STATE_SCHED, &napi->state);
3536 EXPORT_SYMBOL(netif_napi_add);
3538 void netif_napi_del(struct napi_struct *napi)
3540 struct sk_buff *skb, *next;
3542 list_del_init(&napi->dev_list);
3543 napi_free_frags(napi);
3545 for (skb = napi->gro_list; skb; skb = next) {
3551 napi->gro_list = NULL;
3552 napi->gro_count = 0;
3554 EXPORT_SYMBOL(netif_napi_del);
3556 static void net_rx_action(struct softirq_action *h)
3558 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3559 unsigned long time_limit = jiffies + 2;
3560 int budget = netdev_budget;
3563 local_irq_disable();
3565 while (!list_empty(&sd->poll_list)) {
3566 struct napi_struct *n;
3569 /* If softirq window is exhuasted then punt.
3570 * Allow this to run for 2 jiffies since which will allow
3571 * an average latency of 1.5/HZ.
3573 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3578 /* Even though interrupts have been re-enabled, this
3579 * access is safe because interrupts can only add new
3580 * entries to the tail of this list, and only ->poll()
3581 * calls can remove this head entry from the list.
3583 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3585 have = netpoll_poll_lock(n);
3589 /* This NAPI_STATE_SCHED test is for avoiding a race
3590 * with netpoll's poll_napi(). Only the entity which
3591 * obtains the lock and sees NAPI_STATE_SCHED set will
3592 * actually make the ->poll() call. Therefore we avoid
3593 * accidently calling ->poll() when NAPI is not scheduled.
3596 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3597 work = n->poll(n, weight);
3601 WARN_ON_ONCE(work > weight);
3605 local_irq_disable();
3607 /* Drivers must not modify the NAPI state if they
3608 * consume the entire weight. In such cases this code
3609 * still "owns" the NAPI instance and therefore can
3610 * move the instance around on the list at-will.
3612 if (unlikely(work == weight)) {
3613 if (unlikely(napi_disable_pending(n))) {
3616 local_irq_disable();
3618 list_move_tail(&n->poll_list, &sd->poll_list);
3621 netpoll_poll_unlock(have);
3624 net_rps_action_and_irq_enable(sd);
3626 #ifdef CONFIG_NET_DMA
3628 * There may not be any more sk_buffs coming right now, so push
3629 * any pending DMA copies to hardware
3631 dma_issue_pending_all();
3638 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3642 static gifconf_func_t *gifconf_list[NPROTO];
3645 * register_gifconf - register a SIOCGIF handler
3646 * @family: Address family
3647 * @gifconf: Function handler
3649 * Register protocol dependent address dumping routines. The handler
3650 * that is passed must not be freed or reused until it has been replaced
3651 * by another handler.
3653 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3655 if (family >= NPROTO)
3657 gifconf_list[family] = gifconf;
3660 EXPORT_SYMBOL(register_gifconf);
3664 * Map an interface index to its name (SIOCGIFNAME)
3668 * We need this ioctl for efficient implementation of the
3669 * if_indextoname() function required by the IPv6 API. Without
3670 * it, we would have to search all the interfaces to find a
3674 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3676 struct net_device *dev;
3680 * Fetch the caller's info block.
3683 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3687 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3693 strcpy(ifr.ifr_name, dev->name);
3696 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3702 * Perform a SIOCGIFCONF call. This structure will change
3703 * size eventually, and there is nothing I can do about it.
3704 * Thus we will need a 'compatibility mode'.
3707 static int dev_ifconf(struct net *net, char __user *arg)
3710 struct net_device *dev;
3717 * Fetch the caller's info block.
3720 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3727 * Loop over the interfaces, and write an info block for each.
3731 for_each_netdev(net, dev) {
3732 for (i = 0; i < NPROTO; i++) {
3733 if (gifconf_list[i]) {
3736 done = gifconf_list[i](dev, NULL, 0);
3738 done = gifconf_list[i](dev, pos + total,
3748 * All done. Write the updated control block back to the caller.
3750 ifc.ifc_len = total;
3753 * Both BSD and Solaris return 0 here, so we do too.
3755 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3758 #ifdef CONFIG_PROC_FS
3760 * This is invoked by the /proc filesystem handler to display a device
3763 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3766 struct net *net = seq_file_net(seq);
3768 struct net_device *dev;
3772 return SEQ_START_TOKEN;
3775 for_each_netdev_rcu(net, dev)
3782 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3784 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3785 first_net_device(seq_file_net(seq)) :
3786 next_net_device((struct net_device *)v);
3789 return rcu_dereference(dev);
3792 void dev_seq_stop(struct seq_file *seq, void *v)
3798 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3800 struct rtnl_link_stats64 temp;
3801 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3803 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3804 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3805 dev->name, stats->rx_bytes, stats->rx_packets,
3807 stats->rx_dropped + stats->rx_missed_errors,
3808 stats->rx_fifo_errors,
3809 stats->rx_length_errors + stats->rx_over_errors +
3810 stats->rx_crc_errors + stats->rx_frame_errors,
3811 stats->rx_compressed, stats->multicast,
3812 stats->tx_bytes, stats->tx_packets,
3813 stats->tx_errors, stats->tx_dropped,
3814 stats->tx_fifo_errors, stats->collisions,
3815 stats->tx_carrier_errors +
3816 stats->tx_aborted_errors +
3817 stats->tx_window_errors +
3818 stats->tx_heartbeat_errors,
3819 stats->tx_compressed);
3823 * Called from the PROCfs module. This now uses the new arbitrary sized
3824 * /proc/net interface to create /proc/net/dev
3826 static int dev_seq_show(struct seq_file *seq, void *v)
3828 if (v == SEQ_START_TOKEN)
3829 seq_puts(seq, "Inter-| Receive "
3831 " face |bytes packets errs drop fifo frame "
3832 "compressed multicast|bytes packets errs "
3833 "drop fifo colls carrier compressed\n");
3835 dev_seq_printf_stats(seq, v);
3839 static struct softnet_data *softnet_get_online(loff_t *pos)
3841 struct softnet_data *sd = NULL;
3843 while (*pos < nr_cpu_ids)
3844 if (cpu_online(*pos)) {
3845 sd = &per_cpu(softnet_data, *pos);
3852 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3854 return softnet_get_online(pos);
3857 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3860 return softnet_get_online(pos);
3863 static void softnet_seq_stop(struct seq_file *seq, void *v)
3867 static int softnet_seq_show(struct seq_file *seq, void *v)
3869 struct softnet_data *sd = v;
3871 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3872 sd->processed, sd->dropped, sd->time_squeeze, 0,
3873 0, 0, 0, 0, /* was fastroute */
3874 sd->cpu_collision, sd->received_rps);
3878 static const struct seq_operations dev_seq_ops = {
3879 .start = dev_seq_start,
3880 .next = dev_seq_next,
3881 .stop = dev_seq_stop,
3882 .show = dev_seq_show,
3885 static int dev_seq_open(struct inode *inode, struct file *file)
3887 return seq_open_net(inode, file, &dev_seq_ops,
3888 sizeof(struct seq_net_private));
3891 static const struct file_operations dev_seq_fops = {
3892 .owner = THIS_MODULE,
3893 .open = dev_seq_open,
3895 .llseek = seq_lseek,
3896 .release = seq_release_net,
3899 static const struct seq_operations softnet_seq_ops = {
3900 .start = softnet_seq_start,
3901 .next = softnet_seq_next,
3902 .stop = softnet_seq_stop,
3903 .show = softnet_seq_show,
3906 static int softnet_seq_open(struct inode *inode, struct file *file)
3908 return seq_open(file, &softnet_seq_ops);
3911 static const struct file_operations softnet_seq_fops = {
3912 .owner = THIS_MODULE,
3913 .open = softnet_seq_open,
3915 .llseek = seq_lseek,
3916 .release = seq_release,
3919 static void *ptype_get_idx(loff_t pos)
3921 struct packet_type *pt = NULL;
3925 list_for_each_entry_rcu(pt, &ptype_all, list) {
3931 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3932 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3941 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3945 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3948 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3950 struct packet_type *pt;
3951 struct list_head *nxt;
3955 if (v == SEQ_START_TOKEN)
3956 return ptype_get_idx(0);
3959 nxt = pt->list.next;
3960 if (pt->type == htons(ETH_P_ALL)) {
3961 if (nxt != &ptype_all)
3964 nxt = ptype_base[0].next;
3966 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3968 while (nxt == &ptype_base[hash]) {
3969 if (++hash >= PTYPE_HASH_SIZE)
3971 nxt = ptype_base[hash].next;
3974 return list_entry(nxt, struct packet_type, list);
3977 static void ptype_seq_stop(struct seq_file *seq, void *v)
3983 static int ptype_seq_show(struct seq_file *seq, void *v)
3985 struct packet_type *pt = v;
3987 if (v == SEQ_START_TOKEN)
3988 seq_puts(seq, "Type Device Function\n");
3989 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3990 if (pt->type == htons(ETH_P_ALL))
3991 seq_puts(seq, "ALL ");
3993 seq_printf(seq, "%04x", ntohs(pt->type));
3995 seq_printf(seq, " %-8s %pF\n",
3996 pt->dev ? pt->dev->name : "", pt->func);
4002 static const struct seq_operations ptype_seq_ops = {
4003 .start = ptype_seq_start,
4004 .next = ptype_seq_next,
4005 .stop = ptype_seq_stop,
4006 .show = ptype_seq_show,
4009 static int ptype_seq_open(struct inode *inode, struct file *file)
4011 return seq_open_net(inode, file, &ptype_seq_ops,
4012 sizeof(struct seq_net_private));
4015 static const struct file_operations ptype_seq_fops = {
4016 .owner = THIS_MODULE,
4017 .open = ptype_seq_open,
4019 .llseek = seq_lseek,
4020 .release = seq_release_net,
4024 static int __net_init dev_proc_net_init(struct net *net)
4028 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4030 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4032 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4035 if (wext_proc_init(net))
4041 proc_net_remove(net, "ptype");
4043 proc_net_remove(net, "softnet_stat");
4045 proc_net_remove(net, "dev");
4049 static void __net_exit dev_proc_net_exit(struct net *net)
4051 wext_proc_exit(net);
4053 proc_net_remove(net, "ptype");
4054 proc_net_remove(net, "softnet_stat");
4055 proc_net_remove(net, "dev");
4058 static struct pernet_operations __net_initdata dev_proc_ops = {
4059 .init = dev_proc_net_init,
4060 .exit = dev_proc_net_exit,
4063 static int __init dev_proc_init(void)
4065 return register_pernet_subsys(&dev_proc_ops);
4068 #define dev_proc_init() 0
4069 #endif /* CONFIG_PROC_FS */
4073 * netdev_set_master - set up master/slave pair
4074 * @slave: slave device
4075 * @master: new master device
4077 * Changes the master device of the slave. Pass %NULL to break the
4078 * bonding. The caller must hold the RTNL semaphore. On a failure
4079 * a negative errno code is returned. On success the reference counts
4080 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4081 * function returns zero.
4083 int netdev_set_master(struct net_device *slave, struct net_device *master)
4085 struct net_device *old = slave->master;
4095 slave->master = master;
4102 slave->flags |= IFF_SLAVE;
4104 slave->flags &= ~IFF_SLAVE;
4106 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4109 EXPORT_SYMBOL(netdev_set_master);
4111 static void dev_change_rx_flags(struct net_device *dev, int flags)
4113 const struct net_device_ops *ops = dev->netdev_ops;
4115 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4116 ops->ndo_change_rx_flags(dev, flags);
4119 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4121 unsigned short old_flags = dev->flags;
4127 dev->flags |= IFF_PROMISC;
4128 dev->promiscuity += inc;
4129 if (dev->promiscuity == 0) {
4132 * If inc causes overflow, untouch promisc and return error.
4135 dev->flags &= ~IFF_PROMISC;
4137 dev->promiscuity -= inc;
4138 printk(KERN_WARNING "%s: promiscuity touches roof, "
4139 "set promiscuity failed, promiscuity feature "
4140 "of device might be broken.\n", dev->name);
4144 if (dev->flags != old_flags) {
4145 printk(KERN_INFO "device %s %s promiscuous mode\n",
4146 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4148 if (audit_enabled) {
4149 current_uid_gid(&uid, &gid);
4150 audit_log(current->audit_context, GFP_ATOMIC,
4151 AUDIT_ANOM_PROMISCUOUS,
4152 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4153 dev->name, (dev->flags & IFF_PROMISC),
4154 (old_flags & IFF_PROMISC),
4155 audit_get_loginuid(current),
4157 audit_get_sessionid(current));
4160 dev_change_rx_flags(dev, IFF_PROMISC);
4166 * dev_set_promiscuity - update promiscuity count on a device
4170 * Add or remove promiscuity from a device. While the count in the device
4171 * remains above zero the interface remains promiscuous. Once it hits zero
4172 * the device reverts back to normal filtering operation. A negative inc
4173 * value is used to drop promiscuity on the device.
4174 * Return 0 if successful or a negative errno code on error.
4176 int dev_set_promiscuity(struct net_device *dev, int inc)
4178 unsigned short old_flags = dev->flags;
4181 err = __dev_set_promiscuity(dev, inc);
4184 if (dev->flags != old_flags)
4185 dev_set_rx_mode(dev);
4188 EXPORT_SYMBOL(dev_set_promiscuity);
4191 * dev_set_allmulti - update allmulti count on a device
4195 * Add or remove reception of all multicast frames to a device. While the
4196 * count in the device remains above zero the interface remains listening
4197 * to all interfaces. Once it hits zero the device reverts back to normal
4198 * filtering operation. A negative @inc value is used to drop the counter
4199 * when releasing a resource needing all multicasts.
4200 * Return 0 if successful or a negative errno code on error.
4203 int dev_set_allmulti(struct net_device *dev, int inc)
4205 unsigned short old_flags = dev->flags;
4209 dev->flags |= IFF_ALLMULTI;
4210 dev->allmulti += inc;
4211 if (dev->allmulti == 0) {
4214 * If inc causes overflow, untouch allmulti and return error.
4217 dev->flags &= ~IFF_ALLMULTI;
4219 dev->allmulti -= inc;
4220 printk(KERN_WARNING "%s: allmulti touches roof, "
4221 "set allmulti failed, allmulti feature of "
4222 "device might be broken.\n", dev->name);
4226 if (dev->flags ^ old_flags) {
4227 dev_change_rx_flags(dev, IFF_ALLMULTI);
4228 dev_set_rx_mode(dev);
4232 EXPORT_SYMBOL(dev_set_allmulti);
4235 * Upload unicast and multicast address lists to device and
4236 * configure RX filtering. When the device doesn't support unicast
4237 * filtering it is put in promiscuous mode while unicast addresses
4240 void __dev_set_rx_mode(struct net_device *dev)
4242 const struct net_device_ops *ops = dev->netdev_ops;
4244 /* dev_open will call this function so the list will stay sane. */
4245 if (!(dev->flags&IFF_UP))
4248 if (!netif_device_present(dev))
4251 if (ops->ndo_set_rx_mode)
4252 ops->ndo_set_rx_mode(dev);
4254 /* Unicast addresses changes may only happen under the rtnl,
4255 * therefore calling __dev_set_promiscuity here is safe.
4257 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4258 __dev_set_promiscuity(dev, 1);
4259 dev->uc_promisc = 1;
4260 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4261 __dev_set_promiscuity(dev, -1);
4262 dev->uc_promisc = 0;
4265 if (ops->ndo_set_multicast_list)
4266 ops->ndo_set_multicast_list(dev);
4270 void dev_set_rx_mode(struct net_device *dev)
4272 netif_addr_lock_bh(dev);
4273 __dev_set_rx_mode(dev);
4274 netif_addr_unlock_bh(dev);
4278 * dev_get_flags - get flags reported to userspace
4281 * Get the combination of flag bits exported through APIs to userspace.
4283 unsigned dev_get_flags(const struct net_device *dev)
4287 flags = (dev->flags & ~(IFF_PROMISC |
4292 (dev->gflags & (IFF_PROMISC |
4295 if (netif_running(dev)) {
4296 if (netif_oper_up(dev))
4297 flags |= IFF_RUNNING;
4298 if (netif_carrier_ok(dev))
4299 flags |= IFF_LOWER_UP;
4300 if (netif_dormant(dev))
4301 flags |= IFF_DORMANT;
4306 EXPORT_SYMBOL(dev_get_flags);
4308 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4310 int old_flags = dev->flags;
4316 * Set the flags on our device.
4319 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4320 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4322 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4326 * Load in the correct multicast list now the flags have changed.
4329 if ((old_flags ^ flags) & IFF_MULTICAST)
4330 dev_change_rx_flags(dev, IFF_MULTICAST);
4332 dev_set_rx_mode(dev);
4335 * Have we downed the interface. We handle IFF_UP ourselves
4336 * according to user attempts to set it, rather than blindly
4341 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4342 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4345 dev_set_rx_mode(dev);
4348 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4349 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4351 dev->gflags ^= IFF_PROMISC;
4352 dev_set_promiscuity(dev, inc);
4355 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4356 is important. Some (broken) drivers set IFF_PROMISC, when
4357 IFF_ALLMULTI is requested not asking us and not reporting.
4359 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4360 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4362 dev->gflags ^= IFF_ALLMULTI;
4363 dev_set_allmulti(dev, inc);
4369 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4371 unsigned int changes = dev->flags ^ old_flags;
4373 if (changes & IFF_UP) {
4374 if (dev->flags & IFF_UP)
4375 call_netdevice_notifiers(NETDEV_UP, dev);
4377 call_netdevice_notifiers(NETDEV_DOWN, dev);
4380 if (dev->flags & IFF_UP &&
4381 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4382 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4386 * dev_change_flags - change device settings
4388 * @flags: device state flags
4390 * Change settings on device based state flags. The flags are
4391 * in the userspace exported format.
4393 int dev_change_flags(struct net_device *dev, unsigned flags)
4396 int old_flags = dev->flags;
4398 ret = __dev_change_flags(dev, flags);
4402 changes = old_flags ^ dev->flags;
4404 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4406 __dev_notify_flags(dev, old_flags);
4409 EXPORT_SYMBOL(dev_change_flags);
4412 * dev_set_mtu - Change maximum transfer unit
4414 * @new_mtu: new transfer unit
4416 * Change the maximum transfer size of the network device.
4418 int dev_set_mtu(struct net_device *dev, int new_mtu)
4420 const struct net_device_ops *ops = dev->netdev_ops;
4423 if (new_mtu == dev->mtu)
4426 /* MTU must be positive. */
4430 if (!netif_device_present(dev))
4434 if (ops->ndo_change_mtu)
4435 err = ops->ndo_change_mtu(dev, new_mtu);
4439 if (!err && dev->flags & IFF_UP)
4440 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4443 EXPORT_SYMBOL(dev_set_mtu);
4446 * dev_set_mac_address - Change Media Access Control Address
4450 * Change the hardware (MAC) address of the device
4452 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4454 const struct net_device_ops *ops = dev->netdev_ops;
4457 if (!ops->ndo_set_mac_address)
4459 if (sa->sa_family != dev->type)
4461 if (!netif_device_present(dev))
4463 err = ops->ndo_set_mac_address(dev, sa);
4465 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4468 EXPORT_SYMBOL(dev_set_mac_address);
4471 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4473 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4476 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4482 case SIOCGIFFLAGS: /* Get interface flags */
4483 ifr->ifr_flags = (short) dev_get_flags(dev);
4486 case SIOCGIFMETRIC: /* Get the metric on the interface
4487 (currently unused) */
4488 ifr->ifr_metric = 0;
4491 case SIOCGIFMTU: /* Get the MTU of a device */
4492 ifr->ifr_mtu = dev->mtu;
4497 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4499 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4500 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4501 ifr->ifr_hwaddr.sa_family = dev->type;
4509 ifr->ifr_map.mem_start = dev->mem_start;
4510 ifr->ifr_map.mem_end = dev->mem_end;
4511 ifr->ifr_map.base_addr = dev->base_addr;
4512 ifr->ifr_map.irq = dev->irq;
4513 ifr->ifr_map.dma = dev->dma;
4514 ifr->ifr_map.port = dev->if_port;
4518 ifr->ifr_ifindex = dev->ifindex;
4522 ifr->ifr_qlen = dev->tx_queue_len;
4526 /* dev_ioctl() should ensure this case
4538 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4540 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4543 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4544 const struct net_device_ops *ops;
4549 ops = dev->netdev_ops;
4552 case SIOCSIFFLAGS: /* Set interface flags */
4553 return dev_change_flags(dev, ifr->ifr_flags);
4555 case SIOCSIFMETRIC: /* Set the metric on the interface
4556 (currently unused) */
4559 case SIOCSIFMTU: /* Set the MTU of a device */
4560 return dev_set_mtu(dev, ifr->ifr_mtu);
4563 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4565 case SIOCSIFHWBROADCAST:
4566 if (ifr->ifr_hwaddr.sa_family != dev->type)
4568 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4569 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4570 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4574 if (ops->ndo_set_config) {
4575 if (!netif_device_present(dev))
4577 return ops->ndo_set_config(dev, &ifr->ifr_map);
4582 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4583 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4585 if (!netif_device_present(dev))
4587 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4590 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4591 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4593 if (!netif_device_present(dev))
4595 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4598 if (ifr->ifr_qlen < 0)
4600 dev->tx_queue_len = ifr->ifr_qlen;
4604 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4605 return dev_change_name(dev, ifr->ifr_newname);
4608 * Unknown or private ioctl
4611 if ((cmd >= SIOCDEVPRIVATE &&
4612 cmd <= SIOCDEVPRIVATE + 15) ||
4613 cmd == SIOCBONDENSLAVE ||
4614 cmd == SIOCBONDRELEASE ||
4615 cmd == SIOCBONDSETHWADDR ||
4616 cmd == SIOCBONDSLAVEINFOQUERY ||
4617 cmd == SIOCBONDINFOQUERY ||
4618 cmd == SIOCBONDCHANGEACTIVE ||
4619 cmd == SIOCGMIIPHY ||
4620 cmd == SIOCGMIIREG ||
4621 cmd == SIOCSMIIREG ||
4622 cmd == SIOCBRADDIF ||
4623 cmd == SIOCBRDELIF ||
4624 cmd == SIOCSHWTSTAMP ||
4625 cmd == SIOCWANDEV) {
4627 if (ops->ndo_do_ioctl) {
4628 if (netif_device_present(dev))
4629 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4641 * This function handles all "interface"-type I/O control requests. The actual
4642 * 'doing' part of this is dev_ifsioc above.
4646 * dev_ioctl - network device ioctl
4647 * @net: the applicable net namespace
4648 * @cmd: command to issue
4649 * @arg: pointer to a struct ifreq in user space
4651 * Issue ioctl functions to devices. This is normally called by the
4652 * user space syscall interfaces but can sometimes be useful for
4653 * other purposes. The return value is the return from the syscall if
4654 * positive or a negative errno code on error.
4657 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4663 /* One special case: SIOCGIFCONF takes ifconf argument
4664 and requires shared lock, because it sleeps writing
4668 if (cmd == SIOCGIFCONF) {
4670 ret = dev_ifconf(net, (char __user *) arg);
4674 if (cmd == SIOCGIFNAME)
4675 return dev_ifname(net, (struct ifreq __user *)arg);
4677 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4680 ifr.ifr_name[IFNAMSIZ-1] = 0;
4682 colon = strchr(ifr.ifr_name, ':');
4687 * See which interface the caller is talking about.
4692 * These ioctl calls:
4693 * - can be done by all.
4694 * - atomic and do not require locking.
4705 dev_load(net, ifr.ifr_name);
4707 ret = dev_ifsioc_locked(net, &ifr, cmd);
4712 if (copy_to_user(arg, &ifr,
4713 sizeof(struct ifreq)))
4719 dev_load(net, ifr.ifr_name);
4721 ret = dev_ethtool(net, &ifr);
4726 if (copy_to_user(arg, &ifr,
4727 sizeof(struct ifreq)))
4733 * These ioctl calls:
4734 * - require superuser power.
4735 * - require strict serialization.
4741 if (!capable(CAP_NET_ADMIN))
4743 dev_load(net, ifr.ifr_name);
4745 ret = dev_ifsioc(net, &ifr, cmd);
4750 if (copy_to_user(arg, &ifr,
4751 sizeof(struct ifreq)))
4757 * These ioctl calls:
4758 * - require superuser power.
4759 * - require strict serialization.
4760 * - do not return a value
4770 case SIOCSIFHWBROADCAST:
4773 case SIOCBONDENSLAVE:
4774 case SIOCBONDRELEASE:
4775 case SIOCBONDSETHWADDR:
4776 case SIOCBONDCHANGEACTIVE:
4780 if (!capable(CAP_NET_ADMIN))
4783 case SIOCBONDSLAVEINFOQUERY:
4784 case SIOCBONDINFOQUERY:
4785 dev_load(net, ifr.ifr_name);
4787 ret = dev_ifsioc(net, &ifr, cmd);
4792 /* Get the per device memory space. We can add this but
4793 * currently do not support it */
4795 /* Set the per device memory buffer space.
4796 * Not applicable in our case */
4801 * Unknown or private ioctl.
4804 if (cmd == SIOCWANDEV ||
4805 (cmd >= SIOCDEVPRIVATE &&
4806 cmd <= SIOCDEVPRIVATE + 15)) {
4807 dev_load(net, ifr.ifr_name);
4809 ret = dev_ifsioc(net, &ifr, cmd);
4811 if (!ret && copy_to_user(arg, &ifr,
4812 sizeof(struct ifreq)))
4816 /* Take care of Wireless Extensions */
4817 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4818 return wext_handle_ioctl(net, &ifr, cmd, arg);
4825 * dev_new_index - allocate an ifindex
4826 * @net: the applicable net namespace
4828 * Returns a suitable unique value for a new device interface
4829 * number. The caller must hold the rtnl semaphore or the
4830 * dev_base_lock to be sure it remains unique.
4832 static int dev_new_index(struct net *net)
4838 if (!__dev_get_by_index(net, ifindex))
4843 /* Delayed registration/unregisteration */
4844 static LIST_HEAD(net_todo_list);
4846 static void net_set_todo(struct net_device *dev)
4848 list_add_tail(&dev->todo_list, &net_todo_list);
4851 static void rollback_registered_many(struct list_head *head)
4853 struct net_device *dev, *tmp;
4855 BUG_ON(dev_boot_phase);
4858 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4859 /* Some devices call without registering
4860 * for initialization unwind. Remove those
4861 * devices and proceed with the remaining.
4863 if (dev->reg_state == NETREG_UNINITIALIZED) {
4864 pr_debug("unregister_netdevice: device %s/%p never "
4865 "was registered\n", dev->name, dev);
4868 list_del(&dev->unreg_list);
4872 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4874 /* If device is running, close it first. */
4877 /* And unlink it from device chain. */
4878 unlist_netdevice(dev);
4880 dev->reg_state = NETREG_UNREGISTERING;
4885 list_for_each_entry(dev, head, unreg_list) {
4886 /* Shutdown queueing discipline. */
4890 /* Notify protocols, that we are about to destroy
4891 this device. They should clean all the things.
4893 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4895 if (!dev->rtnl_link_ops ||
4896 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4897 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4900 * Flush the unicast and multicast chains
4905 if (dev->netdev_ops->ndo_uninit)
4906 dev->netdev_ops->ndo_uninit(dev);
4908 /* Notifier chain MUST detach us from master device. */
4909 WARN_ON(dev->master);
4911 /* Remove entries from kobject tree */
4912 netdev_unregister_kobject(dev);
4915 /* Process any work delayed until the end of the batch */
4916 dev = list_first_entry(head, struct net_device, unreg_list);
4917 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4921 list_for_each_entry(dev, head, unreg_list)
4925 static void rollback_registered(struct net_device *dev)
4929 list_add(&dev->unreg_list, &single);
4930 rollback_registered_many(&single);
4933 unsigned long netdev_fix_features(unsigned long features, const char *name)
4935 /* Fix illegal SG+CSUM combinations. */
4936 if ((features & NETIF_F_SG) &&
4937 !(features & NETIF_F_ALL_CSUM)) {
4939 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4940 "checksum feature.\n", name);
4941 features &= ~NETIF_F_SG;
4944 /* TSO requires that SG is present as well. */
4945 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4947 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4948 "SG feature.\n", name);
4949 features &= ~NETIF_F_TSO;
4952 if (features & NETIF_F_UFO) {
4953 if (!(features & NETIF_F_GEN_CSUM)) {
4955 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4956 "since no NETIF_F_HW_CSUM feature.\n",
4958 features &= ~NETIF_F_UFO;
4961 if (!(features & NETIF_F_SG)) {
4963 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4964 "since no NETIF_F_SG feature.\n", name);
4965 features &= ~NETIF_F_UFO;
4971 EXPORT_SYMBOL(netdev_fix_features);
4974 * netif_stacked_transfer_operstate - transfer operstate
4975 * @rootdev: the root or lower level device to transfer state from
4976 * @dev: the device to transfer operstate to
4978 * Transfer operational state from root to device. This is normally
4979 * called when a stacking relationship exists between the root
4980 * device and the device(a leaf device).
4982 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4983 struct net_device *dev)
4985 if (rootdev->operstate == IF_OPER_DORMANT)
4986 netif_dormant_on(dev);
4988 netif_dormant_off(dev);
4990 if (netif_carrier_ok(rootdev)) {
4991 if (!netif_carrier_ok(dev))
4992 netif_carrier_on(dev);
4994 if (netif_carrier_ok(dev))
4995 netif_carrier_off(dev);
4998 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5000 static int netif_alloc_rx_queues(struct net_device *dev)
5003 unsigned int i, count = dev->num_rx_queues;
5004 struct netdev_rx_queue *rx;
5008 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5010 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5016 * Set a pointer to first element in the array which holds the
5019 for (i = 0; i < count; i++)
5025 static int netif_alloc_netdev_queues(struct net_device *dev)
5027 unsigned int count = dev->num_tx_queues;
5028 struct netdev_queue *tx;
5032 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5034 pr_err("netdev: Unable to allocate %u tx queues.\n",
5042 static void netdev_init_one_queue(struct net_device *dev,
5043 struct netdev_queue *queue,
5048 /* Initialize queue lock */
5049 spin_lock_init(&queue->_xmit_lock);
5050 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5051 queue->xmit_lock_owner = -1;
5054 static void netdev_init_queues(struct net_device *dev)
5056 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5057 spin_lock_init(&dev->tx_global_lock);
5061 * register_netdevice - register a network device
5062 * @dev: device to register
5064 * Take a completed network device structure and add it to the kernel
5065 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5066 * chain. 0 is returned on success. A negative errno code is returned
5067 * on a failure to set up the device, or if the name is a duplicate.
5069 * Callers must hold the rtnl semaphore. You may want
5070 * register_netdev() instead of this.
5073 * The locking appears insufficient to guarantee two parallel registers
5074 * will not get the same name.
5077 int register_netdevice(struct net_device *dev)
5080 struct net *net = dev_net(dev);
5082 BUG_ON(dev_boot_phase);
5087 /* When net_device's are persistent, this will be fatal. */
5088 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5091 spin_lock_init(&dev->addr_list_lock);
5092 netdev_set_addr_lockdep_class(dev);
5096 ret = netif_alloc_rx_queues(dev);
5100 ret = netif_alloc_netdev_queues(dev);
5104 netdev_init_queues(dev);
5106 /* Init, if this function is available */
5107 if (dev->netdev_ops->ndo_init) {
5108 ret = dev->netdev_ops->ndo_init(dev);
5116 ret = dev_get_valid_name(dev, dev->name, 0);
5120 dev->ifindex = dev_new_index(net);
5121 if (dev->iflink == -1)
5122 dev->iflink = dev->ifindex;
5124 /* Fix illegal checksum combinations */
5125 if ((dev->features & NETIF_F_HW_CSUM) &&
5126 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5127 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5129 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5132 if ((dev->features & NETIF_F_NO_CSUM) &&
5133 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5134 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5136 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5139 dev->features = netdev_fix_features(dev->features, dev->name);
5141 /* Enable software GSO if SG is supported. */
5142 if (dev->features & NETIF_F_SG)
5143 dev->features |= NETIF_F_GSO;
5145 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5146 * vlan_dev_init() will do the dev->features check, so these features
5147 * are enabled only if supported by underlying device.
5149 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5151 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5152 ret = notifier_to_errno(ret);
5156 ret = netdev_register_kobject(dev);
5159 dev->reg_state = NETREG_REGISTERED;
5162 * Default initial state at registry is that the
5163 * device is present.
5166 set_bit(__LINK_STATE_PRESENT, &dev->state);
5168 dev_init_scheduler(dev);
5170 list_netdevice(dev);
5172 /* Notify protocols, that a new device appeared. */
5173 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5174 ret = notifier_to_errno(ret);
5176 rollback_registered(dev);
5177 dev->reg_state = NETREG_UNREGISTERED;
5180 * Prevent userspace races by waiting until the network
5181 * device is fully setup before sending notifications.
5183 if (!dev->rtnl_link_ops ||
5184 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5185 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5191 if (dev->netdev_ops->ndo_uninit)
5192 dev->netdev_ops->ndo_uninit(dev);
5195 EXPORT_SYMBOL(register_netdevice);
5198 * init_dummy_netdev - init a dummy network device for NAPI
5199 * @dev: device to init
5201 * This takes a network device structure and initialize the minimum
5202 * amount of fields so it can be used to schedule NAPI polls without
5203 * registering a full blown interface. This is to be used by drivers
5204 * that need to tie several hardware interfaces to a single NAPI
5205 * poll scheduler due to HW limitations.
5207 int init_dummy_netdev(struct net_device *dev)
5209 /* Clear everything. Note we don't initialize spinlocks
5210 * are they aren't supposed to be taken by any of the
5211 * NAPI code and this dummy netdev is supposed to be
5212 * only ever used for NAPI polls
5214 memset(dev, 0, sizeof(struct net_device));
5216 /* make sure we BUG if trying to hit standard
5217 * register/unregister code path
5219 dev->reg_state = NETREG_DUMMY;
5221 /* NAPI wants this */
5222 INIT_LIST_HEAD(&dev->napi_list);
5224 /* a dummy interface is started by default */
5225 set_bit(__LINK_STATE_PRESENT, &dev->state);
5226 set_bit(__LINK_STATE_START, &dev->state);
5228 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5229 * because users of this 'device' dont need to change
5235 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5239 * register_netdev - register a network device
5240 * @dev: device to register
5242 * Take a completed network device structure and add it to the kernel
5243 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5244 * chain. 0 is returned on success. A negative errno code is returned
5245 * on a failure to set up the device, or if the name is a duplicate.
5247 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5248 * and expands the device name if you passed a format string to
5251 int register_netdev(struct net_device *dev)
5258 * If the name is a format string the caller wants us to do a
5261 if (strchr(dev->name, '%')) {
5262 err = dev_alloc_name(dev, dev->name);
5267 err = register_netdevice(dev);
5272 EXPORT_SYMBOL(register_netdev);
5274 int netdev_refcnt_read(const struct net_device *dev)
5278 for_each_possible_cpu(i)
5279 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5282 EXPORT_SYMBOL(netdev_refcnt_read);
5285 * netdev_wait_allrefs - wait until all references are gone.
5287 * This is called when unregistering network devices.
5289 * Any protocol or device that holds a reference should register
5290 * for netdevice notification, and cleanup and put back the
5291 * reference if they receive an UNREGISTER event.
5292 * We can get stuck here if buggy protocols don't correctly
5295 static void netdev_wait_allrefs(struct net_device *dev)
5297 unsigned long rebroadcast_time, warning_time;
5300 linkwatch_forget_dev(dev);
5302 rebroadcast_time = warning_time = jiffies;
5303 refcnt = netdev_refcnt_read(dev);
5305 while (refcnt != 0) {
5306 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5309 /* Rebroadcast unregister notification */
5310 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5311 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5312 * should have already handle it the first time */
5314 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5316 /* We must not have linkwatch events
5317 * pending on unregister. If this
5318 * happens, we simply run the queue
5319 * unscheduled, resulting in a noop
5322 linkwatch_run_queue();
5327 rebroadcast_time = jiffies;
5332 refcnt = netdev_refcnt_read(dev);
5334 if (time_after(jiffies, warning_time + 10 * HZ)) {
5335 printk(KERN_EMERG "unregister_netdevice: "
5336 "waiting for %s to become free. Usage "
5339 warning_time = jiffies;
5348 * register_netdevice(x1);
5349 * register_netdevice(x2);
5351 * unregister_netdevice(y1);
5352 * unregister_netdevice(y2);
5358 * We are invoked by rtnl_unlock().
5359 * This allows us to deal with problems:
5360 * 1) We can delete sysfs objects which invoke hotplug
5361 * without deadlocking with linkwatch via keventd.
5362 * 2) Since we run with the RTNL semaphore not held, we can sleep
5363 * safely in order to wait for the netdev refcnt to drop to zero.
5365 * We must not return until all unregister events added during
5366 * the interval the lock was held have been completed.
5368 void netdev_run_todo(void)
5370 struct list_head list;
5372 /* Snapshot list, allow later requests */
5373 list_replace_init(&net_todo_list, &list);
5377 while (!list_empty(&list)) {
5378 struct net_device *dev
5379 = list_first_entry(&list, struct net_device, todo_list);
5380 list_del(&dev->todo_list);
5382 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5383 printk(KERN_ERR "network todo '%s' but state %d\n",
5384 dev->name, dev->reg_state);
5389 dev->reg_state = NETREG_UNREGISTERED;
5391 on_each_cpu(flush_backlog, dev, 1);
5393 netdev_wait_allrefs(dev);
5396 BUG_ON(netdev_refcnt_read(dev));
5397 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5398 WARN_ON(dev->ip6_ptr);
5399 WARN_ON(dev->dn_ptr);
5401 if (dev->destructor)
5402 dev->destructor(dev);
5404 /* Free network device */
5405 kobject_put(&dev->dev.kobj);
5410 * dev_txq_stats_fold - fold tx_queues stats
5411 * @dev: device to get statistics from
5412 * @stats: struct rtnl_link_stats64 to hold results
5414 void dev_txq_stats_fold(const struct net_device *dev,
5415 struct rtnl_link_stats64 *stats)
5417 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5419 struct netdev_queue *txq;
5421 for (i = 0; i < dev->num_tx_queues; i++) {
5422 txq = netdev_get_tx_queue(dev, i);
5423 spin_lock_bh(&txq->_xmit_lock);
5424 tx_bytes += txq->tx_bytes;
5425 tx_packets += txq->tx_packets;
5426 tx_dropped += txq->tx_dropped;
5427 spin_unlock_bh(&txq->_xmit_lock);
5429 if (tx_bytes || tx_packets || tx_dropped) {
5430 stats->tx_bytes = tx_bytes;
5431 stats->tx_packets = tx_packets;
5432 stats->tx_dropped = tx_dropped;
5435 EXPORT_SYMBOL(dev_txq_stats_fold);
5437 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5438 * fields in the same order, with only the type differing.
5440 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5441 const struct net_device_stats *netdev_stats)
5443 #if BITS_PER_LONG == 64
5444 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5445 memcpy(stats64, netdev_stats, sizeof(*stats64));
5447 size_t i, n = sizeof(*stats64) / sizeof(u64);
5448 const unsigned long *src = (const unsigned long *)netdev_stats;
5449 u64 *dst = (u64 *)stats64;
5451 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5452 sizeof(*stats64) / sizeof(u64));
5453 for (i = 0; i < n; i++)
5459 * dev_get_stats - get network device statistics
5460 * @dev: device to get statistics from
5461 * @storage: place to store stats
5463 * Get network statistics from device. Return @storage.
5464 * The device driver may provide its own method by setting
5465 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5466 * otherwise the internal statistics structure is used.
5468 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5469 struct rtnl_link_stats64 *storage)
5471 const struct net_device_ops *ops = dev->netdev_ops;
5473 if (ops->ndo_get_stats64) {
5474 memset(storage, 0, sizeof(*storage));
5475 ops->ndo_get_stats64(dev, storage);
5476 } else if (ops->ndo_get_stats) {
5477 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5479 netdev_stats_to_stats64(storage, &dev->stats);
5480 dev_txq_stats_fold(dev, storage);
5482 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5485 EXPORT_SYMBOL(dev_get_stats);
5487 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5489 struct netdev_queue *queue = dev_ingress_queue(dev);
5491 #ifdef CONFIG_NET_CLS_ACT
5494 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5497 netdev_init_one_queue(dev, queue, NULL);
5498 queue->qdisc = &noop_qdisc;
5499 queue->qdisc_sleeping = &noop_qdisc;
5500 rcu_assign_pointer(dev->ingress_queue, queue);
5506 * alloc_netdev_mq - allocate network device
5507 * @sizeof_priv: size of private data to allocate space for
5508 * @name: device name format string
5509 * @setup: callback to initialize device
5510 * @queue_count: the number of subqueues to allocate
5512 * Allocates a struct net_device with private data area for driver use
5513 * and performs basic initialization. Also allocates subquue structs
5514 * for each queue on the device at the end of the netdevice.
5516 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5517 void (*setup)(struct net_device *), unsigned int queue_count)
5519 struct net_device *dev;
5521 struct net_device *p;
5523 BUG_ON(strlen(name) >= sizeof(dev->name));
5525 if (queue_count < 1) {
5526 pr_err("alloc_netdev: Unable to allocate device "
5527 "with zero queues.\n");
5531 alloc_size = sizeof(struct net_device);
5533 /* ensure 32-byte alignment of private area */
5534 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5535 alloc_size += sizeof_priv;
5537 /* ensure 32-byte alignment of whole construct */
5538 alloc_size += NETDEV_ALIGN - 1;
5540 p = kzalloc(alloc_size, GFP_KERNEL);
5542 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5546 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5547 dev->padded = (char *)dev - (char *)p;
5549 dev->pcpu_refcnt = alloc_percpu(int);
5550 if (!dev->pcpu_refcnt)
5553 if (dev_addr_init(dev))
5559 dev_net_set(dev, &init_net);
5561 dev->num_tx_queues = queue_count;
5562 dev->real_num_tx_queues = queue_count;
5565 dev->num_rx_queues = queue_count;
5566 dev->real_num_rx_queues = queue_count;
5569 dev->gso_max_size = GSO_MAX_SIZE;
5571 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5572 dev->ethtool_ntuple_list.count = 0;
5573 INIT_LIST_HEAD(&dev->napi_list);
5574 INIT_LIST_HEAD(&dev->unreg_list);
5575 INIT_LIST_HEAD(&dev->link_watch_list);
5576 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5578 strcpy(dev->name, name);
5582 free_percpu(dev->pcpu_refcnt);
5587 EXPORT_SYMBOL(alloc_netdev_mq);
5590 * free_netdev - free network device
5593 * This function does the last stage of destroying an allocated device
5594 * interface. The reference to the device object is released.
5595 * If this is the last reference then it will be freed.
5597 void free_netdev(struct net_device *dev)
5599 struct napi_struct *p, *n;
5601 release_net(dev_net(dev));
5605 kfree(rcu_dereference_raw(dev->ingress_queue));
5607 /* Flush device addresses */
5608 dev_addr_flush(dev);
5610 /* Clear ethtool n-tuple list */
5611 ethtool_ntuple_flush(dev);
5613 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5616 free_percpu(dev->pcpu_refcnt);
5617 dev->pcpu_refcnt = NULL;
5619 /* Compatibility with error handling in drivers */
5620 if (dev->reg_state == NETREG_UNINITIALIZED) {
5621 kfree((char *)dev - dev->padded);
5625 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5626 dev->reg_state = NETREG_RELEASED;
5628 /* will free via device release */
5629 put_device(&dev->dev);
5631 EXPORT_SYMBOL(free_netdev);
5634 * synchronize_net - Synchronize with packet receive processing
5636 * Wait for packets currently being received to be done.
5637 * Does not block later packets from starting.
5639 void synchronize_net(void)
5644 EXPORT_SYMBOL(synchronize_net);
5647 * unregister_netdevice_queue - remove device from the kernel
5651 * This function shuts down a device interface and removes it
5652 * from the kernel tables.
5653 * If head not NULL, device is queued to be unregistered later.
5655 * Callers must hold the rtnl semaphore. You may want
5656 * unregister_netdev() instead of this.
5659 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5664 list_move_tail(&dev->unreg_list, head);
5666 rollback_registered(dev);
5667 /* Finish processing unregister after unlock */
5671 EXPORT_SYMBOL(unregister_netdevice_queue);
5674 * unregister_netdevice_many - unregister many devices
5675 * @head: list of devices
5677 void unregister_netdevice_many(struct list_head *head)
5679 struct net_device *dev;
5681 if (!list_empty(head)) {
5682 rollback_registered_many(head);
5683 list_for_each_entry(dev, head, unreg_list)
5687 EXPORT_SYMBOL(unregister_netdevice_many);
5690 * unregister_netdev - remove device from the kernel
5693 * This function shuts down a device interface and removes it
5694 * from the kernel tables.
5696 * This is just a wrapper for unregister_netdevice that takes
5697 * the rtnl semaphore. In general you want to use this and not
5698 * unregister_netdevice.
5700 void unregister_netdev(struct net_device *dev)
5703 unregister_netdevice(dev);
5706 EXPORT_SYMBOL(unregister_netdev);
5709 * dev_change_net_namespace - move device to different nethost namespace
5711 * @net: network namespace
5712 * @pat: If not NULL name pattern to try if the current device name
5713 * is already taken in the destination network namespace.
5715 * This function shuts down a device interface and moves it
5716 * to a new network namespace. On success 0 is returned, on
5717 * a failure a netagive errno code is returned.
5719 * Callers must hold the rtnl semaphore.
5722 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5728 /* Don't allow namespace local devices to be moved. */
5730 if (dev->features & NETIF_F_NETNS_LOCAL)
5733 /* Ensure the device has been registrered */
5735 if (dev->reg_state != NETREG_REGISTERED)
5738 /* Get out if there is nothing todo */
5740 if (net_eq(dev_net(dev), net))
5743 /* Pick the destination device name, and ensure
5744 * we can use it in the destination network namespace.
5747 if (__dev_get_by_name(net, dev->name)) {
5748 /* We get here if we can't use the current device name */
5751 if (dev_get_valid_name(dev, pat, 1))
5756 * And now a mini version of register_netdevice unregister_netdevice.
5759 /* If device is running close it first. */
5762 /* And unlink it from device chain */
5764 unlist_netdevice(dev);
5768 /* Shutdown queueing discipline. */
5771 /* Notify protocols, that we are about to destroy
5772 this device. They should clean all the things.
5774 Note that dev->reg_state stays at NETREG_REGISTERED.
5775 This is wanted because this way 8021q and macvlan know
5776 the device is just moving and can keep their slaves up.
5778 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5779 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5782 * Flush the unicast and multicast chains
5787 /* Actually switch the network namespace */
5788 dev_net_set(dev, net);
5790 /* If there is an ifindex conflict assign a new one */
5791 if (__dev_get_by_index(net, dev->ifindex)) {
5792 int iflink = (dev->iflink == dev->ifindex);
5793 dev->ifindex = dev_new_index(net);
5795 dev->iflink = dev->ifindex;
5798 /* Fixup kobjects */
5799 err = device_rename(&dev->dev, dev->name);
5802 /* Add the device back in the hashes */
5803 list_netdevice(dev);
5805 /* Notify protocols, that a new device appeared. */
5806 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5809 * Prevent userspace races by waiting until the network
5810 * device is fully setup before sending notifications.
5812 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5819 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5821 static int dev_cpu_callback(struct notifier_block *nfb,
5822 unsigned long action,
5825 struct sk_buff **list_skb;
5826 struct sk_buff *skb;
5827 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5828 struct softnet_data *sd, *oldsd;
5830 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5833 local_irq_disable();
5834 cpu = smp_processor_id();
5835 sd = &per_cpu(softnet_data, cpu);
5836 oldsd = &per_cpu(softnet_data, oldcpu);
5838 /* Find end of our completion_queue. */
5839 list_skb = &sd->completion_queue;
5841 list_skb = &(*list_skb)->next;
5842 /* Append completion queue from offline CPU. */
5843 *list_skb = oldsd->completion_queue;
5844 oldsd->completion_queue = NULL;
5846 /* Append output queue from offline CPU. */
5847 if (oldsd->output_queue) {
5848 *sd->output_queue_tailp = oldsd->output_queue;
5849 sd->output_queue_tailp = oldsd->output_queue_tailp;
5850 oldsd->output_queue = NULL;
5851 oldsd->output_queue_tailp = &oldsd->output_queue;
5854 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5857 /* Process offline CPU's input_pkt_queue */
5858 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5860 input_queue_head_incr(oldsd);
5862 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5864 input_queue_head_incr(oldsd);
5872 * netdev_increment_features - increment feature set by one
5873 * @all: current feature set
5874 * @one: new feature set
5875 * @mask: mask feature set
5877 * Computes a new feature set after adding a device with feature set
5878 * @one to the master device with current feature set @all. Will not
5879 * enable anything that is off in @mask. Returns the new feature set.
5881 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5884 /* If device needs checksumming, downgrade to it. */
5885 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5886 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5887 else if (mask & NETIF_F_ALL_CSUM) {
5888 /* If one device supports v4/v6 checksumming, set for all. */
5889 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5890 !(all & NETIF_F_GEN_CSUM)) {
5891 all &= ~NETIF_F_ALL_CSUM;
5892 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5895 /* If one device supports hw checksumming, set for all. */
5896 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5897 all &= ~NETIF_F_ALL_CSUM;
5898 all |= NETIF_F_HW_CSUM;
5902 one |= NETIF_F_ALL_CSUM;
5904 one |= all & NETIF_F_ONE_FOR_ALL;
5905 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5906 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5910 EXPORT_SYMBOL(netdev_increment_features);
5912 static struct hlist_head *netdev_create_hash(void)
5915 struct hlist_head *hash;
5917 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5919 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5920 INIT_HLIST_HEAD(&hash[i]);
5925 /* Initialize per network namespace state */
5926 static int __net_init netdev_init(struct net *net)
5928 INIT_LIST_HEAD(&net->dev_base_head);
5930 net->dev_name_head = netdev_create_hash();
5931 if (net->dev_name_head == NULL)
5934 net->dev_index_head = netdev_create_hash();
5935 if (net->dev_index_head == NULL)
5941 kfree(net->dev_name_head);
5947 * netdev_drivername - network driver for the device
5948 * @dev: network device
5949 * @buffer: buffer for resulting name
5950 * @len: size of buffer
5952 * Determine network driver for device.
5954 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5956 const struct device_driver *driver;
5957 const struct device *parent;
5959 if (len <= 0 || !buffer)
5963 parent = dev->dev.parent;
5968 driver = parent->driver;
5969 if (driver && driver->name)
5970 strlcpy(buffer, driver->name, len);
5974 static int __netdev_printk(const char *level, const struct net_device *dev,
5975 struct va_format *vaf)
5979 if (dev && dev->dev.parent)
5980 r = dev_printk(level, dev->dev.parent, "%s: %pV",
5981 netdev_name(dev), vaf);
5983 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
5985 r = printk("%s(NULL net_device): %pV", level, vaf);
5990 int netdev_printk(const char *level, const struct net_device *dev,
5991 const char *format, ...)
5993 struct va_format vaf;
5997 va_start(args, format);
6002 r = __netdev_printk(level, dev, &vaf);
6007 EXPORT_SYMBOL(netdev_printk);
6009 #define define_netdev_printk_level(func, level) \
6010 int func(const struct net_device *dev, const char *fmt, ...) \
6013 struct va_format vaf; \
6016 va_start(args, fmt); \
6021 r = __netdev_printk(level, dev, &vaf); \
6026 EXPORT_SYMBOL(func);
6028 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6029 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6030 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6031 define_netdev_printk_level(netdev_err, KERN_ERR);
6032 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6033 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6034 define_netdev_printk_level(netdev_info, KERN_INFO);
6036 static void __net_exit netdev_exit(struct net *net)
6038 kfree(net->dev_name_head);
6039 kfree(net->dev_index_head);
6042 static struct pernet_operations __net_initdata netdev_net_ops = {
6043 .init = netdev_init,
6044 .exit = netdev_exit,
6047 static void __net_exit default_device_exit(struct net *net)
6049 struct net_device *dev, *aux;
6051 * Push all migratable network devices back to the
6052 * initial network namespace
6055 for_each_netdev_safe(net, dev, aux) {
6057 char fb_name[IFNAMSIZ];
6059 /* Ignore unmoveable devices (i.e. loopback) */
6060 if (dev->features & NETIF_F_NETNS_LOCAL)
6063 /* Leave virtual devices for the generic cleanup */
6064 if (dev->rtnl_link_ops)
6067 /* Push remaing network devices to init_net */
6068 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6069 err = dev_change_net_namespace(dev, &init_net, fb_name);
6071 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6072 __func__, dev->name, err);
6079 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6081 /* At exit all network devices most be removed from a network
6082 * namespace. Do this in the reverse order of registeration.
6083 * Do this across as many network namespaces as possible to
6084 * improve batching efficiency.
6086 struct net_device *dev;
6088 LIST_HEAD(dev_kill_list);
6091 list_for_each_entry(net, net_list, exit_list) {
6092 for_each_netdev_reverse(net, dev) {
6093 if (dev->rtnl_link_ops)
6094 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6096 unregister_netdevice_queue(dev, &dev_kill_list);
6099 unregister_netdevice_many(&dev_kill_list);
6103 static struct pernet_operations __net_initdata default_device_ops = {
6104 .exit = default_device_exit,
6105 .exit_batch = default_device_exit_batch,
6109 * Initialize the DEV module. At boot time this walks the device list and
6110 * unhooks any devices that fail to initialise (normally hardware not
6111 * present) and leaves us with a valid list of present and active devices.
6116 * This is called single threaded during boot, so no need
6117 * to take the rtnl semaphore.
6119 static int __init net_dev_init(void)
6121 int i, rc = -ENOMEM;
6123 BUG_ON(!dev_boot_phase);
6125 if (dev_proc_init())
6128 if (netdev_kobject_init())
6131 INIT_LIST_HEAD(&ptype_all);
6132 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6133 INIT_LIST_HEAD(&ptype_base[i]);
6135 if (register_pernet_subsys(&netdev_net_ops))
6139 * Initialise the packet receive queues.
6142 for_each_possible_cpu(i) {
6143 struct softnet_data *sd = &per_cpu(softnet_data, i);
6145 memset(sd, 0, sizeof(*sd));
6146 skb_queue_head_init(&sd->input_pkt_queue);
6147 skb_queue_head_init(&sd->process_queue);
6148 sd->completion_queue = NULL;
6149 INIT_LIST_HEAD(&sd->poll_list);
6150 sd->output_queue = NULL;
6151 sd->output_queue_tailp = &sd->output_queue;
6153 sd->csd.func = rps_trigger_softirq;
6159 sd->backlog.poll = process_backlog;
6160 sd->backlog.weight = weight_p;
6161 sd->backlog.gro_list = NULL;
6162 sd->backlog.gro_count = 0;
6167 /* The loopback device is special if any other network devices
6168 * is present in a network namespace the loopback device must
6169 * be present. Since we now dynamically allocate and free the
6170 * loopback device ensure this invariant is maintained by
6171 * keeping the loopback device as the first device on the
6172 * list of network devices. Ensuring the loopback devices
6173 * is the first device that appears and the last network device
6176 if (register_pernet_device(&loopback_net_ops))
6179 if (register_pernet_device(&default_device_ops))
6182 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6183 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6185 hotcpu_notifier(dev_cpu_callback, 0);
6193 subsys_initcall(net_dev_init);
6195 static int __init initialize_hashrnd(void)
6197 get_random_bytes(&hashrnd, sizeof(hashrnd));
6201 late_initcall_sync(initialize_hashrnd);