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.
753 * The caller must hold RCU or RTNL.
754 * The returned device has not had its ref count increased
755 * and the caller must therefore be careful about locking
759 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
762 struct net_device *dev;
764 for_each_netdev_rcu(net, dev)
765 if (dev->type == type &&
766 !memcmp(dev->dev_addr, ha, dev->addr_len))
771 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
773 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
775 struct net_device *dev;
778 for_each_netdev(net, dev)
779 if (dev->type == type)
784 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
786 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
788 struct net_device *dev, *ret = NULL;
791 for_each_netdev_rcu(net, dev)
792 if (dev->type == type) {
800 EXPORT_SYMBOL(dev_getfirstbyhwtype);
803 * dev_get_by_flags_rcu - find any device with given flags
804 * @net: the applicable net namespace
805 * @if_flags: IFF_* values
806 * @mask: bitmask of bits in if_flags to check
808 * Search for any interface with the given flags. Returns NULL if a device
809 * is not found or a pointer to the device. Must be called inside
810 * rcu_read_lock(), and result refcount is unchanged.
813 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
816 struct net_device *dev, *ret;
819 for_each_netdev_rcu(net, dev) {
820 if (((dev->flags ^ if_flags) & mask) == 0) {
827 EXPORT_SYMBOL(dev_get_by_flags_rcu);
830 * dev_valid_name - check if name is okay for network device
833 * Network device names need to be valid file names to
834 * to allow sysfs to work. We also disallow any kind of
837 int dev_valid_name(const char *name)
841 if (strlen(name) >= IFNAMSIZ)
843 if (!strcmp(name, ".") || !strcmp(name, ".."))
847 if (*name == '/' || isspace(*name))
853 EXPORT_SYMBOL(dev_valid_name);
856 * __dev_alloc_name - allocate a name for a device
857 * @net: network namespace to allocate the device name in
858 * @name: name format string
859 * @buf: scratch buffer and result name string
861 * Passed a format string - eg "lt%d" it will try and find a suitable
862 * id. It scans list of devices to build up a free map, then chooses
863 * the first empty slot. The caller must hold the dev_base or rtnl lock
864 * while allocating the name and adding the device in order to avoid
866 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
867 * Returns the number of the unit assigned or a negative errno code.
870 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
874 const int max_netdevices = 8*PAGE_SIZE;
875 unsigned long *inuse;
876 struct net_device *d;
878 p = strnchr(name, IFNAMSIZ-1, '%');
881 * Verify the string as this thing may have come from
882 * the user. There must be either one "%d" and no other "%"
885 if (p[1] != 'd' || strchr(p + 2, '%'))
888 /* Use one page as a bit array of possible slots */
889 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
893 for_each_netdev(net, d) {
894 if (!sscanf(d->name, name, &i))
896 if (i < 0 || i >= max_netdevices)
899 /* avoid cases where sscanf is not exact inverse of printf */
900 snprintf(buf, IFNAMSIZ, name, i);
901 if (!strncmp(buf, d->name, IFNAMSIZ))
905 i = find_first_zero_bit(inuse, max_netdevices);
906 free_page((unsigned long) inuse);
910 snprintf(buf, IFNAMSIZ, name, i);
911 if (!__dev_get_by_name(net, buf))
914 /* It is possible to run out of possible slots
915 * when the name is long and there isn't enough space left
916 * for the digits, or if all bits are used.
922 * dev_alloc_name - allocate a name for a device
924 * @name: name format string
926 * Passed a format string - eg "lt%d" it will try and find a suitable
927 * id. It scans list of devices to build up a free map, then chooses
928 * the first empty slot. The caller must hold the dev_base or rtnl lock
929 * while allocating the name and adding the device in order to avoid
931 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
932 * Returns the number of the unit assigned or a negative errno code.
935 int dev_alloc_name(struct net_device *dev, const char *name)
941 BUG_ON(!dev_net(dev));
943 ret = __dev_alloc_name(net, name, buf);
945 strlcpy(dev->name, buf, IFNAMSIZ);
948 EXPORT_SYMBOL(dev_alloc_name);
950 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
954 BUG_ON(!dev_net(dev));
957 if (!dev_valid_name(name))
960 if (fmt && strchr(name, '%'))
961 return dev_alloc_name(dev, name);
962 else if (__dev_get_by_name(net, name))
964 else if (dev->name != name)
965 strlcpy(dev->name, name, IFNAMSIZ);
971 * dev_change_name - change name of a device
973 * @newname: name (or format string) must be at least IFNAMSIZ
975 * Change name of a device, can pass format strings "eth%d".
978 int dev_change_name(struct net_device *dev, const char *newname)
980 char oldname[IFNAMSIZ];
986 BUG_ON(!dev_net(dev));
989 if (dev->flags & IFF_UP)
992 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
995 memcpy(oldname, dev->name, IFNAMSIZ);
997 err = dev_get_valid_name(dev, newname, 1);
1002 ret = device_rename(&dev->dev, dev->name);
1004 memcpy(dev->name, oldname, IFNAMSIZ);
1008 write_lock_bh(&dev_base_lock);
1009 hlist_del(&dev->name_hlist);
1010 write_unlock_bh(&dev_base_lock);
1014 write_lock_bh(&dev_base_lock);
1015 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1016 write_unlock_bh(&dev_base_lock);
1018 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1019 ret = notifier_to_errno(ret);
1022 /* err >= 0 after dev_alloc_name() or stores the first errno */
1025 memcpy(dev->name, oldname, IFNAMSIZ);
1029 "%s: name change rollback failed: %d.\n",
1038 * dev_set_alias - change ifalias of a device
1040 * @alias: name up to IFALIASZ
1041 * @len: limit of bytes to copy from info
1043 * Set ifalias for a device,
1045 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1049 if (len >= IFALIASZ)
1054 kfree(dev->ifalias);
1055 dev->ifalias = NULL;
1060 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1064 strlcpy(dev->ifalias, alias, len+1);
1070 * netdev_features_change - device changes features
1071 * @dev: device to cause notification
1073 * Called to indicate a device has changed features.
1075 void netdev_features_change(struct net_device *dev)
1077 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1079 EXPORT_SYMBOL(netdev_features_change);
1082 * netdev_state_change - device changes state
1083 * @dev: device to cause notification
1085 * Called to indicate a device has changed state. This function calls
1086 * the notifier chains for netdev_chain and sends a NEWLINK message
1087 * to the routing socket.
1089 void netdev_state_change(struct net_device *dev)
1091 if (dev->flags & IFF_UP) {
1092 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1093 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1096 EXPORT_SYMBOL(netdev_state_change);
1098 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1100 return call_netdevice_notifiers(event, dev);
1102 EXPORT_SYMBOL(netdev_bonding_change);
1105 * dev_load - load a network module
1106 * @net: the applicable net namespace
1107 * @name: name of interface
1109 * If a network interface is not present and the process has suitable
1110 * privileges this function loads the module. If module loading is not
1111 * available in this kernel then it becomes a nop.
1114 void dev_load(struct net *net, const char *name)
1116 struct net_device *dev;
1119 dev = dev_get_by_name_rcu(net, name);
1122 if (!dev && capable(CAP_NET_ADMIN))
1123 request_module("%s", name);
1125 EXPORT_SYMBOL(dev_load);
1127 static int __dev_open(struct net_device *dev)
1129 const struct net_device_ops *ops = dev->netdev_ops;
1135 * Is it even present?
1137 if (!netif_device_present(dev))
1140 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1141 ret = notifier_to_errno(ret);
1146 * Call device private open method
1148 set_bit(__LINK_STATE_START, &dev->state);
1150 if (ops->ndo_validate_addr)
1151 ret = ops->ndo_validate_addr(dev);
1153 if (!ret && ops->ndo_open)
1154 ret = ops->ndo_open(dev);
1157 * If it went open OK then:
1161 clear_bit(__LINK_STATE_START, &dev->state);
1166 dev->flags |= IFF_UP;
1171 net_dmaengine_get();
1174 * Initialize multicasting status
1176 dev_set_rx_mode(dev);
1179 * Wakeup transmit queue engine
1188 * dev_open - prepare an interface for use.
1189 * @dev: device to open
1191 * Takes a device from down to up state. The device's private open
1192 * function is invoked and then the multicast lists are loaded. Finally
1193 * the device is moved into the up state and a %NETDEV_UP message is
1194 * sent to the netdev notifier chain.
1196 * Calling this function on an active interface is a nop. On a failure
1197 * a negative errno code is returned.
1199 int dev_open(struct net_device *dev)
1206 if (dev->flags & IFF_UP)
1212 ret = __dev_open(dev);
1217 * ... and announce new interface.
1219 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1220 call_netdevice_notifiers(NETDEV_UP, dev);
1224 EXPORT_SYMBOL(dev_open);
1226 static int __dev_close_many(struct list_head *head)
1228 struct net_device *dev;
1233 list_for_each_entry(dev, head, unreg_list) {
1235 * Tell people we are going down, so that they can
1236 * prepare to death, when device is still operating.
1238 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1240 clear_bit(__LINK_STATE_START, &dev->state);
1242 /* Synchronize to scheduled poll. We cannot touch poll list, it
1243 * can be even on different cpu. So just clear netif_running().
1245 * dev->stop() will invoke napi_disable() on all of it's
1246 * napi_struct instances on this device.
1248 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1251 dev_deactivate_many(head);
1253 list_for_each_entry(dev, head, unreg_list) {
1254 const struct net_device_ops *ops = dev->netdev_ops;
1257 * Call the device specific close. This cannot fail.
1258 * Only if device is UP
1260 * We allow it to be called even after a DETACH hot-plug
1267 * Device is now down.
1270 dev->flags &= ~IFF_UP;
1275 net_dmaengine_put();
1281 static int __dev_close(struct net_device *dev)
1285 list_add(&dev->unreg_list, &single);
1286 return __dev_close_many(&single);
1289 int dev_close_many(struct list_head *head)
1291 struct net_device *dev, *tmp;
1292 LIST_HEAD(tmp_list);
1294 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1295 if (!(dev->flags & IFF_UP))
1296 list_move(&dev->unreg_list, &tmp_list);
1298 __dev_close_many(head);
1301 * Tell people we are down
1303 list_for_each_entry(dev, head, unreg_list) {
1304 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1305 call_netdevice_notifiers(NETDEV_DOWN, dev);
1308 /* rollback_registered_many needs the complete original list */
1309 list_splice(&tmp_list, head);
1314 * dev_close - shutdown an interface.
1315 * @dev: device to shutdown
1317 * This function moves an active device into down state. A
1318 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1319 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1322 int dev_close(struct net_device *dev)
1326 list_add(&dev->unreg_list, &single);
1327 dev_close_many(&single);
1331 EXPORT_SYMBOL(dev_close);
1335 * dev_disable_lro - disable Large Receive Offload on a device
1338 * Disable Large Receive Offload (LRO) on a net device. Must be
1339 * called under RTNL. This is needed if received packets may be
1340 * forwarded to another interface.
1342 void dev_disable_lro(struct net_device *dev)
1344 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1345 dev->ethtool_ops->set_flags) {
1346 u32 flags = dev->ethtool_ops->get_flags(dev);
1347 if (flags & ETH_FLAG_LRO) {
1348 flags &= ~ETH_FLAG_LRO;
1349 dev->ethtool_ops->set_flags(dev, flags);
1352 WARN_ON(dev->features & NETIF_F_LRO);
1354 EXPORT_SYMBOL(dev_disable_lro);
1357 static int dev_boot_phase = 1;
1360 * Device change register/unregister. These are not inline or static
1361 * as we export them to the world.
1365 * register_netdevice_notifier - register a network notifier block
1368 * Register a notifier to be called when network device events occur.
1369 * The notifier passed is linked into the kernel structures and must
1370 * not be reused until it has been unregistered. A negative errno code
1371 * is returned on a failure.
1373 * When registered all registration and up events are replayed
1374 * to the new notifier to allow device to have a race free
1375 * view of the network device list.
1378 int register_netdevice_notifier(struct notifier_block *nb)
1380 struct net_device *dev;
1381 struct net_device *last;
1386 err = raw_notifier_chain_register(&netdev_chain, nb);
1392 for_each_netdev(net, dev) {
1393 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1394 err = notifier_to_errno(err);
1398 if (!(dev->flags & IFF_UP))
1401 nb->notifier_call(nb, NETDEV_UP, dev);
1412 for_each_netdev(net, dev) {
1416 if (dev->flags & IFF_UP) {
1417 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1418 nb->notifier_call(nb, NETDEV_DOWN, dev);
1420 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1421 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1425 raw_notifier_chain_unregister(&netdev_chain, nb);
1428 EXPORT_SYMBOL(register_netdevice_notifier);
1431 * unregister_netdevice_notifier - unregister a network notifier block
1434 * Unregister a notifier previously registered by
1435 * register_netdevice_notifier(). The notifier is unlinked into the
1436 * kernel structures and may then be reused. A negative errno code
1437 * is returned on a failure.
1440 int unregister_netdevice_notifier(struct notifier_block *nb)
1445 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1449 EXPORT_SYMBOL(unregister_netdevice_notifier);
1452 * call_netdevice_notifiers - call all network notifier blocks
1453 * @val: value passed unmodified to notifier function
1454 * @dev: net_device pointer passed unmodified to notifier function
1456 * Call all network notifier blocks. Parameters and return value
1457 * are as for raw_notifier_call_chain().
1460 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1463 return raw_notifier_call_chain(&netdev_chain, val, dev);
1466 /* When > 0 there are consumers of rx skb time stamps */
1467 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1469 void net_enable_timestamp(void)
1471 atomic_inc(&netstamp_needed);
1473 EXPORT_SYMBOL(net_enable_timestamp);
1475 void net_disable_timestamp(void)
1477 atomic_dec(&netstamp_needed);
1479 EXPORT_SYMBOL(net_disable_timestamp);
1481 static inline void net_timestamp_set(struct sk_buff *skb)
1483 if (atomic_read(&netstamp_needed))
1484 __net_timestamp(skb);
1486 skb->tstamp.tv64 = 0;
1489 static inline void net_timestamp_check(struct sk_buff *skb)
1491 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1492 __net_timestamp(skb);
1496 * dev_forward_skb - loopback an skb to another netif
1498 * @dev: destination network device
1499 * @skb: buffer to forward
1502 * NET_RX_SUCCESS (no congestion)
1503 * NET_RX_DROP (packet was dropped, but freed)
1505 * dev_forward_skb can be used for injecting an skb from the
1506 * start_xmit function of one device into the receive queue
1507 * of another device.
1509 * The receiving device may be in another namespace, so
1510 * we have to clear all information in the skb that could
1511 * impact namespace isolation.
1513 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1518 if (unlikely(!(dev->flags & IFF_UP) ||
1519 (skb->len > (dev->mtu + dev->hard_header_len + VLAN_HLEN)))) {
1520 atomic_long_inc(&dev->rx_dropped);
1524 skb_set_dev(skb, dev);
1525 skb->tstamp.tv64 = 0;
1526 skb->pkt_type = PACKET_HOST;
1527 skb->protocol = eth_type_trans(skb, dev);
1528 return netif_rx(skb);
1530 EXPORT_SYMBOL_GPL(dev_forward_skb);
1532 static inline int deliver_skb(struct sk_buff *skb,
1533 struct packet_type *pt_prev,
1534 struct net_device *orig_dev)
1536 atomic_inc(&skb->users);
1537 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1541 * Support routine. Sends outgoing frames to any network
1542 * taps currently in use.
1545 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1547 struct packet_type *ptype;
1548 struct sk_buff *skb2 = NULL;
1549 struct packet_type *pt_prev = NULL;
1552 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1553 /* Never send packets back to the socket
1554 * they originated from - MvS (miquels@drinkel.ow.org)
1556 if ((ptype->dev == dev || !ptype->dev) &&
1557 (ptype->af_packet_priv == NULL ||
1558 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1560 deliver_skb(skb2, pt_prev, skb->dev);
1565 skb2 = skb_clone(skb, GFP_ATOMIC);
1569 net_timestamp_set(skb2);
1571 /* skb->nh should be correctly
1572 set by sender, so that the second statement is
1573 just protection against buggy protocols.
1575 skb_reset_mac_header(skb2);
1577 if (skb_network_header(skb2) < skb2->data ||
1578 skb2->network_header > skb2->tail) {
1579 if (net_ratelimit())
1580 printk(KERN_CRIT "protocol %04x is "
1582 ntohs(skb2->protocol),
1584 skb_reset_network_header(skb2);
1587 skb2->transport_header = skb2->network_header;
1588 skb2->pkt_type = PACKET_OUTGOING;
1593 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1598 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1599 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1601 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1605 if (txq < 1 || txq > dev->num_tx_queues)
1608 if (dev->reg_state == NETREG_REGISTERED) {
1611 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1616 if (txq < dev->real_num_tx_queues)
1617 qdisc_reset_all_tx_gt(dev, txq);
1620 dev->real_num_tx_queues = txq;
1623 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1627 * netif_set_real_num_rx_queues - set actual number of RX queues used
1628 * @dev: Network device
1629 * @rxq: Actual number of RX queues
1631 * This must be called either with the rtnl_lock held or before
1632 * registration of the net device. Returns 0 on success, or a
1633 * negative error code. If called before registration, it always
1636 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1640 if (rxq < 1 || rxq > dev->num_rx_queues)
1643 if (dev->reg_state == NETREG_REGISTERED) {
1646 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1652 dev->real_num_rx_queues = rxq;
1655 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1658 static inline void __netif_reschedule(struct Qdisc *q)
1660 struct softnet_data *sd;
1661 unsigned long flags;
1663 local_irq_save(flags);
1664 sd = &__get_cpu_var(softnet_data);
1665 q->next_sched = NULL;
1666 *sd->output_queue_tailp = q;
1667 sd->output_queue_tailp = &q->next_sched;
1668 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1669 local_irq_restore(flags);
1672 void __netif_schedule(struct Qdisc *q)
1674 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1675 __netif_reschedule(q);
1677 EXPORT_SYMBOL(__netif_schedule);
1679 void dev_kfree_skb_irq(struct sk_buff *skb)
1681 if (atomic_dec_and_test(&skb->users)) {
1682 struct softnet_data *sd;
1683 unsigned long flags;
1685 local_irq_save(flags);
1686 sd = &__get_cpu_var(softnet_data);
1687 skb->next = sd->completion_queue;
1688 sd->completion_queue = skb;
1689 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1690 local_irq_restore(flags);
1693 EXPORT_SYMBOL(dev_kfree_skb_irq);
1695 void dev_kfree_skb_any(struct sk_buff *skb)
1697 if (in_irq() || irqs_disabled())
1698 dev_kfree_skb_irq(skb);
1702 EXPORT_SYMBOL(dev_kfree_skb_any);
1706 * netif_device_detach - mark device as removed
1707 * @dev: network device
1709 * Mark device as removed from system and therefore no longer available.
1711 void netif_device_detach(struct net_device *dev)
1713 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1714 netif_running(dev)) {
1715 netif_tx_stop_all_queues(dev);
1718 EXPORT_SYMBOL(netif_device_detach);
1721 * netif_device_attach - mark device as attached
1722 * @dev: network device
1724 * Mark device as attached from system and restart if needed.
1726 void netif_device_attach(struct net_device *dev)
1728 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1729 netif_running(dev)) {
1730 netif_tx_wake_all_queues(dev);
1731 __netdev_watchdog_up(dev);
1734 EXPORT_SYMBOL(netif_device_attach);
1737 * skb_dev_set -- assign a new device to a buffer
1738 * @skb: buffer for the new device
1739 * @dev: network device
1741 * If an skb is owned by a device already, we have to reset
1742 * all data private to the namespace a device belongs to
1743 * before assigning it a new device.
1745 #ifdef CONFIG_NET_NS
1746 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1749 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1752 skb_init_secmark(skb);
1756 skb->ipvs_property = 0;
1757 #ifdef CONFIG_NET_SCHED
1763 EXPORT_SYMBOL(skb_set_dev);
1764 #endif /* CONFIG_NET_NS */
1767 * Invalidate hardware checksum when packet is to be mangled, and
1768 * complete checksum manually on outgoing path.
1770 int skb_checksum_help(struct sk_buff *skb)
1773 int ret = 0, offset;
1775 if (skb->ip_summed == CHECKSUM_COMPLETE)
1776 goto out_set_summed;
1778 if (unlikely(skb_shinfo(skb)->gso_size)) {
1779 /* Let GSO fix up the checksum. */
1780 goto out_set_summed;
1783 offset = skb_checksum_start_offset(skb);
1784 BUG_ON(offset >= skb_headlen(skb));
1785 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1787 offset += skb->csum_offset;
1788 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1790 if (skb_cloned(skb) &&
1791 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1792 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1797 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1799 skb->ip_summed = CHECKSUM_NONE;
1803 EXPORT_SYMBOL(skb_checksum_help);
1806 * skb_gso_segment - Perform segmentation on skb.
1807 * @skb: buffer to segment
1808 * @features: features for the output path (see dev->features)
1810 * This function segments the given skb and returns a list of segments.
1812 * It may return NULL if the skb requires no segmentation. This is
1813 * only possible when GSO is used for verifying header integrity.
1815 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1817 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1818 struct packet_type *ptype;
1819 __be16 type = skb->protocol;
1820 int vlan_depth = ETH_HLEN;
1823 while (type == htons(ETH_P_8021Q)) {
1824 struct vlan_hdr *vh;
1826 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1827 return ERR_PTR(-EINVAL);
1829 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1830 type = vh->h_vlan_encapsulated_proto;
1831 vlan_depth += VLAN_HLEN;
1834 skb_reset_mac_header(skb);
1835 skb->mac_len = skb->network_header - skb->mac_header;
1836 __skb_pull(skb, skb->mac_len);
1838 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1839 struct net_device *dev = skb->dev;
1840 struct ethtool_drvinfo info = {};
1842 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1843 dev->ethtool_ops->get_drvinfo(dev, &info);
1845 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1846 info.driver, dev ? dev->features : 0L,
1847 skb->sk ? skb->sk->sk_route_caps : 0L,
1848 skb->len, skb->data_len, skb->ip_summed);
1850 if (skb_header_cloned(skb) &&
1851 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1852 return ERR_PTR(err);
1856 list_for_each_entry_rcu(ptype,
1857 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1858 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1859 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1860 err = ptype->gso_send_check(skb);
1861 segs = ERR_PTR(err);
1862 if (err || skb_gso_ok(skb, features))
1864 __skb_push(skb, (skb->data -
1865 skb_network_header(skb)));
1867 segs = ptype->gso_segment(skb, features);
1873 __skb_push(skb, skb->data - skb_mac_header(skb));
1877 EXPORT_SYMBOL(skb_gso_segment);
1879 /* Take action when hardware reception checksum errors are detected. */
1881 void netdev_rx_csum_fault(struct net_device *dev)
1883 if (net_ratelimit()) {
1884 printk(KERN_ERR "%s: hw csum failure.\n",
1885 dev ? dev->name : "<unknown>");
1889 EXPORT_SYMBOL(netdev_rx_csum_fault);
1892 /* Actually, we should eliminate this check as soon as we know, that:
1893 * 1. IOMMU is present and allows to map all the memory.
1894 * 2. No high memory really exists on this machine.
1897 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1899 #ifdef CONFIG_HIGHMEM
1901 if (!(dev->features & NETIF_F_HIGHDMA)) {
1902 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1903 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1907 if (PCI_DMA_BUS_IS_PHYS) {
1908 struct device *pdev = dev->dev.parent;
1912 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1913 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1914 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1923 void (*destructor)(struct sk_buff *skb);
1926 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1928 static void dev_gso_skb_destructor(struct sk_buff *skb)
1930 struct dev_gso_cb *cb;
1933 struct sk_buff *nskb = skb->next;
1935 skb->next = nskb->next;
1938 } while (skb->next);
1940 cb = DEV_GSO_CB(skb);
1942 cb->destructor(skb);
1946 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1947 * @skb: buffer to segment
1948 * @features: device features as applicable to this skb
1950 * This function segments the given skb and stores the list of segments
1953 static int dev_gso_segment(struct sk_buff *skb, int features)
1955 struct sk_buff *segs;
1957 segs = skb_gso_segment(skb, features);
1959 /* Verifying header integrity only. */
1964 return PTR_ERR(segs);
1967 DEV_GSO_CB(skb)->destructor = skb->destructor;
1968 skb->destructor = dev_gso_skb_destructor;
1974 * Try to orphan skb early, right before transmission by the device.
1975 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1976 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1978 static inline void skb_orphan_try(struct sk_buff *skb)
1980 struct sock *sk = skb->sk;
1982 if (sk && !skb_shinfo(skb)->tx_flags) {
1983 /* skb_tx_hash() wont be able to get sk.
1984 * We copy sk_hash into skb->rxhash
1987 skb->rxhash = sk->sk_hash;
1992 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1994 return ((features & NETIF_F_GEN_CSUM) ||
1995 ((features & NETIF_F_V4_CSUM) &&
1996 protocol == htons(ETH_P_IP)) ||
1997 ((features & NETIF_F_V6_CSUM) &&
1998 protocol == htons(ETH_P_IPV6)) ||
1999 ((features & NETIF_F_FCOE_CRC) &&
2000 protocol == htons(ETH_P_FCOE)));
2003 static int harmonize_features(struct sk_buff *skb, __be16 protocol, int features)
2005 if (!can_checksum_protocol(features, protocol)) {
2006 features &= ~NETIF_F_ALL_CSUM;
2007 features &= ~NETIF_F_SG;
2008 } else if (illegal_highdma(skb->dev, skb)) {
2009 features &= ~NETIF_F_SG;
2015 int netif_skb_features(struct sk_buff *skb)
2017 __be16 protocol = skb->protocol;
2018 int features = skb->dev->features;
2020 if (protocol == htons(ETH_P_8021Q)) {
2021 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2022 protocol = veh->h_vlan_encapsulated_proto;
2023 } else if (!vlan_tx_tag_present(skb)) {
2024 return harmonize_features(skb, protocol, features);
2027 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2029 if (protocol != htons(ETH_P_8021Q)) {
2030 return harmonize_features(skb, protocol, features);
2032 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2033 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2034 return harmonize_features(skb, protocol, features);
2037 EXPORT_SYMBOL(netif_skb_features);
2040 * Returns true if either:
2041 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2042 * 2. skb is fragmented and the device does not support SG, or if
2043 * at least one of fragments is in highmem and device does not
2044 * support DMA from it.
2046 static inline int skb_needs_linearize(struct sk_buff *skb,
2049 return skb_is_nonlinear(skb) &&
2050 ((skb_has_frag_list(skb) &&
2051 !(features & NETIF_F_FRAGLIST)) ||
2052 (skb_shinfo(skb)->nr_frags &&
2053 !(features & NETIF_F_SG)));
2056 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2057 struct netdev_queue *txq)
2059 const struct net_device_ops *ops = dev->netdev_ops;
2060 int rc = NETDEV_TX_OK;
2062 if (likely(!skb->next)) {
2066 * If device doesnt need skb->dst, release it right now while
2067 * its hot in this cpu cache
2069 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2072 if (!list_empty(&ptype_all))
2073 dev_queue_xmit_nit(skb, dev);
2075 skb_orphan_try(skb);
2077 features = netif_skb_features(skb);
2079 if (vlan_tx_tag_present(skb) &&
2080 !(features & NETIF_F_HW_VLAN_TX)) {
2081 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2088 if (netif_needs_gso(skb, features)) {
2089 if (unlikely(dev_gso_segment(skb, features)))
2094 if (skb_needs_linearize(skb, features) &&
2095 __skb_linearize(skb))
2098 /* If packet is not checksummed and device does not
2099 * support checksumming for this protocol, complete
2100 * checksumming here.
2102 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2103 skb_set_transport_header(skb,
2104 skb_checksum_start_offset(skb));
2105 if (!(features & NETIF_F_ALL_CSUM) &&
2106 skb_checksum_help(skb))
2111 rc = ops->ndo_start_xmit(skb, dev);
2112 trace_net_dev_xmit(skb, rc);
2113 if (rc == NETDEV_TX_OK)
2114 txq_trans_update(txq);
2120 struct sk_buff *nskb = skb->next;
2122 skb->next = nskb->next;
2126 * If device doesnt need nskb->dst, release it right now while
2127 * its hot in this cpu cache
2129 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2132 rc = ops->ndo_start_xmit(nskb, dev);
2133 trace_net_dev_xmit(nskb, rc);
2134 if (unlikely(rc != NETDEV_TX_OK)) {
2135 if (rc & ~NETDEV_TX_MASK)
2136 goto out_kfree_gso_skb;
2137 nskb->next = skb->next;
2141 txq_trans_update(txq);
2142 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2143 return NETDEV_TX_BUSY;
2144 } while (skb->next);
2147 if (likely(skb->next == NULL))
2148 skb->destructor = DEV_GSO_CB(skb)->destructor;
2155 static u32 hashrnd __read_mostly;
2158 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2159 * to be used as a distribution range.
2161 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2162 unsigned int num_tx_queues)
2166 if (skb_rx_queue_recorded(skb)) {
2167 hash = skb_get_rx_queue(skb);
2168 while (unlikely(hash >= num_tx_queues))
2169 hash -= num_tx_queues;
2173 if (skb->sk && skb->sk->sk_hash)
2174 hash = skb->sk->sk_hash;
2176 hash = (__force u16) skb->protocol ^ skb->rxhash;
2177 hash = jhash_1word(hash, hashrnd);
2179 return (u16) (((u64) hash * num_tx_queues) >> 32);
2181 EXPORT_SYMBOL(__skb_tx_hash);
2183 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2185 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2186 if (net_ratelimit()) {
2187 pr_warning("%s selects TX queue %d, but "
2188 "real number of TX queues is %d\n",
2189 dev->name, queue_index, dev->real_num_tx_queues);
2196 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2199 struct xps_dev_maps *dev_maps;
2200 struct xps_map *map;
2201 int queue_index = -1;
2204 dev_maps = rcu_dereference(dev->xps_maps);
2206 map = rcu_dereference(
2207 dev_maps->cpu_map[raw_smp_processor_id()]);
2210 queue_index = map->queues[0];
2213 if (skb->sk && skb->sk->sk_hash)
2214 hash = skb->sk->sk_hash;
2216 hash = (__force u16) skb->protocol ^
2218 hash = jhash_1word(hash, hashrnd);
2219 queue_index = map->queues[
2220 ((u64)hash * map->len) >> 32];
2222 if (unlikely(queue_index >= dev->real_num_tx_queues))
2234 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2235 struct sk_buff *skb)
2238 const struct net_device_ops *ops = dev->netdev_ops;
2240 if (dev->real_num_tx_queues == 1)
2242 else if (ops->ndo_select_queue) {
2243 queue_index = ops->ndo_select_queue(dev, skb);
2244 queue_index = dev_cap_txqueue(dev, queue_index);
2246 struct sock *sk = skb->sk;
2247 queue_index = sk_tx_queue_get(sk);
2249 if (queue_index < 0 || skb->ooo_okay ||
2250 queue_index >= dev->real_num_tx_queues) {
2251 int old_index = queue_index;
2253 queue_index = get_xps_queue(dev, skb);
2254 if (queue_index < 0)
2255 queue_index = skb_tx_hash(dev, skb);
2257 if (queue_index != old_index && sk) {
2258 struct dst_entry *dst =
2259 rcu_dereference_check(sk->sk_dst_cache, 1);
2261 if (dst && skb_dst(skb) == dst)
2262 sk_tx_queue_set(sk, queue_index);
2267 skb_set_queue_mapping(skb, queue_index);
2268 return netdev_get_tx_queue(dev, queue_index);
2271 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2272 struct net_device *dev,
2273 struct netdev_queue *txq)
2275 spinlock_t *root_lock = qdisc_lock(q);
2276 bool contended = qdisc_is_running(q);
2280 * Heuristic to force contended enqueues to serialize on a
2281 * separate lock before trying to get qdisc main lock.
2282 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2283 * and dequeue packets faster.
2285 if (unlikely(contended))
2286 spin_lock(&q->busylock);
2288 spin_lock(root_lock);
2289 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2292 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2293 qdisc_run_begin(q)) {
2295 * This is a work-conserving queue; there are no old skbs
2296 * waiting to be sent out; and the qdisc is not running -
2297 * xmit the skb directly.
2299 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2302 qdisc_skb_cb(skb)->pkt_len = skb->len;
2303 qdisc_bstats_update(q, skb);
2305 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2306 if (unlikely(contended)) {
2307 spin_unlock(&q->busylock);
2314 rc = NET_XMIT_SUCCESS;
2317 rc = qdisc_enqueue_root(skb, q);
2318 if (qdisc_run_begin(q)) {
2319 if (unlikely(contended)) {
2320 spin_unlock(&q->busylock);
2326 spin_unlock(root_lock);
2327 if (unlikely(contended))
2328 spin_unlock(&q->busylock);
2332 static DEFINE_PER_CPU(int, xmit_recursion);
2333 #define RECURSION_LIMIT 10
2336 * dev_queue_xmit - transmit a buffer
2337 * @skb: buffer to transmit
2339 * Queue a buffer for transmission to a network device. The caller must
2340 * have set the device and priority and built the buffer before calling
2341 * this function. The function can be called from an interrupt.
2343 * A negative errno code is returned on a failure. A success does not
2344 * guarantee the frame will be transmitted as it may be dropped due
2345 * to congestion or traffic shaping.
2347 * -----------------------------------------------------------------------------------
2348 * I notice this method can also return errors from the queue disciplines,
2349 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2352 * Regardless of the return value, the skb is consumed, so it is currently
2353 * difficult to retry a send to this method. (You can bump the ref count
2354 * before sending to hold a reference for retry if you are careful.)
2356 * When calling this method, interrupts MUST be enabled. This is because
2357 * the BH enable code must have IRQs enabled so that it will not deadlock.
2360 int dev_queue_xmit(struct sk_buff *skb)
2362 struct net_device *dev = skb->dev;
2363 struct netdev_queue *txq;
2367 /* Disable soft irqs for various locks below. Also
2368 * stops preemption for RCU.
2372 txq = dev_pick_tx(dev, skb);
2373 q = rcu_dereference_bh(txq->qdisc);
2375 #ifdef CONFIG_NET_CLS_ACT
2376 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2378 trace_net_dev_queue(skb);
2380 rc = __dev_xmit_skb(skb, q, dev, txq);
2384 /* The device has no queue. Common case for software devices:
2385 loopback, all the sorts of tunnels...
2387 Really, it is unlikely that netif_tx_lock protection is necessary
2388 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2390 However, it is possible, that they rely on protection
2393 Check this and shot the lock. It is not prone from deadlocks.
2394 Either shot noqueue qdisc, it is even simpler 8)
2396 if (dev->flags & IFF_UP) {
2397 int cpu = smp_processor_id(); /* ok because BHs are off */
2399 if (txq->xmit_lock_owner != cpu) {
2401 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2402 goto recursion_alert;
2404 HARD_TX_LOCK(dev, txq, cpu);
2406 if (!netif_tx_queue_stopped(txq)) {
2407 __this_cpu_inc(xmit_recursion);
2408 rc = dev_hard_start_xmit(skb, dev, txq);
2409 __this_cpu_dec(xmit_recursion);
2410 if (dev_xmit_complete(rc)) {
2411 HARD_TX_UNLOCK(dev, txq);
2415 HARD_TX_UNLOCK(dev, txq);
2416 if (net_ratelimit())
2417 printk(KERN_CRIT "Virtual device %s asks to "
2418 "queue packet!\n", dev->name);
2420 /* Recursion is detected! It is possible,
2424 if (net_ratelimit())
2425 printk(KERN_CRIT "Dead loop on virtual device "
2426 "%s, fix it urgently!\n", dev->name);
2431 rcu_read_unlock_bh();
2436 rcu_read_unlock_bh();
2439 EXPORT_SYMBOL(dev_queue_xmit);
2442 /*=======================================================================
2444 =======================================================================*/
2446 int netdev_max_backlog __read_mostly = 1000;
2447 int netdev_tstamp_prequeue __read_mostly = 1;
2448 int netdev_budget __read_mostly = 300;
2449 int weight_p __read_mostly = 64; /* old backlog weight */
2451 /* Called with irq disabled */
2452 static inline void ____napi_schedule(struct softnet_data *sd,
2453 struct napi_struct *napi)
2455 list_add_tail(&napi->poll_list, &sd->poll_list);
2456 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2460 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2461 * and src/dst port numbers. Returns a non-zero hash number on success
2464 __u32 __skb_get_rxhash(struct sk_buff *skb)
2466 int nhoff, hash = 0, poff;
2467 struct ipv6hdr *ip6;
2470 u32 addr1, addr2, ihl;
2476 nhoff = skb_network_offset(skb);
2478 switch (skb->protocol) {
2479 case __constant_htons(ETH_P_IP):
2480 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2483 ip = (struct iphdr *) (skb->data + nhoff);
2484 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2487 ip_proto = ip->protocol;
2488 addr1 = (__force u32) ip->saddr;
2489 addr2 = (__force u32) ip->daddr;
2492 case __constant_htons(ETH_P_IPV6):
2493 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2496 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2497 ip_proto = ip6->nexthdr;
2498 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2499 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2507 poff = proto_ports_offset(ip_proto);
2509 nhoff += ihl * 4 + poff;
2510 if (pskb_may_pull(skb, nhoff + 4)) {
2511 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2512 if (ports.v16[1] < ports.v16[0])
2513 swap(ports.v16[0], ports.v16[1]);
2517 /* get a consistent hash (same value on both flow directions) */
2521 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2528 EXPORT_SYMBOL(__skb_get_rxhash);
2532 /* One global table that all flow-based protocols share. */
2533 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2534 EXPORT_SYMBOL(rps_sock_flow_table);
2537 * get_rps_cpu is called from netif_receive_skb and returns the target
2538 * CPU from the RPS map of the receiving queue for a given skb.
2539 * rcu_read_lock must be held on entry.
2541 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2542 struct rps_dev_flow **rflowp)
2544 struct netdev_rx_queue *rxqueue;
2545 struct rps_map *map;
2546 struct rps_dev_flow_table *flow_table;
2547 struct rps_sock_flow_table *sock_flow_table;
2551 if (skb_rx_queue_recorded(skb)) {
2552 u16 index = skb_get_rx_queue(skb);
2553 if (unlikely(index >= dev->real_num_rx_queues)) {
2554 WARN_ONCE(dev->real_num_rx_queues > 1,
2555 "%s received packet on queue %u, but number "
2556 "of RX queues is %u\n",
2557 dev->name, index, dev->real_num_rx_queues);
2560 rxqueue = dev->_rx + index;
2564 map = rcu_dereference(rxqueue->rps_map);
2566 if (map->len == 1) {
2567 tcpu = map->cpus[0];
2568 if (cpu_online(tcpu))
2572 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2576 skb_reset_network_header(skb);
2577 if (!skb_get_rxhash(skb))
2580 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2581 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2582 if (flow_table && sock_flow_table) {
2584 struct rps_dev_flow *rflow;
2586 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2589 next_cpu = sock_flow_table->ents[skb->rxhash &
2590 sock_flow_table->mask];
2593 * If the desired CPU (where last recvmsg was done) is
2594 * different from current CPU (one in the rx-queue flow
2595 * table entry), switch if one of the following holds:
2596 * - Current CPU is unset (equal to RPS_NO_CPU).
2597 * - Current CPU is offline.
2598 * - The current CPU's queue tail has advanced beyond the
2599 * last packet that was enqueued using this table entry.
2600 * This guarantees that all previous packets for the flow
2601 * have been dequeued, thus preserving in order delivery.
2603 if (unlikely(tcpu != next_cpu) &&
2604 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2605 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2606 rflow->last_qtail)) >= 0)) {
2607 tcpu = rflow->cpu = next_cpu;
2608 if (tcpu != RPS_NO_CPU)
2609 rflow->last_qtail = per_cpu(softnet_data,
2610 tcpu).input_queue_head;
2612 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2620 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2622 if (cpu_online(tcpu)) {
2632 /* Called from hardirq (IPI) context */
2633 static void rps_trigger_softirq(void *data)
2635 struct softnet_data *sd = data;
2637 ____napi_schedule(sd, &sd->backlog);
2641 #endif /* CONFIG_RPS */
2644 * Check if this softnet_data structure is another cpu one
2645 * If yes, queue it to our IPI list and return 1
2648 static int rps_ipi_queued(struct softnet_data *sd)
2651 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2654 sd->rps_ipi_next = mysd->rps_ipi_list;
2655 mysd->rps_ipi_list = sd;
2657 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2660 #endif /* CONFIG_RPS */
2665 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2666 * queue (may be a remote CPU queue).
2668 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2669 unsigned int *qtail)
2671 struct softnet_data *sd;
2672 unsigned long flags;
2674 sd = &per_cpu(softnet_data, cpu);
2676 local_irq_save(flags);
2679 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2680 if (skb_queue_len(&sd->input_pkt_queue)) {
2682 __skb_queue_tail(&sd->input_pkt_queue, skb);
2683 input_queue_tail_incr_save(sd, qtail);
2685 local_irq_restore(flags);
2686 return NET_RX_SUCCESS;
2689 /* Schedule NAPI for backlog device
2690 * We can use non atomic operation since we own the queue lock
2692 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2693 if (!rps_ipi_queued(sd))
2694 ____napi_schedule(sd, &sd->backlog);
2702 local_irq_restore(flags);
2704 atomic_long_inc(&skb->dev->rx_dropped);
2710 * netif_rx - post buffer to the network code
2711 * @skb: buffer to post
2713 * This function receives a packet from a device driver and queues it for
2714 * the upper (protocol) levels to process. It always succeeds. The buffer
2715 * may be dropped during processing for congestion control or by the
2719 * NET_RX_SUCCESS (no congestion)
2720 * NET_RX_DROP (packet was dropped)
2724 int netif_rx(struct sk_buff *skb)
2728 /* if netpoll wants it, pretend we never saw it */
2729 if (netpoll_rx(skb))
2732 if (netdev_tstamp_prequeue)
2733 net_timestamp_check(skb);
2735 trace_netif_rx(skb);
2738 struct rps_dev_flow voidflow, *rflow = &voidflow;
2744 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2746 cpu = smp_processor_id();
2748 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2756 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2762 EXPORT_SYMBOL(netif_rx);
2764 int netif_rx_ni(struct sk_buff *skb)
2769 err = netif_rx(skb);
2770 if (local_softirq_pending())
2776 EXPORT_SYMBOL(netif_rx_ni);
2778 static void net_tx_action(struct softirq_action *h)
2780 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2782 if (sd->completion_queue) {
2783 struct sk_buff *clist;
2785 local_irq_disable();
2786 clist = sd->completion_queue;
2787 sd->completion_queue = NULL;
2791 struct sk_buff *skb = clist;
2792 clist = clist->next;
2794 WARN_ON(atomic_read(&skb->users));
2795 trace_kfree_skb(skb, net_tx_action);
2800 if (sd->output_queue) {
2803 local_irq_disable();
2804 head = sd->output_queue;
2805 sd->output_queue = NULL;
2806 sd->output_queue_tailp = &sd->output_queue;
2810 struct Qdisc *q = head;
2811 spinlock_t *root_lock;
2813 head = head->next_sched;
2815 root_lock = qdisc_lock(q);
2816 if (spin_trylock(root_lock)) {
2817 smp_mb__before_clear_bit();
2818 clear_bit(__QDISC_STATE_SCHED,
2821 spin_unlock(root_lock);
2823 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2825 __netif_reschedule(q);
2827 smp_mb__before_clear_bit();
2828 clear_bit(__QDISC_STATE_SCHED,
2836 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2837 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2838 /* This hook is defined here for ATM LANE */
2839 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2840 unsigned char *addr) __read_mostly;
2841 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2844 #ifdef CONFIG_NET_CLS_ACT
2845 /* TODO: Maybe we should just force sch_ingress to be compiled in
2846 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2847 * a compare and 2 stores extra right now if we dont have it on
2848 * but have CONFIG_NET_CLS_ACT
2849 * NOTE: This doesnt stop any functionality; if you dont have
2850 * the ingress scheduler, you just cant add policies on ingress.
2853 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2855 struct net_device *dev = skb->dev;
2856 u32 ttl = G_TC_RTTL(skb->tc_verd);
2857 int result = TC_ACT_OK;
2860 if (unlikely(MAX_RED_LOOP < ttl++)) {
2861 if (net_ratelimit())
2862 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2863 skb->skb_iif, dev->ifindex);
2867 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2868 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2871 if (q != &noop_qdisc) {
2872 spin_lock(qdisc_lock(q));
2873 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2874 result = qdisc_enqueue_root(skb, q);
2875 spin_unlock(qdisc_lock(q));
2881 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2882 struct packet_type **pt_prev,
2883 int *ret, struct net_device *orig_dev)
2885 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
2887 if (!rxq || rxq->qdisc == &noop_qdisc)
2891 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2895 switch (ing_filter(skb, rxq)) {
2909 * netdev_rx_handler_register - register receive handler
2910 * @dev: device to register a handler for
2911 * @rx_handler: receive handler to register
2912 * @rx_handler_data: data pointer that is used by rx handler
2914 * Register a receive hander for a device. This handler will then be
2915 * called from __netif_receive_skb. A negative errno code is returned
2918 * The caller must hold the rtnl_mutex.
2920 int netdev_rx_handler_register(struct net_device *dev,
2921 rx_handler_func_t *rx_handler,
2922 void *rx_handler_data)
2926 if (dev->rx_handler)
2929 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2930 rcu_assign_pointer(dev->rx_handler, rx_handler);
2934 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2937 * netdev_rx_handler_unregister - unregister receive handler
2938 * @dev: device to unregister a handler from
2940 * Unregister a receive hander from a device.
2942 * The caller must hold the rtnl_mutex.
2944 void netdev_rx_handler_unregister(struct net_device *dev)
2948 rcu_assign_pointer(dev->rx_handler, NULL);
2949 rcu_assign_pointer(dev->rx_handler_data, NULL);
2951 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2953 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2954 struct net_device *master)
2956 if (skb->pkt_type == PACKET_HOST) {
2957 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2959 memcpy(dest, master->dev_addr, ETH_ALEN);
2963 /* On bonding slaves other than the currently active slave, suppress
2964 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2965 * ARP on active-backup slaves with arp_validate enabled.
2967 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2969 struct net_device *dev = skb->dev;
2971 if (master->priv_flags & IFF_MASTER_ARPMON)
2972 dev->last_rx = jiffies;
2974 if ((master->priv_flags & IFF_MASTER_ALB) &&
2975 (master->priv_flags & IFF_BRIDGE_PORT)) {
2976 /* Do address unmangle. The local destination address
2977 * will be always the one master has. Provides the right
2978 * functionality in a bridge.
2980 skb_bond_set_mac_by_master(skb, master);
2983 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2984 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2985 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2988 if (master->priv_flags & IFF_MASTER_ALB) {
2989 if (skb->pkt_type != PACKET_BROADCAST &&
2990 skb->pkt_type != PACKET_MULTICAST)
2993 if (master->priv_flags & IFF_MASTER_8023AD &&
2994 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
3001 EXPORT_SYMBOL(__skb_bond_should_drop);
3003 static int __netif_receive_skb(struct sk_buff *skb)
3005 struct packet_type *ptype, *pt_prev;
3006 rx_handler_func_t *rx_handler;
3007 struct net_device *orig_dev;
3008 struct net_device *master;
3009 struct net_device *null_or_orig;
3010 struct net_device *orig_or_bond;
3011 int ret = NET_RX_DROP;
3014 if (!netdev_tstamp_prequeue)
3015 net_timestamp_check(skb);
3017 trace_netif_receive_skb(skb);
3019 /* if we've gotten here through NAPI, check netpoll */
3020 if (netpoll_receive_skb(skb))
3024 skb->skb_iif = skb->dev->ifindex;
3027 * bonding note: skbs received on inactive slaves should only
3028 * be delivered to pkt handlers that are exact matches. Also
3029 * the deliver_no_wcard flag will be set. If packet handlers
3030 * are sensitive to duplicate packets these skbs will need to
3031 * be dropped at the handler.
3033 null_or_orig = NULL;
3034 orig_dev = skb->dev;
3035 master = ACCESS_ONCE(orig_dev->master);
3036 if (skb->deliver_no_wcard)
3037 null_or_orig = orig_dev;
3039 if (skb_bond_should_drop(skb, master)) {
3040 skb->deliver_no_wcard = 1;
3041 null_or_orig = orig_dev; /* deliver only exact match */
3046 __this_cpu_inc(softnet_data.processed);
3047 skb_reset_network_header(skb);
3048 skb_reset_transport_header(skb);
3049 skb->mac_len = skb->network_header - skb->mac_header;
3055 #ifdef CONFIG_NET_CLS_ACT
3056 if (skb->tc_verd & TC_NCLS) {
3057 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3062 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3063 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
3064 ptype->dev == orig_dev) {
3066 ret = deliver_skb(skb, pt_prev, orig_dev);
3071 #ifdef CONFIG_NET_CLS_ACT
3072 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3078 /* Handle special case of bridge or macvlan */
3079 rx_handler = rcu_dereference(skb->dev->rx_handler);
3082 ret = deliver_skb(skb, pt_prev, orig_dev);
3085 skb = rx_handler(skb);
3090 if (vlan_tx_tag_present(skb)) {
3092 ret = deliver_skb(skb, pt_prev, orig_dev);
3095 if (vlan_hwaccel_do_receive(&skb)) {
3096 ret = __netif_receive_skb(skb);
3098 } else if (unlikely(!skb))
3103 * Make sure frames received on VLAN interfaces stacked on
3104 * bonding interfaces still make their way to any base bonding
3105 * device that may have registered for a specific ptype. The
3106 * handler may have to adjust skb->dev and orig_dev.
3108 orig_or_bond = orig_dev;
3109 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
3110 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
3111 orig_or_bond = vlan_dev_real_dev(skb->dev);
3114 type = skb->protocol;
3115 list_for_each_entry_rcu(ptype,
3116 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3117 if (ptype->type == type && (ptype->dev == null_or_orig ||
3118 ptype->dev == skb->dev || ptype->dev == orig_dev ||
3119 ptype->dev == orig_or_bond)) {
3121 ret = deliver_skb(skb, pt_prev, orig_dev);
3127 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3129 atomic_long_inc(&skb->dev->rx_dropped);
3131 /* Jamal, now you will not able to escape explaining
3132 * me how you were going to use this. :-)
3143 * netif_receive_skb - process receive buffer from network
3144 * @skb: buffer to process
3146 * netif_receive_skb() is the main receive data processing function.
3147 * It always succeeds. The buffer may be dropped during processing
3148 * for congestion control or by the protocol layers.
3150 * This function may only be called from softirq context and interrupts
3151 * should be enabled.
3153 * Return values (usually ignored):
3154 * NET_RX_SUCCESS: no congestion
3155 * NET_RX_DROP: packet was dropped
3157 int netif_receive_skb(struct sk_buff *skb)
3159 if (netdev_tstamp_prequeue)
3160 net_timestamp_check(skb);
3162 if (skb_defer_rx_timestamp(skb))
3163 return NET_RX_SUCCESS;
3167 struct rps_dev_flow voidflow, *rflow = &voidflow;
3172 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3175 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3179 ret = __netif_receive_skb(skb);
3185 return __netif_receive_skb(skb);
3188 EXPORT_SYMBOL(netif_receive_skb);
3190 /* Network device is going away, flush any packets still pending
3191 * Called with irqs disabled.
3193 static void flush_backlog(void *arg)
3195 struct net_device *dev = arg;
3196 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3197 struct sk_buff *skb, *tmp;
3200 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3201 if (skb->dev == dev) {
3202 __skb_unlink(skb, &sd->input_pkt_queue);
3204 input_queue_head_incr(sd);
3209 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3210 if (skb->dev == dev) {
3211 __skb_unlink(skb, &sd->process_queue);
3213 input_queue_head_incr(sd);
3218 static int napi_gro_complete(struct sk_buff *skb)
3220 struct packet_type *ptype;
3221 __be16 type = skb->protocol;
3222 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3225 if (NAPI_GRO_CB(skb)->count == 1) {
3226 skb_shinfo(skb)->gso_size = 0;
3231 list_for_each_entry_rcu(ptype, head, list) {
3232 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3235 err = ptype->gro_complete(skb);
3241 WARN_ON(&ptype->list == head);
3243 return NET_RX_SUCCESS;
3247 return netif_receive_skb(skb);
3250 inline void napi_gro_flush(struct napi_struct *napi)
3252 struct sk_buff *skb, *next;
3254 for (skb = napi->gro_list; skb; skb = next) {
3257 napi_gro_complete(skb);
3260 napi->gro_count = 0;
3261 napi->gro_list = NULL;
3263 EXPORT_SYMBOL(napi_gro_flush);
3265 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3267 struct sk_buff **pp = NULL;
3268 struct packet_type *ptype;
3269 __be16 type = skb->protocol;
3270 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3273 enum gro_result ret;
3275 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3278 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3282 list_for_each_entry_rcu(ptype, head, list) {
3283 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3286 skb_set_network_header(skb, skb_gro_offset(skb));
3287 mac_len = skb->network_header - skb->mac_header;
3288 skb->mac_len = mac_len;
3289 NAPI_GRO_CB(skb)->same_flow = 0;
3290 NAPI_GRO_CB(skb)->flush = 0;
3291 NAPI_GRO_CB(skb)->free = 0;
3293 pp = ptype->gro_receive(&napi->gro_list, skb);
3298 if (&ptype->list == head)
3301 same_flow = NAPI_GRO_CB(skb)->same_flow;
3302 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3305 struct sk_buff *nskb = *pp;
3309 napi_gro_complete(nskb);
3316 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3320 NAPI_GRO_CB(skb)->count = 1;
3321 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3322 skb->next = napi->gro_list;
3323 napi->gro_list = skb;
3327 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3328 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3330 BUG_ON(skb->end - skb->tail < grow);
3332 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3335 skb->data_len -= grow;
3337 skb_shinfo(skb)->frags[0].page_offset += grow;
3338 skb_shinfo(skb)->frags[0].size -= grow;
3340 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3341 put_page(skb_shinfo(skb)->frags[0].page);
3342 memmove(skb_shinfo(skb)->frags,
3343 skb_shinfo(skb)->frags + 1,
3344 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3355 EXPORT_SYMBOL(dev_gro_receive);
3357 static inline gro_result_t
3358 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3362 for (p = napi->gro_list; p; p = p->next) {
3363 unsigned long diffs;
3365 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3366 diffs |= p->vlan_tci ^ skb->vlan_tci;
3367 diffs |= compare_ether_header(skb_mac_header(p),
3368 skb_gro_mac_header(skb));
3369 NAPI_GRO_CB(p)->same_flow = !diffs;
3370 NAPI_GRO_CB(p)->flush = 0;
3373 return dev_gro_receive(napi, skb);
3376 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3380 if (netif_receive_skb(skb))
3385 case GRO_MERGED_FREE:
3396 EXPORT_SYMBOL(napi_skb_finish);
3398 void skb_gro_reset_offset(struct sk_buff *skb)
3400 NAPI_GRO_CB(skb)->data_offset = 0;
3401 NAPI_GRO_CB(skb)->frag0 = NULL;
3402 NAPI_GRO_CB(skb)->frag0_len = 0;
3404 if (skb->mac_header == skb->tail &&
3405 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3406 NAPI_GRO_CB(skb)->frag0 =
3407 page_address(skb_shinfo(skb)->frags[0].page) +
3408 skb_shinfo(skb)->frags[0].page_offset;
3409 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3412 EXPORT_SYMBOL(skb_gro_reset_offset);
3414 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3416 skb_gro_reset_offset(skb);
3418 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3420 EXPORT_SYMBOL(napi_gro_receive);
3422 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3424 __skb_pull(skb, skb_headlen(skb));
3425 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3427 skb->dev = napi->dev;
3432 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3434 struct sk_buff *skb = napi->skb;
3437 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3443 EXPORT_SYMBOL(napi_get_frags);
3445 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3451 skb->protocol = eth_type_trans(skb, skb->dev);
3453 if (ret == GRO_HELD)
3454 skb_gro_pull(skb, -ETH_HLEN);
3455 else if (netif_receive_skb(skb))
3460 case GRO_MERGED_FREE:
3461 napi_reuse_skb(napi, skb);
3470 EXPORT_SYMBOL(napi_frags_finish);
3472 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3474 struct sk_buff *skb = napi->skb;
3481 skb_reset_mac_header(skb);
3482 skb_gro_reset_offset(skb);
3484 off = skb_gro_offset(skb);
3485 hlen = off + sizeof(*eth);
3486 eth = skb_gro_header_fast(skb, off);
3487 if (skb_gro_header_hard(skb, hlen)) {
3488 eth = skb_gro_header_slow(skb, hlen, off);
3489 if (unlikely(!eth)) {
3490 napi_reuse_skb(napi, skb);
3496 skb_gro_pull(skb, sizeof(*eth));
3499 * This works because the only protocols we care about don't require
3500 * special handling. We'll fix it up properly at the end.
3502 skb->protocol = eth->h_proto;
3507 EXPORT_SYMBOL(napi_frags_skb);
3509 gro_result_t napi_gro_frags(struct napi_struct *napi)
3511 struct sk_buff *skb = napi_frags_skb(napi);
3516 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3518 EXPORT_SYMBOL(napi_gro_frags);
3521 * net_rps_action sends any pending IPI's for rps.
3522 * Note: called with local irq disabled, but exits with local irq enabled.
3524 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3527 struct softnet_data *remsd = sd->rps_ipi_list;
3530 sd->rps_ipi_list = NULL;
3534 /* Send pending IPI's to kick RPS processing on remote cpus. */
3536 struct softnet_data *next = remsd->rps_ipi_next;
3538 if (cpu_online(remsd->cpu))
3539 __smp_call_function_single(remsd->cpu,
3548 static int process_backlog(struct napi_struct *napi, int quota)
3551 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3554 /* Check if we have pending ipi, its better to send them now,
3555 * not waiting net_rx_action() end.
3557 if (sd->rps_ipi_list) {
3558 local_irq_disable();
3559 net_rps_action_and_irq_enable(sd);
3562 napi->weight = weight_p;
3563 local_irq_disable();
3564 while (work < quota) {
3565 struct sk_buff *skb;
3568 while ((skb = __skb_dequeue(&sd->process_queue))) {
3570 __netif_receive_skb(skb);
3571 local_irq_disable();
3572 input_queue_head_incr(sd);
3573 if (++work >= quota) {
3580 qlen = skb_queue_len(&sd->input_pkt_queue);
3582 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3583 &sd->process_queue);
3585 if (qlen < quota - work) {
3587 * Inline a custom version of __napi_complete().
3588 * only current cpu owns and manipulates this napi,
3589 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3590 * we can use a plain write instead of clear_bit(),
3591 * and we dont need an smp_mb() memory barrier.
3593 list_del(&napi->poll_list);
3596 quota = work + qlen;
3606 * __napi_schedule - schedule for receive
3607 * @n: entry to schedule
3609 * The entry's receive function will be scheduled to run
3611 void __napi_schedule(struct napi_struct *n)
3613 unsigned long flags;
3615 local_irq_save(flags);
3616 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3617 local_irq_restore(flags);
3619 EXPORT_SYMBOL(__napi_schedule);
3621 void __napi_complete(struct napi_struct *n)
3623 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3624 BUG_ON(n->gro_list);
3626 list_del(&n->poll_list);
3627 smp_mb__before_clear_bit();
3628 clear_bit(NAPI_STATE_SCHED, &n->state);
3630 EXPORT_SYMBOL(__napi_complete);
3632 void napi_complete(struct napi_struct *n)
3634 unsigned long flags;
3637 * don't let napi dequeue from the cpu poll list
3638 * just in case its running on a different cpu
3640 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3644 local_irq_save(flags);
3646 local_irq_restore(flags);
3648 EXPORT_SYMBOL(napi_complete);
3650 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3651 int (*poll)(struct napi_struct *, int), int weight)
3653 INIT_LIST_HEAD(&napi->poll_list);
3654 napi->gro_count = 0;
3655 napi->gro_list = NULL;
3658 napi->weight = weight;
3659 list_add(&napi->dev_list, &dev->napi_list);
3661 #ifdef CONFIG_NETPOLL
3662 spin_lock_init(&napi->poll_lock);
3663 napi->poll_owner = -1;
3665 set_bit(NAPI_STATE_SCHED, &napi->state);
3667 EXPORT_SYMBOL(netif_napi_add);
3669 void netif_napi_del(struct napi_struct *napi)
3671 struct sk_buff *skb, *next;
3673 list_del_init(&napi->dev_list);
3674 napi_free_frags(napi);
3676 for (skb = napi->gro_list; skb; skb = next) {
3682 napi->gro_list = NULL;
3683 napi->gro_count = 0;
3685 EXPORT_SYMBOL(netif_napi_del);
3687 static void net_rx_action(struct softirq_action *h)
3689 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3690 unsigned long time_limit = jiffies + 2;
3691 int budget = netdev_budget;
3694 local_irq_disable();
3696 while (!list_empty(&sd->poll_list)) {
3697 struct napi_struct *n;
3700 /* If softirq window is exhuasted then punt.
3701 * Allow this to run for 2 jiffies since which will allow
3702 * an average latency of 1.5/HZ.
3704 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3709 /* Even though interrupts have been re-enabled, this
3710 * access is safe because interrupts can only add new
3711 * entries to the tail of this list, and only ->poll()
3712 * calls can remove this head entry from the list.
3714 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3716 have = netpoll_poll_lock(n);
3720 /* This NAPI_STATE_SCHED test is for avoiding a race
3721 * with netpoll's poll_napi(). Only the entity which
3722 * obtains the lock and sees NAPI_STATE_SCHED set will
3723 * actually make the ->poll() call. Therefore we avoid
3724 * accidently calling ->poll() when NAPI is not scheduled.
3727 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3728 work = n->poll(n, weight);
3732 WARN_ON_ONCE(work > weight);
3736 local_irq_disable();
3738 /* Drivers must not modify the NAPI state if they
3739 * consume the entire weight. In such cases this code
3740 * still "owns" the NAPI instance and therefore can
3741 * move the instance around on the list at-will.
3743 if (unlikely(work == weight)) {
3744 if (unlikely(napi_disable_pending(n))) {
3747 local_irq_disable();
3749 list_move_tail(&n->poll_list, &sd->poll_list);
3752 netpoll_poll_unlock(have);
3755 net_rps_action_and_irq_enable(sd);
3757 #ifdef CONFIG_NET_DMA
3759 * There may not be any more sk_buffs coming right now, so push
3760 * any pending DMA copies to hardware
3762 dma_issue_pending_all();
3769 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3773 static gifconf_func_t *gifconf_list[NPROTO];
3776 * register_gifconf - register a SIOCGIF handler
3777 * @family: Address family
3778 * @gifconf: Function handler
3780 * Register protocol dependent address dumping routines. The handler
3781 * that is passed must not be freed or reused until it has been replaced
3782 * by another handler.
3784 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3786 if (family >= NPROTO)
3788 gifconf_list[family] = gifconf;
3791 EXPORT_SYMBOL(register_gifconf);
3795 * Map an interface index to its name (SIOCGIFNAME)
3799 * We need this ioctl for efficient implementation of the
3800 * if_indextoname() function required by the IPv6 API. Without
3801 * it, we would have to search all the interfaces to find a
3805 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3807 struct net_device *dev;
3811 * Fetch the caller's info block.
3814 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3818 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3824 strcpy(ifr.ifr_name, dev->name);
3827 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3833 * Perform a SIOCGIFCONF call. This structure will change
3834 * size eventually, and there is nothing I can do about it.
3835 * Thus we will need a 'compatibility mode'.
3838 static int dev_ifconf(struct net *net, char __user *arg)
3841 struct net_device *dev;
3848 * Fetch the caller's info block.
3851 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3858 * Loop over the interfaces, and write an info block for each.
3862 for_each_netdev(net, dev) {
3863 for (i = 0; i < NPROTO; i++) {
3864 if (gifconf_list[i]) {
3867 done = gifconf_list[i](dev, NULL, 0);
3869 done = gifconf_list[i](dev, pos + total,
3879 * All done. Write the updated control block back to the caller.
3881 ifc.ifc_len = total;
3884 * Both BSD and Solaris return 0 here, so we do too.
3886 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3889 #ifdef CONFIG_PROC_FS
3891 * This is invoked by the /proc filesystem handler to display a device
3894 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3897 struct net *net = seq_file_net(seq);
3899 struct net_device *dev;
3903 return SEQ_START_TOKEN;
3906 for_each_netdev_rcu(net, dev)
3913 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3915 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3916 first_net_device(seq_file_net(seq)) :
3917 next_net_device((struct net_device *)v);
3920 return rcu_dereference(dev);
3923 void dev_seq_stop(struct seq_file *seq, void *v)
3929 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3931 struct rtnl_link_stats64 temp;
3932 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3934 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3935 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3936 dev->name, stats->rx_bytes, stats->rx_packets,
3938 stats->rx_dropped + stats->rx_missed_errors,
3939 stats->rx_fifo_errors,
3940 stats->rx_length_errors + stats->rx_over_errors +
3941 stats->rx_crc_errors + stats->rx_frame_errors,
3942 stats->rx_compressed, stats->multicast,
3943 stats->tx_bytes, stats->tx_packets,
3944 stats->tx_errors, stats->tx_dropped,
3945 stats->tx_fifo_errors, stats->collisions,
3946 stats->tx_carrier_errors +
3947 stats->tx_aborted_errors +
3948 stats->tx_window_errors +
3949 stats->tx_heartbeat_errors,
3950 stats->tx_compressed);
3954 * Called from the PROCfs module. This now uses the new arbitrary sized
3955 * /proc/net interface to create /proc/net/dev
3957 static int dev_seq_show(struct seq_file *seq, void *v)
3959 if (v == SEQ_START_TOKEN)
3960 seq_puts(seq, "Inter-| Receive "
3962 " face |bytes packets errs drop fifo frame "
3963 "compressed multicast|bytes packets errs "
3964 "drop fifo colls carrier compressed\n");
3966 dev_seq_printf_stats(seq, v);
3970 static struct softnet_data *softnet_get_online(loff_t *pos)
3972 struct softnet_data *sd = NULL;
3974 while (*pos < nr_cpu_ids)
3975 if (cpu_online(*pos)) {
3976 sd = &per_cpu(softnet_data, *pos);
3983 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3985 return softnet_get_online(pos);
3988 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3991 return softnet_get_online(pos);
3994 static void softnet_seq_stop(struct seq_file *seq, void *v)
3998 static int softnet_seq_show(struct seq_file *seq, void *v)
4000 struct softnet_data *sd = v;
4002 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4003 sd->processed, sd->dropped, sd->time_squeeze, 0,
4004 0, 0, 0, 0, /* was fastroute */
4005 sd->cpu_collision, sd->received_rps);
4009 static const struct seq_operations dev_seq_ops = {
4010 .start = dev_seq_start,
4011 .next = dev_seq_next,
4012 .stop = dev_seq_stop,
4013 .show = dev_seq_show,
4016 static int dev_seq_open(struct inode *inode, struct file *file)
4018 return seq_open_net(inode, file, &dev_seq_ops,
4019 sizeof(struct seq_net_private));
4022 static const struct file_operations dev_seq_fops = {
4023 .owner = THIS_MODULE,
4024 .open = dev_seq_open,
4026 .llseek = seq_lseek,
4027 .release = seq_release_net,
4030 static const struct seq_operations softnet_seq_ops = {
4031 .start = softnet_seq_start,
4032 .next = softnet_seq_next,
4033 .stop = softnet_seq_stop,
4034 .show = softnet_seq_show,
4037 static int softnet_seq_open(struct inode *inode, struct file *file)
4039 return seq_open(file, &softnet_seq_ops);
4042 static const struct file_operations softnet_seq_fops = {
4043 .owner = THIS_MODULE,
4044 .open = softnet_seq_open,
4046 .llseek = seq_lseek,
4047 .release = seq_release,
4050 static void *ptype_get_idx(loff_t pos)
4052 struct packet_type *pt = NULL;
4056 list_for_each_entry_rcu(pt, &ptype_all, list) {
4062 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4063 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4072 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4076 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4079 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4081 struct packet_type *pt;
4082 struct list_head *nxt;
4086 if (v == SEQ_START_TOKEN)
4087 return ptype_get_idx(0);
4090 nxt = pt->list.next;
4091 if (pt->type == htons(ETH_P_ALL)) {
4092 if (nxt != &ptype_all)
4095 nxt = ptype_base[0].next;
4097 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4099 while (nxt == &ptype_base[hash]) {
4100 if (++hash >= PTYPE_HASH_SIZE)
4102 nxt = ptype_base[hash].next;
4105 return list_entry(nxt, struct packet_type, list);
4108 static void ptype_seq_stop(struct seq_file *seq, void *v)
4114 static int ptype_seq_show(struct seq_file *seq, void *v)
4116 struct packet_type *pt = v;
4118 if (v == SEQ_START_TOKEN)
4119 seq_puts(seq, "Type Device Function\n");
4120 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4121 if (pt->type == htons(ETH_P_ALL))
4122 seq_puts(seq, "ALL ");
4124 seq_printf(seq, "%04x", ntohs(pt->type));
4126 seq_printf(seq, " %-8s %pF\n",
4127 pt->dev ? pt->dev->name : "", pt->func);
4133 static const struct seq_operations ptype_seq_ops = {
4134 .start = ptype_seq_start,
4135 .next = ptype_seq_next,
4136 .stop = ptype_seq_stop,
4137 .show = ptype_seq_show,
4140 static int ptype_seq_open(struct inode *inode, struct file *file)
4142 return seq_open_net(inode, file, &ptype_seq_ops,
4143 sizeof(struct seq_net_private));
4146 static const struct file_operations ptype_seq_fops = {
4147 .owner = THIS_MODULE,
4148 .open = ptype_seq_open,
4150 .llseek = seq_lseek,
4151 .release = seq_release_net,
4155 static int __net_init dev_proc_net_init(struct net *net)
4159 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4161 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4163 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4166 if (wext_proc_init(net))
4172 proc_net_remove(net, "ptype");
4174 proc_net_remove(net, "softnet_stat");
4176 proc_net_remove(net, "dev");
4180 static void __net_exit dev_proc_net_exit(struct net *net)
4182 wext_proc_exit(net);
4184 proc_net_remove(net, "ptype");
4185 proc_net_remove(net, "softnet_stat");
4186 proc_net_remove(net, "dev");
4189 static struct pernet_operations __net_initdata dev_proc_ops = {
4190 .init = dev_proc_net_init,
4191 .exit = dev_proc_net_exit,
4194 static int __init dev_proc_init(void)
4196 return register_pernet_subsys(&dev_proc_ops);
4199 #define dev_proc_init() 0
4200 #endif /* CONFIG_PROC_FS */
4204 * netdev_set_master - set up master/slave pair
4205 * @slave: slave device
4206 * @master: new master device
4208 * Changes the master device of the slave. Pass %NULL to break the
4209 * bonding. The caller must hold the RTNL semaphore. On a failure
4210 * a negative errno code is returned. On success the reference counts
4211 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4212 * function returns zero.
4214 int netdev_set_master(struct net_device *slave, struct net_device *master)
4216 struct net_device *old = slave->master;
4226 slave->master = master;
4233 slave->flags |= IFF_SLAVE;
4235 slave->flags &= ~IFF_SLAVE;
4237 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4240 EXPORT_SYMBOL(netdev_set_master);
4242 static void dev_change_rx_flags(struct net_device *dev, int flags)
4244 const struct net_device_ops *ops = dev->netdev_ops;
4246 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4247 ops->ndo_change_rx_flags(dev, flags);
4250 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4252 unsigned short old_flags = dev->flags;
4258 dev->flags |= IFF_PROMISC;
4259 dev->promiscuity += inc;
4260 if (dev->promiscuity == 0) {
4263 * If inc causes overflow, untouch promisc and return error.
4266 dev->flags &= ~IFF_PROMISC;
4268 dev->promiscuity -= inc;
4269 printk(KERN_WARNING "%s: promiscuity touches roof, "
4270 "set promiscuity failed, promiscuity feature "
4271 "of device might be broken.\n", dev->name);
4275 if (dev->flags != old_flags) {
4276 printk(KERN_INFO "device %s %s promiscuous mode\n",
4277 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4279 if (audit_enabled) {
4280 current_uid_gid(&uid, &gid);
4281 audit_log(current->audit_context, GFP_ATOMIC,
4282 AUDIT_ANOM_PROMISCUOUS,
4283 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4284 dev->name, (dev->flags & IFF_PROMISC),
4285 (old_flags & IFF_PROMISC),
4286 audit_get_loginuid(current),
4288 audit_get_sessionid(current));
4291 dev_change_rx_flags(dev, IFF_PROMISC);
4297 * dev_set_promiscuity - update promiscuity count on a device
4301 * Add or remove promiscuity from a device. While the count in the device
4302 * remains above zero the interface remains promiscuous. Once it hits zero
4303 * the device reverts back to normal filtering operation. A negative inc
4304 * value is used to drop promiscuity on the device.
4305 * Return 0 if successful or a negative errno code on error.
4307 int dev_set_promiscuity(struct net_device *dev, int inc)
4309 unsigned short old_flags = dev->flags;
4312 err = __dev_set_promiscuity(dev, inc);
4315 if (dev->flags != old_flags)
4316 dev_set_rx_mode(dev);
4319 EXPORT_SYMBOL(dev_set_promiscuity);
4322 * dev_set_allmulti - update allmulti count on a device
4326 * Add or remove reception of all multicast frames to a device. While the
4327 * count in the device remains above zero the interface remains listening
4328 * to all interfaces. Once it hits zero the device reverts back to normal
4329 * filtering operation. A negative @inc value is used to drop the counter
4330 * when releasing a resource needing all multicasts.
4331 * Return 0 if successful or a negative errno code on error.
4334 int dev_set_allmulti(struct net_device *dev, int inc)
4336 unsigned short old_flags = dev->flags;
4340 dev->flags |= IFF_ALLMULTI;
4341 dev->allmulti += inc;
4342 if (dev->allmulti == 0) {
4345 * If inc causes overflow, untouch allmulti and return error.
4348 dev->flags &= ~IFF_ALLMULTI;
4350 dev->allmulti -= inc;
4351 printk(KERN_WARNING "%s: allmulti touches roof, "
4352 "set allmulti failed, allmulti feature of "
4353 "device might be broken.\n", dev->name);
4357 if (dev->flags ^ old_flags) {
4358 dev_change_rx_flags(dev, IFF_ALLMULTI);
4359 dev_set_rx_mode(dev);
4363 EXPORT_SYMBOL(dev_set_allmulti);
4366 * Upload unicast and multicast address lists to device and
4367 * configure RX filtering. When the device doesn't support unicast
4368 * filtering it is put in promiscuous mode while unicast addresses
4371 void __dev_set_rx_mode(struct net_device *dev)
4373 const struct net_device_ops *ops = dev->netdev_ops;
4375 /* dev_open will call this function so the list will stay sane. */
4376 if (!(dev->flags&IFF_UP))
4379 if (!netif_device_present(dev))
4382 if (ops->ndo_set_rx_mode)
4383 ops->ndo_set_rx_mode(dev);
4385 /* Unicast addresses changes may only happen under the rtnl,
4386 * therefore calling __dev_set_promiscuity here is safe.
4388 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4389 __dev_set_promiscuity(dev, 1);
4390 dev->uc_promisc = 1;
4391 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4392 __dev_set_promiscuity(dev, -1);
4393 dev->uc_promisc = 0;
4396 if (ops->ndo_set_multicast_list)
4397 ops->ndo_set_multicast_list(dev);
4401 void dev_set_rx_mode(struct net_device *dev)
4403 netif_addr_lock_bh(dev);
4404 __dev_set_rx_mode(dev);
4405 netif_addr_unlock_bh(dev);
4409 * dev_get_flags - get flags reported to userspace
4412 * Get the combination of flag bits exported through APIs to userspace.
4414 unsigned dev_get_flags(const struct net_device *dev)
4418 flags = (dev->flags & ~(IFF_PROMISC |
4423 (dev->gflags & (IFF_PROMISC |
4426 if (netif_running(dev)) {
4427 if (netif_oper_up(dev))
4428 flags |= IFF_RUNNING;
4429 if (netif_carrier_ok(dev))
4430 flags |= IFF_LOWER_UP;
4431 if (netif_dormant(dev))
4432 flags |= IFF_DORMANT;
4437 EXPORT_SYMBOL(dev_get_flags);
4439 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4441 int old_flags = dev->flags;
4447 * Set the flags on our device.
4450 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4451 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4453 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4457 * Load in the correct multicast list now the flags have changed.
4460 if ((old_flags ^ flags) & IFF_MULTICAST)
4461 dev_change_rx_flags(dev, IFF_MULTICAST);
4463 dev_set_rx_mode(dev);
4466 * Have we downed the interface. We handle IFF_UP ourselves
4467 * according to user attempts to set it, rather than blindly
4472 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4473 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4476 dev_set_rx_mode(dev);
4479 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4480 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4482 dev->gflags ^= IFF_PROMISC;
4483 dev_set_promiscuity(dev, inc);
4486 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4487 is important. Some (broken) drivers set IFF_PROMISC, when
4488 IFF_ALLMULTI is requested not asking us and not reporting.
4490 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4491 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4493 dev->gflags ^= IFF_ALLMULTI;
4494 dev_set_allmulti(dev, inc);
4500 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4502 unsigned int changes = dev->flags ^ old_flags;
4504 if (changes & IFF_UP) {
4505 if (dev->flags & IFF_UP)
4506 call_netdevice_notifiers(NETDEV_UP, dev);
4508 call_netdevice_notifiers(NETDEV_DOWN, dev);
4511 if (dev->flags & IFF_UP &&
4512 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4513 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4517 * dev_change_flags - change device settings
4519 * @flags: device state flags
4521 * Change settings on device based state flags. The flags are
4522 * in the userspace exported format.
4524 int dev_change_flags(struct net_device *dev, unsigned flags)
4527 int old_flags = dev->flags;
4529 ret = __dev_change_flags(dev, flags);
4533 changes = old_flags ^ dev->flags;
4535 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4537 __dev_notify_flags(dev, old_flags);
4540 EXPORT_SYMBOL(dev_change_flags);
4543 * dev_set_mtu - Change maximum transfer unit
4545 * @new_mtu: new transfer unit
4547 * Change the maximum transfer size of the network device.
4549 int dev_set_mtu(struct net_device *dev, int new_mtu)
4551 const struct net_device_ops *ops = dev->netdev_ops;
4554 if (new_mtu == dev->mtu)
4557 /* MTU must be positive. */
4561 if (!netif_device_present(dev))
4565 if (ops->ndo_change_mtu)
4566 err = ops->ndo_change_mtu(dev, new_mtu);
4570 if (!err && dev->flags & IFF_UP)
4571 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4574 EXPORT_SYMBOL(dev_set_mtu);
4577 * dev_set_mac_address - Change Media Access Control Address
4581 * Change the hardware (MAC) address of the device
4583 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4585 const struct net_device_ops *ops = dev->netdev_ops;
4588 if (!ops->ndo_set_mac_address)
4590 if (sa->sa_family != dev->type)
4592 if (!netif_device_present(dev))
4594 err = ops->ndo_set_mac_address(dev, sa);
4596 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4599 EXPORT_SYMBOL(dev_set_mac_address);
4602 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4604 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4607 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4613 case SIOCGIFFLAGS: /* Get interface flags */
4614 ifr->ifr_flags = (short) dev_get_flags(dev);
4617 case SIOCGIFMETRIC: /* Get the metric on the interface
4618 (currently unused) */
4619 ifr->ifr_metric = 0;
4622 case SIOCGIFMTU: /* Get the MTU of a device */
4623 ifr->ifr_mtu = dev->mtu;
4628 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4630 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4631 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4632 ifr->ifr_hwaddr.sa_family = dev->type;
4640 ifr->ifr_map.mem_start = dev->mem_start;
4641 ifr->ifr_map.mem_end = dev->mem_end;
4642 ifr->ifr_map.base_addr = dev->base_addr;
4643 ifr->ifr_map.irq = dev->irq;
4644 ifr->ifr_map.dma = dev->dma;
4645 ifr->ifr_map.port = dev->if_port;
4649 ifr->ifr_ifindex = dev->ifindex;
4653 ifr->ifr_qlen = dev->tx_queue_len;
4657 /* dev_ioctl() should ensure this case
4669 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4671 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4674 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4675 const struct net_device_ops *ops;
4680 ops = dev->netdev_ops;
4683 case SIOCSIFFLAGS: /* Set interface flags */
4684 return dev_change_flags(dev, ifr->ifr_flags);
4686 case SIOCSIFMETRIC: /* Set the metric on the interface
4687 (currently unused) */
4690 case SIOCSIFMTU: /* Set the MTU of a device */
4691 return dev_set_mtu(dev, ifr->ifr_mtu);
4694 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4696 case SIOCSIFHWBROADCAST:
4697 if (ifr->ifr_hwaddr.sa_family != dev->type)
4699 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4700 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4701 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4705 if (ops->ndo_set_config) {
4706 if (!netif_device_present(dev))
4708 return ops->ndo_set_config(dev, &ifr->ifr_map);
4713 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4714 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4716 if (!netif_device_present(dev))
4718 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4721 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4722 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4724 if (!netif_device_present(dev))
4726 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4729 if (ifr->ifr_qlen < 0)
4731 dev->tx_queue_len = ifr->ifr_qlen;
4735 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4736 return dev_change_name(dev, ifr->ifr_newname);
4739 * Unknown or private ioctl
4742 if ((cmd >= SIOCDEVPRIVATE &&
4743 cmd <= SIOCDEVPRIVATE + 15) ||
4744 cmd == SIOCBONDENSLAVE ||
4745 cmd == SIOCBONDRELEASE ||
4746 cmd == SIOCBONDSETHWADDR ||
4747 cmd == SIOCBONDSLAVEINFOQUERY ||
4748 cmd == SIOCBONDINFOQUERY ||
4749 cmd == SIOCBONDCHANGEACTIVE ||
4750 cmd == SIOCGMIIPHY ||
4751 cmd == SIOCGMIIREG ||
4752 cmd == SIOCSMIIREG ||
4753 cmd == SIOCBRADDIF ||
4754 cmd == SIOCBRDELIF ||
4755 cmd == SIOCSHWTSTAMP ||
4756 cmd == SIOCWANDEV) {
4758 if (ops->ndo_do_ioctl) {
4759 if (netif_device_present(dev))
4760 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4772 * This function handles all "interface"-type I/O control requests. The actual
4773 * 'doing' part of this is dev_ifsioc above.
4777 * dev_ioctl - network device ioctl
4778 * @net: the applicable net namespace
4779 * @cmd: command to issue
4780 * @arg: pointer to a struct ifreq in user space
4782 * Issue ioctl functions to devices. This is normally called by the
4783 * user space syscall interfaces but can sometimes be useful for
4784 * other purposes. The return value is the return from the syscall if
4785 * positive or a negative errno code on error.
4788 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4794 /* One special case: SIOCGIFCONF takes ifconf argument
4795 and requires shared lock, because it sleeps writing
4799 if (cmd == SIOCGIFCONF) {
4801 ret = dev_ifconf(net, (char __user *) arg);
4805 if (cmd == SIOCGIFNAME)
4806 return dev_ifname(net, (struct ifreq __user *)arg);
4808 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4811 ifr.ifr_name[IFNAMSIZ-1] = 0;
4813 colon = strchr(ifr.ifr_name, ':');
4818 * See which interface the caller is talking about.
4823 * These ioctl calls:
4824 * - can be done by all.
4825 * - atomic and do not require locking.
4836 dev_load(net, ifr.ifr_name);
4838 ret = dev_ifsioc_locked(net, &ifr, cmd);
4843 if (copy_to_user(arg, &ifr,
4844 sizeof(struct ifreq)))
4850 dev_load(net, ifr.ifr_name);
4852 ret = dev_ethtool(net, &ifr);
4857 if (copy_to_user(arg, &ifr,
4858 sizeof(struct ifreq)))
4864 * These ioctl calls:
4865 * - require superuser power.
4866 * - require strict serialization.
4872 if (!capable(CAP_NET_ADMIN))
4874 dev_load(net, ifr.ifr_name);
4876 ret = dev_ifsioc(net, &ifr, cmd);
4881 if (copy_to_user(arg, &ifr,
4882 sizeof(struct ifreq)))
4888 * These ioctl calls:
4889 * - require superuser power.
4890 * - require strict serialization.
4891 * - do not return a value
4901 case SIOCSIFHWBROADCAST:
4904 case SIOCBONDENSLAVE:
4905 case SIOCBONDRELEASE:
4906 case SIOCBONDSETHWADDR:
4907 case SIOCBONDCHANGEACTIVE:
4911 if (!capable(CAP_NET_ADMIN))
4914 case SIOCBONDSLAVEINFOQUERY:
4915 case SIOCBONDINFOQUERY:
4916 dev_load(net, ifr.ifr_name);
4918 ret = dev_ifsioc(net, &ifr, cmd);
4923 /* Get the per device memory space. We can add this but
4924 * currently do not support it */
4926 /* Set the per device memory buffer space.
4927 * Not applicable in our case */
4932 * Unknown or private ioctl.
4935 if (cmd == SIOCWANDEV ||
4936 (cmd >= SIOCDEVPRIVATE &&
4937 cmd <= SIOCDEVPRIVATE + 15)) {
4938 dev_load(net, ifr.ifr_name);
4940 ret = dev_ifsioc(net, &ifr, cmd);
4942 if (!ret && copy_to_user(arg, &ifr,
4943 sizeof(struct ifreq)))
4947 /* Take care of Wireless Extensions */
4948 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4949 return wext_handle_ioctl(net, &ifr, cmd, arg);
4956 * dev_new_index - allocate an ifindex
4957 * @net: the applicable net namespace
4959 * Returns a suitable unique value for a new device interface
4960 * number. The caller must hold the rtnl semaphore or the
4961 * dev_base_lock to be sure it remains unique.
4963 static int dev_new_index(struct net *net)
4969 if (!__dev_get_by_index(net, ifindex))
4974 /* Delayed registration/unregisteration */
4975 static LIST_HEAD(net_todo_list);
4977 static void net_set_todo(struct net_device *dev)
4979 list_add_tail(&dev->todo_list, &net_todo_list);
4982 static void rollback_registered_many(struct list_head *head)
4984 struct net_device *dev, *tmp;
4986 BUG_ON(dev_boot_phase);
4989 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4990 /* Some devices call without registering
4991 * for initialization unwind. Remove those
4992 * devices and proceed with the remaining.
4994 if (dev->reg_state == NETREG_UNINITIALIZED) {
4995 pr_debug("unregister_netdevice: device %s/%p never "
4996 "was registered\n", dev->name, dev);
4999 list_del(&dev->unreg_list);
5003 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5006 /* If device is running, close it first. */
5007 dev_close_many(head);
5009 list_for_each_entry(dev, head, unreg_list) {
5010 /* And unlink it from device chain. */
5011 unlist_netdevice(dev);
5013 dev->reg_state = NETREG_UNREGISTERING;
5018 list_for_each_entry(dev, head, unreg_list) {
5019 /* Shutdown queueing discipline. */
5023 /* Notify protocols, that we are about to destroy
5024 this device. They should clean all the things.
5026 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5028 if (!dev->rtnl_link_ops ||
5029 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5030 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5033 * Flush the unicast and multicast chains
5038 if (dev->netdev_ops->ndo_uninit)
5039 dev->netdev_ops->ndo_uninit(dev);
5041 /* Notifier chain MUST detach us from master device. */
5042 WARN_ON(dev->master);
5044 /* Remove entries from kobject tree */
5045 netdev_unregister_kobject(dev);
5048 /* Process any work delayed until the end of the batch */
5049 dev = list_first_entry(head, struct net_device, unreg_list);
5050 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5054 list_for_each_entry(dev, head, unreg_list)
5058 static void rollback_registered(struct net_device *dev)
5062 list_add(&dev->unreg_list, &single);
5063 rollback_registered_many(&single);
5066 unsigned long netdev_fix_features(unsigned long features, const char *name)
5068 /* Fix illegal SG+CSUM combinations. */
5069 if ((features & NETIF_F_SG) &&
5070 !(features & NETIF_F_ALL_CSUM)) {
5072 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
5073 "checksum feature.\n", name);
5074 features &= ~NETIF_F_SG;
5077 /* TSO requires that SG is present as well. */
5078 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
5080 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
5081 "SG feature.\n", name);
5082 features &= ~NETIF_F_TSO;
5085 if (features & NETIF_F_UFO) {
5086 /* maybe split UFO into V4 and V6? */
5087 if (!((features & NETIF_F_GEN_CSUM) ||
5088 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5089 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5091 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5092 "since no checksum offload features.\n",
5094 features &= ~NETIF_F_UFO;
5097 if (!(features & NETIF_F_SG)) {
5099 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5100 "since no NETIF_F_SG feature.\n", name);
5101 features &= ~NETIF_F_UFO;
5107 EXPORT_SYMBOL(netdev_fix_features);
5110 * netif_stacked_transfer_operstate - transfer operstate
5111 * @rootdev: the root or lower level device to transfer state from
5112 * @dev: the device to transfer operstate to
5114 * Transfer operational state from root to device. This is normally
5115 * called when a stacking relationship exists between the root
5116 * device and the device(a leaf device).
5118 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5119 struct net_device *dev)
5121 if (rootdev->operstate == IF_OPER_DORMANT)
5122 netif_dormant_on(dev);
5124 netif_dormant_off(dev);
5126 if (netif_carrier_ok(rootdev)) {
5127 if (!netif_carrier_ok(dev))
5128 netif_carrier_on(dev);
5130 if (netif_carrier_ok(dev))
5131 netif_carrier_off(dev);
5134 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5137 static int netif_alloc_rx_queues(struct net_device *dev)
5139 unsigned int i, count = dev->num_rx_queues;
5140 struct netdev_rx_queue *rx;
5144 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5146 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5151 for (i = 0; i < count; i++)
5157 static void netdev_init_one_queue(struct net_device *dev,
5158 struct netdev_queue *queue, void *_unused)
5160 /* Initialize queue lock */
5161 spin_lock_init(&queue->_xmit_lock);
5162 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5163 queue->xmit_lock_owner = -1;
5164 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5168 static int netif_alloc_netdev_queues(struct net_device *dev)
5170 unsigned int count = dev->num_tx_queues;
5171 struct netdev_queue *tx;
5175 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5177 pr_err("netdev: Unable to allocate %u tx queues.\n",
5183 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5184 spin_lock_init(&dev->tx_global_lock);
5190 * register_netdevice - register a network device
5191 * @dev: device to register
5193 * Take a completed network device structure and add it to the kernel
5194 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5195 * chain. 0 is returned on success. A negative errno code is returned
5196 * on a failure to set up the device, or if the name is a duplicate.
5198 * Callers must hold the rtnl semaphore. You may want
5199 * register_netdev() instead of this.
5202 * The locking appears insufficient to guarantee two parallel registers
5203 * will not get the same name.
5206 int register_netdevice(struct net_device *dev)
5209 struct net *net = dev_net(dev);
5211 BUG_ON(dev_boot_phase);
5216 /* When net_device's are persistent, this will be fatal. */
5217 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5220 spin_lock_init(&dev->addr_list_lock);
5221 netdev_set_addr_lockdep_class(dev);
5225 /* Init, if this function is available */
5226 if (dev->netdev_ops->ndo_init) {
5227 ret = dev->netdev_ops->ndo_init(dev);
5235 ret = dev_get_valid_name(dev, dev->name, 0);
5239 dev->ifindex = dev_new_index(net);
5240 if (dev->iflink == -1)
5241 dev->iflink = dev->ifindex;
5243 /* Fix illegal checksum combinations */
5244 if ((dev->features & NETIF_F_HW_CSUM) &&
5245 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5246 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5248 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5251 if ((dev->features & NETIF_F_NO_CSUM) &&
5252 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5253 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5255 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5258 dev->features = netdev_fix_features(dev->features, dev->name);
5260 /* Enable software GSO if SG is supported. */
5261 if (dev->features & NETIF_F_SG)
5262 dev->features |= NETIF_F_GSO;
5264 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5265 * vlan_dev_init() will do the dev->features check, so these features
5266 * are enabled only if supported by underlying device.
5268 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5270 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5271 ret = notifier_to_errno(ret);
5275 ret = netdev_register_kobject(dev);
5278 dev->reg_state = NETREG_REGISTERED;
5281 * Default initial state at registry is that the
5282 * device is present.
5285 set_bit(__LINK_STATE_PRESENT, &dev->state);
5287 dev_init_scheduler(dev);
5289 list_netdevice(dev);
5291 /* Notify protocols, that a new device appeared. */
5292 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5293 ret = notifier_to_errno(ret);
5295 rollback_registered(dev);
5296 dev->reg_state = NETREG_UNREGISTERED;
5299 * Prevent userspace races by waiting until the network
5300 * device is fully setup before sending notifications.
5302 if (!dev->rtnl_link_ops ||
5303 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5304 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5310 if (dev->netdev_ops->ndo_uninit)
5311 dev->netdev_ops->ndo_uninit(dev);
5314 EXPORT_SYMBOL(register_netdevice);
5317 * init_dummy_netdev - init a dummy network device for NAPI
5318 * @dev: device to init
5320 * This takes a network device structure and initialize the minimum
5321 * amount of fields so it can be used to schedule NAPI polls without
5322 * registering a full blown interface. This is to be used by drivers
5323 * that need to tie several hardware interfaces to a single NAPI
5324 * poll scheduler due to HW limitations.
5326 int init_dummy_netdev(struct net_device *dev)
5328 /* Clear everything. Note we don't initialize spinlocks
5329 * are they aren't supposed to be taken by any of the
5330 * NAPI code and this dummy netdev is supposed to be
5331 * only ever used for NAPI polls
5333 memset(dev, 0, sizeof(struct net_device));
5335 /* make sure we BUG if trying to hit standard
5336 * register/unregister code path
5338 dev->reg_state = NETREG_DUMMY;
5340 /* NAPI wants this */
5341 INIT_LIST_HEAD(&dev->napi_list);
5343 /* a dummy interface is started by default */
5344 set_bit(__LINK_STATE_PRESENT, &dev->state);
5345 set_bit(__LINK_STATE_START, &dev->state);
5347 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5348 * because users of this 'device' dont need to change
5354 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5358 * register_netdev - register a network device
5359 * @dev: device to register
5361 * Take a completed network device structure and add it to the kernel
5362 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5363 * chain. 0 is returned on success. A negative errno code is returned
5364 * on a failure to set up the device, or if the name is a duplicate.
5366 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5367 * and expands the device name if you passed a format string to
5370 int register_netdev(struct net_device *dev)
5377 * If the name is a format string the caller wants us to do a
5380 if (strchr(dev->name, '%')) {
5381 err = dev_alloc_name(dev, dev->name);
5386 err = register_netdevice(dev);
5391 EXPORT_SYMBOL(register_netdev);
5393 int netdev_refcnt_read(const struct net_device *dev)
5397 for_each_possible_cpu(i)
5398 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5401 EXPORT_SYMBOL(netdev_refcnt_read);
5404 * netdev_wait_allrefs - wait until all references are gone.
5406 * This is called when unregistering network devices.
5408 * Any protocol or device that holds a reference should register
5409 * for netdevice notification, and cleanup and put back the
5410 * reference if they receive an UNREGISTER event.
5411 * We can get stuck here if buggy protocols don't correctly
5414 static void netdev_wait_allrefs(struct net_device *dev)
5416 unsigned long rebroadcast_time, warning_time;
5419 linkwatch_forget_dev(dev);
5421 rebroadcast_time = warning_time = jiffies;
5422 refcnt = netdev_refcnt_read(dev);
5424 while (refcnt != 0) {
5425 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5428 /* Rebroadcast unregister notification */
5429 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5430 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5431 * should have already handle it the first time */
5433 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5435 /* We must not have linkwatch events
5436 * pending on unregister. If this
5437 * happens, we simply run the queue
5438 * unscheduled, resulting in a noop
5441 linkwatch_run_queue();
5446 rebroadcast_time = jiffies;
5451 refcnt = netdev_refcnt_read(dev);
5453 if (time_after(jiffies, warning_time + 10 * HZ)) {
5454 printk(KERN_EMERG "unregister_netdevice: "
5455 "waiting for %s to become free. Usage "
5458 warning_time = jiffies;
5467 * register_netdevice(x1);
5468 * register_netdevice(x2);
5470 * unregister_netdevice(y1);
5471 * unregister_netdevice(y2);
5477 * We are invoked by rtnl_unlock().
5478 * This allows us to deal with problems:
5479 * 1) We can delete sysfs objects which invoke hotplug
5480 * without deadlocking with linkwatch via keventd.
5481 * 2) Since we run with the RTNL semaphore not held, we can sleep
5482 * safely in order to wait for the netdev refcnt to drop to zero.
5484 * We must not return until all unregister events added during
5485 * the interval the lock was held have been completed.
5487 void netdev_run_todo(void)
5489 struct list_head list;
5491 /* Snapshot list, allow later requests */
5492 list_replace_init(&net_todo_list, &list);
5496 while (!list_empty(&list)) {
5497 struct net_device *dev
5498 = list_first_entry(&list, struct net_device, todo_list);
5499 list_del(&dev->todo_list);
5501 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5502 printk(KERN_ERR "network todo '%s' but state %d\n",
5503 dev->name, dev->reg_state);
5508 dev->reg_state = NETREG_UNREGISTERED;
5510 on_each_cpu(flush_backlog, dev, 1);
5512 netdev_wait_allrefs(dev);
5515 BUG_ON(netdev_refcnt_read(dev));
5516 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5517 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5518 WARN_ON(dev->dn_ptr);
5520 if (dev->destructor)
5521 dev->destructor(dev);
5523 /* Free network device */
5524 kobject_put(&dev->dev.kobj);
5528 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5529 * fields in the same order, with only the type differing.
5531 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5532 const struct net_device_stats *netdev_stats)
5534 #if BITS_PER_LONG == 64
5535 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5536 memcpy(stats64, netdev_stats, sizeof(*stats64));
5538 size_t i, n = sizeof(*stats64) / sizeof(u64);
5539 const unsigned long *src = (const unsigned long *)netdev_stats;
5540 u64 *dst = (u64 *)stats64;
5542 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5543 sizeof(*stats64) / sizeof(u64));
5544 for (i = 0; i < n; i++)
5550 * dev_get_stats - get network device statistics
5551 * @dev: device to get statistics from
5552 * @storage: place to store stats
5554 * Get network statistics from device. Return @storage.
5555 * The device driver may provide its own method by setting
5556 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5557 * otherwise the internal statistics structure is used.
5559 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5560 struct rtnl_link_stats64 *storage)
5562 const struct net_device_ops *ops = dev->netdev_ops;
5564 if (ops->ndo_get_stats64) {
5565 memset(storage, 0, sizeof(*storage));
5566 ops->ndo_get_stats64(dev, storage);
5567 } else if (ops->ndo_get_stats) {
5568 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5570 netdev_stats_to_stats64(storage, &dev->stats);
5572 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5575 EXPORT_SYMBOL(dev_get_stats);
5577 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5579 struct netdev_queue *queue = dev_ingress_queue(dev);
5581 #ifdef CONFIG_NET_CLS_ACT
5584 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5587 netdev_init_one_queue(dev, queue, NULL);
5588 queue->qdisc = &noop_qdisc;
5589 queue->qdisc_sleeping = &noop_qdisc;
5590 rcu_assign_pointer(dev->ingress_queue, queue);
5596 * alloc_netdev_mqs - allocate network device
5597 * @sizeof_priv: size of private data to allocate space for
5598 * @name: device name format string
5599 * @setup: callback to initialize device
5600 * @txqs: the number of TX subqueues to allocate
5601 * @rxqs: the number of RX subqueues to allocate
5603 * Allocates a struct net_device with private data area for driver use
5604 * and performs basic initialization. Also allocates subquue structs
5605 * for each queue on the device.
5607 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5608 void (*setup)(struct net_device *),
5609 unsigned int txqs, unsigned int rxqs)
5611 struct net_device *dev;
5613 struct net_device *p;
5615 BUG_ON(strlen(name) >= sizeof(dev->name));
5618 pr_err("alloc_netdev: Unable to allocate device "
5619 "with zero queues.\n");
5625 pr_err("alloc_netdev: Unable to allocate device "
5626 "with zero RX queues.\n");
5631 alloc_size = sizeof(struct net_device);
5633 /* ensure 32-byte alignment of private area */
5634 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5635 alloc_size += sizeof_priv;
5637 /* ensure 32-byte alignment of whole construct */
5638 alloc_size += NETDEV_ALIGN - 1;
5640 p = kzalloc(alloc_size, GFP_KERNEL);
5642 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5646 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5647 dev->padded = (char *)dev - (char *)p;
5649 dev->pcpu_refcnt = alloc_percpu(int);
5650 if (!dev->pcpu_refcnt)
5653 if (dev_addr_init(dev))
5659 dev_net_set(dev, &init_net);
5661 dev->num_tx_queues = txqs;
5662 dev->real_num_tx_queues = txqs;
5663 if (netif_alloc_netdev_queues(dev))
5667 dev->num_rx_queues = rxqs;
5668 dev->real_num_rx_queues = rxqs;
5669 if (netif_alloc_rx_queues(dev))
5673 dev->gso_max_size = GSO_MAX_SIZE;
5675 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5676 dev->ethtool_ntuple_list.count = 0;
5677 INIT_LIST_HEAD(&dev->napi_list);
5678 INIT_LIST_HEAD(&dev->unreg_list);
5679 INIT_LIST_HEAD(&dev->link_watch_list);
5680 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5682 strcpy(dev->name, name);
5686 free_percpu(dev->pcpu_refcnt);
5696 EXPORT_SYMBOL(alloc_netdev_mqs);
5699 * free_netdev - free network device
5702 * This function does the last stage of destroying an allocated device
5703 * interface. The reference to the device object is released.
5704 * If this is the last reference then it will be freed.
5706 void free_netdev(struct net_device *dev)
5708 struct napi_struct *p, *n;
5710 release_net(dev_net(dev));
5717 kfree(rcu_dereference_raw(dev->ingress_queue));
5719 /* Flush device addresses */
5720 dev_addr_flush(dev);
5722 /* Clear ethtool n-tuple list */
5723 ethtool_ntuple_flush(dev);
5725 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5728 free_percpu(dev->pcpu_refcnt);
5729 dev->pcpu_refcnt = NULL;
5731 /* Compatibility with error handling in drivers */
5732 if (dev->reg_state == NETREG_UNINITIALIZED) {
5733 kfree((char *)dev - dev->padded);
5737 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5738 dev->reg_state = NETREG_RELEASED;
5740 /* will free via device release */
5741 put_device(&dev->dev);
5743 EXPORT_SYMBOL(free_netdev);
5746 * synchronize_net - Synchronize with packet receive processing
5748 * Wait for packets currently being received to be done.
5749 * Does not block later packets from starting.
5751 void synchronize_net(void)
5756 EXPORT_SYMBOL(synchronize_net);
5759 * unregister_netdevice_queue - remove device from the kernel
5763 * This function shuts down a device interface and removes it
5764 * from the kernel tables.
5765 * If head not NULL, device is queued to be unregistered later.
5767 * Callers must hold the rtnl semaphore. You may want
5768 * unregister_netdev() instead of this.
5771 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5776 list_move_tail(&dev->unreg_list, head);
5778 rollback_registered(dev);
5779 /* Finish processing unregister after unlock */
5783 EXPORT_SYMBOL(unregister_netdevice_queue);
5786 * unregister_netdevice_many - unregister many devices
5787 * @head: list of devices
5789 void unregister_netdevice_many(struct list_head *head)
5791 struct net_device *dev;
5793 if (!list_empty(head)) {
5794 rollback_registered_many(head);
5795 list_for_each_entry(dev, head, unreg_list)
5799 EXPORT_SYMBOL(unregister_netdevice_many);
5802 * unregister_netdev - remove device from the kernel
5805 * This function shuts down a device interface and removes it
5806 * from the kernel tables.
5808 * This is just a wrapper for unregister_netdevice that takes
5809 * the rtnl semaphore. In general you want to use this and not
5810 * unregister_netdevice.
5812 void unregister_netdev(struct net_device *dev)
5815 unregister_netdevice(dev);
5818 EXPORT_SYMBOL(unregister_netdev);
5821 * dev_change_net_namespace - move device to different nethost namespace
5823 * @net: network namespace
5824 * @pat: If not NULL name pattern to try if the current device name
5825 * is already taken in the destination network namespace.
5827 * This function shuts down a device interface and moves it
5828 * to a new network namespace. On success 0 is returned, on
5829 * a failure a netagive errno code is returned.
5831 * Callers must hold the rtnl semaphore.
5834 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5840 /* Don't allow namespace local devices to be moved. */
5842 if (dev->features & NETIF_F_NETNS_LOCAL)
5845 /* Ensure the device has been registrered */
5847 if (dev->reg_state != NETREG_REGISTERED)
5850 /* Get out if there is nothing todo */
5852 if (net_eq(dev_net(dev), net))
5855 /* Pick the destination device name, and ensure
5856 * we can use it in the destination network namespace.
5859 if (__dev_get_by_name(net, dev->name)) {
5860 /* We get here if we can't use the current device name */
5863 if (dev_get_valid_name(dev, pat, 1))
5868 * And now a mini version of register_netdevice unregister_netdevice.
5871 /* If device is running close it first. */
5874 /* And unlink it from device chain */
5876 unlist_netdevice(dev);
5880 /* Shutdown queueing discipline. */
5883 /* Notify protocols, that we are about to destroy
5884 this device. They should clean all the things.
5886 Note that dev->reg_state stays at NETREG_REGISTERED.
5887 This is wanted because this way 8021q and macvlan know
5888 the device is just moving and can keep their slaves up.
5890 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5891 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5894 * Flush the unicast and multicast chains
5899 /* Actually switch the network namespace */
5900 dev_net_set(dev, net);
5902 /* If there is an ifindex conflict assign a new one */
5903 if (__dev_get_by_index(net, dev->ifindex)) {
5904 int iflink = (dev->iflink == dev->ifindex);
5905 dev->ifindex = dev_new_index(net);
5907 dev->iflink = dev->ifindex;
5910 /* Fixup kobjects */
5911 err = device_rename(&dev->dev, dev->name);
5914 /* Add the device back in the hashes */
5915 list_netdevice(dev);
5917 /* Notify protocols, that a new device appeared. */
5918 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5921 * Prevent userspace races by waiting until the network
5922 * device is fully setup before sending notifications.
5924 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5931 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5933 static int dev_cpu_callback(struct notifier_block *nfb,
5934 unsigned long action,
5937 struct sk_buff **list_skb;
5938 struct sk_buff *skb;
5939 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5940 struct softnet_data *sd, *oldsd;
5942 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5945 local_irq_disable();
5946 cpu = smp_processor_id();
5947 sd = &per_cpu(softnet_data, cpu);
5948 oldsd = &per_cpu(softnet_data, oldcpu);
5950 /* Find end of our completion_queue. */
5951 list_skb = &sd->completion_queue;
5953 list_skb = &(*list_skb)->next;
5954 /* Append completion queue from offline CPU. */
5955 *list_skb = oldsd->completion_queue;
5956 oldsd->completion_queue = NULL;
5958 /* Append output queue from offline CPU. */
5959 if (oldsd->output_queue) {
5960 *sd->output_queue_tailp = oldsd->output_queue;
5961 sd->output_queue_tailp = oldsd->output_queue_tailp;
5962 oldsd->output_queue = NULL;
5963 oldsd->output_queue_tailp = &oldsd->output_queue;
5966 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5969 /* Process offline CPU's input_pkt_queue */
5970 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5972 input_queue_head_incr(oldsd);
5974 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5976 input_queue_head_incr(oldsd);
5984 * netdev_increment_features - increment feature set by one
5985 * @all: current feature set
5986 * @one: new feature set
5987 * @mask: mask feature set
5989 * Computes a new feature set after adding a device with feature set
5990 * @one to the master device with current feature set @all. Will not
5991 * enable anything that is off in @mask. Returns the new feature set.
5993 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5996 /* If device needs checksumming, downgrade to it. */
5997 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5998 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5999 else if (mask & NETIF_F_ALL_CSUM) {
6000 /* If one device supports v4/v6 checksumming, set for all. */
6001 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
6002 !(all & NETIF_F_GEN_CSUM)) {
6003 all &= ~NETIF_F_ALL_CSUM;
6004 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
6007 /* If one device supports hw checksumming, set for all. */
6008 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
6009 all &= ~NETIF_F_ALL_CSUM;
6010 all |= NETIF_F_HW_CSUM;
6014 one |= NETIF_F_ALL_CSUM;
6016 one |= all & NETIF_F_ONE_FOR_ALL;
6017 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
6018 all |= one & mask & NETIF_F_ONE_FOR_ALL;
6022 EXPORT_SYMBOL(netdev_increment_features);
6024 static struct hlist_head *netdev_create_hash(void)
6027 struct hlist_head *hash;
6029 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6031 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6032 INIT_HLIST_HEAD(&hash[i]);
6037 /* Initialize per network namespace state */
6038 static int __net_init netdev_init(struct net *net)
6040 INIT_LIST_HEAD(&net->dev_base_head);
6042 net->dev_name_head = netdev_create_hash();
6043 if (net->dev_name_head == NULL)
6046 net->dev_index_head = netdev_create_hash();
6047 if (net->dev_index_head == NULL)
6053 kfree(net->dev_name_head);
6059 * netdev_drivername - network driver for the device
6060 * @dev: network device
6061 * @buffer: buffer for resulting name
6062 * @len: size of buffer
6064 * Determine network driver for device.
6066 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6068 const struct device_driver *driver;
6069 const struct device *parent;
6071 if (len <= 0 || !buffer)
6075 parent = dev->dev.parent;
6080 driver = parent->driver;
6081 if (driver && driver->name)
6082 strlcpy(buffer, driver->name, len);
6086 static int __netdev_printk(const char *level, const struct net_device *dev,
6087 struct va_format *vaf)
6091 if (dev && dev->dev.parent)
6092 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6093 netdev_name(dev), vaf);
6095 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6097 r = printk("%s(NULL net_device): %pV", level, vaf);
6102 int netdev_printk(const char *level, const struct net_device *dev,
6103 const char *format, ...)
6105 struct va_format vaf;
6109 va_start(args, format);
6114 r = __netdev_printk(level, dev, &vaf);
6119 EXPORT_SYMBOL(netdev_printk);
6121 #define define_netdev_printk_level(func, level) \
6122 int func(const struct net_device *dev, const char *fmt, ...) \
6125 struct va_format vaf; \
6128 va_start(args, fmt); \
6133 r = __netdev_printk(level, dev, &vaf); \
6138 EXPORT_SYMBOL(func);
6140 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6141 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6142 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6143 define_netdev_printk_level(netdev_err, KERN_ERR);
6144 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6145 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6146 define_netdev_printk_level(netdev_info, KERN_INFO);
6148 static void __net_exit netdev_exit(struct net *net)
6150 kfree(net->dev_name_head);
6151 kfree(net->dev_index_head);
6154 static struct pernet_operations __net_initdata netdev_net_ops = {
6155 .init = netdev_init,
6156 .exit = netdev_exit,
6159 static void __net_exit default_device_exit(struct net *net)
6161 struct net_device *dev, *aux;
6163 * Push all migratable network devices back to the
6164 * initial network namespace
6167 for_each_netdev_safe(net, dev, aux) {
6169 char fb_name[IFNAMSIZ];
6171 /* Ignore unmoveable devices (i.e. loopback) */
6172 if (dev->features & NETIF_F_NETNS_LOCAL)
6175 /* Leave virtual devices for the generic cleanup */
6176 if (dev->rtnl_link_ops)
6179 /* Push remaing network devices to init_net */
6180 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6181 err = dev_change_net_namespace(dev, &init_net, fb_name);
6183 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6184 __func__, dev->name, err);
6191 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6193 /* At exit all network devices most be removed from a network
6194 * namespace. Do this in the reverse order of registration.
6195 * Do this across as many network namespaces as possible to
6196 * improve batching efficiency.
6198 struct net_device *dev;
6200 LIST_HEAD(dev_kill_list);
6203 list_for_each_entry(net, net_list, exit_list) {
6204 for_each_netdev_reverse(net, dev) {
6205 if (dev->rtnl_link_ops)
6206 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6208 unregister_netdevice_queue(dev, &dev_kill_list);
6211 unregister_netdevice_many(&dev_kill_list);
6215 static struct pernet_operations __net_initdata default_device_ops = {
6216 .exit = default_device_exit,
6217 .exit_batch = default_device_exit_batch,
6221 * Initialize the DEV module. At boot time this walks the device list and
6222 * unhooks any devices that fail to initialise (normally hardware not
6223 * present) and leaves us with a valid list of present and active devices.
6228 * This is called single threaded during boot, so no need
6229 * to take the rtnl semaphore.
6231 static int __init net_dev_init(void)
6233 int i, rc = -ENOMEM;
6235 BUG_ON(!dev_boot_phase);
6237 if (dev_proc_init())
6240 if (netdev_kobject_init())
6243 INIT_LIST_HEAD(&ptype_all);
6244 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6245 INIT_LIST_HEAD(&ptype_base[i]);
6247 if (register_pernet_subsys(&netdev_net_ops))
6251 * Initialise the packet receive queues.
6254 for_each_possible_cpu(i) {
6255 struct softnet_data *sd = &per_cpu(softnet_data, i);
6257 memset(sd, 0, sizeof(*sd));
6258 skb_queue_head_init(&sd->input_pkt_queue);
6259 skb_queue_head_init(&sd->process_queue);
6260 sd->completion_queue = NULL;
6261 INIT_LIST_HEAD(&sd->poll_list);
6262 sd->output_queue = NULL;
6263 sd->output_queue_tailp = &sd->output_queue;
6265 sd->csd.func = rps_trigger_softirq;
6271 sd->backlog.poll = process_backlog;
6272 sd->backlog.weight = weight_p;
6273 sd->backlog.gro_list = NULL;
6274 sd->backlog.gro_count = 0;
6279 /* The loopback device is special if any other network devices
6280 * is present in a network namespace the loopback device must
6281 * be present. Since we now dynamically allocate and free the
6282 * loopback device ensure this invariant is maintained by
6283 * keeping the loopback device as the first device on the
6284 * list of network devices. Ensuring the loopback devices
6285 * is the first device that appears and the last network device
6288 if (register_pernet_device(&loopback_net_ops))
6291 if (register_pernet_device(&default_device_ops))
6294 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6295 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6297 hotcpu_notifier(dev_cpu_callback, 0);
6305 subsys_initcall(net_dev_init);
6307 static int __init initialize_hashrnd(void)
6309 get_random_bytes(&hashrnd, sizeof(hashrnd));
6313 late_initcall_sync(initialize_hashrnd);