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 <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
145 * The list of packet types we will receive (as opposed to discard)
146 * and the routines to invoke.
148 * Why 16. Because with 16 the only overlap we get on a hash of the
149 * low nibble of the protocol value is RARP/SNAP/X.25.
151 * NOTE: That is no longer true with the addition of VLAN tags. Not
152 * sure which should go first, but I bet it won't make much
153 * difference if we are running VLANs. The good news is that
154 * this protocol won't be in the list unless compiled in, so
155 * the average user (w/out VLANs) will not be adversely affected.
172 #define PTYPE_HASH_SIZE (16)
173 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
175 static DEFINE_SPINLOCK(ptype_lock);
176 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
177 static struct list_head ptype_all __read_mostly; /* Taps */
180 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
183 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
185 * Writers must hold the rtnl semaphore while they loop through the
186 * dev_base_head list, and hold dev_base_lock for writing when they do the
187 * actual updates. This allows pure readers to access the list even
188 * while a writer is preparing to update it.
190 * To put it another way, dev_base_lock is held for writing only to
191 * protect against pure readers; the rtnl semaphore provides the
192 * protection against other writers.
194 * See, for example usages, register_netdevice() and
195 * unregister_netdevice(), which must be called with the rtnl
198 DEFINE_RWLOCK(dev_base_lock);
199 EXPORT_SYMBOL(dev_base_lock);
201 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
203 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
204 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
207 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
209 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
212 static inline void rps_lock(struct softnet_data *sd)
215 spin_lock(&sd->input_pkt_queue.lock);
219 static inline void rps_unlock(struct softnet_data *sd)
222 spin_unlock(&sd->input_pkt_queue.lock);
226 /* Device list insertion */
227 static int list_netdevice(struct net_device *dev)
229 struct net *net = dev_net(dev);
233 write_lock_bh(&dev_base_lock);
234 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
235 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
236 hlist_add_head_rcu(&dev->index_hlist,
237 dev_index_hash(net, dev->ifindex));
238 write_unlock_bh(&dev_base_lock);
242 /* Device list removal
243 * caller must respect a RCU grace period before freeing/reusing dev
245 static void unlist_netdevice(struct net_device *dev)
249 /* Unlink dev from the device chain */
250 write_lock_bh(&dev_base_lock);
251 list_del_rcu(&dev->dev_list);
252 hlist_del_rcu(&dev->name_hlist);
253 hlist_del_rcu(&dev->index_hlist);
254 write_unlock_bh(&dev_base_lock);
261 static RAW_NOTIFIER_HEAD(netdev_chain);
264 * Device drivers call our routines to queue packets here. We empty the
265 * queue in the local softnet handler.
268 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
269 EXPORT_PER_CPU_SYMBOL(softnet_data);
271 #ifdef CONFIG_LOCKDEP
273 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
274 * according to dev->type
276 static const unsigned short netdev_lock_type[] =
277 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
278 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
279 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
280 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
281 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
282 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
283 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
284 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
285 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
286 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
287 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
288 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
289 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
290 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
291 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
292 ARPHRD_VOID, ARPHRD_NONE};
294 static const char *const netdev_lock_name[] =
295 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
296 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
297 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
298 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
299 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
300 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
301 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
302 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
303 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
304 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
305 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
306 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
307 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
308 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
309 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
310 "_xmit_VOID", "_xmit_NONE"};
312 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
313 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
315 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
319 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
320 if (netdev_lock_type[i] == dev_type)
322 /* the last key is used by default */
323 return ARRAY_SIZE(netdev_lock_type) - 1;
326 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
327 unsigned short dev_type)
331 i = netdev_lock_pos(dev_type);
332 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
333 netdev_lock_name[i]);
336 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
340 i = netdev_lock_pos(dev->type);
341 lockdep_set_class_and_name(&dev->addr_list_lock,
342 &netdev_addr_lock_key[i],
343 netdev_lock_name[i]);
346 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
347 unsigned short dev_type)
350 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
355 /*******************************************************************************
357 Protocol management and registration routines
359 *******************************************************************************/
362 * Add a protocol ID to the list. Now that the input handler is
363 * smarter we can dispense with all the messy stuff that used to be
366 * BEWARE!!! Protocol handlers, mangling input packets,
367 * MUST BE last in hash buckets and checking protocol handlers
368 * MUST start from promiscuous ptype_all chain in net_bh.
369 * It is true now, do not change it.
370 * Explanation follows: if protocol handler, mangling packet, will
371 * be the first on list, it is not able to sense, that packet
372 * is cloned and should be copied-on-write, so that it will
373 * change it and subsequent readers will get broken packet.
377 static inline struct list_head *ptype_head(const struct packet_type *pt)
379 if (pt->type == htons(ETH_P_ALL))
382 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
386 * dev_add_pack - add packet handler
387 * @pt: packet type declaration
389 * Add a protocol handler to the networking stack. The passed &packet_type
390 * is linked into kernel lists and may not be freed until it has been
391 * removed from the kernel lists.
393 * This call does not sleep therefore it can not
394 * guarantee all CPU's that are in middle of receiving packets
395 * will see the new packet type (until the next received packet).
398 void dev_add_pack(struct packet_type *pt)
400 struct list_head *head = ptype_head(pt);
402 spin_lock(&ptype_lock);
403 list_add_rcu(&pt->list, head);
404 spin_unlock(&ptype_lock);
406 EXPORT_SYMBOL(dev_add_pack);
409 * __dev_remove_pack - remove packet handler
410 * @pt: packet type declaration
412 * Remove a protocol handler that was previously added to the kernel
413 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
414 * from the kernel lists and can be freed or reused once this function
417 * The packet type might still be in use by receivers
418 * and must not be freed until after all the CPU's have gone
419 * through a quiescent state.
421 void __dev_remove_pack(struct packet_type *pt)
423 struct list_head *head = ptype_head(pt);
424 struct packet_type *pt1;
426 spin_lock(&ptype_lock);
428 list_for_each_entry(pt1, head, list) {
430 list_del_rcu(&pt->list);
435 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
437 spin_unlock(&ptype_lock);
439 EXPORT_SYMBOL(__dev_remove_pack);
442 * dev_remove_pack - remove packet handler
443 * @pt: packet type declaration
445 * Remove a protocol handler that was previously added to the kernel
446 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
447 * from the kernel lists and can be freed or reused once this function
450 * This call sleeps to guarantee that no CPU is looking at the packet
453 void dev_remove_pack(struct packet_type *pt)
455 __dev_remove_pack(pt);
459 EXPORT_SYMBOL(dev_remove_pack);
461 /******************************************************************************
463 Device Boot-time Settings Routines
465 *******************************************************************************/
467 /* Boot time configuration table */
468 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
471 * netdev_boot_setup_add - add new setup entry
472 * @name: name of the device
473 * @map: configured settings for the device
475 * Adds new setup entry to the dev_boot_setup list. The function
476 * returns 0 on error and 1 on success. This is a generic routine to
479 static int netdev_boot_setup_add(char *name, struct ifmap *map)
481 struct netdev_boot_setup *s;
485 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
486 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
487 memset(s[i].name, 0, sizeof(s[i].name));
488 strlcpy(s[i].name, name, IFNAMSIZ);
489 memcpy(&s[i].map, map, sizeof(s[i].map));
494 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
498 * netdev_boot_setup_check - check boot time settings
499 * @dev: the netdevice
501 * Check boot time settings for the device.
502 * The found settings are set for the device to be used
503 * later in the device probing.
504 * Returns 0 if no settings found, 1 if they are.
506 int netdev_boot_setup_check(struct net_device *dev)
508 struct netdev_boot_setup *s = dev_boot_setup;
511 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
512 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
513 !strcmp(dev->name, s[i].name)) {
514 dev->irq = s[i].map.irq;
515 dev->base_addr = s[i].map.base_addr;
516 dev->mem_start = s[i].map.mem_start;
517 dev->mem_end = s[i].map.mem_end;
523 EXPORT_SYMBOL(netdev_boot_setup_check);
527 * netdev_boot_base - get address from boot time settings
528 * @prefix: prefix for network device
529 * @unit: id for network device
531 * Check boot time settings for the base address of device.
532 * The found settings are set for the device to be used
533 * later in the device probing.
534 * Returns 0 if no settings found.
536 unsigned long netdev_boot_base(const char *prefix, int unit)
538 const struct netdev_boot_setup *s = dev_boot_setup;
542 sprintf(name, "%s%d", prefix, unit);
545 * If device already registered then return base of 1
546 * to indicate not to probe for this interface
548 if (__dev_get_by_name(&init_net, name))
551 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
552 if (!strcmp(name, s[i].name))
553 return s[i].map.base_addr;
558 * Saves at boot time configured settings for any netdevice.
560 int __init netdev_boot_setup(char *str)
565 str = get_options(str, ARRAY_SIZE(ints), ints);
570 memset(&map, 0, sizeof(map));
574 map.base_addr = ints[2];
576 map.mem_start = ints[3];
578 map.mem_end = ints[4];
580 /* Add new entry to the list */
581 return netdev_boot_setup_add(str, &map);
584 __setup("netdev=", netdev_boot_setup);
586 /*******************************************************************************
588 Device Interface Subroutines
590 *******************************************************************************/
593 * __dev_get_by_name - find a device by its name
594 * @net: the applicable net namespace
595 * @name: name to find
597 * Find an interface by name. Must be called under RTNL semaphore
598 * or @dev_base_lock. If the name is found a pointer to the device
599 * is returned. If the name is not found then %NULL is returned. The
600 * reference counters are not incremented so the caller must be
601 * careful with locks.
604 struct net_device *__dev_get_by_name(struct net *net, const char *name)
606 struct hlist_node *p;
607 struct net_device *dev;
608 struct hlist_head *head = dev_name_hash(net, name);
610 hlist_for_each_entry(dev, p, head, name_hlist)
611 if (!strncmp(dev->name, name, IFNAMSIZ))
616 EXPORT_SYMBOL(__dev_get_by_name);
619 * dev_get_by_name_rcu - find a device by its name
620 * @net: the applicable net namespace
621 * @name: name to find
623 * Find an interface by name.
624 * If the name is found a pointer to the device is returned.
625 * If the name is not found then %NULL is returned.
626 * The reference counters are not incremented so the caller must be
627 * careful with locks. The caller must hold RCU lock.
630 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
632 struct hlist_node *p;
633 struct net_device *dev;
634 struct hlist_head *head = dev_name_hash(net, name);
636 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
637 if (!strncmp(dev->name, name, IFNAMSIZ))
642 EXPORT_SYMBOL(dev_get_by_name_rcu);
645 * dev_get_by_name - find a device by its name
646 * @net: the applicable net namespace
647 * @name: name to find
649 * Find an interface by name. This can be called from any
650 * context and does its own locking. The returned handle has
651 * the usage count incremented and the caller must use dev_put() to
652 * release it when it is no longer needed. %NULL is returned if no
653 * matching device is found.
656 struct net_device *dev_get_by_name(struct net *net, const char *name)
658 struct net_device *dev;
661 dev = dev_get_by_name_rcu(net, name);
667 EXPORT_SYMBOL(dev_get_by_name);
670 * __dev_get_by_index - find a device by its ifindex
671 * @net: the applicable net namespace
672 * @ifindex: index of device
674 * Search for an interface by index. Returns %NULL if the device
675 * is not found or a pointer to the device. The device has not
676 * had its reference counter increased so the caller must be careful
677 * about locking. The caller must hold either the RTNL semaphore
681 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
683 struct hlist_node *p;
684 struct net_device *dev;
685 struct hlist_head *head = dev_index_hash(net, ifindex);
687 hlist_for_each_entry(dev, p, head, index_hlist)
688 if (dev->ifindex == ifindex)
693 EXPORT_SYMBOL(__dev_get_by_index);
696 * dev_get_by_index_rcu - find a device by its ifindex
697 * @net: the applicable net namespace
698 * @ifindex: index of device
700 * Search for an interface by index. Returns %NULL if the device
701 * is not found or a pointer to the device. The device has not
702 * had its reference counter increased so the caller must be careful
703 * about locking. The caller must hold RCU lock.
706 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
708 struct hlist_node *p;
709 struct net_device *dev;
710 struct hlist_head *head = dev_index_hash(net, ifindex);
712 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
713 if (dev->ifindex == ifindex)
718 EXPORT_SYMBOL(dev_get_by_index_rcu);
722 * dev_get_by_index - find a device by its ifindex
723 * @net: the applicable net namespace
724 * @ifindex: index of device
726 * Search for an interface by index. Returns NULL if the device
727 * is not found or a pointer to the device. The device returned has
728 * had a reference added and the pointer is safe until the user calls
729 * dev_put to indicate they have finished with it.
732 struct net_device *dev_get_by_index(struct net *net, int ifindex)
734 struct net_device *dev;
737 dev = dev_get_by_index_rcu(net, ifindex);
743 EXPORT_SYMBOL(dev_get_by_index);
746 * dev_getbyhwaddr_rcu - find a device by its hardware address
747 * @net: the applicable net namespace
748 * @type: media type of device
749 * @ha: hardware address
751 * Search for an interface by MAC address. Returns NULL if the device
752 * is not found or a pointer to the device. The caller must hold RCU
753 * The returned device has not had its ref count increased
754 * and the caller must therefore be careful about locking
758 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
761 struct net_device *dev;
763 for_each_netdev_rcu(net, dev)
764 if (dev->type == type &&
765 !memcmp(dev->dev_addr, ha, dev->addr_len))
770 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
772 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
774 struct net_device *dev;
777 for_each_netdev(net, dev)
778 if (dev->type == type)
783 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
785 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
787 struct net_device *dev, *ret = NULL;
790 for_each_netdev_rcu(net, dev)
791 if (dev->type == type) {
799 EXPORT_SYMBOL(dev_getfirstbyhwtype);
802 * dev_get_by_flags_rcu - find any device with given flags
803 * @net: the applicable net namespace
804 * @if_flags: IFF_* values
805 * @mask: bitmask of bits in if_flags to check
807 * Search for any interface with the given flags. Returns NULL if a device
808 * is not found or a pointer to the device. Must be called inside
809 * rcu_read_lock(), and result refcount is unchanged.
812 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
815 struct net_device *dev, *ret;
818 for_each_netdev_rcu(net, dev) {
819 if (((dev->flags ^ if_flags) & mask) == 0) {
826 EXPORT_SYMBOL(dev_get_by_flags_rcu);
829 * dev_valid_name - check if name is okay for network device
832 * Network device names need to be valid file names to
833 * to allow sysfs to work. We also disallow any kind of
836 int dev_valid_name(const char *name)
840 if (strlen(name) >= IFNAMSIZ)
842 if (!strcmp(name, ".") || !strcmp(name, ".."))
846 if (*name == '/' || isspace(*name))
852 EXPORT_SYMBOL(dev_valid_name);
855 * __dev_alloc_name - allocate a name for a device
856 * @net: network namespace to allocate the device name in
857 * @name: name format string
858 * @buf: scratch buffer and result name string
860 * Passed a format string - eg "lt%d" it will try and find a suitable
861 * id. It scans list of devices to build up a free map, then chooses
862 * the first empty slot. The caller must hold the dev_base or rtnl lock
863 * while allocating the name and adding the device in order to avoid
865 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
866 * Returns the number of the unit assigned or a negative errno code.
869 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
873 const int max_netdevices = 8*PAGE_SIZE;
874 unsigned long *inuse;
875 struct net_device *d;
877 p = strnchr(name, IFNAMSIZ-1, '%');
880 * Verify the string as this thing may have come from
881 * the user. There must be either one "%d" and no other "%"
884 if (p[1] != 'd' || strchr(p + 2, '%'))
887 /* Use one page as a bit array of possible slots */
888 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
892 for_each_netdev(net, d) {
893 if (!sscanf(d->name, name, &i))
895 if (i < 0 || i >= max_netdevices)
898 /* avoid cases where sscanf is not exact inverse of printf */
899 snprintf(buf, IFNAMSIZ, name, i);
900 if (!strncmp(buf, d->name, IFNAMSIZ))
904 i = find_first_zero_bit(inuse, max_netdevices);
905 free_page((unsigned long) inuse);
909 snprintf(buf, IFNAMSIZ, name, i);
910 if (!__dev_get_by_name(net, buf))
913 /* It is possible to run out of possible slots
914 * when the name is long and there isn't enough space left
915 * for the digits, or if all bits are used.
921 * dev_alloc_name - allocate a name for a device
923 * @name: name format string
925 * Passed a format string - eg "lt%d" it will try and find a suitable
926 * id. It scans list of devices to build up a free map, then chooses
927 * the first empty slot. The caller must hold the dev_base or rtnl lock
928 * while allocating the name and adding the device in order to avoid
930 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
931 * Returns the number of the unit assigned or a negative errno code.
934 int dev_alloc_name(struct net_device *dev, const char *name)
940 BUG_ON(!dev_net(dev));
942 ret = __dev_alloc_name(net, name, buf);
944 strlcpy(dev->name, buf, IFNAMSIZ);
947 EXPORT_SYMBOL(dev_alloc_name);
949 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
953 BUG_ON(!dev_net(dev));
956 if (!dev_valid_name(name))
959 if (fmt && strchr(name, '%'))
960 return dev_alloc_name(dev, name);
961 else if (__dev_get_by_name(net, name))
963 else if (dev->name != name)
964 strlcpy(dev->name, name, IFNAMSIZ);
970 * dev_change_name - change name of a device
972 * @newname: name (or format string) must be at least IFNAMSIZ
974 * Change name of a device, can pass format strings "eth%d".
977 int dev_change_name(struct net_device *dev, const char *newname)
979 char oldname[IFNAMSIZ];
985 BUG_ON(!dev_net(dev));
988 if (dev->flags & IFF_UP)
991 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
994 memcpy(oldname, dev->name, IFNAMSIZ);
996 err = dev_get_valid_name(dev, newname, 1);
1001 ret = device_rename(&dev->dev, dev->name);
1003 memcpy(dev->name, oldname, IFNAMSIZ);
1007 write_lock_bh(&dev_base_lock);
1008 hlist_del(&dev->name_hlist);
1009 write_unlock_bh(&dev_base_lock);
1013 write_lock_bh(&dev_base_lock);
1014 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1015 write_unlock_bh(&dev_base_lock);
1017 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1018 ret = notifier_to_errno(ret);
1021 /* err >= 0 after dev_alloc_name() or stores the first errno */
1024 memcpy(dev->name, oldname, IFNAMSIZ);
1028 "%s: name change rollback failed: %d.\n",
1037 * dev_set_alias - change ifalias of a device
1039 * @alias: name up to IFALIASZ
1040 * @len: limit of bytes to copy from info
1042 * Set ifalias for a device,
1044 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1048 if (len >= IFALIASZ)
1053 kfree(dev->ifalias);
1054 dev->ifalias = NULL;
1059 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1063 strlcpy(dev->ifalias, alias, len+1);
1069 * netdev_features_change - device changes features
1070 * @dev: device to cause notification
1072 * Called to indicate a device has changed features.
1074 void netdev_features_change(struct net_device *dev)
1076 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1078 EXPORT_SYMBOL(netdev_features_change);
1081 * netdev_state_change - device changes state
1082 * @dev: device to cause notification
1084 * Called to indicate a device has changed state. This function calls
1085 * the notifier chains for netdev_chain and sends a NEWLINK message
1086 * to the routing socket.
1088 void netdev_state_change(struct net_device *dev)
1090 if (dev->flags & IFF_UP) {
1091 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1092 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1095 EXPORT_SYMBOL(netdev_state_change);
1097 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1099 return call_netdevice_notifiers(event, dev);
1101 EXPORT_SYMBOL(netdev_bonding_change);
1104 * dev_load - load a network module
1105 * @net: the applicable net namespace
1106 * @name: name of interface
1108 * If a network interface is not present and the process has suitable
1109 * privileges this function loads the module. If module loading is not
1110 * available in this kernel then it becomes a nop.
1113 void dev_load(struct net *net, const char *name)
1115 struct net_device *dev;
1118 dev = dev_get_by_name_rcu(net, name);
1121 if (!dev && capable(CAP_NET_ADMIN))
1122 request_module("%s", name);
1124 EXPORT_SYMBOL(dev_load);
1126 static int __dev_open(struct net_device *dev)
1128 const struct net_device_ops *ops = dev->netdev_ops;
1134 * Is it even present?
1136 if (!netif_device_present(dev))
1139 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1140 ret = notifier_to_errno(ret);
1145 * Call device private open method
1147 set_bit(__LINK_STATE_START, &dev->state);
1149 if (ops->ndo_validate_addr)
1150 ret = ops->ndo_validate_addr(dev);
1152 if (!ret && ops->ndo_open)
1153 ret = ops->ndo_open(dev);
1156 * If it went open OK then:
1160 clear_bit(__LINK_STATE_START, &dev->state);
1165 dev->flags |= IFF_UP;
1170 net_dmaengine_get();
1173 * Initialize multicasting status
1175 dev_set_rx_mode(dev);
1178 * Wakeup transmit queue engine
1187 * dev_open - prepare an interface for use.
1188 * @dev: device to open
1190 * Takes a device from down to up state. The device's private open
1191 * function is invoked and then the multicast lists are loaded. Finally
1192 * the device is moved into the up state and a %NETDEV_UP message is
1193 * sent to the netdev notifier chain.
1195 * Calling this function on an active interface is a nop. On a failure
1196 * a negative errno code is returned.
1198 int dev_open(struct net_device *dev)
1205 if (dev->flags & IFF_UP)
1211 ret = __dev_open(dev);
1216 * ... and announce new interface.
1218 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1219 call_netdevice_notifiers(NETDEV_UP, dev);
1223 EXPORT_SYMBOL(dev_open);
1225 static int __dev_close_many(struct list_head *head)
1227 struct net_device *dev;
1232 list_for_each_entry(dev, head, unreg_list) {
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, it
1242 * 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(). */
1250 dev_deactivate_many(head);
1252 list_for_each_entry(dev, head, unreg_list) {
1253 const struct net_device_ops *ops = dev->netdev_ops;
1256 * Call the device specific close. This cannot fail.
1257 * Only if device is UP
1259 * We allow it to be called even after a DETACH hot-plug
1266 * Device is now down.
1269 dev->flags &= ~IFF_UP;
1274 net_dmaengine_put();
1280 static int __dev_close(struct net_device *dev)
1284 list_add(&dev->unreg_list, &single);
1285 return __dev_close_many(&single);
1288 int dev_close_many(struct list_head *head)
1290 struct net_device *dev, *tmp;
1291 LIST_HEAD(tmp_list);
1293 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1294 if (!(dev->flags & IFF_UP))
1295 list_move(&dev->unreg_list, &tmp_list);
1297 __dev_close_many(head);
1300 * Tell people we are down
1302 list_for_each_entry(dev, head, unreg_list) {
1303 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1304 call_netdevice_notifiers(NETDEV_DOWN, dev);
1307 /* rollback_registered_many needs the complete original list */
1308 list_splice(&tmp_list, head);
1313 * dev_close - shutdown an interface.
1314 * @dev: device to shutdown
1316 * This function moves an active device into down state. A
1317 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1318 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1321 int dev_close(struct net_device *dev)
1325 list_add(&dev->unreg_list, &single);
1326 dev_close_many(&single);
1330 EXPORT_SYMBOL(dev_close);
1334 * dev_disable_lro - disable Large Receive Offload on a device
1337 * Disable Large Receive Offload (LRO) on a net device. Must be
1338 * called under RTNL. This is needed if received packets may be
1339 * forwarded to another interface.
1341 void dev_disable_lro(struct net_device *dev)
1343 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1344 dev->ethtool_ops->set_flags) {
1345 u32 flags = dev->ethtool_ops->get_flags(dev);
1346 if (flags & ETH_FLAG_LRO) {
1347 flags &= ~ETH_FLAG_LRO;
1348 dev->ethtool_ops->set_flags(dev, flags);
1351 WARN_ON(dev->features & NETIF_F_LRO);
1353 EXPORT_SYMBOL(dev_disable_lro);
1356 static int dev_boot_phase = 1;
1359 * Device change register/unregister. These are not inline or static
1360 * as we export them to the world.
1364 * register_netdevice_notifier - register a network notifier block
1367 * Register a notifier to be called when network device events occur.
1368 * The notifier passed is linked into the kernel structures and must
1369 * not be reused until it has been unregistered. A negative errno code
1370 * is returned on a failure.
1372 * When registered all registration and up events are replayed
1373 * to the new notifier to allow device to have a race free
1374 * view of the network device list.
1377 int register_netdevice_notifier(struct notifier_block *nb)
1379 struct net_device *dev;
1380 struct net_device *last;
1385 err = raw_notifier_chain_register(&netdev_chain, nb);
1391 for_each_netdev(net, dev) {
1392 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1393 err = notifier_to_errno(err);
1397 if (!(dev->flags & IFF_UP))
1400 nb->notifier_call(nb, NETDEV_UP, dev);
1411 for_each_netdev(net, dev) {
1415 if (dev->flags & IFF_UP) {
1416 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1417 nb->notifier_call(nb, NETDEV_DOWN, dev);
1419 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1420 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1424 raw_notifier_chain_unregister(&netdev_chain, nb);
1427 EXPORT_SYMBOL(register_netdevice_notifier);
1430 * unregister_netdevice_notifier - unregister a network notifier block
1433 * Unregister a notifier previously registered by
1434 * register_netdevice_notifier(). The notifier is unlinked into the
1435 * kernel structures and may then be reused. A negative errno code
1436 * is returned on a failure.
1439 int unregister_netdevice_notifier(struct notifier_block *nb)
1444 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1448 EXPORT_SYMBOL(unregister_netdevice_notifier);
1451 * call_netdevice_notifiers - call all network notifier blocks
1452 * @val: value passed unmodified to notifier function
1453 * @dev: net_device pointer passed unmodified to notifier function
1455 * Call all network notifier blocks. Parameters and return value
1456 * are as for raw_notifier_call_chain().
1459 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1462 return raw_notifier_call_chain(&netdev_chain, val, dev);
1465 /* When > 0 there are consumers of rx skb time stamps */
1466 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1468 void net_enable_timestamp(void)
1470 atomic_inc(&netstamp_needed);
1472 EXPORT_SYMBOL(net_enable_timestamp);
1474 void net_disable_timestamp(void)
1476 atomic_dec(&netstamp_needed);
1478 EXPORT_SYMBOL(net_disable_timestamp);
1480 static inline void net_timestamp_set(struct sk_buff *skb)
1482 if (atomic_read(&netstamp_needed))
1483 __net_timestamp(skb);
1485 skb->tstamp.tv64 = 0;
1488 static inline void net_timestamp_check(struct sk_buff *skb)
1490 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1491 __net_timestamp(skb);
1495 * dev_forward_skb - loopback an skb to another netif
1497 * @dev: destination network device
1498 * @skb: buffer to forward
1501 * NET_RX_SUCCESS (no congestion)
1502 * NET_RX_DROP (packet was dropped, but freed)
1504 * dev_forward_skb can be used for injecting an skb from the
1505 * start_xmit function of one device into the receive queue
1506 * of another device.
1508 * The receiving device may be in another namespace, so
1509 * we have to clear all information in the skb that could
1510 * impact namespace isolation.
1512 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1517 if (unlikely(!(dev->flags & IFF_UP) ||
1518 (skb->len > (dev->mtu + dev->hard_header_len + VLAN_HLEN)))) {
1519 atomic_long_inc(&dev->rx_dropped);
1523 skb_set_dev(skb, dev);
1524 skb->tstamp.tv64 = 0;
1525 skb->pkt_type = PACKET_HOST;
1526 skb->protocol = eth_type_trans(skb, dev);
1527 return netif_rx(skb);
1529 EXPORT_SYMBOL_GPL(dev_forward_skb);
1531 static inline int deliver_skb(struct sk_buff *skb,
1532 struct packet_type *pt_prev,
1533 struct net_device *orig_dev)
1535 atomic_inc(&skb->users);
1536 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1540 * Support routine. Sends outgoing frames to any network
1541 * taps currently in use.
1544 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1546 struct packet_type *ptype;
1547 struct sk_buff *skb2 = NULL;
1548 struct packet_type *pt_prev = NULL;
1550 #ifdef CONFIG_NET_CLS_ACT
1551 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1552 net_timestamp_set(skb);
1554 net_timestamp_set(skb);
1558 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1559 /* Never send packets back to the socket
1560 * they originated from - MvS (miquels@drinkel.ow.org)
1562 if ((ptype->dev == dev || !ptype->dev) &&
1563 (ptype->af_packet_priv == NULL ||
1564 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1566 deliver_skb(skb2, pt_prev, skb->dev);
1571 skb2 = skb_clone(skb, GFP_ATOMIC);
1575 /* skb->nh should be correctly
1576 set by sender, so that the second statement is
1577 just protection against buggy protocols.
1579 skb_reset_mac_header(skb2);
1581 if (skb_network_header(skb2) < skb2->data ||
1582 skb2->network_header > skb2->tail) {
1583 if (net_ratelimit())
1584 printk(KERN_CRIT "protocol %04x is "
1586 ntohs(skb2->protocol),
1588 skb_reset_network_header(skb2);
1591 skb2->transport_header = skb2->network_header;
1592 skb2->pkt_type = PACKET_OUTGOING;
1597 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1602 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1603 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1605 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1609 if (txq < 1 || txq > dev->num_tx_queues)
1612 if (dev->reg_state == NETREG_REGISTERED) {
1615 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1620 if (txq < dev->real_num_tx_queues)
1621 qdisc_reset_all_tx_gt(dev, txq);
1624 dev->real_num_tx_queues = txq;
1627 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1631 * netif_set_real_num_rx_queues - set actual number of RX queues used
1632 * @dev: Network device
1633 * @rxq: Actual number of RX queues
1635 * This must be called either with the rtnl_lock held or before
1636 * registration of the net device. Returns 0 on success, or a
1637 * negative error code. If called before registration, it always
1640 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1644 if (rxq < 1 || rxq > dev->num_rx_queues)
1647 if (dev->reg_state == NETREG_REGISTERED) {
1650 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1656 dev->real_num_rx_queues = rxq;
1659 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1662 static inline void __netif_reschedule(struct Qdisc *q)
1664 struct softnet_data *sd;
1665 unsigned long flags;
1667 local_irq_save(flags);
1668 sd = &__get_cpu_var(softnet_data);
1669 q->next_sched = NULL;
1670 *sd->output_queue_tailp = q;
1671 sd->output_queue_tailp = &q->next_sched;
1672 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1673 local_irq_restore(flags);
1676 void __netif_schedule(struct Qdisc *q)
1678 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1679 __netif_reschedule(q);
1681 EXPORT_SYMBOL(__netif_schedule);
1683 void dev_kfree_skb_irq(struct sk_buff *skb)
1685 if (atomic_dec_and_test(&skb->users)) {
1686 struct softnet_data *sd;
1687 unsigned long flags;
1689 local_irq_save(flags);
1690 sd = &__get_cpu_var(softnet_data);
1691 skb->next = sd->completion_queue;
1692 sd->completion_queue = skb;
1693 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1694 local_irq_restore(flags);
1697 EXPORT_SYMBOL(dev_kfree_skb_irq);
1699 void dev_kfree_skb_any(struct sk_buff *skb)
1701 if (in_irq() || irqs_disabled())
1702 dev_kfree_skb_irq(skb);
1706 EXPORT_SYMBOL(dev_kfree_skb_any);
1710 * netif_device_detach - mark device as removed
1711 * @dev: network device
1713 * Mark device as removed from system and therefore no longer available.
1715 void netif_device_detach(struct net_device *dev)
1717 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1718 netif_running(dev)) {
1719 netif_tx_stop_all_queues(dev);
1722 EXPORT_SYMBOL(netif_device_detach);
1725 * netif_device_attach - mark device as attached
1726 * @dev: network device
1728 * Mark device as attached from system and restart if needed.
1730 void netif_device_attach(struct net_device *dev)
1732 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1733 netif_running(dev)) {
1734 netif_tx_wake_all_queues(dev);
1735 __netdev_watchdog_up(dev);
1738 EXPORT_SYMBOL(netif_device_attach);
1740 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1742 return ((features & NETIF_F_NO_CSUM) ||
1743 ((features & NETIF_F_V4_CSUM) &&
1744 protocol == htons(ETH_P_IP)) ||
1745 ((features & NETIF_F_V6_CSUM) &&
1746 protocol == htons(ETH_P_IPV6)) ||
1747 ((features & NETIF_F_FCOE_CRC) &&
1748 protocol == htons(ETH_P_FCOE)));
1751 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1753 __be16 protocol = skb->protocol;
1754 int features = dev->features;
1756 if (vlan_tx_tag_present(skb)) {
1757 features &= dev->vlan_features;
1758 } else if (protocol == htons(ETH_P_8021Q)) {
1759 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1760 protocol = veh->h_vlan_encapsulated_proto;
1761 features &= dev->vlan_features;
1764 return can_checksum_protocol(features, protocol);
1768 * skb_dev_set -- assign a new device to a buffer
1769 * @skb: buffer for the new device
1770 * @dev: network device
1772 * If an skb is owned by a device already, we have to reset
1773 * all data private to the namespace a device belongs to
1774 * before assigning it a new device.
1776 #ifdef CONFIG_NET_NS
1777 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1780 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1783 skb_init_secmark(skb);
1787 skb->ipvs_property = 0;
1788 #ifdef CONFIG_NET_SCHED
1794 EXPORT_SYMBOL(skb_set_dev);
1795 #endif /* CONFIG_NET_NS */
1798 * Invalidate hardware checksum when packet is to be mangled, and
1799 * complete checksum manually on outgoing path.
1801 int skb_checksum_help(struct sk_buff *skb)
1804 int ret = 0, offset;
1806 if (skb->ip_summed == CHECKSUM_COMPLETE)
1807 goto out_set_summed;
1809 if (unlikely(skb_shinfo(skb)->gso_size)) {
1810 /* Let GSO fix up the checksum. */
1811 goto out_set_summed;
1814 offset = skb_checksum_start_offset(skb);
1815 BUG_ON(offset >= skb_headlen(skb));
1816 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1818 offset += skb->csum_offset;
1819 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1821 if (skb_cloned(skb) &&
1822 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1823 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1828 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1830 skb->ip_summed = CHECKSUM_NONE;
1834 EXPORT_SYMBOL(skb_checksum_help);
1837 * skb_gso_segment - Perform segmentation on skb.
1838 * @skb: buffer to segment
1839 * @features: features for the output path (see dev->features)
1841 * This function segments the given skb and returns a list of segments.
1843 * It may return NULL if the skb requires no segmentation. This is
1844 * only possible when GSO is used for verifying header integrity.
1846 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1848 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1849 struct packet_type *ptype;
1850 __be16 type = skb->protocol;
1851 int vlan_depth = ETH_HLEN;
1854 while (type == htons(ETH_P_8021Q)) {
1855 struct vlan_hdr *vh;
1857 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1858 return ERR_PTR(-EINVAL);
1860 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1861 type = vh->h_vlan_encapsulated_proto;
1862 vlan_depth += VLAN_HLEN;
1865 skb_reset_mac_header(skb);
1866 skb->mac_len = skb->network_header - skb->mac_header;
1867 __skb_pull(skb, skb->mac_len);
1869 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1870 struct net_device *dev = skb->dev;
1871 struct ethtool_drvinfo info = {};
1873 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1874 dev->ethtool_ops->get_drvinfo(dev, &info);
1876 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1877 info.driver, dev ? dev->features : 0L,
1878 skb->sk ? skb->sk->sk_route_caps : 0L,
1879 skb->len, skb->data_len, skb->ip_summed);
1881 if (skb_header_cloned(skb) &&
1882 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1883 return ERR_PTR(err);
1887 list_for_each_entry_rcu(ptype,
1888 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1889 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1890 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1891 err = ptype->gso_send_check(skb);
1892 segs = ERR_PTR(err);
1893 if (err || skb_gso_ok(skb, features))
1895 __skb_push(skb, (skb->data -
1896 skb_network_header(skb)));
1898 segs = ptype->gso_segment(skb, features);
1904 __skb_push(skb, skb->data - skb_mac_header(skb));
1908 EXPORT_SYMBOL(skb_gso_segment);
1910 /* Take action when hardware reception checksum errors are detected. */
1912 void netdev_rx_csum_fault(struct net_device *dev)
1914 if (net_ratelimit()) {
1915 printk(KERN_ERR "%s: hw csum failure.\n",
1916 dev ? dev->name : "<unknown>");
1920 EXPORT_SYMBOL(netdev_rx_csum_fault);
1923 /* Actually, we should eliminate this check as soon as we know, that:
1924 * 1. IOMMU is present and allows to map all the memory.
1925 * 2. No high memory really exists on this machine.
1928 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1930 #ifdef CONFIG_HIGHMEM
1932 if (!(dev->features & NETIF_F_HIGHDMA)) {
1933 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1934 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1938 if (PCI_DMA_BUS_IS_PHYS) {
1939 struct device *pdev = dev->dev.parent;
1943 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1944 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1945 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1954 void (*destructor)(struct sk_buff *skb);
1957 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1959 static void dev_gso_skb_destructor(struct sk_buff *skb)
1961 struct dev_gso_cb *cb;
1964 struct sk_buff *nskb = skb->next;
1966 skb->next = nskb->next;
1969 } while (skb->next);
1971 cb = DEV_GSO_CB(skb);
1973 cb->destructor(skb);
1977 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1978 * @skb: buffer to segment
1980 * This function segments the given skb and stores the list of segments
1983 static int dev_gso_segment(struct sk_buff *skb)
1985 struct net_device *dev = skb->dev;
1986 struct sk_buff *segs;
1987 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1990 segs = skb_gso_segment(skb, features);
1992 /* Verifying header integrity only. */
1997 return PTR_ERR(segs);
2000 DEV_GSO_CB(skb)->destructor = skb->destructor;
2001 skb->destructor = dev_gso_skb_destructor;
2007 * Try to orphan skb early, right before transmission by the device.
2008 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2009 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2011 static inline void skb_orphan_try(struct sk_buff *skb)
2013 struct sock *sk = skb->sk;
2015 if (sk && !skb_shinfo(skb)->tx_flags) {
2016 /* skb_tx_hash() wont be able to get sk.
2017 * We copy sk_hash into skb->rxhash
2020 skb->rxhash = sk->sk_hash;
2025 int netif_get_vlan_features(struct sk_buff *skb, struct net_device *dev)
2027 __be16 protocol = skb->protocol;
2029 if (protocol == htons(ETH_P_8021Q)) {
2030 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2031 protocol = veh->h_vlan_encapsulated_proto;
2032 } else if (!skb->vlan_tci)
2033 return dev->features;
2035 if (protocol != htons(ETH_P_8021Q))
2036 return dev->features & dev->vlan_features;
2040 EXPORT_SYMBOL(netif_get_vlan_features);
2043 * Returns true if either:
2044 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2045 * 2. skb is fragmented and the device does not support SG, or if
2046 * at least one of fragments is in highmem and device does not
2047 * support DMA from it.
2049 static inline int skb_needs_linearize(struct sk_buff *skb,
2050 struct net_device *dev)
2052 if (skb_is_nonlinear(skb)) {
2053 int features = dev->features;
2055 if (vlan_tx_tag_present(skb))
2056 features &= dev->vlan_features;
2058 return (skb_has_frag_list(skb) &&
2059 !(features & NETIF_F_FRAGLIST)) ||
2060 (skb_shinfo(skb)->nr_frags &&
2061 (!(features & NETIF_F_SG) ||
2062 illegal_highdma(dev, skb)));
2068 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2069 struct netdev_queue *txq)
2071 const struct net_device_ops *ops = dev->netdev_ops;
2072 int rc = NETDEV_TX_OK;
2074 if (likely(!skb->next)) {
2076 * If device doesnt need skb->dst, release it right now while
2077 * its hot in this cpu cache
2079 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2082 if (!list_empty(&ptype_all))
2083 dev_queue_xmit_nit(skb, dev);
2085 skb_orphan_try(skb);
2087 if (vlan_tx_tag_present(skb) &&
2088 !(dev->features & NETIF_F_HW_VLAN_TX)) {
2089 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2096 if (netif_needs_gso(dev, skb)) {
2097 if (unlikely(dev_gso_segment(skb)))
2102 if (skb_needs_linearize(skb, dev) &&
2103 __skb_linearize(skb))
2106 /* If packet is not checksummed and device does not
2107 * support checksumming for this protocol, complete
2108 * checksumming here.
2110 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2111 skb_set_transport_header(skb,
2112 skb_checksum_start_offset(skb));
2113 if (!dev_can_checksum(dev, skb) &&
2114 skb_checksum_help(skb))
2119 rc = ops->ndo_start_xmit(skb, dev);
2120 trace_net_dev_xmit(skb, rc);
2121 if (rc == NETDEV_TX_OK)
2122 txq_trans_update(txq);
2128 struct sk_buff *nskb = skb->next;
2130 skb->next = nskb->next;
2134 * If device doesnt need nskb->dst, release it right now while
2135 * its hot in this cpu cache
2137 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2140 rc = ops->ndo_start_xmit(nskb, dev);
2141 trace_net_dev_xmit(nskb, rc);
2142 if (unlikely(rc != NETDEV_TX_OK)) {
2143 if (rc & ~NETDEV_TX_MASK)
2144 goto out_kfree_gso_skb;
2145 nskb->next = skb->next;
2149 txq_trans_update(txq);
2150 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2151 return NETDEV_TX_BUSY;
2152 } while (skb->next);
2155 if (likely(skb->next == NULL))
2156 skb->destructor = DEV_GSO_CB(skb)->destructor;
2163 static u32 hashrnd __read_mostly;
2166 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2167 * to be used as a distribution range.
2169 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2170 unsigned int num_tx_queues)
2174 if (skb_rx_queue_recorded(skb)) {
2175 hash = skb_get_rx_queue(skb);
2176 while (unlikely(hash >= num_tx_queues))
2177 hash -= num_tx_queues;
2181 if (skb->sk && skb->sk->sk_hash)
2182 hash = skb->sk->sk_hash;
2184 hash = (__force u16) skb->protocol ^ skb->rxhash;
2185 hash = jhash_1word(hash, hashrnd);
2187 return (u16) (((u64) hash * num_tx_queues) >> 32);
2189 EXPORT_SYMBOL(__skb_tx_hash);
2191 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2193 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2194 if (net_ratelimit()) {
2195 pr_warning("%s selects TX queue %d, but "
2196 "real number of TX queues is %d\n",
2197 dev->name, queue_index, dev->real_num_tx_queues);
2204 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2207 struct xps_dev_maps *dev_maps;
2208 struct xps_map *map;
2209 int queue_index = -1;
2212 dev_maps = rcu_dereference(dev->xps_maps);
2214 map = rcu_dereference(
2215 dev_maps->cpu_map[raw_smp_processor_id()]);
2218 queue_index = map->queues[0];
2221 if (skb->sk && skb->sk->sk_hash)
2222 hash = skb->sk->sk_hash;
2224 hash = (__force u16) skb->protocol ^
2226 hash = jhash_1word(hash, hashrnd);
2227 queue_index = map->queues[
2228 ((u64)hash * map->len) >> 32];
2230 if (unlikely(queue_index >= dev->real_num_tx_queues))
2242 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2243 struct sk_buff *skb)
2246 const struct net_device_ops *ops = dev->netdev_ops;
2248 if (dev->real_num_tx_queues == 1)
2250 else if (ops->ndo_select_queue) {
2251 queue_index = ops->ndo_select_queue(dev, skb);
2252 queue_index = dev_cap_txqueue(dev, queue_index);
2254 struct sock *sk = skb->sk;
2255 queue_index = sk_tx_queue_get(sk);
2257 if (queue_index < 0 || skb->ooo_okay ||
2258 queue_index >= dev->real_num_tx_queues) {
2259 int old_index = queue_index;
2261 queue_index = get_xps_queue(dev, skb);
2262 if (queue_index < 0)
2263 queue_index = skb_tx_hash(dev, skb);
2265 if (queue_index != old_index && sk) {
2266 struct dst_entry *dst =
2267 rcu_dereference_check(sk->sk_dst_cache, 1);
2269 if (dst && skb_dst(skb) == dst)
2270 sk_tx_queue_set(sk, queue_index);
2275 skb_set_queue_mapping(skb, queue_index);
2276 return netdev_get_tx_queue(dev, queue_index);
2279 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2280 struct net_device *dev,
2281 struct netdev_queue *txq)
2283 spinlock_t *root_lock = qdisc_lock(q);
2284 bool contended = qdisc_is_running(q);
2288 * Heuristic to force contended enqueues to serialize on a
2289 * separate lock before trying to get qdisc main lock.
2290 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2291 * and dequeue packets faster.
2293 if (unlikely(contended))
2294 spin_lock(&q->busylock);
2296 spin_lock(root_lock);
2297 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2300 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2301 qdisc_run_begin(q)) {
2303 * This is a work-conserving queue; there are no old skbs
2304 * waiting to be sent out; and the qdisc is not running -
2305 * xmit the skb directly.
2307 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2309 __qdisc_update_bstats(q, skb->len);
2310 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2311 if (unlikely(contended)) {
2312 spin_unlock(&q->busylock);
2319 rc = NET_XMIT_SUCCESS;
2322 rc = qdisc_enqueue_root(skb, q);
2323 if (qdisc_run_begin(q)) {
2324 if (unlikely(contended)) {
2325 spin_unlock(&q->busylock);
2331 spin_unlock(root_lock);
2332 if (unlikely(contended))
2333 spin_unlock(&q->busylock);
2337 static DEFINE_PER_CPU(int, xmit_recursion);
2338 #define RECURSION_LIMIT 10
2341 * dev_queue_xmit - transmit a buffer
2342 * @skb: buffer to transmit
2344 * Queue a buffer for transmission to a network device. The caller must
2345 * have set the device and priority and built the buffer before calling
2346 * this function. The function can be called from an interrupt.
2348 * A negative errno code is returned on a failure. A success does not
2349 * guarantee the frame will be transmitted as it may be dropped due
2350 * to congestion or traffic shaping.
2352 * -----------------------------------------------------------------------------------
2353 * I notice this method can also return errors from the queue disciplines,
2354 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2357 * Regardless of the return value, the skb is consumed, so it is currently
2358 * difficult to retry a send to this method. (You can bump the ref count
2359 * before sending to hold a reference for retry if you are careful.)
2361 * When calling this method, interrupts MUST be enabled. This is because
2362 * the BH enable code must have IRQs enabled so that it will not deadlock.
2365 int dev_queue_xmit(struct sk_buff *skb)
2367 struct net_device *dev = skb->dev;
2368 struct netdev_queue *txq;
2372 /* Disable soft irqs for various locks below. Also
2373 * stops preemption for RCU.
2377 txq = dev_pick_tx(dev, skb);
2378 q = rcu_dereference_bh(txq->qdisc);
2380 #ifdef CONFIG_NET_CLS_ACT
2381 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2383 trace_net_dev_queue(skb);
2385 rc = __dev_xmit_skb(skb, q, dev, txq);
2389 /* The device has no queue. Common case for software devices:
2390 loopback, all the sorts of tunnels...
2392 Really, it is unlikely that netif_tx_lock protection is necessary
2393 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2395 However, it is possible, that they rely on protection
2398 Check this and shot the lock. It is not prone from deadlocks.
2399 Either shot noqueue qdisc, it is even simpler 8)
2401 if (dev->flags & IFF_UP) {
2402 int cpu = smp_processor_id(); /* ok because BHs are off */
2404 if (txq->xmit_lock_owner != cpu) {
2406 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2407 goto recursion_alert;
2409 HARD_TX_LOCK(dev, txq, cpu);
2411 if (!netif_tx_queue_stopped(txq)) {
2412 __this_cpu_inc(xmit_recursion);
2413 rc = dev_hard_start_xmit(skb, dev, txq);
2414 __this_cpu_dec(xmit_recursion);
2415 if (dev_xmit_complete(rc)) {
2416 HARD_TX_UNLOCK(dev, txq);
2420 HARD_TX_UNLOCK(dev, txq);
2421 if (net_ratelimit())
2422 printk(KERN_CRIT "Virtual device %s asks to "
2423 "queue packet!\n", dev->name);
2425 /* Recursion is detected! It is possible,
2429 if (net_ratelimit())
2430 printk(KERN_CRIT "Dead loop on virtual device "
2431 "%s, fix it urgently!\n", dev->name);
2436 rcu_read_unlock_bh();
2441 rcu_read_unlock_bh();
2444 EXPORT_SYMBOL(dev_queue_xmit);
2447 /*=======================================================================
2449 =======================================================================*/
2451 int netdev_max_backlog __read_mostly = 1000;
2452 int netdev_tstamp_prequeue __read_mostly = 1;
2453 int netdev_budget __read_mostly = 300;
2454 int weight_p __read_mostly = 64; /* old backlog weight */
2456 /* Called with irq disabled */
2457 static inline void ____napi_schedule(struct softnet_data *sd,
2458 struct napi_struct *napi)
2460 list_add_tail(&napi->poll_list, &sd->poll_list);
2461 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2465 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2466 * and src/dst port numbers. Returns a non-zero hash number on success
2469 __u32 __skb_get_rxhash(struct sk_buff *skb)
2471 int nhoff, hash = 0, poff;
2472 struct ipv6hdr *ip6;
2475 u32 addr1, addr2, ihl;
2481 nhoff = skb_network_offset(skb);
2483 switch (skb->protocol) {
2484 case __constant_htons(ETH_P_IP):
2485 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2488 ip = (struct iphdr *) (skb->data + nhoff);
2489 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2492 ip_proto = ip->protocol;
2493 addr1 = (__force u32) ip->saddr;
2494 addr2 = (__force u32) ip->daddr;
2497 case __constant_htons(ETH_P_IPV6):
2498 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2501 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2502 ip_proto = ip6->nexthdr;
2503 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2504 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2512 poff = proto_ports_offset(ip_proto);
2514 nhoff += ihl * 4 + poff;
2515 if (pskb_may_pull(skb, nhoff + 4)) {
2516 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2517 if (ports.v16[1] < ports.v16[0])
2518 swap(ports.v16[0], ports.v16[1]);
2522 /* get a consistent hash (same value on both flow directions) */
2526 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2533 EXPORT_SYMBOL(__skb_get_rxhash);
2537 /* One global table that all flow-based protocols share. */
2538 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2539 EXPORT_SYMBOL(rps_sock_flow_table);
2542 * get_rps_cpu is called from netif_receive_skb and returns the target
2543 * CPU from the RPS map of the receiving queue for a given skb.
2544 * rcu_read_lock must be held on entry.
2546 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2547 struct rps_dev_flow **rflowp)
2549 struct netdev_rx_queue *rxqueue;
2550 struct rps_map *map;
2551 struct rps_dev_flow_table *flow_table;
2552 struct rps_sock_flow_table *sock_flow_table;
2556 if (skb_rx_queue_recorded(skb)) {
2557 u16 index = skb_get_rx_queue(skb);
2558 if (unlikely(index >= dev->real_num_rx_queues)) {
2559 WARN_ONCE(dev->real_num_rx_queues > 1,
2560 "%s received packet on queue %u, but number "
2561 "of RX queues is %u\n",
2562 dev->name, index, dev->real_num_rx_queues);
2565 rxqueue = dev->_rx + index;
2569 map = rcu_dereference(rxqueue->rps_map);
2571 if (map->len == 1) {
2572 tcpu = map->cpus[0];
2573 if (cpu_online(tcpu))
2577 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2581 skb_reset_network_header(skb);
2582 if (!skb_get_rxhash(skb))
2585 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2586 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2587 if (flow_table && sock_flow_table) {
2589 struct rps_dev_flow *rflow;
2591 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2594 next_cpu = sock_flow_table->ents[skb->rxhash &
2595 sock_flow_table->mask];
2598 * If the desired CPU (where last recvmsg was done) is
2599 * different from current CPU (one in the rx-queue flow
2600 * table entry), switch if one of the following holds:
2601 * - Current CPU is unset (equal to RPS_NO_CPU).
2602 * - Current CPU is offline.
2603 * - The current CPU's queue tail has advanced beyond the
2604 * last packet that was enqueued using this table entry.
2605 * This guarantees that all previous packets for the flow
2606 * have been dequeued, thus preserving in order delivery.
2608 if (unlikely(tcpu != next_cpu) &&
2609 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2610 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2611 rflow->last_qtail)) >= 0)) {
2612 tcpu = rflow->cpu = next_cpu;
2613 if (tcpu != RPS_NO_CPU)
2614 rflow->last_qtail = per_cpu(softnet_data,
2615 tcpu).input_queue_head;
2617 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2625 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2627 if (cpu_online(tcpu)) {
2637 /* Called from hardirq (IPI) context */
2638 static void rps_trigger_softirq(void *data)
2640 struct softnet_data *sd = data;
2642 ____napi_schedule(sd, &sd->backlog);
2646 #endif /* CONFIG_RPS */
2649 * Check if this softnet_data structure is another cpu one
2650 * If yes, queue it to our IPI list and return 1
2653 static int rps_ipi_queued(struct softnet_data *sd)
2656 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2659 sd->rps_ipi_next = mysd->rps_ipi_list;
2660 mysd->rps_ipi_list = sd;
2662 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2665 #endif /* CONFIG_RPS */
2670 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2671 * queue (may be a remote CPU queue).
2673 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2674 unsigned int *qtail)
2676 struct softnet_data *sd;
2677 unsigned long flags;
2679 sd = &per_cpu(softnet_data, cpu);
2681 local_irq_save(flags);
2684 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2685 if (skb_queue_len(&sd->input_pkt_queue)) {
2687 __skb_queue_tail(&sd->input_pkt_queue, skb);
2688 input_queue_tail_incr_save(sd, qtail);
2690 local_irq_restore(flags);
2691 return NET_RX_SUCCESS;
2694 /* Schedule NAPI for backlog device
2695 * We can use non atomic operation since we own the queue lock
2697 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2698 if (!rps_ipi_queued(sd))
2699 ____napi_schedule(sd, &sd->backlog);
2707 local_irq_restore(flags);
2709 atomic_long_inc(&skb->dev->rx_dropped);
2715 * netif_rx - post buffer to the network code
2716 * @skb: buffer to post
2718 * This function receives a packet from a device driver and queues it for
2719 * the upper (protocol) levels to process. It always succeeds. The buffer
2720 * may be dropped during processing for congestion control or by the
2724 * NET_RX_SUCCESS (no congestion)
2725 * NET_RX_DROP (packet was dropped)
2729 int netif_rx(struct sk_buff *skb)
2733 /* if netpoll wants it, pretend we never saw it */
2734 if (netpoll_rx(skb))
2737 if (netdev_tstamp_prequeue)
2738 net_timestamp_check(skb);
2740 trace_netif_rx(skb);
2743 struct rps_dev_flow voidflow, *rflow = &voidflow;
2749 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2751 cpu = smp_processor_id();
2753 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2761 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2767 EXPORT_SYMBOL(netif_rx);
2769 int netif_rx_ni(struct sk_buff *skb)
2774 err = netif_rx(skb);
2775 if (local_softirq_pending())
2781 EXPORT_SYMBOL(netif_rx_ni);
2783 static void net_tx_action(struct softirq_action *h)
2785 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2787 if (sd->completion_queue) {
2788 struct sk_buff *clist;
2790 local_irq_disable();
2791 clist = sd->completion_queue;
2792 sd->completion_queue = NULL;
2796 struct sk_buff *skb = clist;
2797 clist = clist->next;
2799 WARN_ON(atomic_read(&skb->users));
2800 trace_kfree_skb(skb, net_tx_action);
2805 if (sd->output_queue) {
2808 local_irq_disable();
2809 head = sd->output_queue;
2810 sd->output_queue = NULL;
2811 sd->output_queue_tailp = &sd->output_queue;
2815 struct Qdisc *q = head;
2816 spinlock_t *root_lock;
2818 head = head->next_sched;
2820 root_lock = qdisc_lock(q);
2821 if (spin_trylock(root_lock)) {
2822 smp_mb__before_clear_bit();
2823 clear_bit(__QDISC_STATE_SCHED,
2826 spin_unlock(root_lock);
2828 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2830 __netif_reschedule(q);
2832 smp_mb__before_clear_bit();
2833 clear_bit(__QDISC_STATE_SCHED,
2841 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2842 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2843 /* This hook is defined here for ATM LANE */
2844 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2845 unsigned char *addr) __read_mostly;
2846 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2849 #ifdef CONFIG_NET_CLS_ACT
2850 /* TODO: Maybe we should just force sch_ingress to be compiled in
2851 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2852 * a compare and 2 stores extra right now if we dont have it on
2853 * but have CONFIG_NET_CLS_ACT
2854 * NOTE: This doesnt stop any functionality; if you dont have
2855 * the ingress scheduler, you just cant add policies on ingress.
2858 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2860 struct net_device *dev = skb->dev;
2861 u32 ttl = G_TC_RTTL(skb->tc_verd);
2862 int result = TC_ACT_OK;
2865 if (unlikely(MAX_RED_LOOP < ttl++)) {
2866 if (net_ratelimit())
2867 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2868 skb->skb_iif, dev->ifindex);
2872 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2873 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2876 if (q != &noop_qdisc) {
2877 spin_lock(qdisc_lock(q));
2878 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2879 result = qdisc_enqueue_root(skb, q);
2880 spin_unlock(qdisc_lock(q));
2886 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2887 struct packet_type **pt_prev,
2888 int *ret, struct net_device *orig_dev)
2890 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
2892 if (!rxq || rxq->qdisc == &noop_qdisc)
2896 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2900 switch (ing_filter(skb, rxq)) {
2914 * netdev_rx_handler_register - register receive handler
2915 * @dev: device to register a handler for
2916 * @rx_handler: receive handler to register
2917 * @rx_handler_data: data pointer that is used by rx handler
2919 * Register a receive hander for a device. This handler will then be
2920 * called from __netif_receive_skb. A negative errno code is returned
2923 * The caller must hold the rtnl_mutex.
2925 int netdev_rx_handler_register(struct net_device *dev,
2926 rx_handler_func_t *rx_handler,
2927 void *rx_handler_data)
2931 if (dev->rx_handler)
2934 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2935 rcu_assign_pointer(dev->rx_handler, rx_handler);
2939 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2942 * netdev_rx_handler_unregister - unregister receive handler
2943 * @dev: device to unregister a handler from
2945 * Unregister a receive hander from a device.
2947 * The caller must hold the rtnl_mutex.
2949 void netdev_rx_handler_unregister(struct net_device *dev)
2953 rcu_assign_pointer(dev->rx_handler, NULL);
2954 rcu_assign_pointer(dev->rx_handler_data, NULL);
2956 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2958 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2959 struct net_device *master)
2961 if (skb->pkt_type == PACKET_HOST) {
2962 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2964 memcpy(dest, master->dev_addr, ETH_ALEN);
2968 /* On bonding slaves other than the currently active slave, suppress
2969 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2970 * ARP on active-backup slaves with arp_validate enabled.
2972 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2974 struct net_device *dev = skb->dev;
2976 if (master->priv_flags & IFF_MASTER_ARPMON)
2977 dev->last_rx = jiffies;
2979 if ((master->priv_flags & IFF_MASTER_ALB) &&
2980 (master->priv_flags & IFF_BRIDGE_PORT)) {
2981 /* Do address unmangle. The local destination address
2982 * will be always the one master has. Provides the right
2983 * functionality in a bridge.
2985 skb_bond_set_mac_by_master(skb, master);
2988 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2989 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2990 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2993 if (master->priv_flags & IFF_MASTER_ALB) {
2994 if (skb->pkt_type != PACKET_BROADCAST &&
2995 skb->pkt_type != PACKET_MULTICAST)
2998 if (master->priv_flags & IFF_MASTER_8023AD &&
2999 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
3006 EXPORT_SYMBOL(__skb_bond_should_drop);
3008 static int __netif_receive_skb(struct sk_buff *skb)
3010 struct packet_type *ptype, *pt_prev;
3011 rx_handler_func_t *rx_handler;
3012 struct net_device *orig_dev;
3013 struct net_device *master;
3014 struct net_device *null_or_orig;
3015 struct net_device *orig_or_bond;
3016 int ret = NET_RX_DROP;
3019 if (!netdev_tstamp_prequeue)
3020 net_timestamp_check(skb);
3022 trace_netif_receive_skb(skb);
3024 /* if we've gotten here through NAPI, check netpoll */
3025 if (netpoll_receive_skb(skb))
3029 skb->skb_iif = skb->dev->ifindex;
3032 * bonding note: skbs received on inactive slaves should only
3033 * be delivered to pkt handlers that are exact matches. Also
3034 * the deliver_no_wcard flag will be set. If packet handlers
3035 * are sensitive to duplicate packets these skbs will need to
3036 * be dropped at the handler.
3038 null_or_orig = NULL;
3039 orig_dev = skb->dev;
3040 master = ACCESS_ONCE(orig_dev->master);
3041 if (skb->deliver_no_wcard)
3042 null_or_orig = orig_dev;
3044 if (skb_bond_should_drop(skb, master)) {
3045 skb->deliver_no_wcard = 1;
3046 null_or_orig = orig_dev; /* deliver only exact match */
3051 __this_cpu_inc(softnet_data.processed);
3052 skb_reset_network_header(skb);
3053 skb_reset_transport_header(skb);
3054 skb->mac_len = skb->network_header - skb->mac_header;
3060 #ifdef CONFIG_NET_CLS_ACT
3061 if (skb->tc_verd & TC_NCLS) {
3062 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3067 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3068 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
3069 ptype->dev == orig_dev) {
3071 ret = deliver_skb(skb, pt_prev, orig_dev);
3076 #ifdef CONFIG_NET_CLS_ACT
3077 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3083 /* Handle special case of bridge or macvlan */
3084 rx_handler = rcu_dereference(skb->dev->rx_handler);
3087 ret = deliver_skb(skb, pt_prev, orig_dev);
3090 skb = rx_handler(skb);
3095 if (vlan_tx_tag_present(skb)) {
3097 ret = deliver_skb(skb, pt_prev, orig_dev);
3100 if (vlan_hwaccel_do_receive(&skb)) {
3101 ret = __netif_receive_skb(skb);
3103 } else if (unlikely(!skb))
3108 * Make sure frames received on VLAN interfaces stacked on
3109 * bonding interfaces still make their way to any base bonding
3110 * device that may have registered for a specific ptype. The
3111 * handler may have to adjust skb->dev and orig_dev.
3113 orig_or_bond = orig_dev;
3114 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
3115 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
3116 orig_or_bond = vlan_dev_real_dev(skb->dev);
3119 type = skb->protocol;
3120 list_for_each_entry_rcu(ptype,
3121 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3122 if (ptype->type == type && (ptype->dev == null_or_orig ||
3123 ptype->dev == skb->dev || ptype->dev == orig_dev ||
3124 ptype->dev == orig_or_bond)) {
3126 ret = deliver_skb(skb, pt_prev, orig_dev);
3132 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3134 atomic_long_inc(&skb->dev->rx_dropped);
3136 /* Jamal, now you will not able to escape explaining
3137 * me how you were going to use this. :-)
3148 * netif_receive_skb - process receive buffer from network
3149 * @skb: buffer to process
3151 * netif_receive_skb() is the main receive data processing function.
3152 * It always succeeds. The buffer may be dropped during processing
3153 * for congestion control or by the protocol layers.
3155 * This function may only be called from softirq context and interrupts
3156 * should be enabled.
3158 * Return values (usually ignored):
3159 * NET_RX_SUCCESS: no congestion
3160 * NET_RX_DROP: packet was dropped
3162 int netif_receive_skb(struct sk_buff *skb)
3164 if (netdev_tstamp_prequeue)
3165 net_timestamp_check(skb);
3167 if (skb_defer_rx_timestamp(skb))
3168 return NET_RX_SUCCESS;
3172 struct rps_dev_flow voidflow, *rflow = &voidflow;
3177 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3180 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3184 ret = __netif_receive_skb(skb);
3190 return __netif_receive_skb(skb);
3193 EXPORT_SYMBOL(netif_receive_skb);
3195 /* Network device is going away, flush any packets still pending
3196 * Called with irqs disabled.
3198 static void flush_backlog(void *arg)
3200 struct net_device *dev = arg;
3201 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3202 struct sk_buff *skb, *tmp;
3205 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3206 if (skb->dev == dev) {
3207 __skb_unlink(skb, &sd->input_pkt_queue);
3209 input_queue_head_incr(sd);
3214 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3215 if (skb->dev == dev) {
3216 __skb_unlink(skb, &sd->process_queue);
3218 input_queue_head_incr(sd);
3223 static int napi_gro_complete(struct sk_buff *skb)
3225 struct packet_type *ptype;
3226 __be16 type = skb->protocol;
3227 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3230 if (NAPI_GRO_CB(skb)->count == 1) {
3231 skb_shinfo(skb)->gso_size = 0;
3236 list_for_each_entry_rcu(ptype, head, list) {
3237 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3240 err = ptype->gro_complete(skb);
3246 WARN_ON(&ptype->list == head);
3248 return NET_RX_SUCCESS;
3252 return netif_receive_skb(skb);
3255 inline void napi_gro_flush(struct napi_struct *napi)
3257 struct sk_buff *skb, *next;
3259 for (skb = napi->gro_list; skb; skb = next) {
3262 napi_gro_complete(skb);
3265 napi->gro_count = 0;
3266 napi->gro_list = NULL;
3268 EXPORT_SYMBOL(napi_gro_flush);
3270 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3272 struct sk_buff **pp = NULL;
3273 struct packet_type *ptype;
3274 __be16 type = skb->protocol;
3275 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3278 enum gro_result ret;
3280 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3283 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3287 list_for_each_entry_rcu(ptype, head, list) {
3288 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3291 skb_set_network_header(skb, skb_gro_offset(skb));
3292 mac_len = skb->network_header - skb->mac_header;
3293 skb->mac_len = mac_len;
3294 NAPI_GRO_CB(skb)->same_flow = 0;
3295 NAPI_GRO_CB(skb)->flush = 0;
3296 NAPI_GRO_CB(skb)->free = 0;
3298 pp = ptype->gro_receive(&napi->gro_list, skb);
3303 if (&ptype->list == head)
3306 same_flow = NAPI_GRO_CB(skb)->same_flow;
3307 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3310 struct sk_buff *nskb = *pp;
3314 napi_gro_complete(nskb);
3321 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3325 NAPI_GRO_CB(skb)->count = 1;
3326 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3327 skb->next = napi->gro_list;
3328 napi->gro_list = skb;
3332 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3333 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3335 BUG_ON(skb->end - skb->tail < grow);
3337 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3340 skb->data_len -= grow;
3342 skb_shinfo(skb)->frags[0].page_offset += grow;
3343 skb_shinfo(skb)->frags[0].size -= grow;
3345 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3346 put_page(skb_shinfo(skb)->frags[0].page);
3347 memmove(skb_shinfo(skb)->frags,
3348 skb_shinfo(skb)->frags + 1,
3349 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3360 EXPORT_SYMBOL(dev_gro_receive);
3362 static inline gro_result_t
3363 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3367 for (p = napi->gro_list; p; p = p->next) {
3368 unsigned long diffs;
3370 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3371 diffs |= p->vlan_tci ^ skb->vlan_tci;
3372 diffs |= compare_ether_header(skb_mac_header(p),
3373 skb_gro_mac_header(skb));
3374 NAPI_GRO_CB(p)->same_flow = !diffs;
3375 NAPI_GRO_CB(p)->flush = 0;
3378 return dev_gro_receive(napi, skb);
3381 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3385 if (netif_receive_skb(skb))
3390 case GRO_MERGED_FREE:
3401 EXPORT_SYMBOL(napi_skb_finish);
3403 void skb_gro_reset_offset(struct sk_buff *skb)
3405 NAPI_GRO_CB(skb)->data_offset = 0;
3406 NAPI_GRO_CB(skb)->frag0 = NULL;
3407 NAPI_GRO_CB(skb)->frag0_len = 0;
3409 if (skb->mac_header == skb->tail &&
3410 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3411 NAPI_GRO_CB(skb)->frag0 =
3412 page_address(skb_shinfo(skb)->frags[0].page) +
3413 skb_shinfo(skb)->frags[0].page_offset;
3414 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3417 EXPORT_SYMBOL(skb_gro_reset_offset);
3419 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3421 skb_gro_reset_offset(skb);
3423 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3425 EXPORT_SYMBOL(napi_gro_receive);
3427 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3429 __skb_pull(skb, skb_headlen(skb));
3430 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3436 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3438 struct sk_buff *skb = napi->skb;
3441 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3447 EXPORT_SYMBOL(napi_get_frags);
3449 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3455 skb->protocol = eth_type_trans(skb, skb->dev);
3457 if (ret == GRO_HELD)
3458 skb_gro_pull(skb, -ETH_HLEN);
3459 else if (netif_receive_skb(skb))
3464 case GRO_MERGED_FREE:
3465 napi_reuse_skb(napi, skb);
3474 EXPORT_SYMBOL(napi_frags_finish);
3476 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3478 struct sk_buff *skb = napi->skb;
3485 skb_reset_mac_header(skb);
3486 skb_gro_reset_offset(skb);
3488 off = skb_gro_offset(skb);
3489 hlen = off + sizeof(*eth);
3490 eth = skb_gro_header_fast(skb, off);
3491 if (skb_gro_header_hard(skb, hlen)) {
3492 eth = skb_gro_header_slow(skb, hlen, off);
3493 if (unlikely(!eth)) {
3494 napi_reuse_skb(napi, skb);
3500 skb_gro_pull(skb, sizeof(*eth));
3503 * This works because the only protocols we care about don't require
3504 * special handling. We'll fix it up properly at the end.
3506 skb->protocol = eth->h_proto;
3511 EXPORT_SYMBOL(napi_frags_skb);
3513 gro_result_t napi_gro_frags(struct napi_struct *napi)
3515 struct sk_buff *skb = napi_frags_skb(napi);
3520 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3522 EXPORT_SYMBOL(napi_gro_frags);
3525 * net_rps_action sends any pending IPI's for rps.
3526 * Note: called with local irq disabled, but exits with local irq enabled.
3528 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3531 struct softnet_data *remsd = sd->rps_ipi_list;
3534 sd->rps_ipi_list = NULL;
3538 /* Send pending IPI's to kick RPS processing on remote cpus. */
3540 struct softnet_data *next = remsd->rps_ipi_next;
3542 if (cpu_online(remsd->cpu))
3543 __smp_call_function_single(remsd->cpu,
3552 static int process_backlog(struct napi_struct *napi, int quota)
3555 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3558 /* Check if we have pending ipi, its better to send them now,
3559 * not waiting net_rx_action() end.
3561 if (sd->rps_ipi_list) {
3562 local_irq_disable();
3563 net_rps_action_and_irq_enable(sd);
3566 napi->weight = weight_p;
3567 local_irq_disable();
3568 while (work < quota) {
3569 struct sk_buff *skb;
3572 while ((skb = __skb_dequeue(&sd->process_queue))) {
3574 __netif_receive_skb(skb);
3575 local_irq_disable();
3576 input_queue_head_incr(sd);
3577 if (++work >= quota) {
3584 qlen = skb_queue_len(&sd->input_pkt_queue);
3586 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3587 &sd->process_queue);
3589 if (qlen < quota - work) {
3591 * Inline a custom version of __napi_complete().
3592 * only current cpu owns and manipulates this napi,
3593 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3594 * we can use a plain write instead of clear_bit(),
3595 * and we dont need an smp_mb() memory barrier.
3597 list_del(&napi->poll_list);
3600 quota = work + qlen;
3610 * __napi_schedule - schedule for receive
3611 * @n: entry to schedule
3613 * The entry's receive function will be scheduled to run
3615 void __napi_schedule(struct napi_struct *n)
3617 unsigned long flags;
3619 local_irq_save(flags);
3620 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3621 local_irq_restore(flags);
3623 EXPORT_SYMBOL(__napi_schedule);
3625 void __napi_complete(struct napi_struct *n)
3627 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3628 BUG_ON(n->gro_list);
3630 list_del(&n->poll_list);
3631 smp_mb__before_clear_bit();
3632 clear_bit(NAPI_STATE_SCHED, &n->state);
3634 EXPORT_SYMBOL(__napi_complete);
3636 void napi_complete(struct napi_struct *n)
3638 unsigned long flags;
3641 * don't let napi dequeue from the cpu poll list
3642 * just in case its running on a different cpu
3644 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3648 local_irq_save(flags);
3650 local_irq_restore(flags);
3652 EXPORT_SYMBOL(napi_complete);
3654 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3655 int (*poll)(struct napi_struct *, int), int weight)
3657 INIT_LIST_HEAD(&napi->poll_list);
3658 napi->gro_count = 0;
3659 napi->gro_list = NULL;
3662 napi->weight = weight;
3663 list_add(&napi->dev_list, &dev->napi_list);
3665 #ifdef CONFIG_NETPOLL
3666 spin_lock_init(&napi->poll_lock);
3667 napi->poll_owner = -1;
3669 set_bit(NAPI_STATE_SCHED, &napi->state);
3671 EXPORT_SYMBOL(netif_napi_add);
3673 void netif_napi_del(struct napi_struct *napi)
3675 struct sk_buff *skb, *next;
3677 list_del_init(&napi->dev_list);
3678 napi_free_frags(napi);
3680 for (skb = napi->gro_list; skb; skb = next) {
3686 napi->gro_list = NULL;
3687 napi->gro_count = 0;
3689 EXPORT_SYMBOL(netif_napi_del);
3691 static void net_rx_action(struct softirq_action *h)
3693 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3694 unsigned long time_limit = jiffies + 2;
3695 int budget = netdev_budget;
3698 local_irq_disable();
3700 while (!list_empty(&sd->poll_list)) {
3701 struct napi_struct *n;
3704 /* If softirq window is exhuasted then punt.
3705 * Allow this to run for 2 jiffies since which will allow
3706 * an average latency of 1.5/HZ.
3708 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3713 /* Even though interrupts have been re-enabled, this
3714 * access is safe because interrupts can only add new
3715 * entries to the tail of this list, and only ->poll()
3716 * calls can remove this head entry from the list.
3718 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3720 have = netpoll_poll_lock(n);
3724 /* This NAPI_STATE_SCHED test is for avoiding a race
3725 * with netpoll's poll_napi(). Only the entity which
3726 * obtains the lock and sees NAPI_STATE_SCHED set will
3727 * actually make the ->poll() call. Therefore we avoid
3728 * accidently calling ->poll() when NAPI is not scheduled.
3731 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3732 work = n->poll(n, weight);
3736 WARN_ON_ONCE(work > weight);
3740 local_irq_disable();
3742 /* Drivers must not modify the NAPI state if they
3743 * consume the entire weight. In such cases this code
3744 * still "owns" the NAPI instance and therefore can
3745 * move the instance around on the list at-will.
3747 if (unlikely(work == weight)) {
3748 if (unlikely(napi_disable_pending(n))) {
3751 local_irq_disable();
3753 list_move_tail(&n->poll_list, &sd->poll_list);
3756 netpoll_poll_unlock(have);
3759 net_rps_action_and_irq_enable(sd);
3761 #ifdef CONFIG_NET_DMA
3763 * There may not be any more sk_buffs coming right now, so push
3764 * any pending DMA copies to hardware
3766 dma_issue_pending_all();
3773 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3777 static gifconf_func_t *gifconf_list[NPROTO];
3780 * register_gifconf - register a SIOCGIF handler
3781 * @family: Address family
3782 * @gifconf: Function handler
3784 * Register protocol dependent address dumping routines. The handler
3785 * that is passed must not be freed or reused until it has been replaced
3786 * by another handler.
3788 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3790 if (family >= NPROTO)
3792 gifconf_list[family] = gifconf;
3795 EXPORT_SYMBOL(register_gifconf);
3799 * Map an interface index to its name (SIOCGIFNAME)
3803 * We need this ioctl for efficient implementation of the
3804 * if_indextoname() function required by the IPv6 API. Without
3805 * it, we would have to search all the interfaces to find a
3809 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3811 struct net_device *dev;
3815 * Fetch the caller's info block.
3818 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3822 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3828 strcpy(ifr.ifr_name, dev->name);
3831 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3837 * Perform a SIOCGIFCONF call. This structure will change
3838 * size eventually, and there is nothing I can do about it.
3839 * Thus we will need a 'compatibility mode'.
3842 static int dev_ifconf(struct net *net, char __user *arg)
3845 struct net_device *dev;
3852 * Fetch the caller's info block.
3855 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3862 * Loop over the interfaces, and write an info block for each.
3866 for_each_netdev(net, dev) {
3867 for (i = 0; i < NPROTO; i++) {
3868 if (gifconf_list[i]) {
3871 done = gifconf_list[i](dev, NULL, 0);
3873 done = gifconf_list[i](dev, pos + total,
3883 * All done. Write the updated control block back to the caller.
3885 ifc.ifc_len = total;
3888 * Both BSD and Solaris return 0 here, so we do too.
3890 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3893 #ifdef CONFIG_PROC_FS
3895 * This is invoked by the /proc filesystem handler to display a device
3898 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3901 struct net *net = seq_file_net(seq);
3903 struct net_device *dev;
3907 return SEQ_START_TOKEN;
3910 for_each_netdev_rcu(net, dev)
3917 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3919 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3920 first_net_device(seq_file_net(seq)) :
3921 next_net_device((struct net_device *)v);
3924 return rcu_dereference(dev);
3927 void dev_seq_stop(struct seq_file *seq, void *v)
3933 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3935 struct rtnl_link_stats64 temp;
3936 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3938 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3939 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3940 dev->name, stats->rx_bytes, stats->rx_packets,
3942 stats->rx_dropped + stats->rx_missed_errors,
3943 stats->rx_fifo_errors,
3944 stats->rx_length_errors + stats->rx_over_errors +
3945 stats->rx_crc_errors + stats->rx_frame_errors,
3946 stats->rx_compressed, stats->multicast,
3947 stats->tx_bytes, stats->tx_packets,
3948 stats->tx_errors, stats->tx_dropped,
3949 stats->tx_fifo_errors, stats->collisions,
3950 stats->tx_carrier_errors +
3951 stats->tx_aborted_errors +
3952 stats->tx_window_errors +
3953 stats->tx_heartbeat_errors,
3954 stats->tx_compressed);
3958 * Called from the PROCfs module. This now uses the new arbitrary sized
3959 * /proc/net interface to create /proc/net/dev
3961 static int dev_seq_show(struct seq_file *seq, void *v)
3963 if (v == SEQ_START_TOKEN)
3964 seq_puts(seq, "Inter-| Receive "
3966 " face |bytes packets errs drop fifo frame "
3967 "compressed multicast|bytes packets errs "
3968 "drop fifo colls carrier compressed\n");
3970 dev_seq_printf_stats(seq, v);
3974 static struct softnet_data *softnet_get_online(loff_t *pos)
3976 struct softnet_data *sd = NULL;
3978 while (*pos < nr_cpu_ids)
3979 if (cpu_online(*pos)) {
3980 sd = &per_cpu(softnet_data, *pos);
3987 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3989 return softnet_get_online(pos);
3992 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3995 return softnet_get_online(pos);
3998 static void softnet_seq_stop(struct seq_file *seq, void *v)
4002 static int softnet_seq_show(struct seq_file *seq, void *v)
4004 struct softnet_data *sd = v;
4006 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4007 sd->processed, sd->dropped, sd->time_squeeze, 0,
4008 0, 0, 0, 0, /* was fastroute */
4009 sd->cpu_collision, sd->received_rps);
4013 static const struct seq_operations dev_seq_ops = {
4014 .start = dev_seq_start,
4015 .next = dev_seq_next,
4016 .stop = dev_seq_stop,
4017 .show = dev_seq_show,
4020 static int dev_seq_open(struct inode *inode, struct file *file)
4022 return seq_open_net(inode, file, &dev_seq_ops,
4023 sizeof(struct seq_net_private));
4026 static const struct file_operations dev_seq_fops = {
4027 .owner = THIS_MODULE,
4028 .open = dev_seq_open,
4030 .llseek = seq_lseek,
4031 .release = seq_release_net,
4034 static const struct seq_operations softnet_seq_ops = {
4035 .start = softnet_seq_start,
4036 .next = softnet_seq_next,
4037 .stop = softnet_seq_stop,
4038 .show = softnet_seq_show,
4041 static int softnet_seq_open(struct inode *inode, struct file *file)
4043 return seq_open(file, &softnet_seq_ops);
4046 static const struct file_operations softnet_seq_fops = {
4047 .owner = THIS_MODULE,
4048 .open = softnet_seq_open,
4050 .llseek = seq_lseek,
4051 .release = seq_release,
4054 static void *ptype_get_idx(loff_t pos)
4056 struct packet_type *pt = NULL;
4060 list_for_each_entry_rcu(pt, &ptype_all, list) {
4066 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4067 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4076 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4080 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4083 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4085 struct packet_type *pt;
4086 struct list_head *nxt;
4090 if (v == SEQ_START_TOKEN)
4091 return ptype_get_idx(0);
4094 nxt = pt->list.next;
4095 if (pt->type == htons(ETH_P_ALL)) {
4096 if (nxt != &ptype_all)
4099 nxt = ptype_base[0].next;
4101 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4103 while (nxt == &ptype_base[hash]) {
4104 if (++hash >= PTYPE_HASH_SIZE)
4106 nxt = ptype_base[hash].next;
4109 return list_entry(nxt, struct packet_type, list);
4112 static void ptype_seq_stop(struct seq_file *seq, void *v)
4118 static int ptype_seq_show(struct seq_file *seq, void *v)
4120 struct packet_type *pt = v;
4122 if (v == SEQ_START_TOKEN)
4123 seq_puts(seq, "Type Device Function\n");
4124 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4125 if (pt->type == htons(ETH_P_ALL))
4126 seq_puts(seq, "ALL ");
4128 seq_printf(seq, "%04x", ntohs(pt->type));
4130 seq_printf(seq, " %-8s %pF\n",
4131 pt->dev ? pt->dev->name : "", pt->func);
4137 static const struct seq_operations ptype_seq_ops = {
4138 .start = ptype_seq_start,
4139 .next = ptype_seq_next,
4140 .stop = ptype_seq_stop,
4141 .show = ptype_seq_show,
4144 static int ptype_seq_open(struct inode *inode, struct file *file)
4146 return seq_open_net(inode, file, &ptype_seq_ops,
4147 sizeof(struct seq_net_private));
4150 static const struct file_operations ptype_seq_fops = {
4151 .owner = THIS_MODULE,
4152 .open = ptype_seq_open,
4154 .llseek = seq_lseek,
4155 .release = seq_release_net,
4159 static int __net_init dev_proc_net_init(struct net *net)
4163 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4165 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4167 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4170 if (wext_proc_init(net))
4176 proc_net_remove(net, "ptype");
4178 proc_net_remove(net, "softnet_stat");
4180 proc_net_remove(net, "dev");
4184 static void __net_exit dev_proc_net_exit(struct net *net)
4186 wext_proc_exit(net);
4188 proc_net_remove(net, "ptype");
4189 proc_net_remove(net, "softnet_stat");
4190 proc_net_remove(net, "dev");
4193 static struct pernet_operations __net_initdata dev_proc_ops = {
4194 .init = dev_proc_net_init,
4195 .exit = dev_proc_net_exit,
4198 static int __init dev_proc_init(void)
4200 return register_pernet_subsys(&dev_proc_ops);
4203 #define dev_proc_init() 0
4204 #endif /* CONFIG_PROC_FS */
4208 * netdev_set_master - set up master/slave pair
4209 * @slave: slave device
4210 * @master: new master device
4212 * Changes the master device of the slave. Pass %NULL to break the
4213 * bonding. The caller must hold the RTNL semaphore. On a failure
4214 * a negative errno code is returned. On success the reference counts
4215 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4216 * function returns zero.
4218 int netdev_set_master(struct net_device *slave, struct net_device *master)
4220 struct net_device *old = slave->master;
4230 slave->master = master;
4237 slave->flags |= IFF_SLAVE;
4239 slave->flags &= ~IFF_SLAVE;
4241 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4244 EXPORT_SYMBOL(netdev_set_master);
4246 static void dev_change_rx_flags(struct net_device *dev, int flags)
4248 const struct net_device_ops *ops = dev->netdev_ops;
4250 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4251 ops->ndo_change_rx_flags(dev, flags);
4254 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4256 unsigned short old_flags = dev->flags;
4262 dev->flags |= IFF_PROMISC;
4263 dev->promiscuity += inc;
4264 if (dev->promiscuity == 0) {
4267 * If inc causes overflow, untouch promisc and return error.
4270 dev->flags &= ~IFF_PROMISC;
4272 dev->promiscuity -= inc;
4273 printk(KERN_WARNING "%s: promiscuity touches roof, "
4274 "set promiscuity failed, promiscuity feature "
4275 "of device might be broken.\n", dev->name);
4279 if (dev->flags != old_flags) {
4280 printk(KERN_INFO "device %s %s promiscuous mode\n",
4281 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4283 if (audit_enabled) {
4284 current_uid_gid(&uid, &gid);
4285 audit_log(current->audit_context, GFP_ATOMIC,
4286 AUDIT_ANOM_PROMISCUOUS,
4287 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4288 dev->name, (dev->flags & IFF_PROMISC),
4289 (old_flags & IFF_PROMISC),
4290 audit_get_loginuid(current),
4292 audit_get_sessionid(current));
4295 dev_change_rx_flags(dev, IFF_PROMISC);
4301 * dev_set_promiscuity - update promiscuity count on a device
4305 * Add or remove promiscuity from a device. While the count in the device
4306 * remains above zero the interface remains promiscuous. Once it hits zero
4307 * the device reverts back to normal filtering operation. A negative inc
4308 * value is used to drop promiscuity on the device.
4309 * Return 0 if successful or a negative errno code on error.
4311 int dev_set_promiscuity(struct net_device *dev, int inc)
4313 unsigned short old_flags = dev->flags;
4316 err = __dev_set_promiscuity(dev, inc);
4319 if (dev->flags != old_flags)
4320 dev_set_rx_mode(dev);
4323 EXPORT_SYMBOL(dev_set_promiscuity);
4326 * dev_set_allmulti - update allmulti count on a device
4330 * Add or remove reception of all multicast frames to a device. While the
4331 * count in the device remains above zero the interface remains listening
4332 * to all interfaces. Once it hits zero the device reverts back to normal
4333 * filtering operation. A negative @inc value is used to drop the counter
4334 * when releasing a resource needing all multicasts.
4335 * Return 0 if successful or a negative errno code on error.
4338 int dev_set_allmulti(struct net_device *dev, int inc)
4340 unsigned short old_flags = dev->flags;
4344 dev->flags |= IFF_ALLMULTI;
4345 dev->allmulti += inc;
4346 if (dev->allmulti == 0) {
4349 * If inc causes overflow, untouch allmulti and return error.
4352 dev->flags &= ~IFF_ALLMULTI;
4354 dev->allmulti -= inc;
4355 printk(KERN_WARNING "%s: allmulti touches roof, "
4356 "set allmulti failed, allmulti feature of "
4357 "device might be broken.\n", dev->name);
4361 if (dev->flags ^ old_flags) {
4362 dev_change_rx_flags(dev, IFF_ALLMULTI);
4363 dev_set_rx_mode(dev);
4367 EXPORT_SYMBOL(dev_set_allmulti);
4370 * Upload unicast and multicast address lists to device and
4371 * configure RX filtering. When the device doesn't support unicast
4372 * filtering it is put in promiscuous mode while unicast addresses
4375 void __dev_set_rx_mode(struct net_device *dev)
4377 const struct net_device_ops *ops = dev->netdev_ops;
4379 /* dev_open will call this function so the list will stay sane. */
4380 if (!(dev->flags&IFF_UP))
4383 if (!netif_device_present(dev))
4386 if (ops->ndo_set_rx_mode)
4387 ops->ndo_set_rx_mode(dev);
4389 /* Unicast addresses changes may only happen under the rtnl,
4390 * therefore calling __dev_set_promiscuity here is safe.
4392 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4393 __dev_set_promiscuity(dev, 1);
4394 dev->uc_promisc = 1;
4395 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4396 __dev_set_promiscuity(dev, -1);
4397 dev->uc_promisc = 0;
4400 if (ops->ndo_set_multicast_list)
4401 ops->ndo_set_multicast_list(dev);
4405 void dev_set_rx_mode(struct net_device *dev)
4407 netif_addr_lock_bh(dev);
4408 __dev_set_rx_mode(dev);
4409 netif_addr_unlock_bh(dev);
4413 * dev_get_flags - get flags reported to userspace
4416 * Get the combination of flag bits exported through APIs to userspace.
4418 unsigned dev_get_flags(const struct net_device *dev)
4422 flags = (dev->flags & ~(IFF_PROMISC |
4427 (dev->gflags & (IFF_PROMISC |
4430 if (netif_running(dev)) {
4431 if (netif_oper_up(dev))
4432 flags |= IFF_RUNNING;
4433 if (netif_carrier_ok(dev))
4434 flags |= IFF_LOWER_UP;
4435 if (netif_dormant(dev))
4436 flags |= IFF_DORMANT;
4441 EXPORT_SYMBOL(dev_get_flags);
4443 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4445 int old_flags = dev->flags;
4451 * Set the flags on our device.
4454 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4455 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4457 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4461 * Load in the correct multicast list now the flags have changed.
4464 if ((old_flags ^ flags) & IFF_MULTICAST)
4465 dev_change_rx_flags(dev, IFF_MULTICAST);
4467 dev_set_rx_mode(dev);
4470 * Have we downed the interface. We handle IFF_UP ourselves
4471 * according to user attempts to set it, rather than blindly
4476 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4477 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4480 dev_set_rx_mode(dev);
4483 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4484 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4486 dev->gflags ^= IFF_PROMISC;
4487 dev_set_promiscuity(dev, inc);
4490 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4491 is important. Some (broken) drivers set IFF_PROMISC, when
4492 IFF_ALLMULTI is requested not asking us and not reporting.
4494 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4495 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4497 dev->gflags ^= IFF_ALLMULTI;
4498 dev_set_allmulti(dev, inc);
4504 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4506 unsigned int changes = dev->flags ^ old_flags;
4508 if (changes & IFF_UP) {
4509 if (dev->flags & IFF_UP)
4510 call_netdevice_notifiers(NETDEV_UP, dev);
4512 call_netdevice_notifiers(NETDEV_DOWN, dev);
4515 if (dev->flags & IFF_UP &&
4516 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4517 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4521 * dev_change_flags - change device settings
4523 * @flags: device state flags
4525 * Change settings on device based state flags. The flags are
4526 * in the userspace exported format.
4528 int dev_change_flags(struct net_device *dev, unsigned flags)
4531 int old_flags = dev->flags;
4533 ret = __dev_change_flags(dev, flags);
4537 changes = old_flags ^ dev->flags;
4539 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4541 __dev_notify_flags(dev, old_flags);
4544 EXPORT_SYMBOL(dev_change_flags);
4547 * dev_set_mtu - Change maximum transfer unit
4549 * @new_mtu: new transfer unit
4551 * Change the maximum transfer size of the network device.
4553 int dev_set_mtu(struct net_device *dev, int new_mtu)
4555 const struct net_device_ops *ops = dev->netdev_ops;
4558 if (new_mtu == dev->mtu)
4561 /* MTU must be positive. */
4565 if (!netif_device_present(dev))
4569 if (ops->ndo_change_mtu)
4570 err = ops->ndo_change_mtu(dev, new_mtu);
4574 if (!err && dev->flags & IFF_UP)
4575 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4578 EXPORT_SYMBOL(dev_set_mtu);
4581 * dev_set_mac_address - Change Media Access Control Address
4585 * Change the hardware (MAC) address of the device
4587 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4589 const struct net_device_ops *ops = dev->netdev_ops;
4592 if (!ops->ndo_set_mac_address)
4594 if (sa->sa_family != dev->type)
4596 if (!netif_device_present(dev))
4598 err = ops->ndo_set_mac_address(dev, sa);
4600 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4603 EXPORT_SYMBOL(dev_set_mac_address);
4606 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4608 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4611 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4617 case SIOCGIFFLAGS: /* Get interface flags */
4618 ifr->ifr_flags = (short) dev_get_flags(dev);
4621 case SIOCGIFMETRIC: /* Get the metric on the interface
4622 (currently unused) */
4623 ifr->ifr_metric = 0;
4626 case SIOCGIFMTU: /* Get the MTU of a device */
4627 ifr->ifr_mtu = dev->mtu;
4632 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4634 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4635 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4636 ifr->ifr_hwaddr.sa_family = dev->type;
4644 ifr->ifr_map.mem_start = dev->mem_start;
4645 ifr->ifr_map.mem_end = dev->mem_end;
4646 ifr->ifr_map.base_addr = dev->base_addr;
4647 ifr->ifr_map.irq = dev->irq;
4648 ifr->ifr_map.dma = dev->dma;
4649 ifr->ifr_map.port = dev->if_port;
4653 ifr->ifr_ifindex = dev->ifindex;
4657 ifr->ifr_qlen = dev->tx_queue_len;
4661 /* dev_ioctl() should ensure this case
4673 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4675 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4678 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4679 const struct net_device_ops *ops;
4684 ops = dev->netdev_ops;
4687 case SIOCSIFFLAGS: /* Set interface flags */
4688 return dev_change_flags(dev, ifr->ifr_flags);
4690 case SIOCSIFMETRIC: /* Set the metric on the interface
4691 (currently unused) */
4694 case SIOCSIFMTU: /* Set the MTU of a device */
4695 return dev_set_mtu(dev, ifr->ifr_mtu);
4698 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4700 case SIOCSIFHWBROADCAST:
4701 if (ifr->ifr_hwaddr.sa_family != dev->type)
4703 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4704 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4705 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4709 if (ops->ndo_set_config) {
4710 if (!netif_device_present(dev))
4712 return ops->ndo_set_config(dev, &ifr->ifr_map);
4717 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4718 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4720 if (!netif_device_present(dev))
4722 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4725 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4726 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4728 if (!netif_device_present(dev))
4730 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4733 if (ifr->ifr_qlen < 0)
4735 dev->tx_queue_len = ifr->ifr_qlen;
4739 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4740 return dev_change_name(dev, ifr->ifr_newname);
4743 * Unknown or private ioctl
4746 if ((cmd >= SIOCDEVPRIVATE &&
4747 cmd <= SIOCDEVPRIVATE + 15) ||
4748 cmd == SIOCBONDENSLAVE ||
4749 cmd == SIOCBONDRELEASE ||
4750 cmd == SIOCBONDSETHWADDR ||
4751 cmd == SIOCBONDSLAVEINFOQUERY ||
4752 cmd == SIOCBONDINFOQUERY ||
4753 cmd == SIOCBONDCHANGEACTIVE ||
4754 cmd == SIOCGMIIPHY ||
4755 cmd == SIOCGMIIREG ||
4756 cmd == SIOCSMIIREG ||
4757 cmd == SIOCBRADDIF ||
4758 cmd == SIOCBRDELIF ||
4759 cmd == SIOCSHWTSTAMP ||
4760 cmd == SIOCWANDEV) {
4762 if (ops->ndo_do_ioctl) {
4763 if (netif_device_present(dev))
4764 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4776 * This function handles all "interface"-type I/O control requests. The actual
4777 * 'doing' part of this is dev_ifsioc above.
4781 * dev_ioctl - network device ioctl
4782 * @net: the applicable net namespace
4783 * @cmd: command to issue
4784 * @arg: pointer to a struct ifreq in user space
4786 * Issue ioctl functions to devices. This is normally called by the
4787 * user space syscall interfaces but can sometimes be useful for
4788 * other purposes. The return value is the return from the syscall if
4789 * positive or a negative errno code on error.
4792 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4798 /* One special case: SIOCGIFCONF takes ifconf argument
4799 and requires shared lock, because it sleeps writing
4803 if (cmd == SIOCGIFCONF) {
4805 ret = dev_ifconf(net, (char __user *) arg);
4809 if (cmd == SIOCGIFNAME)
4810 return dev_ifname(net, (struct ifreq __user *)arg);
4812 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4815 ifr.ifr_name[IFNAMSIZ-1] = 0;
4817 colon = strchr(ifr.ifr_name, ':');
4822 * See which interface the caller is talking about.
4827 * These ioctl calls:
4828 * - can be done by all.
4829 * - atomic and do not require locking.
4840 dev_load(net, ifr.ifr_name);
4842 ret = dev_ifsioc_locked(net, &ifr, cmd);
4847 if (copy_to_user(arg, &ifr,
4848 sizeof(struct ifreq)))
4854 dev_load(net, ifr.ifr_name);
4856 ret = dev_ethtool(net, &ifr);
4861 if (copy_to_user(arg, &ifr,
4862 sizeof(struct ifreq)))
4868 * These ioctl calls:
4869 * - require superuser power.
4870 * - require strict serialization.
4876 if (!capable(CAP_NET_ADMIN))
4878 dev_load(net, ifr.ifr_name);
4880 ret = dev_ifsioc(net, &ifr, cmd);
4885 if (copy_to_user(arg, &ifr,
4886 sizeof(struct ifreq)))
4892 * These ioctl calls:
4893 * - require superuser power.
4894 * - require strict serialization.
4895 * - do not return a value
4905 case SIOCSIFHWBROADCAST:
4908 case SIOCBONDENSLAVE:
4909 case SIOCBONDRELEASE:
4910 case SIOCBONDSETHWADDR:
4911 case SIOCBONDCHANGEACTIVE:
4915 if (!capable(CAP_NET_ADMIN))
4918 case SIOCBONDSLAVEINFOQUERY:
4919 case SIOCBONDINFOQUERY:
4920 dev_load(net, ifr.ifr_name);
4922 ret = dev_ifsioc(net, &ifr, cmd);
4927 /* Get the per device memory space. We can add this but
4928 * currently do not support it */
4930 /* Set the per device memory buffer space.
4931 * Not applicable in our case */
4936 * Unknown or private ioctl.
4939 if (cmd == SIOCWANDEV ||
4940 (cmd >= SIOCDEVPRIVATE &&
4941 cmd <= SIOCDEVPRIVATE + 15)) {
4942 dev_load(net, ifr.ifr_name);
4944 ret = dev_ifsioc(net, &ifr, cmd);
4946 if (!ret && copy_to_user(arg, &ifr,
4947 sizeof(struct ifreq)))
4951 /* Take care of Wireless Extensions */
4952 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4953 return wext_handle_ioctl(net, &ifr, cmd, arg);
4960 * dev_new_index - allocate an ifindex
4961 * @net: the applicable net namespace
4963 * Returns a suitable unique value for a new device interface
4964 * number. The caller must hold the rtnl semaphore or the
4965 * dev_base_lock to be sure it remains unique.
4967 static int dev_new_index(struct net *net)
4973 if (!__dev_get_by_index(net, ifindex))
4978 /* Delayed registration/unregisteration */
4979 static LIST_HEAD(net_todo_list);
4981 static void net_set_todo(struct net_device *dev)
4983 list_add_tail(&dev->todo_list, &net_todo_list);
4986 static void rollback_registered_many(struct list_head *head)
4988 struct net_device *dev, *tmp;
4990 BUG_ON(dev_boot_phase);
4993 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4994 /* Some devices call without registering
4995 * for initialization unwind. Remove those
4996 * devices and proceed with the remaining.
4998 if (dev->reg_state == NETREG_UNINITIALIZED) {
4999 pr_debug("unregister_netdevice: device %s/%p never "
5000 "was registered\n", dev->name, dev);
5003 list_del(&dev->unreg_list);
5007 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5010 /* If device is running, close it first. */
5011 dev_close_many(head);
5013 list_for_each_entry(dev, head, unreg_list) {
5014 /* And unlink it from device chain. */
5015 unlist_netdevice(dev);
5017 dev->reg_state = NETREG_UNREGISTERING;
5022 list_for_each_entry(dev, head, unreg_list) {
5023 /* Shutdown queueing discipline. */
5027 /* Notify protocols, that we are about to destroy
5028 this device. They should clean all the things.
5030 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5032 if (!dev->rtnl_link_ops ||
5033 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5034 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5037 * Flush the unicast and multicast chains
5042 if (dev->netdev_ops->ndo_uninit)
5043 dev->netdev_ops->ndo_uninit(dev);
5045 /* Notifier chain MUST detach us from master device. */
5046 WARN_ON(dev->master);
5048 /* Remove entries from kobject tree */
5049 netdev_unregister_kobject(dev);
5052 /* Process any work delayed until the end of the batch */
5053 dev = list_first_entry(head, struct net_device, unreg_list);
5054 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5058 list_for_each_entry(dev, head, unreg_list)
5062 static void rollback_registered(struct net_device *dev)
5066 list_add(&dev->unreg_list, &single);
5067 rollback_registered_many(&single);
5070 unsigned long netdev_fix_features(unsigned long features, const char *name)
5072 /* Fix illegal SG+CSUM combinations. */
5073 if ((features & NETIF_F_SG) &&
5074 !(features & NETIF_F_ALL_CSUM)) {
5076 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
5077 "checksum feature.\n", name);
5078 features &= ~NETIF_F_SG;
5081 /* TSO requires that SG is present as well. */
5082 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
5084 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
5085 "SG feature.\n", name);
5086 features &= ~NETIF_F_TSO;
5089 if (features & NETIF_F_UFO) {
5090 /* maybe split UFO into V4 and V6? */
5091 if (!((features & NETIF_F_GEN_CSUM) ||
5092 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5093 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5095 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5096 "since no checksum offload features.\n",
5098 features &= ~NETIF_F_UFO;
5101 if (!(features & NETIF_F_SG)) {
5103 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5104 "since no NETIF_F_SG feature.\n", name);
5105 features &= ~NETIF_F_UFO;
5111 EXPORT_SYMBOL(netdev_fix_features);
5114 * netif_stacked_transfer_operstate - transfer operstate
5115 * @rootdev: the root or lower level device to transfer state from
5116 * @dev: the device to transfer operstate to
5118 * Transfer operational state from root to device. This is normally
5119 * called when a stacking relationship exists between the root
5120 * device and the device(a leaf device).
5122 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5123 struct net_device *dev)
5125 if (rootdev->operstate == IF_OPER_DORMANT)
5126 netif_dormant_on(dev);
5128 netif_dormant_off(dev);
5130 if (netif_carrier_ok(rootdev)) {
5131 if (!netif_carrier_ok(dev))
5132 netif_carrier_on(dev);
5134 if (netif_carrier_ok(dev))
5135 netif_carrier_off(dev);
5138 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5141 static int netif_alloc_rx_queues(struct net_device *dev)
5143 unsigned int i, count = dev->num_rx_queues;
5144 struct netdev_rx_queue *rx;
5148 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5150 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5155 for (i = 0; i < count; i++)
5161 static void netdev_init_one_queue(struct net_device *dev,
5162 struct netdev_queue *queue, void *_unused)
5164 /* Initialize queue lock */
5165 spin_lock_init(&queue->_xmit_lock);
5166 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5167 queue->xmit_lock_owner = -1;
5168 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5172 static int netif_alloc_netdev_queues(struct net_device *dev)
5174 unsigned int count = dev->num_tx_queues;
5175 struct netdev_queue *tx;
5179 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5181 pr_err("netdev: Unable to allocate %u tx queues.\n",
5187 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5188 spin_lock_init(&dev->tx_global_lock);
5194 * register_netdevice - register a network device
5195 * @dev: device to register
5197 * Take a completed network device structure and add it to the kernel
5198 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5199 * chain. 0 is returned on success. A negative errno code is returned
5200 * on a failure to set up the device, or if the name is a duplicate.
5202 * Callers must hold the rtnl semaphore. You may want
5203 * register_netdev() instead of this.
5206 * The locking appears insufficient to guarantee two parallel registers
5207 * will not get the same name.
5210 int register_netdevice(struct net_device *dev)
5213 struct net *net = dev_net(dev);
5215 BUG_ON(dev_boot_phase);
5220 /* When net_device's are persistent, this will be fatal. */
5221 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5224 spin_lock_init(&dev->addr_list_lock);
5225 netdev_set_addr_lockdep_class(dev);
5229 /* Init, if this function is available */
5230 if (dev->netdev_ops->ndo_init) {
5231 ret = dev->netdev_ops->ndo_init(dev);
5239 ret = dev_get_valid_name(dev, dev->name, 0);
5243 dev->ifindex = dev_new_index(net);
5244 if (dev->iflink == -1)
5245 dev->iflink = dev->ifindex;
5247 /* Fix illegal checksum combinations */
5248 if ((dev->features & NETIF_F_HW_CSUM) &&
5249 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5250 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5252 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5255 if ((dev->features & NETIF_F_NO_CSUM) &&
5256 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5257 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5259 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5262 dev->features = netdev_fix_features(dev->features, dev->name);
5264 /* Enable software GSO if SG is supported. */
5265 if (dev->features & NETIF_F_SG)
5266 dev->features |= NETIF_F_GSO;
5268 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5269 * vlan_dev_init() will do the dev->features check, so these features
5270 * are enabled only if supported by underlying device.
5272 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5274 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5275 ret = notifier_to_errno(ret);
5279 ret = netdev_register_kobject(dev);
5282 dev->reg_state = NETREG_REGISTERED;
5285 * Default initial state at registry is that the
5286 * device is present.
5289 set_bit(__LINK_STATE_PRESENT, &dev->state);
5291 dev_init_scheduler(dev);
5293 list_netdevice(dev);
5295 /* Notify protocols, that a new device appeared. */
5296 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5297 ret = notifier_to_errno(ret);
5299 rollback_registered(dev);
5300 dev->reg_state = NETREG_UNREGISTERED;
5303 * Prevent userspace races by waiting until the network
5304 * device is fully setup before sending notifications.
5306 if (!dev->rtnl_link_ops ||
5307 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5308 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5314 if (dev->netdev_ops->ndo_uninit)
5315 dev->netdev_ops->ndo_uninit(dev);
5318 EXPORT_SYMBOL(register_netdevice);
5321 * init_dummy_netdev - init a dummy network device for NAPI
5322 * @dev: device to init
5324 * This takes a network device structure and initialize the minimum
5325 * amount of fields so it can be used to schedule NAPI polls without
5326 * registering a full blown interface. This is to be used by drivers
5327 * that need to tie several hardware interfaces to a single NAPI
5328 * poll scheduler due to HW limitations.
5330 int init_dummy_netdev(struct net_device *dev)
5332 /* Clear everything. Note we don't initialize spinlocks
5333 * are they aren't supposed to be taken by any of the
5334 * NAPI code and this dummy netdev is supposed to be
5335 * only ever used for NAPI polls
5337 memset(dev, 0, sizeof(struct net_device));
5339 /* make sure we BUG if trying to hit standard
5340 * register/unregister code path
5342 dev->reg_state = NETREG_DUMMY;
5344 /* NAPI wants this */
5345 INIT_LIST_HEAD(&dev->napi_list);
5347 /* a dummy interface is started by default */
5348 set_bit(__LINK_STATE_PRESENT, &dev->state);
5349 set_bit(__LINK_STATE_START, &dev->state);
5351 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5352 * because users of this 'device' dont need to change
5358 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5362 * register_netdev - register a network device
5363 * @dev: device to register
5365 * Take a completed network device structure and add it to the kernel
5366 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5367 * chain. 0 is returned on success. A negative errno code is returned
5368 * on a failure to set up the device, or if the name is a duplicate.
5370 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5371 * and expands the device name if you passed a format string to
5374 int register_netdev(struct net_device *dev)
5381 * If the name is a format string the caller wants us to do a
5384 if (strchr(dev->name, '%')) {
5385 err = dev_alloc_name(dev, dev->name);
5390 err = register_netdevice(dev);
5395 EXPORT_SYMBOL(register_netdev);
5397 int netdev_refcnt_read(const struct net_device *dev)
5401 for_each_possible_cpu(i)
5402 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5405 EXPORT_SYMBOL(netdev_refcnt_read);
5408 * netdev_wait_allrefs - wait until all references are gone.
5410 * This is called when unregistering network devices.
5412 * Any protocol or device that holds a reference should register
5413 * for netdevice notification, and cleanup and put back the
5414 * reference if they receive an UNREGISTER event.
5415 * We can get stuck here if buggy protocols don't correctly
5418 static void netdev_wait_allrefs(struct net_device *dev)
5420 unsigned long rebroadcast_time, warning_time;
5423 linkwatch_forget_dev(dev);
5425 rebroadcast_time = warning_time = jiffies;
5426 refcnt = netdev_refcnt_read(dev);
5428 while (refcnt != 0) {
5429 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5432 /* Rebroadcast unregister notification */
5433 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5434 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5435 * should have already handle it the first time */
5437 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5439 /* We must not have linkwatch events
5440 * pending on unregister. If this
5441 * happens, we simply run the queue
5442 * unscheduled, resulting in a noop
5445 linkwatch_run_queue();
5450 rebroadcast_time = jiffies;
5455 refcnt = netdev_refcnt_read(dev);
5457 if (time_after(jiffies, warning_time + 10 * HZ)) {
5458 printk(KERN_EMERG "unregister_netdevice: "
5459 "waiting for %s to become free. Usage "
5462 warning_time = jiffies;
5471 * register_netdevice(x1);
5472 * register_netdevice(x2);
5474 * unregister_netdevice(y1);
5475 * unregister_netdevice(y2);
5481 * We are invoked by rtnl_unlock().
5482 * This allows us to deal with problems:
5483 * 1) We can delete sysfs objects which invoke hotplug
5484 * without deadlocking with linkwatch via keventd.
5485 * 2) Since we run with the RTNL semaphore not held, we can sleep
5486 * safely in order to wait for the netdev refcnt to drop to zero.
5488 * We must not return until all unregister events added during
5489 * the interval the lock was held have been completed.
5491 void netdev_run_todo(void)
5493 struct list_head list;
5495 /* Snapshot list, allow later requests */
5496 list_replace_init(&net_todo_list, &list);
5500 while (!list_empty(&list)) {
5501 struct net_device *dev
5502 = list_first_entry(&list, struct net_device, todo_list);
5503 list_del(&dev->todo_list);
5505 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5506 printk(KERN_ERR "network todo '%s' but state %d\n",
5507 dev->name, dev->reg_state);
5512 dev->reg_state = NETREG_UNREGISTERED;
5514 on_each_cpu(flush_backlog, dev, 1);
5516 netdev_wait_allrefs(dev);
5519 BUG_ON(netdev_refcnt_read(dev));
5520 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5521 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5522 WARN_ON(dev->dn_ptr);
5524 if (dev->destructor)
5525 dev->destructor(dev);
5527 /* Free network device */
5528 kobject_put(&dev->dev.kobj);
5533 * dev_txq_stats_fold - fold tx_queues stats
5534 * @dev: device to get statistics from
5535 * @stats: struct rtnl_link_stats64 to hold results
5537 void dev_txq_stats_fold(const struct net_device *dev,
5538 struct rtnl_link_stats64 *stats)
5540 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5542 struct netdev_queue *txq;
5544 for (i = 0; i < dev->num_tx_queues; i++) {
5545 txq = netdev_get_tx_queue(dev, i);
5546 spin_lock_bh(&txq->_xmit_lock);
5547 tx_bytes += txq->tx_bytes;
5548 tx_packets += txq->tx_packets;
5549 tx_dropped += txq->tx_dropped;
5550 spin_unlock_bh(&txq->_xmit_lock);
5552 if (tx_bytes || tx_packets || tx_dropped) {
5553 stats->tx_bytes = tx_bytes;
5554 stats->tx_packets = tx_packets;
5555 stats->tx_dropped = tx_dropped;
5558 EXPORT_SYMBOL(dev_txq_stats_fold);
5560 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5561 * fields in the same order, with only the type differing.
5563 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5564 const struct net_device_stats *netdev_stats)
5566 #if BITS_PER_LONG == 64
5567 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5568 memcpy(stats64, netdev_stats, sizeof(*stats64));
5570 size_t i, n = sizeof(*stats64) / sizeof(u64);
5571 const unsigned long *src = (const unsigned long *)netdev_stats;
5572 u64 *dst = (u64 *)stats64;
5574 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5575 sizeof(*stats64) / sizeof(u64));
5576 for (i = 0; i < n; i++)
5582 * dev_get_stats - get network device statistics
5583 * @dev: device to get statistics from
5584 * @storage: place to store stats
5586 * Get network statistics from device. Return @storage.
5587 * The device driver may provide its own method by setting
5588 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5589 * otherwise the internal statistics structure is used.
5591 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5592 struct rtnl_link_stats64 *storage)
5594 const struct net_device_ops *ops = dev->netdev_ops;
5596 if (ops->ndo_get_stats64) {
5597 memset(storage, 0, sizeof(*storage));
5598 ops->ndo_get_stats64(dev, storage);
5599 } else if (ops->ndo_get_stats) {
5600 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5602 netdev_stats_to_stats64(storage, &dev->stats);
5603 dev_txq_stats_fold(dev, storage);
5605 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5608 EXPORT_SYMBOL(dev_get_stats);
5610 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5612 struct netdev_queue *queue = dev_ingress_queue(dev);
5614 #ifdef CONFIG_NET_CLS_ACT
5617 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5620 netdev_init_one_queue(dev, queue, NULL);
5621 queue->qdisc = &noop_qdisc;
5622 queue->qdisc_sleeping = &noop_qdisc;
5623 rcu_assign_pointer(dev->ingress_queue, queue);
5629 * alloc_netdev_mq - allocate network device
5630 * @sizeof_priv: size of private data to allocate space for
5631 * @name: device name format string
5632 * @setup: callback to initialize device
5633 * @queue_count: the number of subqueues to allocate
5635 * Allocates a struct net_device with private data area for driver use
5636 * and performs basic initialization. Also allocates subquue structs
5637 * for each queue on the device at the end of the netdevice.
5639 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5640 void (*setup)(struct net_device *), unsigned int queue_count)
5642 struct net_device *dev;
5644 struct net_device *p;
5646 BUG_ON(strlen(name) >= sizeof(dev->name));
5648 if (queue_count < 1) {
5649 pr_err("alloc_netdev: Unable to allocate device "
5650 "with zero queues.\n");
5654 alloc_size = sizeof(struct net_device);
5656 /* ensure 32-byte alignment of private area */
5657 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5658 alloc_size += sizeof_priv;
5660 /* ensure 32-byte alignment of whole construct */
5661 alloc_size += NETDEV_ALIGN - 1;
5663 p = kzalloc(alloc_size, GFP_KERNEL);
5665 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5669 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5670 dev->padded = (char *)dev - (char *)p;
5672 dev->pcpu_refcnt = alloc_percpu(int);
5673 if (!dev->pcpu_refcnt)
5676 if (dev_addr_init(dev))
5682 dev_net_set(dev, &init_net);
5684 dev->num_tx_queues = queue_count;
5685 dev->real_num_tx_queues = queue_count;
5686 if (netif_alloc_netdev_queues(dev))
5690 dev->num_rx_queues = queue_count;
5691 dev->real_num_rx_queues = queue_count;
5692 if (netif_alloc_rx_queues(dev))
5696 dev->gso_max_size = GSO_MAX_SIZE;
5698 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5699 dev->ethtool_ntuple_list.count = 0;
5700 INIT_LIST_HEAD(&dev->napi_list);
5701 INIT_LIST_HEAD(&dev->unreg_list);
5702 INIT_LIST_HEAD(&dev->link_watch_list);
5703 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5705 strcpy(dev->name, name);
5709 free_percpu(dev->pcpu_refcnt);
5719 EXPORT_SYMBOL(alloc_netdev_mq);
5722 * free_netdev - free network device
5725 * This function does the last stage of destroying an allocated device
5726 * interface. The reference to the device object is released.
5727 * If this is the last reference then it will be freed.
5729 void free_netdev(struct net_device *dev)
5731 struct napi_struct *p, *n;
5733 release_net(dev_net(dev));
5740 kfree(rcu_dereference_raw(dev->ingress_queue));
5742 /* Flush device addresses */
5743 dev_addr_flush(dev);
5745 /* Clear ethtool n-tuple list */
5746 ethtool_ntuple_flush(dev);
5748 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5751 free_percpu(dev->pcpu_refcnt);
5752 dev->pcpu_refcnt = NULL;
5754 /* Compatibility with error handling in drivers */
5755 if (dev->reg_state == NETREG_UNINITIALIZED) {
5756 kfree((char *)dev - dev->padded);
5760 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5761 dev->reg_state = NETREG_RELEASED;
5763 /* will free via device release */
5764 put_device(&dev->dev);
5766 EXPORT_SYMBOL(free_netdev);
5769 * synchronize_net - Synchronize with packet receive processing
5771 * Wait for packets currently being received to be done.
5772 * Does not block later packets from starting.
5774 void synchronize_net(void)
5779 EXPORT_SYMBOL(synchronize_net);
5782 * unregister_netdevice_queue - remove device from the kernel
5786 * This function shuts down a device interface and removes it
5787 * from the kernel tables.
5788 * If head not NULL, device is queued to be unregistered later.
5790 * Callers must hold the rtnl semaphore. You may want
5791 * unregister_netdev() instead of this.
5794 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5799 list_move_tail(&dev->unreg_list, head);
5801 rollback_registered(dev);
5802 /* Finish processing unregister after unlock */
5806 EXPORT_SYMBOL(unregister_netdevice_queue);
5809 * unregister_netdevice_many - unregister many devices
5810 * @head: list of devices
5812 void unregister_netdevice_many(struct list_head *head)
5814 struct net_device *dev;
5816 if (!list_empty(head)) {
5817 rollback_registered_many(head);
5818 list_for_each_entry(dev, head, unreg_list)
5822 EXPORT_SYMBOL(unregister_netdevice_many);
5825 * unregister_netdev - remove device from the kernel
5828 * This function shuts down a device interface and removes it
5829 * from the kernel tables.
5831 * This is just a wrapper for unregister_netdevice that takes
5832 * the rtnl semaphore. In general you want to use this and not
5833 * unregister_netdevice.
5835 void unregister_netdev(struct net_device *dev)
5838 unregister_netdevice(dev);
5841 EXPORT_SYMBOL(unregister_netdev);
5844 * dev_change_net_namespace - move device to different nethost namespace
5846 * @net: network namespace
5847 * @pat: If not NULL name pattern to try if the current device name
5848 * is already taken in the destination network namespace.
5850 * This function shuts down a device interface and moves it
5851 * to a new network namespace. On success 0 is returned, on
5852 * a failure a netagive errno code is returned.
5854 * Callers must hold the rtnl semaphore.
5857 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5863 /* Don't allow namespace local devices to be moved. */
5865 if (dev->features & NETIF_F_NETNS_LOCAL)
5868 /* Ensure the device has been registrered */
5870 if (dev->reg_state != NETREG_REGISTERED)
5873 /* Get out if there is nothing todo */
5875 if (net_eq(dev_net(dev), net))
5878 /* Pick the destination device name, and ensure
5879 * we can use it in the destination network namespace.
5882 if (__dev_get_by_name(net, dev->name)) {
5883 /* We get here if we can't use the current device name */
5886 if (dev_get_valid_name(dev, pat, 1))
5891 * And now a mini version of register_netdevice unregister_netdevice.
5894 /* If device is running close it first. */
5897 /* And unlink it from device chain */
5899 unlist_netdevice(dev);
5903 /* Shutdown queueing discipline. */
5906 /* Notify protocols, that we are about to destroy
5907 this device. They should clean all the things.
5909 Note that dev->reg_state stays at NETREG_REGISTERED.
5910 This is wanted because this way 8021q and macvlan know
5911 the device is just moving and can keep their slaves up.
5913 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5914 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5917 * Flush the unicast and multicast chains
5922 /* Actually switch the network namespace */
5923 dev_net_set(dev, net);
5925 /* If there is an ifindex conflict assign a new one */
5926 if (__dev_get_by_index(net, dev->ifindex)) {
5927 int iflink = (dev->iflink == dev->ifindex);
5928 dev->ifindex = dev_new_index(net);
5930 dev->iflink = dev->ifindex;
5933 /* Fixup kobjects */
5934 err = device_rename(&dev->dev, dev->name);
5937 /* Add the device back in the hashes */
5938 list_netdevice(dev);
5940 /* Notify protocols, that a new device appeared. */
5941 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5944 * Prevent userspace races by waiting until the network
5945 * device is fully setup before sending notifications.
5947 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5954 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5956 static int dev_cpu_callback(struct notifier_block *nfb,
5957 unsigned long action,
5960 struct sk_buff **list_skb;
5961 struct sk_buff *skb;
5962 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5963 struct softnet_data *sd, *oldsd;
5965 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5968 local_irq_disable();
5969 cpu = smp_processor_id();
5970 sd = &per_cpu(softnet_data, cpu);
5971 oldsd = &per_cpu(softnet_data, oldcpu);
5973 /* Find end of our completion_queue. */
5974 list_skb = &sd->completion_queue;
5976 list_skb = &(*list_skb)->next;
5977 /* Append completion queue from offline CPU. */
5978 *list_skb = oldsd->completion_queue;
5979 oldsd->completion_queue = NULL;
5981 /* Append output queue from offline CPU. */
5982 if (oldsd->output_queue) {
5983 *sd->output_queue_tailp = oldsd->output_queue;
5984 sd->output_queue_tailp = oldsd->output_queue_tailp;
5985 oldsd->output_queue = NULL;
5986 oldsd->output_queue_tailp = &oldsd->output_queue;
5989 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5992 /* Process offline CPU's input_pkt_queue */
5993 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5995 input_queue_head_incr(oldsd);
5997 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5999 input_queue_head_incr(oldsd);
6007 * netdev_increment_features - increment feature set by one
6008 * @all: current feature set
6009 * @one: new feature set
6010 * @mask: mask feature set
6012 * Computes a new feature set after adding a device with feature set
6013 * @one to the master device with current feature set @all. Will not
6014 * enable anything that is off in @mask. Returns the new feature set.
6016 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
6019 /* If device needs checksumming, downgrade to it. */
6020 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
6021 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
6022 else if (mask & NETIF_F_ALL_CSUM) {
6023 /* If one device supports v4/v6 checksumming, set for all. */
6024 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
6025 !(all & NETIF_F_GEN_CSUM)) {
6026 all &= ~NETIF_F_ALL_CSUM;
6027 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
6030 /* If one device supports hw checksumming, set for all. */
6031 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
6032 all &= ~NETIF_F_ALL_CSUM;
6033 all |= NETIF_F_HW_CSUM;
6037 one |= NETIF_F_ALL_CSUM;
6039 one |= all & NETIF_F_ONE_FOR_ALL;
6040 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
6041 all |= one & mask & NETIF_F_ONE_FOR_ALL;
6045 EXPORT_SYMBOL(netdev_increment_features);
6047 static struct hlist_head *netdev_create_hash(void)
6050 struct hlist_head *hash;
6052 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6054 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6055 INIT_HLIST_HEAD(&hash[i]);
6060 /* Initialize per network namespace state */
6061 static int __net_init netdev_init(struct net *net)
6063 INIT_LIST_HEAD(&net->dev_base_head);
6065 net->dev_name_head = netdev_create_hash();
6066 if (net->dev_name_head == NULL)
6069 net->dev_index_head = netdev_create_hash();
6070 if (net->dev_index_head == NULL)
6076 kfree(net->dev_name_head);
6082 * netdev_drivername - network driver for the device
6083 * @dev: network device
6084 * @buffer: buffer for resulting name
6085 * @len: size of buffer
6087 * Determine network driver for device.
6089 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6091 const struct device_driver *driver;
6092 const struct device *parent;
6094 if (len <= 0 || !buffer)
6098 parent = dev->dev.parent;
6103 driver = parent->driver;
6104 if (driver && driver->name)
6105 strlcpy(buffer, driver->name, len);
6109 static int __netdev_printk(const char *level, const struct net_device *dev,
6110 struct va_format *vaf)
6114 if (dev && dev->dev.parent)
6115 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6116 netdev_name(dev), vaf);
6118 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6120 r = printk("%s(NULL net_device): %pV", level, vaf);
6125 int netdev_printk(const char *level, const struct net_device *dev,
6126 const char *format, ...)
6128 struct va_format vaf;
6132 va_start(args, format);
6137 r = __netdev_printk(level, dev, &vaf);
6142 EXPORT_SYMBOL(netdev_printk);
6144 #define define_netdev_printk_level(func, level) \
6145 int func(const struct net_device *dev, const char *fmt, ...) \
6148 struct va_format vaf; \
6151 va_start(args, fmt); \
6156 r = __netdev_printk(level, dev, &vaf); \
6161 EXPORT_SYMBOL(func);
6163 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6164 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6165 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6166 define_netdev_printk_level(netdev_err, KERN_ERR);
6167 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6168 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6169 define_netdev_printk_level(netdev_info, KERN_INFO);
6171 static void __net_exit netdev_exit(struct net *net)
6173 kfree(net->dev_name_head);
6174 kfree(net->dev_index_head);
6177 static struct pernet_operations __net_initdata netdev_net_ops = {
6178 .init = netdev_init,
6179 .exit = netdev_exit,
6182 static void __net_exit default_device_exit(struct net *net)
6184 struct net_device *dev, *aux;
6186 * Push all migratable network devices back to the
6187 * initial network namespace
6190 for_each_netdev_safe(net, dev, aux) {
6192 char fb_name[IFNAMSIZ];
6194 /* Ignore unmoveable devices (i.e. loopback) */
6195 if (dev->features & NETIF_F_NETNS_LOCAL)
6198 /* Leave virtual devices for the generic cleanup */
6199 if (dev->rtnl_link_ops)
6202 /* Push remaing network devices to init_net */
6203 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6204 err = dev_change_net_namespace(dev, &init_net, fb_name);
6206 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6207 __func__, dev->name, err);
6214 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6216 /* At exit all network devices most be removed from a network
6217 * namespace. Do this in the reverse order of registeration.
6218 * Do this across as many network namespaces as possible to
6219 * improve batching efficiency.
6221 struct net_device *dev;
6223 LIST_HEAD(dev_kill_list);
6226 list_for_each_entry(net, net_list, exit_list) {
6227 for_each_netdev_reverse(net, dev) {
6228 if (dev->rtnl_link_ops)
6229 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6231 unregister_netdevice_queue(dev, &dev_kill_list);
6234 unregister_netdevice_many(&dev_kill_list);
6238 static struct pernet_operations __net_initdata default_device_ops = {
6239 .exit = default_device_exit,
6240 .exit_batch = default_device_exit_batch,
6244 * Initialize the DEV module. At boot time this walks the device list and
6245 * unhooks any devices that fail to initialise (normally hardware not
6246 * present) and leaves us with a valid list of present and active devices.
6251 * This is called single threaded during boot, so no need
6252 * to take the rtnl semaphore.
6254 static int __init net_dev_init(void)
6256 int i, rc = -ENOMEM;
6258 BUG_ON(!dev_boot_phase);
6260 if (dev_proc_init())
6263 if (netdev_kobject_init())
6266 INIT_LIST_HEAD(&ptype_all);
6267 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6268 INIT_LIST_HEAD(&ptype_base[i]);
6270 if (register_pernet_subsys(&netdev_net_ops))
6274 * Initialise the packet receive queues.
6277 for_each_possible_cpu(i) {
6278 struct softnet_data *sd = &per_cpu(softnet_data, i);
6280 memset(sd, 0, sizeof(*sd));
6281 skb_queue_head_init(&sd->input_pkt_queue);
6282 skb_queue_head_init(&sd->process_queue);
6283 sd->completion_queue = NULL;
6284 INIT_LIST_HEAD(&sd->poll_list);
6285 sd->output_queue = NULL;
6286 sd->output_queue_tailp = &sd->output_queue;
6288 sd->csd.func = rps_trigger_softirq;
6294 sd->backlog.poll = process_backlog;
6295 sd->backlog.weight = weight_p;
6296 sd->backlog.gro_list = NULL;
6297 sd->backlog.gro_count = 0;
6302 /* The loopback device is special if any other network devices
6303 * is present in a network namespace the loopback device must
6304 * be present. Since we now dynamically allocate and free the
6305 * loopback device ensure this invariant is maintained by
6306 * keeping the loopback device as the first device on the
6307 * list of network devices. Ensuring the loopback devices
6308 * is the first device that appears and the last network device
6311 if (register_pernet_device(&loopback_net_ops))
6314 if (register_pernet_device(&default_device_ops))
6317 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6318 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6320 hotcpu_notifier(dev_cpu_callback, 0);
6328 subsys_initcall(net_dev_init);
6330 static int __init initialize_hashrnd(void)
6332 get_random_bytes(&hashrnd, sizeof(hashrnd));
6336 late_initcall_sync(initialize_hashrnd);