2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
145 * The list of packet types we will receive (as opposed to discard)
146 * and the routines to invoke.
148 * Why 16. Because with 16 the only overlap we get on a hash of the
149 * low nibble of the protocol value is RARP/SNAP/X.25.
151 * NOTE: That is no longer true with the addition of VLAN tags. Not
152 * sure which should go first, but I bet it won't make much
153 * difference if we are running VLANs. The good news is that
154 * this protocol won't be in the list unless compiled in, so
155 * the average user (w/out VLANs) will not be adversely affected.
172 #define PTYPE_HASH_SIZE (16)
173 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
175 static DEFINE_SPINLOCK(ptype_lock);
176 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
177 static struct list_head ptype_all __read_mostly; /* Taps */
180 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
183 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
185 * Writers must hold the rtnl semaphore while they loop through the
186 * dev_base_head list, and hold dev_base_lock for writing when they do the
187 * actual updates. This allows pure readers to access the list even
188 * while a writer is preparing to update it.
190 * To put it another way, dev_base_lock is held for writing only to
191 * protect against pure readers; the rtnl semaphore provides the
192 * protection against other writers.
194 * See, for example usages, register_netdevice() and
195 * unregister_netdevice(), which must be called with the rtnl
198 DEFINE_RWLOCK(dev_base_lock);
199 EXPORT_SYMBOL(dev_base_lock);
201 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
203 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
204 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
207 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
209 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
212 static inline void rps_lock(struct softnet_data *sd)
215 spin_lock(&sd->input_pkt_queue.lock);
219 static inline void rps_unlock(struct softnet_data *sd)
222 spin_unlock(&sd->input_pkt_queue.lock);
226 /* Device list insertion */
227 static int list_netdevice(struct net_device *dev)
229 struct net *net = dev_net(dev);
233 write_lock_bh(&dev_base_lock);
234 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
235 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
236 hlist_add_head_rcu(&dev->index_hlist,
237 dev_index_hash(net, dev->ifindex));
238 write_unlock_bh(&dev_base_lock);
242 /* Device list removal
243 * caller must respect a RCU grace period before freeing/reusing dev
245 static void unlist_netdevice(struct net_device *dev)
249 /* Unlink dev from the device chain */
250 write_lock_bh(&dev_base_lock);
251 list_del_rcu(&dev->dev_list);
252 hlist_del_rcu(&dev->name_hlist);
253 hlist_del_rcu(&dev->index_hlist);
254 write_unlock_bh(&dev_base_lock);
261 static RAW_NOTIFIER_HEAD(netdev_chain);
264 * Device drivers call our routines to queue packets here. We empty the
265 * queue in the local softnet handler.
268 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
269 EXPORT_PER_CPU_SYMBOL(softnet_data);
271 #ifdef CONFIG_LOCKDEP
273 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
274 * according to dev->type
276 static const unsigned short netdev_lock_type[] =
277 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
278 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
279 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
280 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
281 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
282 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
283 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
284 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
285 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
286 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
287 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
288 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
289 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
290 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
291 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
292 ARPHRD_VOID, ARPHRD_NONE};
294 static const char *const netdev_lock_name[] =
295 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
296 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
297 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
298 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
299 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
300 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
301 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
302 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
303 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
304 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
305 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
306 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
307 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
308 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
309 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
310 "_xmit_VOID", "_xmit_NONE"};
312 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
313 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
315 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
319 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
320 if (netdev_lock_type[i] == dev_type)
322 /* the last key is used by default */
323 return ARRAY_SIZE(netdev_lock_type) - 1;
326 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
327 unsigned short dev_type)
331 i = netdev_lock_pos(dev_type);
332 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
333 netdev_lock_name[i]);
336 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
340 i = netdev_lock_pos(dev->type);
341 lockdep_set_class_and_name(&dev->addr_list_lock,
342 &netdev_addr_lock_key[i],
343 netdev_lock_name[i]);
346 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
347 unsigned short dev_type)
350 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
355 /*******************************************************************************
357 Protocol management and registration routines
359 *******************************************************************************/
362 * Add a protocol ID to the list. Now that the input handler is
363 * smarter we can dispense with all the messy stuff that used to be
366 * BEWARE!!! Protocol handlers, mangling input packets,
367 * MUST BE last in hash buckets and checking protocol handlers
368 * MUST start from promiscuous ptype_all chain in net_bh.
369 * It is true now, do not change it.
370 * Explanation follows: if protocol handler, mangling packet, will
371 * be the first on list, it is not able to sense, that packet
372 * is cloned and should be copied-on-write, so that it will
373 * change it and subsequent readers will get broken packet.
377 static inline struct list_head *ptype_head(const struct packet_type *pt)
379 if (pt->type == htons(ETH_P_ALL))
382 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
386 * dev_add_pack - add packet handler
387 * @pt: packet type declaration
389 * Add a protocol handler to the networking stack. The passed &packet_type
390 * is linked into kernel lists and may not be freed until it has been
391 * removed from the kernel lists.
393 * This call does not sleep therefore it can not
394 * guarantee all CPU's that are in middle of receiving packets
395 * will see the new packet type (until the next received packet).
398 void dev_add_pack(struct packet_type *pt)
400 struct list_head *head = ptype_head(pt);
402 spin_lock(&ptype_lock);
403 list_add_rcu(&pt->list, head);
404 spin_unlock(&ptype_lock);
406 EXPORT_SYMBOL(dev_add_pack);
409 * __dev_remove_pack - remove packet handler
410 * @pt: packet type declaration
412 * Remove a protocol handler that was previously added to the kernel
413 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
414 * from the kernel lists and can be freed or reused once this function
417 * The packet type might still be in use by receivers
418 * and must not be freed until after all the CPU's have gone
419 * through a quiescent state.
421 void __dev_remove_pack(struct packet_type *pt)
423 struct list_head *head = ptype_head(pt);
424 struct packet_type *pt1;
426 spin_lock(&ptype_lock);
428 list_for_each_entry(pt1, head, list) {
430 list_del_rcu(&pt->list);
435 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
437 spin_unlock(&ptype_lock);
439 EXPORT_SYMBOL(__dev_remove_pack);
442 * dev_remove_pack - remove packet handler
443 * @pt: packet type declaration
445 * Remove a protocol handler that was previously added to the kernel
446 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
447 * from the kernel lists and can be freed or reused once this function
450 * This call sleeps to guarantee that no CPU is looking at the packet
453 void dev_remove_pack(struct packet_type *pt)
455 __dev_remove_pack(pt);
459 EXPORT_SYMBOL(dev_remove_pack);
461 /******************************************************************************
463 Device Boot-time Settings Routines
465 *******************************************************************************/
467 /* Boot time configuration table */
468 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
471 * netdev_boot_setup_add - add new setup entry
472 * @name: name of the device
473 * @map: configured settings for the device
475 * Adds new setup entry to the dev_boot_setup list. The function
476 * returns 0 on error and 1 on success. This is a generic routine to
479 static int netdev_boot_setup_add(char *name, struct ifmap *map)
481 struct netdev_boot_setup *s;
485 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
486 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
487 memset(s[i].name, 0, sizeof(s[i].name));
488 strlcpy(s[i].name, name, IFNAMSIZ);
489 memcpy(&s[i].map, map, sizeof(s[i].map));
494 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
498 * netdev_boot_setup_check - check boot time settings
499 * @dev: the netdevice
501 * Check boot time settings for the device.
502 * The found settings are set for the device to be used
503 * later in the device probing.
504 * Returns 0 if no settings found, 1 if they are.
506 int netdev_boot_setup_check(struct net_device *dev)
508 struct netdev_boot_setup *s = dev_boot_setup;
511 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
512 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
513 !strcmp(dev->name, s[i].name)) {
514 dev->irq = s[i].map.irq;
515 dev->base_addr = s[i].map.base_addr;
516 dev->mem_start = s[i].map.mem_start;
517 dev->mem_end = s[i].map.mem_end;
523 EXPORT_SYMBOL(netdev_boot_setup_check);
527 * netdev_boot_base - get address from boot time settings
528 * @prefix: prefix for network device
529 * @unit: id for network device
531 * Check boot time settings for the base address of device.
532 * The found settings are set for the device to be used
533 * later in the device probing.
534 * Returns 0 if no settings found.
536 unsigned long netdev_boot_base(const char *prefix, int unit)
538 const struct netdev_boot_setup *s = dev_boot_setup;
542 sprintf(name, "%s%d", prefix, unit);
545 * If device already registered then return base of 1
546 * to indicate not to probe for this interface
548 if (__dev_get_by_name(&init_net, name))
551 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
552 if (!strcmp(name, s[i].name))
553 return s[i].map.base_addr;
558 * Saves at boot time configured settings for any netdevice.
560 int __init netdev_boot_setup(char *str)
565 str = get_options(str, ARRAY_SIZE(ints), ints);
570 memset(&map, 0, sizeof(map));
574 map.base_addr = ints[2];
576 map.mem_start = ints[3];
578 map.mem_end = ints[4];
580 /* Add new entry to the list */
581 return netdev_boot_setup_add(str, &map);
584 __setup("netdev=", netdev_boot_setup);
586 /*******************************************************************************
588 Device Interface Subroutines
590 *******************************************************************************/
593 * __dev_get_by_name - find a device by its name
594 * @net: the applicable net namespace
595 * @name: name to find
597 * Find an interface by name. Must be called under RTNL semaphore
598 * or @dev_base_lock. If the name is found a pointer to the device
599 * is returned. If the name is not found then %NULL is returned. The
600 * reference counters are not incremented so the caller must be
601 * careful with locks.
604 struct net_device *__dev_get_by_name(struct net *net, const char *name)
606 struct hlist_node *p;
607 struct net_device *dev;
608 struct hlist_head *head = dev_name_hash(net, name);
610 hlist_for_each_entry(dev, p, head, name_hlist)
611 if (!strncmp(dev->name, name, IFNAMSIZ))
616 EXPORT_SYMBOL(__dev_get_by_name);
619 * dev_get_by_name_rcu - find a device by its name
620 * @net: the applicable net namespace
621 * @name: name to find
623 * Find an interface by name.
624 * If the name is found a pointer to the device is returned.
625 * If the name is not found then %NULL is returned.
626 * The reference counters are not incremented so the caller must be
627 * careful with locks. The caller must hold RCU lock.
630 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
632 struct hlist_node *p;
633 struct net_device *dev;
634 struct hlist_head *head = dev_name_hash(net, name);
636 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
637 if (!strncmp(dev->name, name, IFNAMSIZ))
642 EXPORT_SYMBOL(dev_get_by_name_rcu);
645 * dev_get_by_name - find a device by its name
646 * @net: the applicable net namespace
647 * @name: name to find
649 * Find an interface by name. This can be called from any
650 * context and does its own locking. The returned handle has
651 * the usage count incremented and the caller must use dev_put() to
652 * release it when it is no longer needed. %NULL is returned if no
653 * matching device is found.
656 struct net_device *dev_get_by_name(struct net *net, const char *name)
658 struct net_device *dev;
661 dev = dev_get_by_name_rcu(net, name);
667 EXPORT_SYMBOL(dev_get_by_name);
670 * __dev_get_by_index - find a device by its ifindex
671 * @net: the applicable net namespace
672 * @ifindex: index of device
674 * Search for an interface by index. Returns %NULL if the device
675 * is not found or a pointer to the device. The device has not
676 * had its reference counter increased so the caller must be careful
677 * about locking. The caller must hold either the RTNL semaphore
681 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
683 struct hlist_node *p;
684 struct net_device *dev;
685 struct hlist_head *head = dev_index_hash(net, ifindex);
687 hlist_for_each_entry(dev, p, head, index_hlist)
688 if (dev->ifindex == ifindex)
693 EXPORT_SYMBOL(__dev_get_by_index);
696 * dev_get_by_index_rcu - find a device by its ifindex
697 * @net: the applicable net namespace
698 * @ifindex: index of device
700 * Search for an interface by index. Returns %NULL if the device
701 * is not found or a pointer to the device. The device has not
702 * had its reference counter increased so the caller must be careful
703 * about locking. The caller must hold RCU lock.
706 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
708 struct hlist_node *p;
709 struct net_device *dev;
710 struct hlist_head *head = dev_index_hash(net, ifindex);
712 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
713 if (dev->ifindex == ifindex)
718 EXPORT_SYMBOL(dev_get_by_index_rcu);
722 * dev_get_by_index - find a device by its ifindex
723 * @net: the applicable net namespace
724 * @ifindex: index of device
726 * Search for an interface by index. Returns NULL if the device
727 * is not found or a pointer to the device. The device returned has
728 * had a reference added and the pointer is safe until the user calls
729 * dev_put to indicate they have finished with it.
732 struct net_device *dev_get_by_index(struct net *net, int ifindex)
734 struct net_device *dev;
737 dev = dev_get_by_index_rcu(net, ifindex);
743 EXPORT_SYMBOL(dev_get_by_index);
746 * dev_getbyhwaddr_rcu - find a device by its hardware address
747 * @net: the applicable net namespace
748 * @type: media type of device
749 * @ha: hardware address
751 * Search for an interface by MAC address. Returns NULL if the device
752 * is not found or a pointer to the device. The caller must hold RCU
753 * The returned device has not had its ref count increased
754 * and the caller must therefore be careful about locking
758 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
761 struct net_device *dev;
763 for_each_netdev_rcu(net, dev)
764 if (dev->type == type &&
765 !memcmp(dev->dev_addr, ha, dev->addr_len))
770 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
772 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
774 struct net_device *dev;
777 for_each_netdev(net, dev)
778 if (dev->type == type)
783 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
785 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
787 struct net_device *dev, *ret = NULL;
790 for_each_netdev_rcu(net, dev)
791 if (dev->type == type) {
799 EXPORT_SYMBOL(dev_getfirstbyhwtype);
802 * dev_get_by_flags_rcu - find any device with given flags
803 * @net: the applicable net namespace
804 * @if_flags: IFF_* values
805 * @mask: bitmask of bits in if_flags to check
807 * Search for any interface with the given flags. Returns NULL if a device
808 * is not found or a pointer to the device. Must be called inside
809 * rcu_read_lock(), and result refcount is unchanged.
812 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
815 struct net_device *dev, *ret;
818 for_each_netdev_rcu(net, dev) {
819 if (((dev->flags ^ if_flags) & mask) == 0) {
826 EXPORT_SYMBOL(dev_get_by_flags_rcu);
829 * dev_valid_name - check if name is okay for network device
832 * Network device names need to be valid file names to
833 * to allow sysfs to work. We also disallow any kind of
836 int dev_valid_name(const char *name)
840 if (strlen(name) >= IFNAMSIZ)
842 if (!strcmp(name, ".") || !strcmp(name, ".."))
846 if (*name == '/' || isspace(*name))
852 EXPORT_SYMBOL(dev_valid_name);
855 * __dev_alloc_name - allocate a name for a device
856 * @net: network namespace to allocate the device name in
857 * @name: name format string
858 * @buf: scratch buffer and result name string
860 * Passed a format string - eg "lt%d" it will try and find a suitable
861 * id. It scans list of devices to build up a free map, then chooses
862 * the first empty slot. The caller must hold the dev_base or rtnl lock
863 * while allocating the name and adding the device in order to avoid
865 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
866 * Returns the number of the unit assigned or a negative errno code.
869 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
873 const int max_netdevices = 8*PAGE_SIZE;
874 unsigned long *inuse;
875 struct net_device *d;
877 p = strnchr(name, IFNAMSIZ-1, '%');
880 * Verify the string as this thing may have come from
881 * the user. There must be either one "%d" and no other "%"
884 if (p[1] != 'd' || strchr(p + 2, '%'))
887 /* Use one page as a bit array of possible slots */
888 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
892 for_each_netdev(net, d) {
893 if (!sscanf(d->name, name, &i))
895 if (i < 0 || i >= max_netdevices)
898 /* avoid cases where sscanf is not exact inverse of printf */
899 snprintf(buf, IFNAMSIZ, name, i);
900 if (!strncmp(buf, d->name, IFNAMSIZ))
904 i = find_first_zero_bit(inuse, max_netdevices);
905 free_page((unsigned long) inuse);
909 snprintf(buf, IFNAMSIZ, name, i);
910 if (!__dev_get_by_name(net, buf))
913 /* It is possible to run out of possible slots
914 * when the name is long and there isn't enough space left
915 * for the digits, or if all bits are used.
921 * dev_alloc_name - allocate a name for a device
923 * @name: name format string
925 * Passed a format string - eg "lt%d" it will try and find a suitable
926 * id. It scans list of devices to build up a free map, then chooses
927 * the first empty slot. The caller must hold the dev_base or rtnl lock
928 * while allocating the name and adding the device in order to avoid
930 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
931 * Returns the number of the unit assigned or a negative errno code.
934 int dev_alloc_name(struct net_device *dev, const char *name)
940 BUG_ON(!dev_net(dev));
942 ret = __dev_alloc_name(net, name, buf);
944 strlcpy(dev->name, buf, IFNAMSIZ);
947 EXPORT_SYMBOL(dev_alloc_name);
949 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
953 BUG_ON(!dev_net(dev));
956 if (!dev_valid_name(name))
959 if (fmt && strchr(name, '%'))
960 return dev_alloc_name(dev, name);
961 else if (__dev_get_by_name(net, name))
963 else if (dev->name != name)
964 strlcpy(dev->name, name, IFNAMSIZ);
970 * dev_change_name - change name of a device
972 * @newname: name (or format string) must be at least IFNAMSIZ
974 * Change name of a device, can pass format strings "eth%d".
977 int dev_change_name(struct net_device *dev, const char *newname)
979 char oldname[IFNAMSIZ];
985 BUG_ON(!dev_net(dev));
988 if (dev->flags & IFF_UP)
991 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
994 memcpy(oldname, dev->name, IFNAMSIZ);
996 err = dev_get_valid_name(dev, newname, 1);
1001 ret = device_rename(&dev->dev, dev->name);
1003 memcpy(dev->name, oldname, IFNAMSIZ);
1007 write_lock_bh(&dev_base_lock);
1008 hlist_del(&dev->name_hlist);
1009 write_unlock_bh(&dev_base_lock);
1013 write_lock_bh(&dev_base_lock);
1014 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1015 write_unlock_bh(&dev_base_lock);
1017 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1018 ret = notifier_to_errno(ret);
1021 /* err >= 0 after dev_alloc_name() or stores the first errno */
1024 memcpy(dev->name, oldname, IFNAMSIZ);
1028 "%s: name change rollback failed: %d.\n",
1037 * dev_set_alias - change ifalias of a device
1039 * @alias: name up to IFALIASZ
1040 * @len: limit of bytes to copy from info
1042 * Set ifalias for a device,
1044 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1048 if (len >= IFALIASZ)
1053 kfree(dev->ifalias);
1054 dev->ifalias = NULL;
1059 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1063 strlcpy(dev->ifalias, alias, len+1);
1069 * netdev_features_change - device changes features
1070 * @dev: device to cause notification
1072 * Called to indicate a device has changed features.
1074 void netdev_features_change(struct net_device *dev)
1076 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1078 EXPORT_SYMBOL(netdev_features_change);
1081 * netdev_state_change - device changes state
1082 * @dev: device to cause notification
1084 * Called to indicate a device has changed state. This function calls
1085 * the notifier chains for netdev_chain and sends a NEWLINK message
1086 * to the routing socket.
1088 void netdev_state_change(struct net_device *dev)
1090 if (dev->flags & IFF_UP) {
1091 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1092 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1095 EXPORT_SYMBOL(netdev_state_change);
1097 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1099 return call_netdevice_notifiers(event, dev);
1101 EXPORT_SYMBOL(netdev_bonding_change);
1104 * dev_load - load a network module
1105 * @net: the applicable net namespace
1106 * @name: name of interface
1108 * If a network interface is not present and the process has suitable
1109 * privileges this function loads the module. If module loading is not
1110 * available in this kernel then it becomes a nop.
1113 void dev_load(struct net *net, const char *name)
1115 struct net_device *dev;
1118 dev = dev_get_by_name_rcu(net, name);
1121 if (!dev && capable(CAP_NET_ADMIN))
1122 request_module("%s", name);
1124 EXPORT_SYMBOL(dev_load);
1126 static int __dev_open(struct net_device *dev)
1128 const struct net_device_ops *ops = dev->netdev_ops;
1134 * Is it even present?
1136 if (!netif_device_present(dev))
1139 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1140 ret = notifier_to_errno(ret);
1145 * Call device private open method
1147 set_bit(__LINK_STATE_START, &dev->state);
1149 if (ops->ndo_validate_addr)
1150 ret = ops->ndo_validate_addr(dev);
1152 if (!ret && ops->ndo_open)
1153 ret = ops->ndo_open(dev);
1156 * If it went open OK then:
1160 clear_bit(__LINK_STATE_START, &dev->state);
1165 dev->flags |= IFF_UP;
1170 net_dmaengine_get();
1173 * Initialize multicasting status
1175 dev_set_rx_mode(dev);
1178 * Wakeup transmit queue engine
1187 * dev_open - prepare an interface for use.
1188 * @dev: device to open
1190 * Takes a device from down to up state. The device's private open
1191 * function is invoked and then the multicast lists are loaded. Finally
1192 * the device is moved into the up state and a %NETDEV_UP message is
1193 * sent to the netdev notifier chain.
1195 * Calling this function on an active interface is a nop. On a failure
1196 * a negative errno code is returned.
1198 int dev_open(struct net_device *dev)
1205 if (dev->flags & IFF_UP)
1211 ret = __dev_open(dev);
1216 * ... and announce new interface.
1218 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1219 call_netdevice_notifiers(NETDEV_UP, dev);
1223 EXPORT_SYMBOL(dev_open);
1225 static int __dev_close_many(struct list_head *head)
1227 struct net_device *dev;
1232 list_for_each_entry(dev, head, unreg_list) {
1234 * Tell people we are going down, so that they can
1235 * prepare to death, when device is still operating.
1237 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1239 clear_bit(__LINK_STATE_START, &dev->state);
1241 /* Synchronize to scheduled poll. We cannot touch poll list, it
1242 * can be even on different cpu. So just clear netif_running().
1244 * dev->stop() will invoke napi_disable() on all of it's
1245 * napi_struct instances on this device.
1247 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1250 dev_deactivate_many(head);
1252 list_for_each_entry(dev, head, unreg_list) {
1253 const struct net_device_ops *ops = dev->netdev_ops;
1256 * Call the device specific close. This cannot fail.
1257 * Only if device is UP
1259 * We allow it to be called even after a DETACH hot-plug
1266 * Device is now down.
1269 dev->flags &= ~IFF_UP;
1274 net_dmaengine_put();
1280 static int __dev_close(struct net_device *dev)
1284 list_add(&dev->unreg_list, &single);
1285 return __dev_close_many(&single);
1288 int dev_close_many(struct list_head *head)
1290 struct net_device *dev, *tmp;
1291 LIST_HEAD(tmp_list);
1293 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1294 if (!(dev->flags & IFF_UP))
1295 list_move(&dev->unreg_list, &tmp_list);
1297 __dev_close_many(head);
1300 * Tell people we are down
1302 list_for_each_entry(dev, head, unreg_list) {
1303 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1304 call_netdevice_notifiers(NETDEV_DOWN, dev);
1307 /* rollback_registered_many needs the complete original list */
1308 list_splice(&tmp_list, head);
1313 * dev_close - shutdown an interface.
1314 * @dev: device to shutdown
1316 * This function moves an active device into down state. A
1317 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1318 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1321 int dev_close(struct net_device *dev)
1325 list_add(&dev->unreg_list, &single);
1326 dev_close_many(&single);
1330 EXPORT_SYMBOL(dev_close);
1334 * dev_disable_lro - disable Large Receive Offload on a device
1337 * Disable Large Receive Offload (LRO) on a net device. Must be
1338 * called under RTNL. This is needed if received packets may be
1339 * forwarded to another interface.
1341 void dev_disable_lro(struct net_device *dev)
1343 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1344 dev->ethtool_ops->set_flags) {
1345 u32 flags = dev->ethtool_ops->get_flags(dev);
1346 if (flags & ETH_FLAG_LRO) {
1347 flags &= ~ETH_FLAG_LRO;
1348 dev->ethtool_ops->set_flags(dev, flags);
1351 WARN_ON(dev->features & NETIF_F_LRO);
1353 EXPORT_SYMBOL(dev_disable_lro);
1356 static int dev_boot_phase = 1;
1359 * Device change register/unregister. These are not inline or static
1360 * as we export them to the world.
1364 * register_netdevice_notifier - register a network notifier block
1367 * Register a notifier to be called when network device events occur.
1368 * The notifier passed is linked into the kernel structures and must
1369 * not be reused until it has been unregistered. A negative errno code
1370 * is returned on a failure.
1372 * When registered all registration and up events are replayed
1373 * to the new notifier to allow device to have a race free
1374 * view of the network device list.
1377 int register_netdevice_notifier(struct notifier_block *nb)
1379 struct net_device *dev;
1380 struct net_device *last;
1385 err = raw_notifier_chain_register(&netdev_chain, nb);
1391 for_each_netdev(net, dev) {
1392 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1393 err = notifier_to_errno(err);
1397 if (!(dev->flags & IFF_UP))
1400 nb->notifier_call(nb, NETDEV_UP, dev);
1411 for_each_netdev(net, dev) {
1415 if (dev->flags & IFF_UP) {
1416 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1417 nb->notifier_call(nb, NETDEV_DOWN, dev);
1419 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1420 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1424 raw_notifier_chain_unregister(&netdev_chain, nb);
1427 EXPORT_SYMBOL(register_netdevice_notifier);
1430 * unregister_netdevice_notifier - unregister a network notifier block
1433 * Unregister a notifier previously registered by
1434 * register_netdevice_notifier(). The notifier is unlinked into the
1435 * kernel structures and may then be reused. A negative errno code
1436 * is returned on a failure.
1439 int unregister_netdevice_notifier(struct notifier_block *nb)
1444 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1448 EXPORT_SYMBOL(unregister_netdevice_notifier);
1451 * call_netdevice_notifiers - call all network notifier blocks
1452 * @val: value passed unmodified to notifier function
1453 * @dev: net_device pointer passed unmodified to notifier function
1455 * Call all network notifier blocks. Parameters and return value
1456 * are as for raw_notifier_call_chain().
1459 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1462 return raw_notifier_call_chain(&netdev_chain, val, dev);
1465 /* When > 0 there are consumers of rx skb time stamps */
1466 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1468 void net_enable_timestamp(void)
1470 atomic_inc(&netstamp_needed);
1472 EXPORT_SYMBOL(net_enable_timestamp);
1474 void net_disable_timestamp(void)
1476 atomic_dec(&netstamp_needed);
1478 EXPORT_SYMBOL(net_disable_timestamp);
1480 static inline void net_timestamp_set(struct sk_buff *skb)
1482 if (atomic_read(&netstamp_needed))
1483 __net_timestamp(skb);
1485 skb->tstamp.tv64 = 0;
1488 static inline void net_timestamp_check(struct sk_buff *skb)
1490 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1491 __net_timestamp(skb);
1495 * dev_forward_skb - loopback an skb to another netif
1497 * @dev: destination network device
1498 * @skb: buffer to forward
1501 * NET_RX_SUCCESS (no congestion)
1502 * NET_RX_DROP (packet was dropped, but freed)
1504 * dev_forward_skb can be used for injecting an skb from the
1505 * start_xmit function of one device into the receive queue
1506 * of another device.
1508 * The receiving device may be in another namespace, so
1509 * we have to clear all information in the skb that could
1510 * impact namespace isolation.
1512 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1517 if (unlikely(!(dev->flags & IFF_UP) ||
1518 (skb->len > (dev->mtu + dev->hard_header_len + VLAN_HLEN)))) {
1519 atomic_long_inc(&dev->rx_dropped);
1523 skb_set_dev(skb, dev);
1524 skb->tstamp.tv64 = 0;
1525 skb->pkt_type = PACKET_HOST;
1526 skb->protocol = eth_type_trans(skb, dev);
1527 return netif_rx(skb);
1529 EXPORT_SYMBOL_GPL(dev_forward_skb);
1531 static inline int deliver_skb(struct sk_buff *skb,
1532 struct packet_type *pt_prev,
1533 struct net_device *orig_dev)
1535 atomic_inc(&skb->users);
1536 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1540 * Support routine. Sends outgoing frames to any network
1541 * taps currently in use.
1544 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1546 struct packet_type *ptype;
1547 struct sk_buff *skb2 = NULL;
1548 struct packet_type *pt_prev = NULL;
1551 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1552 /* Never send packets back to the socket
1553 * they originated from - MvS (miquels@drinkel.ow.org)
1555 if ((ptype->dev == dev || !ptype->dev) &&
1556 (ptype->af_packet_priv == NULL ||
1557 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1559 deliver_skb(skb2, pt_prev, skb->dev);
1564 skb2 = skb_clone(skb, GFP_ATOMIC);
1568 net_timestamp_set(skb2);
1570 /* skb->nh should be correctly
1571 set by sender, so that the second statement is
1572 just protection against buggy protocols.
1574 skb_reset_mac_header(skb2);
1576 if (skb_network_header(skb2) < skb2->data ||
1577 skb2->network_header > skb2->tail) {
1578 if (net_ratelimit())
1579 printk(KERN_CRIT "protocol %04x is "
1581 ntohs(skb2->protocol),
1583 skb_reset_network_header(skb2);
1586 skb2->transport_header = skb2->network_header;
1587 skb2->pkt_type = PACKET_OUTGOING;
1592 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1597 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1598 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1600 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1604 if (txq < 1 || txq > dev->num_tx_queues)
1607 if (dev->reg_state == NETREG_REGISTERED) {
1610 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1615 if (txq < dev->real_num_tx_queues)
1616 qdisc_reset_all_tx_gt(dev, txq);
1619 dev->real_num_tx_queues = txq;
1622 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1626 * netif_set_real_num_rx_queues - set actual number of RX queues used
1627 * @dev: Network device
1628 * @rxq: Actual number of RX queues
1630 * This must be called either with the rtnl_lock held or before
1631 * registration of the net device. Returns 0 on success, or a
1632 * negative error code. If called before registration, it always
1635 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1639 if (rxq < 1 || rxq > dev->num_rx_queues)
1642 if (dev->reg_state == NETREG_REGISTERED) {
1645 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1651 dev->real_num_rx_queues = rxq;
1654 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1657 static inline void __netif_reschedule(struct Qdisc *q)
1659 struct softnet_data *sd;
1660 unsigned long flags;
1662 local_irq_save(flags);
1663 sd = &__get_cpu_var(softnet_data);
1664 q->next_sched = NULL;
1665 *sd->output_queue_tailp = q;
1666 sd->output_queue_tailp = &q->next_sched;
1667 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1668 local_irq_restore(flags);
1671 void __netif_schedule(struct Qdisc *q)
1673 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1674 __netif_reschedule(q);
1676 EXPORT_SYMBOL(__netif_schedule);
1678 void dev_kfree_skb_irq(struct sk_buff *skb)
1680 if (atomic_dec_and_test(&skb->users)) {
1681 struct softnet_data *sd;
1682 unsigned long flags;
1684 local_irq_save(flags);
1685 sd = &__get_cpu_var(softnet_data);
1686 skb->next = sd->completion_queue;
1687 sd->completion_queue = skb;
1688 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1689 local_irq_restore(flags);
1692 EXPORT_SYMBOL(dev_kfree_skb_irq);
1694 void dev_kfree_skb_any(struct sk_buff *skb)
1696 if (in_irq() || irqs_disabled())
1697 dev_kfree_skb_irq(skb);
1701 EXPORT_SYMBOL(dev_kfree_skb_any);
1705 * netif_device_detach - mark device as removed
1706 * @dev: network device
1708 * Mark device as removed from system and therefore no longer available.
1710 void netif_device_detach(struct net_device *dev)
1712 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1713 netif_running(dev)) {
1714 netif_tx_stop_all_queues(dev);
1717 EXPORT_SYMBOL(netif_device_detach);
1720 * netif_device_attach - mark device as attached
1721 * @dev: network device
1723 * Mark device as attached from system and restart if needed.
1725 void netif_device_attach(struct net_device *dev)
1727 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1728 netif_running(dev)) {
1729 netif_tx_wake_all_queues(dev);
1730 __netdev_watchdog_up(dev);
1733 EXPORT_SYMBOL(netif_device_attach);
1735 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1737 return ((features & NETIF_F_GEN_CSUM) ||
1738 ((features & NETIF_F_V4_CSUM) &&
1739 protocol == htons(ETH_P_IP)) ||
1740 ((features & NETIF_F_V6_CSUM) &&
1741 protocol == htons(ETH_P_IPV6)) ||
1742 ((features & NETIF_F_FCOE_CRC) &&
1743 protocol == htons(ETH_P_FCOE)));
1746 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1748 __be16 protocol = skb->protocol;
1749 int features = dev->features;
1751 if (vlan_tx_tag_present(skb)) {
1752 features &= dev->vlan_features;
1753 } else if (protocol == htons(ETH_P_8021Q)) {
1754 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1755 protocol = veh->h_vlan_encapsulated_proto;
1756 features &= dev->vlan_features;
1759 return can_checksum_protocol(features, protocol);
1763 * skb_dev_set -- assign a new device to a buffer
1764 * @skb: buffer for the new device
1765 * @dev: network device
1767 * If an skb is owned by a device already, we have to reset
1768 * all data private to the namespace a device belongs to
1769 * before assigning it a new device.
1771 #ifdef CONFIG_NET_NS
1772 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1775 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1778 skb_init_secmark(skb);
1782 skb->ipvs_property = 0;
1783 #ifdef CONFIG_NET_SCHED
1789 EXPORT_SYMBOL(skb_set_dev);
1790 #endif /* CONFIG_NET_NS */
1793 * Invalidate hardware checksum when packet is to be mangled, and
1794 * complete checksum manually on outgoing path.
1796 int skb_checksum_help(struct sk_buff *skb)
1799 int ret = 0, offset;
1801 if (skb->ip_summed == CHECKSUM_COMPLETE)
1802 goto out_set_summed;
1804 if (unlikely(skb_shinfo(skb)->gso_size)) {
1805 /* Let GSO fix up the checksum. */
1806 goto out_set_summed;
1809 offset = skb_checksum_start_offset(skb);
1810 BUG_ON(offset >= skb_headlen(skb));
1811 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1813 offset += skb->csum_offset;
1814 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1816 if (skb_cloned(skb) &&
1817 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1818 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1823 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1825 skb->ip_summed = CHECKSUM_NONE;
1829 EXPORT_SYMBOL(skb_checksum_help);
1832 * skb_gso_segment - Perform segmentation on skb.
1833 * @skb: buffer to segment
1834 * @features: features for the output path (see dev->features)
1836 * This function segments the given skb and returns a list of segments.
1838 * It may return NULL if the skb requires no segmentation. This is
1839 * only possible when GSO is used for verifying header integrity.
1841 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1843 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1844 struct packet_type *ptype;
1845 __be16 type = skb->protocol;
1846 int vlan_depth = ETH_HLEN;
1849 while (type == htons(ETH_P_8021Q)) {
1850 struct vlan_hdr *vh;
1852 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1853 return ERR_PTR(-EINVAL);
1855 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1856 type = vh->h_vlan_encapsulated_proto;
1857 vlan_depth += VLAN_HLEN;
1860 skb_reset_mac_header(skb);
1861 skb->mac_len = skb->network_header - skb->mac_header;
1862 __skb_pull(skb, skb->mac_len);
1864 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1865 struct net_device *dev = skb->dev;
1866 struct ethtool_drvinfo info = {};
1868 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1869 dev->ethtool_ops->get_drvinfo(dev, &info);
1871 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1872 info.driver, dev ? dev->features : 0L,
1873 skb->sk ? skb->sk->sk_route_caps : 0L,
1874 skb->len, skb->data_len, skb->ip_summed);
1876 if (skb_header_cloned(skb) &&
1877 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1878 return ERR_PTR(err);
1882 list_for_each_entry_rcu(ptype,
1883 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1884 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1885 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1886 err = ptype->gso_send_check(skb);
1887 segs = ERR_PTR(err);
1888 if (err || skb_gso_ok(skb, features))
1890 __skb_push(skb, (skb->data -
1891 skb_network_header(skb)));
1893 segs = ptype->gso_segment(skb, features);
1899 __skb_push(skb, skb->data - skb_mac_header(skb));
1903 EXPORT_SYMBOL(skb_gso_segment);
1905 /* Take action when hardware reception checksum errors are detected. */
1907 void netdev_rx_csum_fault(struct net_device *dev)
1909 if (net_ratelimit()) {
1910 printk(KERN_ERR "%s: hw csum failure.\n",
1911 dev ? dev->name : "<unknown>");
1915 EXPORT_SYMBOL(netdev_rx_csum_fault);
1918 /* Actually, we should eliminate this check as soon as we know, that:
1919 * 1. IOMMU is present and allows to map all the memory.
1920 * 2. No high memory really exists on this machine.
1923 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1925 #ifdef CONFIG_HIGHMEM
1927 if (!(dev->features & NETIF_F_HIGHDMA)) {
1928 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1929 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1933 if (PCI_DMA_BUS_IS_PHYS) {
1934 struct device *pdev = dev->dev.parent;
1938 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1939 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1940 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1949 void (*destructor)(struct sk_buff *skb);
1952 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1954 static void dev_gso_skb_destructor(struct sk_buff *skb)
1956 struct dev_gso_cb *cb;
1959 struct sk_buff *nskb = skb->next;
1961 skb->next = nskb->next;
1964 } while (skb->next);
1966 cb = DEV_GSO_CB(skb);
1968 cb->destructor(skb);
1972 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1973 * @skb: buffer to segment
1975 * This function segments the given skb and stores the list of segments
1978 static int dev_gso_segment(struct sk_buff *skb)
1980 struct net_device *dev = skb->dev;
1981 struct sk_buff *segs;
1982 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1985 segs = skb_gso_segment(skb, features);
1987 /* Verifying header integrity only. */
1992 return PTR_ERR(segs);
1995 DEV_GSO_CB(skb)->destructor = skb->destructor;
1996 skb->destructor = dev_gso_skb_destructor;
2002 * Try to orphan skb early, right before transmission by the device.
2003 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2004 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2006 static inline void skb_orphan_try(struct sk_buff *skb)
2008 struct sock *sk = skb->sk;
2010 if (sk && !skb_shinfo(skb)->tx_flags) {
2011 /* skb_tx_hash() wont be able to get sk.
2012 * We copy sk_hash into skb->rxhash
2015 skb->rxhash = sk->sk_hash;
2020 static int harmonize_features(struct sk_buff *skb, __be16 protocol, int features)
2022 if (!can_checksum_protocol(protocol, features)) {
2023 features &= ~NETIF_F_ALL_CSUM;
2024 features &= ~NETIF_F_SG;
2025 } else if (illegal_highdma(skb->dev, skb)) {
2026 features &= ~NETIF_F_SG;
2032 int netif_skb_features(struct sk_buff *skb)
2034 __be16 protocol = skb->protocol;
2035 int features = skb->dev->features;
2037 if (protocol == htons(ETH_P_8021Q)) {
2038 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2039 protocol = veh->h_vlan_encapsulated_proto;
2040 } else if (!vlan_tx_tag_present(skb)) {
2041 return harmonize_features(skb, protocol, features);
2044 features &= skb->dev->vlan_features;
2046 if (protocol != htons(ETH_P_8021Q)) {
2047 return harmonize_features(skb, protocol, features);
2049 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2051 return harmonize_features(skb, protocol, features);
2054 EXPORT_SYMBOL(netif_skb_features);
2057 * Returns true if either:
2058 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2059 * 2. skb is fragmented and the device does not support SG, or if
2060 * at least one of fragments is in highmem and device does not
2061 * support DMA from it.
2063 static inline int skb_needs_linearize(struct sk_buff *skb,
2064 struct net_device *dev)
2066 if (skb_is_nonlinear(skb)) {
2067 int features = dev->features;
2069 if (vlan_tx_tag_present(skb))
2070 features &= dev->vlan_features;
2072 return (skb_has_frag_list(skb) &&
2073 !(features & NETIF_F_FRAGLIST)) ||
2074 (skb_shinfo(skb)->nr_frags &&
2075 (!(features & NETIF_F_SG) ||
2076 illegal_highdma(dev, skb)));
2082 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2083 struct netdev_queue *txq)
2085 const struct net_device_ops *ops = dev->netdev_ops;
2086 int rc = NETDEV_TX_OK;
2088 if (likely(!skb->next)) {
2092 * If device doesnt need skb->dst, release it right now while
2093 * its hot in this cpu cache
2095 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2098 if (!list_empty(&ptype_all))
2099 dev_queue_xmit_nit(skb, dev);
2101 skb_orphan_try(skb);
2103 features = netif_skb_features(skb);
2105 if (vlan_tx_tag_present(skb) &&
2106 !(features & NETIF_F_HW_VLAN_TX)) {
2107 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2114 if (netif_needs_gso(skb, features)) {
2115 if (unlikely(dev_gso_segment(skb)))
2120 if (skb_needs_linearize(skb, dev) &&
2121 __skb_linearize(skb))
2124 /* If packet is not checksummed and device does not
2125 * support checksumming for this protocol, complete
2126 * checksumming here.
2128 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2129 skb_set_transport_header(skb,
2130 skb_checksum_start_offset(skb));
2131 if (!dev_can_checksum(dev, skb) &&
2132 skb_checksum_help(skb))
2137 rc = ops->ndo_start_xmit(skb, dev);
2138 trace_net_dev_xmit(skb, rc);
2139 if (rc == NETDEV_TX_OK)
2140 txq_trans_update(txq);
2146 struct sk_buff *nskb = skb->next;
2148 skb->next = nskb->next;
2152 * If device doesnt need nskb->dst, release it right now while
2153 * its hot in this cpu cache
2155 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2158 rc = ops->ndo_start_xmit(nskb, dev);
2159 trace_net_dev_xmit(nskb, rc);
2160 if (unlikely(rc != NETDEV_TX_OK)) {
2161 if (rc & ~NETDEV_TX_MASK)
2162 goto out_kfree_gso_skb;
2163 nskb->next = skb->next;
2167 txq_trans_update(txq);
2168 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2169 return NETDEV_TX_BUSY;
2170 } while (skb->next);
2173 if (likely(skb->next == NULL))
2174 skb->destructor = DEV_GSO_CB(skb)->destructor;
2181 static u32 hashrnd __read_mostly;
2184 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2185 * to be used as a distribution range.
2187 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2188 unsigned int num_tx_queues)
2192 if (skb_rx_queue_recorded(skb)) {
2193 hash = skb_get_rx_queue(skb);
2194 while (unlikely(hash >= num_tx_queues))
2195 hash -= num_tx_queues;
2199 if (skb->sk && skb->sk->sk_hash)
2200 hash = skb->sk->sk_hash;
2202 hash = (__force u16) skb->protocol ^ skb->rxhash;
2203 hash = jhash_1word(hash, hashrnd);
2205 return (u16) (((u64) hash * num_tx_queues) >> 32);
2207 EXPORT_SYMBOL(__skb_tx_hash);
2209 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2211 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2212 if (net_ratelimit()) {
2213 pr_warning("%s selects TX queue %d, but "
2214 "real number of TX queues is %d\n",
2215 dev->name, queue_index, dev->real_num_tx_queues);
2222 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2225 struct xps_dev_maps *dev_maps;
2226 struct xps_map *map;
2227 int queue_index = -1;
2230 dev_maps = rcu_dereference(dev->xps_maps);
2232 map = rcu_dereference(
2233 dev_maps->cpu_map[raw_smp_processor_id()]);
2236 queue_index = map->queues[0];
2239 if (skb->sk && skb->sk->sk_hash)
2240 hash = skb->sk->sk_hash;
2242 hash = (__force u16) skb->protocol ^
2244 hash = jhash_1word(hash, hashrnd);
2245 queue_index = map->queues[
2246 ((u64)hash * map->len) >> 32];
2248 if (unlikely(queue_index >= dev->real_num_tx_queues))
2260 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2261 struct sk_buff *skb)
2264 const struct net_device_ops *ops = dev->netdev_ops;
2266 if (dev->real_num_tx_queues == 1)
2268 else if (ops->ndo_select_queue) {
2269 queue_index = ops->ndo_select_queue(dev, skb);
2270 queue_index = dev_cap_txqueue(dev, queue_index);
2272 struct sock *sk = skb->sk;
2273 queue_index = sk_tx_queue_get(sk);
2275 if (queue_index < 0 || skb->ooo_okay ||
2276 queue_index >= dev->real_num_tx_queues) {
2277 int old_index = queue_index;
2279 queue_index = get_xps_queue(dev, skb);
2280 if (queue_index < 0)
2281 queue_index = skb_tx_hash(dev, skb);
2283 if (queue_index != old_index && sk) {
2284 struct dst_entry *dst =
2285 rcu_dereference_check(sk->sk_dst_cache, 1);
2287 if (dst && skb_dst(skb) == dst)
2288 sk_tx_queue_set(sk, queue_index);
2293 skb_set_queue_mapping(skb, queue_index);
2294 return netdev_get_tx_queue(dev, queue_index);
2297 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2298 struct net_device *dev,
2299 struct netdev_queue *txq)
2301 spinlock_t *root_lock = qdisc_lock(q);
2302 bool contended = qdisc_is_running(q);
2306 * Heuristic to force contended enqueues to serialize on a
2307 * separate lock before trying to get qdisc main lock.
2308 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2309 * and dequeue packets faster.
2311 if (unlikely(contended))
2312 spin_lock(&q->busylock);
2314 spin_lock(root_lock);
2315 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2318 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2319 qdisc_run_begin(q)) {
2321 * This is a work-conserving queue; there are no old skbs
2322 * waiting to be sent out; and the qdisc is not running -
2323 * xmit the skb directly.
2325 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2327 __qdisc_update_bstats(q, skb->len);
2328 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2329 if (unlikely(contended)) {
2330 spin_unlock(&q->busylock);
2337 rc = NET_XMIT_SUCCESS;
2340 rc = qdisc_enqueue_root(skb, q);
2341 if (qdisc_run_begin(q)) {
2342 if (unlikely(contended)) {
2343 spin_unlock(&q->busylock);
2349 spin_unlock(root_lock);
2350 if (unlikely(contended))
2351 spin_unlock(&q->busylock);
2355 static DEFINE_PER_CPU(int, xmit_recursion);
2356 #define RECURSION_LIMIT 10
2359 * dev_queue_xmit - transmit a buffer
2360 * @skb: buffer to transmit
2362 * Queue a buffer for transmission to a network device. The caller must
2363 * have set the device and priority and built the buffer before calling
2364 * this function. The function can be called from an interrupt.
2366 * A negative errno code is returned on a failure. A success does not
2367 * guarantee the frame will be transmitted as it may be dropped due
2368 * to congestion or traffic shaping.
2370 * -----------------------------------------------------------------------------------
2371 * I notice this method can also return errors from the queue disciplines,
2372 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2375 * Regardless of the return value, the skb is consumed, so it is currently
2376 * difficult to retry a send to this method. (You can bump the ref count
2377 * before sending to hold a reference for retry if you are careful.)
2379 * When calling this method, interrupts MUST be enabled. This is because
2380 * the BH enable code must have IRQs enabled so that it will not deadlock.
2383 int dev_queue_xmit(struct sk_buff *skb)
2385 struct net_device *dev = skb->dev;
2386 struct netdev_queue *txq;
2390 /* Disable soft irqs for various locks below. Also
2391 * stops preemption for RCU.
2395 txq = dev_pick_tx(dev, skb);
2396 q = rcu_dereference_bh(txq->qdisc);
2398 #ifdef CONFIG_NET_CLS_ACT
2399 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2401 trace_net_dev_queue(skb);
2403 rc = __dev_xmit_skb(skb, q, dev, txq);
2407 /* The device has no queue. Common case for software devices:
2408 loopback, all the sorts of tunnels...
2410 Really, it is unlikely that netif_tx_lock protection is necessary
2411 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2413 However, it is possible, that they rely on protection
2416 Check this and shot the lock. It is not prone from deadlocks.
2417 Either shot noqueue qdisc, it is even simpler 8)
2419 if (dev->flags & IFF_UP) {
2420 int cpu = smp_processor_id(); /* ok because BHs are off */
2422 if (txq->xmit_lock_owner != cpu) {
2424 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2425 goto recursion_alert;
2427 HARD_TX_LOCK(dev, txq, cpu);
2429 if (!netif_tx_queue_stopped(txq)) {
2430 __this_cpu_inc(xmit_recursion);
2431 rc = dev_hard_start_xmit(skb, dev, txq);
2432 __this_cpu_dec(xmit_recursion);
2433 if (dev_xmit_complete(rc)) {
2434 HARD_TX_UNLOCK(dev, txq);
2438 HARD_TX_UNLOCK(dev, txq);
2439 if (net_ratelimit())
2440 printk(KERN_CRIT "Virtual device %s asks to "
2441 "queue packet!\n", dev->name);
2443 /* Recursion is detected! It is possible,
2447 if (net_ratelimit())
2448 printk(KERN_CRIT "Dead loop on virtual device "
2449 "%s, fix it urgently!\n", dev->name);
2454 rcu_read_unlock_bh();
2459 rcu_read_unlock_bh();
2462 EXPORT_SYMBOL(dev_queue_xmit);
2465 /*=======================================================================
2467 =======================================================================*/
2469 int netdev_max_backlog __read_mostly = 1000;
2470 int netdev_tstamp_prequeue __read_mostly = 1;
2471 int netdev_budget __read_mostly = 300;
2472 int weight_p __read_mostly = 64; /* old backlog weight */
2474 /* Called with irq disabled */
2475 static inline void ____napi_schedule(struct softnet_data *sd,
2476 struct napi_struct *napi)
2478 list_add_tail(&napi->poll_list, &sd->poll_list);
2479 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2483 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2484 * and src/dst port numbers. Returns a non-zero hash number on success
2487 __u32 __skb_get_rxhash(struct sk_buff *skb)
2489 int nhoff, hash = 0, poff;
2490 struct ipv6hdr *ip6;
2493 u32 addr1, addr2, ihl;
2499 nhoff = skb_network_offset(skb);
2501 switch (skb->protocol) {
2502 case __constant_htons(ETH_P_IP):
2503 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2506 ip = (struct iphdr *) (skb->data + nhoff);
2507 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2510 ip_proto = ip->protocol;
2511 addr1 = (__force u32) ip->saddr;
2512 addr2 = (__force u32) ip->daddr;
2515 case __constant_htons(ETH_P_IPV6):
2516 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2519 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2520 ip_proto = ip6->nexthdr;
2521 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2522 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2530 poff = proto_ports_offset(ip_proto);
2532 nhoff += ihl * 4 + poff;
2533 if (pskb_may_pull(skb, nhoff + 4)) {
2534 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2535 if (ports.v16[1] < ports.v16[0])
2536 swap(ports.v16[0], ports.v16[1]);
2540 /* get a consistent hash (same value on both flow directions) */
2544 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2551 EXPORT_SYMBOL(__skb_get_rxhash);
2555 /* One global table that all flow-based protocols share. */
2556 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2557 EXPORT_SYMBOL(rps_sock_flow_table);
2560 * get_rps_cpu is called from netif_receive_skb and returns the target
2561 * CPU from the RPS map of the receiving queue for a given skb.
2562 * rcu_read_lock must be held on entry.
2564 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2565 struct rps_dev_flow **rflowp)
2567 struct netdev_rx_queue *rxqueue;
2568 struct rps_map *map;
2569 struct rps_dev_flow_table *flow_table;
2570 struct rps_sock_flow_table *sock_flow_table;
2574 if (skb_rx_queue_recorded(skb)) {
2575 u16 index = skb_get_rx_queue(skb);
2576 if (unlikely(index >= dev->real_num_rx_queues)) {
2577 WARN_ONCE(dev->real_num_rx_queues > 1,
2578 "%s received packet on queue %u, but number "
2579 "of RX queues is %u\n",
2580 dev->name, index, dev->real_num_rx_queues);
2583 rxqueue = dev->_rx + index;
2587 map = rcu_dereference(rxqueue->rps_map);
2589 if (map->len == 1) {
2590 tcpu = map->cpus[0];
2591 if (cpu_online(tcpu))
2595 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2599 skb_reset_network_header(skb);
2600 if (!skb_get_rxhash(skb))
2603 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2604 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2605 if (flow_table && sock_flow_table) {
2607 struct rps_dev_flow *rflow;
2609 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2612 next_cpu = sock_flow_table->ents[skb->rxhash &
2613 sock_flow_table->mask];
2616 * If the desired CPU (where last recvmsg was done) is
2617 * different from current CPU (one in the rx-queue flow
2618 * table entry), switch if one of the following holds:
2619 * - Current CPU is unset (equal to RPS_NO_CPU).
2620 * - Current CPU is offline.
2621 * - The current CPU's queue tail has advanced beyond the
2622 * last packet that was enqueued using this table entry.
2623 * This guarantees that all previous packets for the flow
2624 * have been dequeued, thus preserving in order delivery.
2626 if (unlikely(tcpu != next_cpu) &&
2627 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2628 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2629 rflow->last_qtail)) >= 0)) {
2630 tcpu = rflow->cpu = next_cpu;
2631 if (tcpu != RPS_NO_CPU)
2632 rflow->last_qtail = per_cpu(softnet_data,
2633 tcpu).input_queue_head;
2635 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2643 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2645 if (cpu_online(tcpu)) {
2655 /* Called from hardirq (IPI) context */
2656 static void rps_trigger_softirq(void *data)
2658 struct softnet_data *sd = data;
2660 ____napi_schedule(sd, &sd->backlog);
2664 #endif /* CONFIG_RPS */
2667 * Check if this softnet_data structure is another cpu one
2668 * If yes, queue it to our IPI list and return 1
2671 static int rps_ipi_queued(struct softnet_data *sd)
2674 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2677 sd->rps_ipi_next = mysd->rps_ipi_list;
2678 mysd->rps_ipi_list = sd;
2680 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2683 #endif /* CONFIG_RPS */
2688 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2689 * queue (may be a remote CPU queue).
2691 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2692 unsigned int *qtail)
2694 struct softnet_data *sd;
2695 unsigned long flags;
2697 sd = &per_cpu(softnet_data, cpu);
2699 local_irq_save(flags);
2702 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2703 if (skb_queue_len(&sd->input_pkt_queue)) {
2705 __skb_queue_tail(&sd->input_pkt_queue, skb);
2706 input_queue_tail_incr_save(sd, qtail);
2708 local_irq_restore(flags);
2709 return NET_RX_SUCCESS;
2712 /* Schedule NAPI for backlog device
2713 * We can use non atomic operation since we own the queue lock
2715 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2716 if (!rps_ipi_queued(sd))
2717 ____napi_schedule(sd, &sd->backlog);
2725 local_irq_restore(flags);
2727 atomic_long_inc(&skb->dev->rx_dropped);
2733 * netif_rx - post buffer to the network code
2734 * @skb: buffer to post
2736 * This function receives a packet from a device driver and queues it for
2737 * the upper (protocol) levels to process. It always succeeds. The buffer
2738 * may be dropped during processing for congestion control or by the
2742 * NET_RX_SUCCESS (no congestion)
2743 * NET_RX_DROP (packet was dropped)
2747 int netif_rx(struct sk_buff *skb)
2751 /* if netpoll wants it, pretend we never saw it */
2752 if (netpoll_rx(skb))
2755 if (netdev_tstamp_prequeue)
2756 net_timestamp_check(skb);
2758 trace_netif_rx(skb);
2761 struct rps_dev_flow voidflow, *rflow = &voidflow;
2767 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2769 cpu = smp_processor_id();
2771 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2779 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2785 EXPORT_SYMBOL(netif_rx);
2787 int netif_rx_ni(struct sk_buff *skb)
2792 err = netif_rx(skb);
2793 if (local_softirq_pending())
2799 EXPORT_SYMBOL(netif_rx_ni);
2801 static void net_tx_action(struct softirq_action *h)
2803 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2805 if (sd->completion_queue) {
2806 struct sk_buff *clist;
2808 local_irq_disable();
2809 clist = sd->completion_queue;
2810 sd->completion_queue = NULL;
2814 struct sk_buff *skb = clist;
2815 clist = clist->next;
2817 WARN_ON(atomic_read(&skb->users));
2818 trace_kfree_skb(skb, net_tx_action);
2823 if (sd->output_queue) {
2826 local_irq_disable();
2827 head = sd->output_queue;
2828 sd->output_queue = NULL;
2829 sd->output_queue_tailp = &sd->output_queue;
2833 struct Qdisc *q = head;
2834 spinlock_t *root_lock;
2836 head = head->next_sched;
2838 root_lock = qdisc_lock(q);
2839 if (spin_trylock(root_lock)) {
2840 smp_mb__before_clear_bit();
2841 clear_bit(__QDISC_STATE_SCHED,
2844 spin_unlock(root_lock);
2846 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2848 __netif_reschedule(q);
2850 smp_mb__before_clear_bit();
2851 clear_bit(__QDISC_STATE_SCHED,
2859 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2860 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2861 /* This hook is defined here for ATM LANE */
2862 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2863 unsigned char *addr) __read_mostly;
2864 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2867 #ifdef CONFIG_NET_CLS_ACT
2868 /* TODO: Maybe we should just force sch_ingress to be compiled in
2869 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2870 * a compare and 2 stores extra right now if we dont have it on
2871 * but have CONFIG_NET_CLS_ACT
2872 * NOTE: This doesnt stop any functionality; if you dont have
2873 * the ingress scheduler, you just cant add policies on ingress.
2876 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2878 struct net_device *dev = skb->dev;
2879 u32 ttl = G_TC_RTTL(skb->tc_verd);
2880 int result = TC_ACT_OK;
2883 if (unlikely(MAX_RED_LOOP < ttl++)) {
2884 if (net_ratelimit())
2885 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2886 skb->skb_iif, dev->ifindex);
2890 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2891 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2894 if (q != &noop_qdisc) {
2895 spin_lock(qdisc_lock(q));
2896 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2897 result = qdisc_enqueue_root(skb, q);
2898 spin_unlock(qdisc_lock(q));
2904 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2905 struct packet_type **pt_prev,
2906 int *ret, struct net_device *orig_dev)
2908 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
2910 if (!rxq || rxq->qdisc == &noop_qdisc)
2914 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2918 switch (ing_filter(skb, rxq)) {
2932 * netdev_rx_handler_register - register receive handler
2933 * @dev: device to register a handler for
2934 * @rx_handler: receive handler to register
2935 * @rx_handler_data: data pointer that is used by rx handler
2937 * Register a receive hander for a device. This handler will then be
2938 * called from __netif_receive_skb. A negative errno code is returned
2941 * The caller must hold the rtnl_mutex.
2943 int netdev_rx_handler_register(struct net_device *dev,
2944 rx_handler_func_t *rx_handler,
2945 void *rx_handler_data)
2949 if (dev->rx_handler)
2952 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2953 rcu_assign_pointer(dev->rx_handler, rx_handler);
2957 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2960 * netdev_rx_handler_unregister - unregister receive handler
2961 * @dev: device to unregister a handler from
2963 * Unregister a receive hander from a device.
2965 * The caller must hold the rtnl_mutex.
2967 void netdev_rx_handler_unregister(struct net_device *dev)
2971 rcu_assign_pointer(dev->rx_handler, NULL);
2972 rcu_assign_pointer(dev->rx_handler_data, NULL);
2974 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2976 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2977 struct net_device *master)
2979 if (skb->pkt_type == PACKET_HOST) {
2980 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2982 memcpy(dest, master->dev_addr, ETH_ALEN);
2986 /* On bonding slaves other than the currently active slave, suppress
2987 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2988 * ARP on active-backup slaves with arp_validate enabled.
2990 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2992 struct net_device *dev = skb->dev;
2994 if (master->priv_flags & IFF_MASTER_ARPMON)
2995 dev->last_rx = jiffies;
2997 if ((master->priv_flags & IFF_MASTER_ALB) &&
2998 (master->priv_flags & IFF_BRIDGE_PORT)) {
2999 /* Do address unmangle. The local destination address
3000 * will be always the one master has. Provides the right
3001 * functionality in a bridge.
3003 skb_bond_set_mac_by_master(skb, master);
3006 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
3007 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
3008 skb->protocol == __cpu_to_be16(ETH_P_ARP))
3011 if (master->priv_flags & IFF_MASTER_ALB) {
3012 if (skb->pkt_type != PACKET_BROADCAST &&
3013 skb->pkt_type != PACKET_MULTICAST)
3016 if (master->priv_flags & IFF_MASTER_8023AD &&
3017 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
3024 EXPORT_SYMBOL(__skb_bond_should_drop);
3026 static int __netif_receive_skb(struct sk_buff *skb)
3028 struct packet_type *ptype, *pt_prev;
3029 rx_handler_func_t *rx_handler;
3030 struct net_device *orig_dev;
3031 struct net_device *master;
3032 struct net_device *null_or_orig;
3033 struct net_device *orig_or_bond;
3034 int ret = NET_RX_DROP;
3037 if (!netdev_tstamp_prequeue)
3038 net_timestamp_check(skb);
3040 trace_netif_receive_skb(skb);
3042 /* if we've gotten here through NAPI, check netpoll */
3043 if (netpoll_receive_skb(skb))
3047 skb->skb_iif = skb->dev->ifindex;
3050 * bonding note: skbs received on inactive slaves should only
3051 * be delivered to pkt handlers that are exact matches. Also
3052 * the deliver_no_wcard flag will be set. If packet handlers
3053 * are sensitive to duplicate packets these skbs will need to
3054 * be dropped at the handler.
3056 null_or_orig = NULL;
3057 orig_dev = skb->dev;
3058 master = ACCESS_ONCE(orig_dev->master);
3059 if (skb->deliver_no_wcard)
3060 null_or_orig = orig_dev;
3062 if (skb_bond_should_drop(skb, master)) {
3063 skb->deliver_no_wcard = 1;
3064 null_or_orig = orig_dev; /* deliver only exact match */
3069 __this_cpu_inc(softnet_data.processed);
3070 skb_reset_network_header(skb);
3071 skb_reset_transport_header(skb);
3072 skb->mac_len = skb->network_header - skb->mac_header;
3078 #ifdef CONFIG_NET_CLS_ACT
3079 if (skb->tc_verd & TC_NCLS) {
3080 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3085 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3086 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
3087 ptype->dev == orig_dev) {
3089 ret = deliver_skb(skb, pt_prev, orig_dev);
3094 #ifdef CONFIG_NET_CLS_ACT
3095 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3101 /* Handle special case of bridge or macvlan */
3102 rx_handler = rcu_dereference(skb->dev->rx_handler);
3105 ret = deliver_skb(skb, pt_prev, orig_dev);
3108 skb = rx_handler(skb);
3113 if (vlan_tx_tag_present(skb)) {
3115 ret = deliver_skb(skb, pt_prev, orig_dev);
3118 if (vlan_hwaccel_do_receive(&skb)) {
3119 ret = __netif_receive_skb(skb);
3121 } else if (unlikely(!skb))
3126 * Make sure frames received on VLAN interfaces stacked on
3127 * bonding interfaces still make their way to any base bonding
3128 * device that may have registered for a specific ptype. The
3129 * handler may have to adjust skb->dev and orig_dev.
3131 orig_or_bond = orig_dev;
3132 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
3133 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
3134 orig_or_bond = vlan_dev_real_dev(skb->dev);
3137 type = skb->protocol;
3138 list_for_each_entry_rcu(ptype,
3139 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3140 if (ptype->type == type && (ptype->dev == null_or_orig ||
3141 ptype->dev == skb->dev || ptype->dev == orig_dev ||
3142 ptype->dev == orig_or_bond)) {
3144 ret = deliver_skb(skb, pt_prev, orig_dev);
3150 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3152 atomic_long_inc(&skb->dev->rx_dropped);
3154 /* Jamal, now you will not able to escape explaining
3155 * me how you were going to use this. :-)
3166 * netif_receive_skb - process receive buffer from network
3167 * @skb: buffer to process
3169 * netif_receive_skb() is the main receive data processing function.
3170 * It always succeeds. The buffer may be dropped during processing
3171 * for congestion control or by the protocol layers.
3173 * This function may only be called from softirq context and interrupts
3174 * should be enabled.
3176 * Return values (usually ignored):
3177 * NET_RX_SUCCESS: no congestion
3178 * NET_RX_DROP: packet was dropped
3180 int netif_receive_skb(struct sk_buff *skb)
3182 if (netdev_tstamp_prequeue)
3183 net_timestamp_check(skb);
3185 if (skb_defer_rx_timestamp(skb))
3186 return NET_RX_SUCCESS;
3190 struct rps_dev_flow voidflow, *rflow = &voidflow;
3195 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3198 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3202 ret = __netif_receive_skb(skb);
3208 return __netif_receive_skb(skb);
3211 EXPORT_SYMBOL(netif_receive_skb);
3213 /* Network device is going away, flush any packets still pending
3214 * Called with irqs disabled.
3216 static void flush_backlog(void *arg)
3218 struct net_device *dev = arg;
3219 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3220 struct sk_buff *skb, *tmp;
3223 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3224 if (skb->dev == dev) {
3225 __skb_unlink(skb, &sd->input_pkt_queue);
3227 input_queue_head_incr(sd);
3232 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3233 if (skb->dev == dev) {
3234 __skb_unlink(skb, &sd->process_queue);
3236 input_queue_head_incr(sd);
3241 static int napi_gro_complete(struct sk_buff *skb)
3243 struct packet_type *ptype;
3244 __be16 type = skb->protocol;
3245 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3248 if (NAPI_GRO_CB(skb)->count == 1) {
3249 skb_shinfo(skb)->gso_size = 0;
3254 list_for_each_entry_rcu(ptype, head, list) {
3255 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3258 err = ptype->gro_complete(skb);
3264 WARN_ON(&ptype->list == head);
3266 return NET_RX_SUCCESS;
3270 return netif_receive_skb(skb);
3273 inline void napi_gro_flush(struct napi_struct *napi)
3275 struct sk_buff *skb, *next;
3277 for (skb = napi->gro_list; skb; skb = next) {
3280 napi_gro_complete(skb);
3283 napi->gro_count = 0;
3284 napi->gro_list = NULL;
3286 EXPORT_SYMBOL(napi_gro_flush);
3288 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3290 struct sk_buff **pp = NULL;
3291 struct packet_type *ptype;
3292 __be16 type = skb->protocol;
3293 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3296 enum gro_result ret;
3298 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3301 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3305 list_for_each_entry_rcu(ptype, head, list) {
3306 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3309 skb_set_network_header(skb, skb_gro_offset(skb));
3310 mac_len = skb->network_header - skb->mac_header;
3311 skb->mac_len = mac_len;
3312 NAPI_GRO_CB(skb)->same_flow = 0;
3313 NAPI_GRO_CB(skb)->flush = 0;
3314 NAPI_GRO_CB(skb)->free = 0;
3316 pp = ptype->gro_receive(&napi->gro_list, skb);
3321 if (&ptype->list == head)
3324 same_flow = NAPI_GRO_CB(skb)->same_flow;
3325 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3328 struct sk_buff *nskb = *pp;
3332 napi_gro_complete(nskb);
3339 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3343 NAPI_GRO_CB(skb)->count = 1;
3344 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3345 skb->next = napi->gro_list;
3346 napi->gro_list = skb;
3350 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3351 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3353 BUG_ON(skb->end - skb->tail < grow);
3355 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3358 skb->data_len -= grow;
3360 skb_shinfo(skb)->frags[0].page_offset += grow;
3361 skb_shinfo(skb)->frags[0].size -= grow;
3363 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3364 put_page(skb_shinfo(skb)->frags[0].page);
3365 memmove(skb_shinfo(skb)->frags,
3366 skb_shinfo(skb)->frags + 1,
3367 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3378 EXPORT_SYMBOL(dev_gro_receive);
3380 static inline gro_result_t
3381 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3385 for (p = napi->gro_list; p; p = p->next) {
3386 unsigned long diffs;
3388 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3389 diffs |= p->vlan_tci ^ skb->vlan_tci;
3390 diffs |= compare_ether_header(skb_mac_header(p),
3391 skb_gro_mac_header(skb));
3392 NAPI_GRO_CB(p)->same_flow = !diffs;
3393 NAPI_GRO_CB(p)->flush = 0;
3396 return dev_gro_receive(napi, skb);
3399 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3403 if (netif_receive_skb(skb))
3408 case GRO_MERGED_FREE:
3419 EXPORT_SYMBOL(napi_skb_finish);
3421 void skb_gro_reset_offset(struct sk_buff *skb)
3423 NAPI_GRO_CB(skb)->data_offset = 0;
3424 NAPI_GRO_CB(skb)->frag0 = NULL;
3425 NAPI_GRO_CB(skb)->frag0_len = 0;
3427 if (skb->mac_header == skb->tail &&
3428 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3429 NAPI_GRO_CB(skb)->frag0 =
3430 page_address(skb_shinfo(skb)->frags[0].page) +
3431 skb_shinfo(skb)->frags[0].page_offset;
3432 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3435 EXPORT_SYMBOL(skb_gro_reset_offset);
3437 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3439 skb_gro_reset_offset(skb);
3441 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3443 EXPORT_SYMBOL(napi_gro_receive);
3445 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3447 __skb_pull(skb, skb_headlen(skb));
3448 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3454 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3456 struct sk_buff *skb = napi->skb;
3459 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3465 EXPORT_SYMBOL(napi_get_frags);
3467 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3473 skb->protocol = eth_type_trans(skb, skb->dev);
3475 if (ret == GRO_HELD)
3476 skb_gro_pull(skb, -ETH_HLEN);
3477 else if (netif_receive_skb(skb))
3482 case GRO_MERGED_FREE:
3483 napi_reuse_skb(napi, skb);
3492 EXPORT_SYMBOL(napi_frags_finish);
3494 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3496 struct sk_buff *skb = napi->skb;
3503 skb_reset_mac_header(skb);
3504 skb_gro_reset_offset(skb);
3506 off = skb_gro_offset(skb);
3507 hlen = off + sizeof(*eth);
3508 eth = skb_gro_header_fast(skb, off);
3509 if (skb_gro_header_hard(skb, hlen)) {
3510 eth = skb_gro_header_slow(skb, hlen, off);
3511 if (unlikely(!eth)) {
3512 napi_reuse_skb(napi, skb);
3518 skb_gro_pull(skb, sizeof(*eth));
3521 * This works because the only protocols we care about don't require
3522 * special handling. We'll fix it up properly at the end.
3524 skb->protocol = eth->h_proto;
3529 EXPORT_SYMBOL(napi_frags_skb);
3531 gro_result_t napi_gro_frags(struct napi_struct *napi)
3533 struct sk_buff *skb = napi_frags_skb(napi);
3538 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3540 EXPORT_SYMBOL(napi_gro_frags);
3543 * net_rps_action sends any pending IPI's for rps.
3544 * Note: called with local irq disabled, but exits with local irq enabled.
3546 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3549 struct softnet_data *remsd = sd->rps_ipi_list;
3552 sd->rps_ipi_list = NULL;
3556 /* Send pending IPI's to kick RPS processing on remote cpus. */
3558 struct softnet_data *next = remsd->rps_ipi_next;
3560 if (cpu_online(remsd->cpu))
3561 __smp_call_function_single(remsd->cpu,
3570 static int process_backlog(struct napi_struct *napi, int quota)
3573 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3576 /* Check if we have pending ipi, its better to send them now,
3577 * not waiting net_rx_action() end.
3579 if (sd->rps_ipi_list) {
3580 local_irq_disable();
3581 net_rps_action_and_irq_enable(sd);
3584 napi->weight = weight_p;
3585 local_irq_disable();
3586 while (work < quota) {
3587 struct sk_buff *skb;
3590 while ((skb = __skb_dequeue(&sd->process_queue))) {
3592 __netif_receive_skb(skb);
3593 local_irq_disable();
3594 input_queue_head_incr(sd);
3595 if (++work >= quota) {
3602 qlen = skb_queue_len(&sd->input_pkt_queue);
3604 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3605 &sd->process_queue);
3607 if (qlen < quota - work) {
3609 * Inline a custom version of __napi_complete().
3610 * only current cpu owns and manipulates this napi,
3611 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3612 * we can use a plain write instead of clear_bit(),
3613 * and we dont need an smp_mb() memory barrier.
3615 list_del(&napi->poll_list);
3618 quota = work + qlen;
3628 * __napi_schedule - schedule for receive
3629 * @n: entry to schedule
3631 * The entry's receive function will be scheduled to run
3633 void __napi_schedule(struct napi_struct *n)
3635 unsigned long flags;
3637 local_irq_save(flags);
3638 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3639 local_irq_restore(flags);
3641 EXPORT_SYMBOL(__napi_schedule);
3643 void __napi_complete(struct napi_struct *n)
3645 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3646 BUG_ON(n->gro_list);
3648 list_del(&n->poll_list);
3649 smp_mb__before_clear_bit();
3650 clear_bit(NAPI_STATE_SCHED, &n->state);
3652 EXPORT_SYMBOL(__napi_complete);
3654 void napi_complete(struct napi_struct *n)
3656 unsigned long flags;
3659 * don't let napi dequeue from the cpu poll list
3660 * just in case its running on a different cpu
3662 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3666 local_irq_save(flags);
3668 local_irq_restore(flags);
3670 EXPORT_SYMBOL(napi_complete);
3672 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3673 int (*poll)(struct napi_struct *, int), int weight)
3675 INIT_LIST_HEAD(&napi->poll_list);
3676 napi->gro_count = 0;
3677 napi->gro_list = NULL;
3680 napi->weight = weight;
3681 list_add(&napi->dev_list, &dev->napi_list);
3683 #ifdef CONFIG_NETPOLL
3684 spin_lock_init(&napi->poll_lock);
3685 napi->poll_owner = -1;
3687 set_bit(NAPI_STATE_SCHED, &napi->state);
3689 EXPORT_SYMBOL(netif_napi_add);
3691 void netif_napi_del(struct napi_struct *napi)
3693 struct sk_buff *skb, *next;
3695 list_del_init(&napi->dev_list);
3696 napi_free_frags(napi);
3698 for (skb = napi->gro_list; skb; skb = next) {
3704 napi->gro_list = NULL;
3705 napi->gro_count = 0;
3707 EXPORT_SYMBOL(netif_napi_del);
3709 static void net_rx_action(struct softirq_action *h)
3711 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3712 unsigned long time_limit = jiffies + 2;
3713 int budget = netdev_budget;
3716 local_irq_disable();
3718 while (!list_empty(&sd->poll_list)) {
3719 struct napi_struct *n;
3722 /* If softirq window is exhuasted then punt.
3723 * Allow this to run for 2 jiffies since which will allow
3724 * an average latency of 1.5/HZ.
3726 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3731 /* Even though interrupts have been re-enabled, this
3732 * access is safe because interrupts can only add new
3733 * entries to the tail of this list, and only ->poll()
3734 * calls can remove this head entry from the list.
3736 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3738 have = netpoll_poll_lock(n);
3742 /* This NAPI_STATE_SCHED test is for avoiding a race
3743 * with netpoll's poll_napi(). Only the entity which
3744 * obtains the lock and sees NAPI_STATE_SCHED set will
3745 * actually make the ->poll() call. Therefore we avoid
3746 * accidently calling ->poll() when NAPI is not scheduled.
3749 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3750 work = n->poll(n, weight);
3754 WARN_ON_ONCE(work > weight);
3758 local_irq_disable();
3760 /* Drivers must not modify the NAPI state if they
3761 * consume the entire weight. In such cases this code
3762 * still "owns" the NAPI instance and therefore can
3763 * move the instance around on the list at-will.
3765 if (unlikely(work == weight)) {
3766 if (unlikely(napi_disable_pending(n))) {
3769 local_irq_disable();
3771 list_move_tail(&n->poll_list, &sd->poll_list);
3774 netpoll_poll_unlock(have);
3777 net_rps_action_and_irq_enable(sd);
3779 #ifdef CONFIG_NET_DMA
3781 * There may not be any more sk_buffs coming right now, so push
3782 * any pending DMA copies to hardware
3784 dma_issue_pending_all();
3791 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3795 static gifconf_func_t *gifconf_list[NPROTO];
3798 * register_gifconf - register a SIOCGIF handler
3799 * @family: Address family
3800 * @gifconf: Function handler
3802 * Register protocol dependent address dumping routines. The handler
3803 * that is passed must not be freed or reused until it has been replaced
3804 * by another handler.
3806 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3808 if (family >= NPROTO)
3810 gifconf_list[family] = gifconf;
3813 EXPORT_SYMBOL(register_gifconf);
3817 * Map an interface index to its name (SIOCGIFNAME)
3821 * We need this ioctl for efficient implementation of the
3822 * if_indextoname() function required by the IPv6 API. Without
3823 * it, we would have to search all the interfaces to find a
3827 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3829 struct net_device *dev;
3833 * Fetch the caller's info block.
3836 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3840 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3846 strcpy(ifr.ifr_name, dev->name);
3849 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3855 * Perform a SIOCGIFCONF call. This structure will change
3856 * size eventually, and there is nothing I can do about it.
3857 * Thus we will need a 'compatibility mode'.
3860 static int dev_ifconf(struct net *net, char __user *arg)
3863 struct net_device *dev;
3870 * Fetch the caller's info block.
3873 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3880 * Loop over the interfaces, and write an info block for each.
3884 for_each_netdev(net, dev) {
3885 for (i = 0; i < NPROTO; i++) {
3886 if (gifconf_list[i]) {
3889 done = gifconf_list[i](dev, NULL, 0);
3891 done = gifconf_list[i](dev, pos + total,
3901 * All done. Write the updated control block back to the caller.
3903 ifc.ifc_len = total;
3906 * Both BSD and Solaris return 0 here, so we do too.
3908 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3911 #ifdef CONFIG_PROC_FS
3913 * This is invoked by the /proc filesystem handler to display a device
3916 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3919 struct net *net = seq_file_net(seq);
3921 struct net_device *dev;
3925 return SEQ_START_TOKEN;
3928 for_each_netdev_rcu(net, dev)
3935 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3937 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3938 first_net_device(seq_file_net(seq)) :
3939 next_net_device((struct net_device *)v);
3942 return rcu_dereference(dev);
3945 void dev_seq_stop(struct seq_file *seq, void *v)
3951 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3953 struct rtnl_link_stats64 temp;
3954 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3956 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3957 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3958 dev->name, stats->rx_bytes, stats->rx_packets,
3960 stats->rx_dropped + stats->rx_missed_errors,
3961 stats->rx_fifo_errors,
3962 stats->rx_length_errors + stats->rx_over_errors +
3963 stats->rx_crc_errors + stats->rx_frame_errors,
3964 stats->rx_compressed, stats->multicast,
3965 stats->tx_bytes, stats->tx_packets,
3966 stats->tx_errors, stats->tx_dropped,
3967 stats->tx_fifo_errors, stats->collisions,
3968 stats->tx_carrier_errors +
3969 stats->tx_aborted_errors +
3970 stats->tx_window_errors +
3971 stats->tx_heartbeat_errors,
3972 stats->tx_compressed);
3976 * Called from the PROCfs module. This now uses the new arbitrary sized
3977 * /proc/net interface to create /proc/net/dev
3979 static int dev_seq_show(struct seq_file *seq, void *v)
3981 if (v == SEQ_START_TOKEN)
3982 seq_puts(seq, "Inter-| Receive "
3984 " face |bytes packets errs drop fifo frame "
3985 "compressed multicast|bytes packets errs "
3986 "drop fifo colls carrier compressed\n");
3988 dev_seq_printf_stats(seq, v);
3992 static struct softnet_data *softnet_get_online(loff_t *pos)
3994 struct softnet_data *sd = NULL;
3996 while (*pos < nr_cpu_ids)
3997 if (cpu_online(*pos)) {
3998 sd = &per_cpu(softnet_data, *pos);
4005 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4007 return softnet_get_online(pos);
4010 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4013 return softnet_get_online(pos);
4016 static void softnet_seq_stop(struct seq_file *seq, void *v)
4020 static int softnet_seq_show(struct seq_file *seq, void *v)
4022 struct softnet_data *sd = v;
4024 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4025 sd->processed, sd->dropped, sd->time_squeeze, 0,
4026 0, 0, 0, 0, /* was fastroute */
4027 sd->cpu_collision, sd->received_rps);
4031 static const struct seq_operations dev_seq_ops = {
4032 .start = dev_seq_start,
4033 .next = dev_seq_next,
4034 .stop = dev_seq_stop,
4035 .show = dev_seq_show,
4038 static int dev_seq_open(struct inode *inode, struct file *file)
4040 return seq_open_net(inode, file, &dev_seq_ops,
4041 sizeof(struct seq_net_private));
4044 static const struct file_operations dev_seq_fops = {
4045 .owner = THIS_MODULE,
4046 .open = dev_seq_open,
4048 .llseek = seq_lseek,
4049 .release = seq_release_net,
4052 static const struct seq_operations softnet_seq_ops = {
4053 .start = softnet_seq_start,
4054 .next = softnet_seq_next,
4055 .stop = softnet_seq_stop,
4056 .show = softnet_seq_show,
4059 static int softnet_seq_open(struct inode *inode, struct file *file)
4061 return seq_open(file, &softnet_seq_ops);
4064 static const struct file_operations softnet_seq_fops = {
4065 .owner = THIS_MODULE,
4066 .open = softnet_seq_open,
4068 .llseek = seq_lseek,
4069 .release = seq_release,
4072 static void *ptype_get_idx(loff_t pos)
4074 struct packet_type *pt = NULL;
4078 list_for_each_entry_rcu(pt, &ptype_all, list) {
4084 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4085 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4094 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4098 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4101 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4103 struct packet_type *pt;
4104 struct list_head *nxt;
4108 if (v == SEQ_START_TOKEN)
4109 return ptype_get_idx(0);
4112 nxt = pt->list.next;
4113 if (pt->type == htons(ETH_P_ALL)) {
4114 if (nxt != &ptype_all)
4117 nxt = ptype_base[0].next;
4119 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4121 while (nxt == &ptype_base[hash]) {
4122 if (++hash >= PTYPE_HASH_SIZE)
4124 nxt = ptype_base[hash].next;
4127 return list_entry(nxt, struct packet_type, list);
4130 static void ptype_seq_stop(struct seq_file *seq, void *v)
4136 static int ptype_seq_show(struct seq_file *seq, void *v)
4138 struct packet_type *pt = v;
4140 if (v == SEQ_START_TOKEN)
4141 seq_puts(seq, "Type Device Function\n");
4142 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4143 if (pt->type == htons(ETH_P_ALL))
4144 seq_puts(seq, "ALL ");
4146 seq_printf(seq, "%04x", ntohs(pt->type));
4148 seq_printf(seq, " %-8s %pF\n",
4149 pt->dev ? pt->dev->name : "", pt->func);
4155 static const struct seq_operations ptype_seq_ops = {
4156 .start = ptype_seq_start,
4157 .next = ptype_seq_next,
4158 .stop = ptype_seq_stop,
4159 .show = ptype_seq_show,
4162 static int ptype_seq_open(struct inode *inode, struct file *file)
4164 return seq_open_net(inode, file, &ptype_seq_ops,
4165 sizeof(struct seq_net_private));
4168 static const struct file_operations ptype_seq_fops = {
4169 .owner = THIS_MODULE,
4170 .open = ptype_seq_open,
4172 .llseek = seq_lseek,
4173 .release = seq_release_net,
4177 static int __net_init dev_proc_net_init(struct net *net)
4181 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4183 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4185 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4188 if (wext_proc_init(net))
4194 proc_net_remove(net, "ptype");
4196 proc_net_remove(net, "softnet_stat");
4198 proc_net_remove(net, "dev");
4202 static void __net_exit dev_proc_net_exit(struct net *net)
4204 wext_proc_exit(net);
4206 proc_net_remove(net, "ptype");
4207 proc_net_remove(net, "softnet_stat");
4208 proc_net_remove(net, "dev");
4211 static struct pernet_operations __net_initdata dev_proc_ops = {
4212 .init = dev_proc_net_init,
4213 .exit = dev_proc_net_exit,
4216 static int __init dev_proc_init(void)
4218 return register_pernet_subsys(&dev_proc_ops);
4221 #define dev_proc_init() 0
4222 #endif /* CONFIG_PROC_FS */
4226 * netdev_set_master - set up master/slave pair
4227 * @slave: slave device
4228 * @master: new master device
4230 * Changes the master device of the slave. Pass %NULL to break the
4231 * bonding. The caller must hold the RTNL semaphore. On a failure
4232 * a negative errno code is returned. On success the reference counts
4233 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4234 * function returns zero.
4236 int netdev_set_master(struct net_device *slave, struct net_device *master)
4238 struct net_device *old = slave->master;
4248 slave->master = master;
4255 slave->flags |= IFF_SLAVE;
4257 slave->flags &= ~IFF_SLAVE;
4259 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4262 EXPORT_SYMBOL(netdev_set_master);
4264 static void dev_change_rx_flags(struct net_device *dev, int flags)
4266 const struct net_device_ops *ops = dev->netdev_ops;
4268 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4269 ops->ndo_change_rx_flags(dev, flags);
4272 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4274 unsigned short old_flags = dev->flags;
4280 dev->flags |= IFF_PROMISC;
4281 dev->promiscuity += inc;
4282 if (dev->promiscuity == 0) {
4285 * If inc causes overflow, untouch promisc and return error.
4288 dev->flags &= ~IFF_PROMISC;
4290 dev->promiscuity -= inc;
4291 printk(KERN_WARNING "%s: promiscuity touches roof, "
4292 "set promiscuity failed, promiscuity feature "
4293 "of device might be broken.\n", dev->name);
4297 if (dev->flags != old_flags) {
4298 printk(KERN_INFO "device %s %s promiscuous mode\n",
4299 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4301 if (audit_enabled) {
4302 current_uid_gid(&uid, &gid);
4303 audit_log(current->audit_context, GFP_ATOMIC,
4304 AUDIT_ANOM_PROMISCUOUS,
4305 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4306 dev->name, (dev->flags & IFF_PROMISC),
4307 (old_flags & IFF_PROMISC),
4308 audit_get_loginuid(current),
4310 audit_get_sessionid(current));
4313 dev_change_rx_flags(dev, IFF_PROMISC);
4319 * dev_set_promiscuity - update promiscuity count on a device
4323 * Add or remove promiscuity from a device. While the count in the device
4324 * remains above zero the interface remains promiscuous. Once it hits zero
4325 * the device reverts back to normal filtering operation. A negative inc
4326 * value is used to drop promiscuity on the device.
4327 * Return 0 if successful or a negative errno code on error.
4329 int dev_set_promiscuity(struct net_device *dev, int inc)
4331 unsigned short old_flags = dev->flags;
4334 err = __dev_set_promiscuity(dev, inc);
4337 if (dev->flags != old_flags)
4338 dev_set_rx_mode(dev);
4341 EXPORT_SYMBOL(dev_set_promiscuity);
4344 * dev_set_allmulti - update allmulti count on a device
4348 * Add or remove reception of all multicast frames to a device. While the
4349 * count in the device remains above zero the interface remains listening
4350 * to all interfaces. Once it hits zero the device reverts back to normal
4351 * filtering operation. A negative @inc value is used to drop the counter
4352 * when releasing a resource needing all multicasts.
4353 * Return 0 if successful or a negative errno code on error.
4356 int dev_set_allmulti(struct net_device *dev, int inc)
4358 unsigned short old_flags = dev->flags;
4362 dev->flags |= IFF_ALLMULTI;
4363 dev->allmulti += inc;
4364 if (dev->allmulti == 0) {
4367 * If inc causes overflow, untouch allmulti and return error.
4370 dev->flags &= ~IFF_ALLMULTI;
4372 dev->allmulti -= inc;
4373 printk(KERN_WARNING "%s: allmulti touches roof, "
4374 "set allmulti failed, allmulti feature of "
4375 "device might be broken.\n", dev->name);
4379 if (dev->flags ^ old_flags) {
4380 dev_change_rx_flags(dev, IFF_ALLMULTI);
4381 dev_set_rx_mode(dev);
4385 EXPORT_SYMBOL(dev_set_allmulti);
4388 * Upload unicast and multicast address lists to device and
4389 * configure RX filtering. When the device doesn't support unicast
4390 * filtering it is put in promiscuous mode while unicast addresses
4393 void __dev_set_rx_mode(struct net_device *dev)
4395 const struct net_device_ops *ops = dev->netdev_ops;
4397 /* dev_open will call this function so the list will stay sane. */
4398 if (!(dev->flags&IFF_UP))
4401 if (!netif_device_present(dev))
4404 if (ops->ndo_set_rx_mode)
4405 ops->ndo_set_rx_mode(dev);
4407 /* Unicast addresses changes may only happen under the rtnl,
4408 * therefore calling __dev_set_promiscuity here is safe.
4410 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4411 __dev_set_promiscuity(dev, 1);
4412 dev->uc_promisc = 1;
4413 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4414 __dev_set_promiscuity(dev, -1);
4415 dev->uc_promisc = 0;
4418 if (ops->ndo_set_multicast_list)
4419 ops->ndo_set_multicast_list(dev);
4423 void dev_set_rx_mode(struct net_device *dev)
4425 netif_addr_lock_bh(dev);
4426 __dev_set_rx_mode(dev);
4427 netif_addr_unlock_bh(dev);
4431 * dev_get_flags - get flags reported to userspace
4434 * Get the combination of flag bits exported through APIs to userspace.
4436 unsigned dev_get_flags(const struct net_device *dev)
4440 flags = (dev->flags & ~(IFF_PROMISC |
4445 (dev->gflags & (IFF_PROMISC |
4448 if (netif_running(dev)) {
4449 if (netif_oper_up(dev))
4450 flags |= IFF_RUNNING;
4451 if (netif_carrier_ok(dev))
4452 flags |= IFF_LOWER_UP;
4453 if (netif_dormant(dev))
4454 flags |= IFF_DORMANT;
4459 EXPORT_SYMBOL(dev_get_flags);
4461 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4463 int old_flags = dev->flags;
4469 * Set the flags on our device.
4472 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4473 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4475 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4479 * Load in the correct multicast list now the flags have changed.
4482 if ((old_flags ^ flags) & IFF_MULTICAST)
4483 dev_change_rx_flags(dev, IFF_MULTICAST);
4485 dev_set_rx_mode(dev);
4488 * Have we downed the interface. We handle IFF_UP ourselves
4489 * according to user attempts to set it, rather than blindly
4494 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4495 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4498 dev_set_rx_mode(dev);
4501 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4502 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4504 dev->gflags ^= IFF_PROMISC;
4505 dev_set_promiscuity(dev, inc);
4508 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4509 is important. Some (broken) drivers set IFF_PROMISC, when
4510 IFF_ALLMULTI is requested not asking us and not reporting.
4512 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4513 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4515 dev->gflags ^= IFF_ALLMULTI;
4516 dev_set_allmulti(dev, inc);
4522 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4524 unsigned int changes = dev->flags ^ old_flags;
4526 if (changes & IFF_UP) {
4527 if (dev->flags & IFF_UP)
4528 call_netdevice_notifiers(NETDEV_UP, dev);
4530 call_netdevice_notifiers(NETDEV_DOWN, dev);
4533 if (dev->flags & IFF_UP &&
4534 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4535 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4539 * dev_change_flags - change device settings
4541 * @flags: device state flags
4543 * Change settings on device based state flags. The flags are
4544 * in the userspace exported format.
4546 int dev_change_flags(struct net_device *dev, unsigned flags)
4549 int old_flags = dev->flags;
4551 ret = __dev_change_flags(dev, flags);
4555 changes = old_flags ^ dev->flags;
4557 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4559 __dev_notify_flags(dev, old_flags);
4562 EXPORT_SYMBOL(dev_change_flags);
4565 * dev_set_mtu - Change maximum transfer unit
4567 * @new_mtu: new transfer unit
4569 * Change the maximum transfer size of the network device.
4571 int dev_set_mtu(struct net_device *dev, int new_mtu)
4573 const struct net_device_ops *ops = dev->netdev_ops;
4576 if (new_mtu == dev->mtu)
4579 /* MTU must be positive. */
4583 if (!netif_device_present(dev))
4587 if (ops->ndo_change_mtu)
4588 err = ops->ndo_change_mtu(dev, new_mtu);
4592 if (!err && dev->flags & IFF_UP)
4593 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4596 EXPORT_SYMBOL(dev_set_mtu);
4599 * dev_set_mac_address - Change Media Access Control Address
4603 * Change the hardware (MAC) address of the device
4605 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4607 const struct net_device_ops *ops = dev->netdev_ops;
4610 if (!ops->ndo_set_mac_address)
4612 if (sa->sa_family != dev->type)
4614 if (!netif_device_present(dev))
4616 err = ops->ndo_set_mac_address(dev, sa);
4618 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4621 EXPORT_SYMBOL(dev_set_mac_address);
4624 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4626 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4629 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4635 case SIOCGIFFLAGS: /* Get interface flags */
4636 ifr->ifr_flags = (short) dev_get_flags(dev);
4639 case SIOCGIFMETRIC: /* Get the metric on the interface
4640 (currently unused) */
4641 ifr->ifr_metric = 0;
4644 case SIOCGIFMTU: /* Get the MTU of a device */
4645 ifr->ifr_mtu = dev->mtu;
4650 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4652 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4653 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4654 ifr->ifr_hwaddr.sa_family = dev->type;
4662 ifr->ifr_map.mem_start = dev->mem_start;
4663 ifr->ifr_map.mem_end = dev->mem_end;
4664 ifr->ifr_map.base_addr = dev->base_addr;
4665 ifr->ifr_map.irq = dev->irq;
4666 ifr->ifr_map.dma = dev->dma;
4667 ifr->ifr_map.port = dev->if_port;
4671 ifr->ifr_ifindex = dev->ifindex;
4675 ifr->ifr_qlen = dev->tx_queue_len;
4679 /* dev_ioctl() should ensure this case
4691 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4693 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4696 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4697 const struct net_device_ops *ops;
4702 ops = dev->netdev_ops;
4705 case SIOCSIFFLAGS: /* Set interface flags */
4706 return dev_change_flags(dev, ifr->ifr_flags);
4708 case SIOCSIFMETRIC: /* Set the metric on the interface
4709 (currently unused) */
4712 case SIOCSIFMTU: /* Set the MTU of a device */
4713 return dev_set_mtu(dev, ifr->ifr_mtu);
4716 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4718 case SIOCSIFHWBROADCAST:
4719 if (ifr->ifr_hwaddr.sa_family != dev->type)
4721 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4722 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4723 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4727 if (ops->ndo_set_config) {
4728 if (!netif_device_present(dev))
4730 return ops->ndo_set_config(dev, &ifr->ifr_map);
4735 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4736 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4738 if (!netif_device_present(dev))
4740 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4743 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4744 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4746 if (!netif_device_present(dev))
4748 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4751 if (ifr->ifr_qlen < 0)
4753 dev->tx_queue_len = ifr->ifr_qlen;
4757 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4758 return dev_change_name(dev, ifr->ifr_newname);
4761 * Unknown or private ioctl
4764 if ((cmd >= SIOCDEVPRIVATE &&
4765 cmd <= SIOCDEVPRIVATE + 15) ||
4766 cmd == SIOCBONDENSLAVE ||
4767 cmd == SIOCBONDRELEASE ||
4768 cmd == SIOCBONDSETHWADDR ||
4769 cmd == SIOCBONDSLAVEINFOQUERY ||
4770 cmd == SIOCBONDINFOQUERY ||
4771 cmd == SIOCBONDCHANGEACTIVE ||
4772 cmd == SIOCGMIIPHY ||
4773 cmd == SIOCGMIIREG ||
4774 cmd == SIOCSMIIREG ||
4775 cmd == SIOCBRADDIF ||
4776 cmd == SIOCBRDELIF ||
4777 cmd == SIOCSHWTSTAMP ||
4778 cmd == SIOCWANDEV) {
4780 if (ops->ndo_do_ioctl) {
4781 if (netif_device_present(dev))
4782 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4794 * This function handles all "interface"-type I/O control requests. The actual
4795 * 'doing' part of this is dev_ifsioc above.
4799 * dev_ioctl - network device ioctl
4800 * @net: the applicable net namespace
4801 * @cmd: command to issue
4802 * @arg: pointer to a struct ifreq in user space
4804 * Issue ioctl functions to devices. This is normally called by the
4805 * user space syscall interfaces but can sometimes be useful for
4806 * other purposes. The return value is the return from the syscall if
4807 * positive or a negative errno code on error.
4810 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4816 /* One special case: SIOCGIFCONF takes ifconf argument
4817 and requires shared lock, because it sleeps writing
4821 if (cmd == SIOCGIFCONF) {
4823 ret = dev_ifconf(net, (char __user *) arg);
4827 if (cmd == SIOCGIFNAME)
4828 return dev_ifname(net, (struct ifreq __user *)arg);
4830 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4833 ifr.ifr_name[IFNAMSIZ-1] = 0;
4835 colon = strchr(ifr.ifr_name, ':');
4840 * See which interface the caller is talking about.
4845 * These ioctl calls:
4846 * - can be done by all.
4847 * - atomic and do not require locking.
4858 dev_load(net, ifr.ifr_name);
4860 ret = dev_ifsioc_locked(net, &ifr, cmd);
4865 if (copy_to_user(arg, &ifr,
4866 sizeof(struct ifreq)))
4872 dev_load(net, ifr.ifr_name);
4874 ret = dev_ethtool(net, &ifr);
4879 if (copy_to_user(arg, &ifr,
4880 sizeof(struct ifreq)))
4886 * These ioctl calls:
4887 * - require superuser power.
4888 * - require strict serialization.
4894 if (!capable(CAP_NET_ADMIN))
4896 dev_load(net, ifr.ifr_name);
4898 ret = dev_ifsioc(net, &ifr, cmd);
4903 if (copy_to_user(arg, &ifr,
4904 sizeof(struct ifreq)))
4910 * These ioctl calls:
4911 * - require superuser power.
4912 * - require strict serialization.
4913 * - do not return a value
4923 case SIOCSIFHWBROADCAST:
4926 case SIOCBONDENSLAVE:
4927 case SIOCBONDRELEASE:
4928 case SIOCBONDSETHWADDR:
4929 case SIOCBONDCHANGEACTIVE:
4933 if (!capable(CAP_NET_ADMIN))
4936 case SIOCBONDSLAVEINFOQUERY:
4937 case SIOCBONDINFOQUERY:
4938 dev_load(net, ifr.ifr_name);
4940 ret = dev_ifsioc(net, &ifr, cmd);
4945 /* Get the per device memory space. We can add this but
4946 * currently do not support it */
4948 /* Set the per device memory buffer space.
4949 * Not applicable in our case */
4954 * Unknown or private ioctl.
4957 if (cmd == SIOCWANDEV ||
4958 (cmd >= SIOCDEVPRIVATE &&
4959 cmd <= SIOCDEVPRIVATE + 15)) {
4960 dev_load(net, ifr.ifr_name);
4962 ret = dev_ifsioc(net, &ifr, cmd);
4964 if (!ret && copy_to_user(arg, &ifr,
4965 sizeof(struct ifreq)))
4969 /* Take care of Wireless Extensions */
4970 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4971 return wext_handle_ioctl(net, &ifr, cmd, arg);
4978 * dev_new_index - allocate an ifindex
4979 * @net: the applicable net namespace
4981 * Returns a suitable unique value for a new device interface
4982 * number. The caller must hold the rtnl semaphore or the
4983 * dev_base_lock to be sure it remains unique.
4985 static int dev_new_index(struct net *net)
4991 if (!__dev_get_by_index(net, ifindex))
4996 /* Delayed registration/unregisteration */
4997 static LIST_HEAD(net_todo_list);
4999 static void net_set_todo(struct net_device *dev)
5001 list_add_tail(&dev->todo_list, &net_todo_list);
5004 static void rollback_registered_many(struct list_head *head)
5006 struct net_device *dev, *tmp;
5008 BUG_ON(dev_boot_phase);
5011 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5012 /* Some devices call without registering
5013 * for initialization unwind. Remove those
5014 * devices and proceed with the remaining.
5016 if (dev->reg_state == NETREG_UNINITIALIZED) {
5017 pr_debug("unregister_netdevice: device %s/%p never "
5018 "was registered\n", dev->name, dev);
5021 list_del(&dev->unreg_list);
5025 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5028 /* If device is running, close it first. */
5029 dev_close_many(head);
5031 list_for_each_entry(dev, head, unreg_list) {
5032 /* And unlink it from device chain. */
5033 unlist_netdevice(dev);
5035 dev->reg_state = NETREG_UNREGISTERING;
5040 list_for_each_entry(dev, head, unreg_list) {
5041 /* Shutdown queueing discipline. */
5045 /* Notify protocols, that we are about to destroy
5046 this device. They should clean all the things.
5048 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5050 if (!dev->rtnl_link_ops ||
5051 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5052 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5055 * Flush the unicast and multicast chains
5060 if (dev->netdev_ops->ndo_uninit)
5061 dev->netdev_ops->ndo_uninit(dev);
5063 /* Notifier chain MUST detach us from master device. */
5064 WARN_ON(dev->master);
5066 /* Remove entries from kobject tree */
5067 netdev_unregister_kobject(dev);
5070 /* Process any work delayed until the end of the batch */
5071 dev = list_first_entry(head, struct net_device, unreg_list);
5072 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5076 list_for_each_entry(dev, head, unreg_list)
5080 static void rollback_registered(struct net_device *dev)
5084 list_add(&dev->unreg_list, &single);
5085 rollback_registered_many(&single);
5088 unsigned long netdev_fix_features(unsigned long features, const char *name)
5090 /* Fix illegal SG+CSUM combinations. */
5091 if ((features & NETIF_F_SG) &&
5092 !(features & NETIF_F_ALL_CSUM)) {
5094 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
5095 "checksum feature.\n", name);
5096 features &= ~NETIF_F_SG;
5099 /* TSO requires that SG is present as well. */
5100 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
5102 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
5103 "SG feature.\n", name);
5104 features &= ~NETIF_F_TSO;
5107 if (features & NETIF_F_UFO) {
5108 /* maybe split UFO into V4 and V6? */
5109 if (!((features & NETIF_F_GEN_CSUM) ||
5110 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5111 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5113 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5114 "since no checksum offload features.\n",
5116 features &= ~NETIF_F_UFO;
5119 if (!(features & NETIF_F_SG)) {
5121 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5122 "since no NETIF_F_SG feature.\n", name);
5123 features &= ~NETIF_F_UFO;
5129 EXPORT_SYMBOL(netdev_fix_features);
5132 * netif_stacked_transfer_operstate - transfer operstate
5133 * @rootdev: the root or lower level device to transfer state from
5134 * @dev: the device to transfer operstate to
5136 * Transfer operational state from root to device. This is normally
5137 * called when a stacking relationship exists between the root
5138 * device and the device(a leaf device).
5140 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5141 struct net_device *dev)
5143 if (rootdev->operstate == IF_OPER_DORMANT)
5144 netif_dormant_on(dev);
5146 netif_dormant_off(dev);
5148 if (netif_carrier_ok(rootdev)) {
5149 if (!netif_carrier_ok(dev))
5150 netif_carrier_on(dev);
5152 if (netif_carrier_ok(dev))
5153 netif_carrier_off(dev);
5156 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5159 static int netif_alloc_rx_queues(struct net_device *dev)
5161 unsigned int i, count = dev->num_rx_queues;
5162 struct netdev_rx_queue *rx;
5166 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5168 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5173 for (i = 0; i < count; i++)
5179 static void netdev_init_one_queue(struct net_device *dev,
5180 struct netdev_queue *queue, void *_unused)
5182 /* Initialize queue lock */
5183 spin_lock_init(&queue->_xmit_lock);
5184 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5185 queue->xmit_lock_owner = -1;
5186 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5190 static int netif_alloc_netdev_queues(struct net_device *dev)
5192 unsigned int count = dev->num_tx_queues;
5193 struct netdev_queue *tx;
5197 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5199 pr_err("netdev: Unable to allocate %u tx queues.\n",
5205 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5206 spin_lock_init(&dev->tx_global_lock);
5212 * register_netdevice - register a network device
5213 * @dev: device to register
5215 * Take a completed network device structure and add it to the kernel
5216 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5217 * chain. 0 is returned on success. A negative errno code is returned
5218 * on a failure to set up the device, or if the name is a duplicate.
5220 * Callers must hold the rtnl semaphore. You may want
5221 * register_netdev() instead of this.
5224 * The locking appears insufficient to guarantee two parallel registers
5225 * will not get the same name.
5228 int register_netdevice(struct net_device *dev)
5231 struct net *net = dev_net(dev);
5233 BUG_ON(dev_boot_phase);
5238 /* When net_device's are persistent, this will be fatal. */
5239 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5242 spin_lock_init(&dev->addr_list_lock);
5243 netdev_set_addr_lockdep_class(dev);
5247 /* Init, if this function is available */
5248 if (dev->netdev_ops->ndo_init) {
5249 ret = dev->netdev_ops->ndo_init(dev);
5257 ret = dev_get_valid_name(dev, dev->name, 0);
5261 dev->ifindex = dev_new_index(net);
5262 if (dev->iflink == -1)
5263 dev->iflink = dev->ifindex;
5265 /* Fix illegal checksum combinations */
5266 if ((dev->features & NETIF_F_HW_CSUM) &&
5267 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5268 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5270 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5273 if ((dev->features & NETIF_F_NO_CSUM) &&
5274 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5275 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5277 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5280 dev->features = netdev_fix_features(dev->features, dev->name);
5282 /* Enable software GSO if SG is supported. */
5283 if (dev->features & NETIF_F_SG)
5284 dev->features |= NETIF_F_GSO;
5286 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5287 * vlan_dev_init() will do the dev->features check, so these features
5288 * are enabled only if supported by underlying device.
5290 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5292 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5293 ret = notifier_to_errno(ret);
5297 ret = netdev_register_kobject(dev);
5300 dev->reg_state = NETREG_REGISTERED;
5303 * Default initial state at registry is that the
5304 * device is present.
5307 set_bit(__LINK_STATE_PRESENT, &dev->state);
5309 dev_init_scheduler(dev);
5311 list_netdevice(dev);
5313 /* Notify protocols, that a new device appeared. */
5314 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5315 ret = notifier_to_errno(ret);
5317 rollback_registered(dev);
5318 dev->reg_state = NETREG_UNREGISTERED;
5321 * Prevent userspace races by waiting until the network
5322 * device is fully setup before sending notifications.
5324 if (!dev->rtnl_link_ops ||
5325 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5326 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5332 if (dev->netdev_ops->ndo_uninit)
5333 dev->netdev_ops->ndo_uninit(dev);
5336 EXPORT_SYMBOL(register_netdevice);
5339 * init_dummy_netdev - init a dummy network device for NAPI
5340 * @dev: device to init
5342 * This takes a network device structure and initialize the minimum
5343 * amount of fields so it can be used to schedule NAPI polls without
5344 * registering a full blown interface. This is to be used by drivers
5345 * that need to tie several hardware interfaces to a single NAPI
5346 * poll scheduler due to HW limitations.
5348 int init_dummy_netdev(struct net_device *dev)
5350 /* Clear everything. Note we don't initialize spinlocks
5351 * are they aren't supposed to be taken by any of the
5352 * NAPI code and this dummy netdev is supposed to be
5353 * only ever used for NAPI polls
5355 memset(dev, 0, sizeof(struct net_device));
5357 /* make sure we BUG if trying to hit standard
5358 * register/unregister code path
5360 dev->reg_state = NETREG_DUMMY;
5362 /* NAPI wants this */
5363 INIT_LIST_HEAD(&dev->napi_list);
5365 /* a dummy interface is started by default */
5366 set_bit(__LINK_STATE_PRESENT, &dev->state);
5367 set_bit(__LINK_STATE_START, &dev->state);
5369 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5370 * because users of this 'device' dont need to change
5376 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5380 * register_netdev - register a network device
5381 * @dev: device to register
5383 * Take a completed network device structure and add it to the kernel
5384 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5385 * chain. 0 is returned on success. A negative errno code is returned
5386 * on a failure to set up the device, or if the name is a duplicate.
5388 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5389 * and expands the device name if you passed a format string to
5392 int register_netdev(struct net_device *dev)
5399 * If the name is a format string the caller wants us to do a
5402 if (strchr(dev->name, '%')) {
5403 err = dev_alloc_name(dev, dev->name);
5408 err = register_netdevice(dev);
5413 EXPORT_SYMBOL(register_netdev);
5415 int netdev_refcnt_read(const struct net_device *dev)
5419 for_each_possible_cpu(i)
5420 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5423 EXPORT_SYMBOL(netdev_refcnt_read);
5426 * netdev_wait_allrefs - wait until all references are gone.
5428 * This is called when unregistering network devices.
5430 * Any protocol or device that holds a reference should register
5431 * for netdevice notification, and cleanup and put back the
5432 * reference if they receive an UNREGISTER event.
5433 * We can get stuck here if buggy protocols don't correctly
5436 static void netdev_wait_allrefs(struct net_device *dev)
5438 unsigned long rebroadcast_time, warning_time;
5441 linkwatch_forget_dev(dev);
5443 rebroadcast_time = warning_time = jiffies;
5444 refcnt = netdev_refcnt_read(dev);
5446 while (refcnt != 0) {
5447 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5450 /* Rebroadcast unregister notification */
5451 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5452 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5453 * should have already handle it the first time */
5455 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5457 /* We must not have linkwatch events
5458 * pending on unregister. If this
5459 * happens, we simply run the queue
5460 * unscheduled, resulting in a noop
5463 linkwatch_run_queue();
5468 rebroadcast_time = jiffies;
5473 refcnt = netdev_refcnt_read(dev);
5475 if (time_after(jiffies, warning_time + 10 * HZ)) {
5476 printk(KERN_EMERG "unregister_netdevice: "
5477 "waiting for %s to become free. Usage "
5480 warning_time = jiffies;
5489 * register_netdevice(x1);
5490 * register_netdevice(x2);
5492 * unregister_netdevice(y1);
5493 * unregister_netdevice(y2);
5499 * We are invoked by rtnl_unlock().
5500 * This allows us to deal with problems:
5501 * 1) We can delete sysfs objects which invoke hotplug
5502 * without deadlocking with linkwatch via keventd.
5503 * 2) Since we run with the RTNL semaphore not held, we can sleep
5504 * safely in order to wait for the netdev refcnt to drop to zero.
5506 * We must not return until all unregister events added during
5507 * the interval the lock was held have been completed.
5509 void netdev_run_todo(void)
5511 struct list_head list;
5513 /* Snapshot list, allow later requests */
5514 list_replace_init(&net_todo_list, &list);
5518 while (!list_empty(&list)) {
5519 struct net_device *dev
5520 = list_first_entry(&list, struct net_device, todo_list);
5521 list_del(&dev->todo_list);
5523 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5524 printk(KERN_ERR "network todo '%s' but state %d\n",
5525 dev->name, dev->reg_state);
5530 dev->reg_state = NETREG_UNREGISTERED;
5532 on_each_cpu(flush_backlog, dev, 1);
5534 netdev_wait_allrefs(dev);
5537 BUG_ON(netdev_refcnt_read(dev));
5538 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5539 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5540 WARN_ON(dev->dn_ptr);
5542 if (dev->destructor)
5543 dev->destructor(dev);
5545 /* Free network device */
5546 kobject_put(&dev->dev.kobj);
5551 * dev_txq_stats_fold - fold tx_queues stats
5552 * @dev: device to get statistics from
5553 * @stats: struct rtnl_link_stats64 to hold results
5555 void dev_txq_stats_fold(const struct net_device *dev,
5556 struct rtnl_link_stats64 *stats)
5558 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5560 struct netdev_queue *txq;
5562 for (i = 0; i < dev->num_tx_queues; i++) {
5563 txq = netdev_get_tx_queue(dev, i);
5564 spin_lock_bh(&txq->_xmit_lock);
5565 tx_bytes += txq->tx_bytes;
5566 tx_packets += txq->tx_packets;
5567 tx_dropped += txq->tx_dropped;
5568 spin_unlock_bh(&txq->_xmit_lock);
5570 if (tx_bytes || tx_packets || tx_dropped) {
5571 stats->tx_bytes = tx_bytes;
5572 stats->tx_packets = tx_packets;
5573 stats->tx_dropped = tx_dropped;
5576 EXPORT_SYMBOL(dev_txq_stats_fold);
5578 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5579 * fields in the same order, with only the type differing.
5581 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5582 const struct net_device_stats *netdev_stats)
5584 #if BITS_PER_LONG == 64
5585 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5586 memcpy(stats64, netdev_stats, sizeof(*stats64));
5588 size_t i, n = sizeof(*stats64) / sizeof(u64);
5589 const unsigned long *src = (const unsigned long *)netdev_stats;
5590 u64 *dst = (u64 *)stats64;
5592 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5593 sizeof(*stats64) / sizeof(u64));
5594 for (i = 0; i < n; i++)
5600 * dev_get_stats - get network device statistics
5601 * @dev: device to get statistics from
5602 * @storage: place to store stats
5604 * Get network statistics from device. Return @storage.
5605 * The device driver may provide its own method by setting
5606 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5607 * otherwise the internal statistics structure is used.
5609 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5610 struct rtnl_link_stats64 *storage)
5612 const struct net_device_ops *ops = dev->netdev_ops;
5614 if (ops->ndo_get_stats64) {
5615 memset(storage, 0, sizeof(*storage));
5616 ops->ndo_get_stats64(dev, storage);
5617 } else if (ops->ndo_get_stats) {
5618 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5620 netdev_stats_to_stats64(storage, &dev->stats);
5621 dev_txq_stats_fold(dev, storage);
5623 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5626 EXPORT_SYMBOL(dev_get_stats);
5628 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5630 struct netdev_queue *queue = dev_ingress_queue(dev);
5632 #ifdef CONFIG_NET_CLS_ACT
5635 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5638 netdev_init_one_queue(dev, queue, NULL);
5639 queue->qdisc = &noop_qdisc;
5640 queue->qdisc_sleeping = &noop_qdisc;
5641 rcu_assign_pointer(dev->ingress_queue, queue);
5647 * alloc_netdev_mq - allocate network device
5648 * @sizeof_priv: size of private data to allocate space for
5649 * @name: device name format string
5650 * @setup: callback to initialize device
5651 * @queue_count: the number of subqueues to allocate
5653 * Allocates a struct net_device with private data area for driver use
5654 * and performs basic initialization. Also allocates subquue structs
5655 * for each queue on the device at the end of the netdevice.
5657 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5658 void (*setup)(struct net_device *), unsigned int queue_count)
5660 struct net_device *dev;
5662 struct net_device *p;
5664 BUG_ON(strlen(name) >= sizeof(dev->name));
5666 if (queue_count < 1) {
5667 pr_err("alloc_netdev: Unable to allocate device "
5668 "with zero queues.\n");
5672 alloc_size = sizeof(struct net_device);
5674 /* ensure 32-byte alignment of private area */
5675 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5676 alloc_size += sizeof_priv;
5678 /* ensure 32-byte alignment of whole construct */
5679 alloc_size += NETDEV_ALIGN - 1;
5681 p = kzalloc(alloc_size, GFP_KERNEL);
5683 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5687 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5688 dev->padded = (char *)dev - (char *)p;
5690 dev->pcpu_refcnt = alloc_percpu(int);
5691 if (!dev->pcpu_refcnt)
5694 if (dev_addr_init(dev))
5700 dev_net_set(dev, &init_net);
5702 dev->num_tx_queues = queue_count;
5703 dev->real_num_tx_queues = queue_count;
5704 if (netif_alloc_netdev_queues(dev))
5708 dev->num_rx_queues = queue_count;
5709 dev->real_num_rx_queues = queue_count;
5710 if (netif_alloc_rx_queues(dev))
5714 dev->gso_max_size = GSO_MAX_SIZE;
5716 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5717 dev->ethtool_ntuple_list.count = 0;
5718 INIT_LIST_HEAD(&dev->napi_list);
5719 INIT_LIST_HEAD(&dev->unreg_list);
5720 INIT_LIST_HEAD(&dev->link_watch_list);
5721 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5723 strcpy(dev->name, name);
5727 free_percpu(dev->pcpu_refcnt);
5737 EXPORT_SYMBOL(alloc_netdev_mq);
5740 * free_netdev - free network device
5743 * This function does the last stage of destroying an allocated device
5744 * interface. The reference to the device object is released.
5745 * If this is the last reference then it will be freed.
5747 void free_netdev(struct net_device *dev)
5749 struct napi_struct *p, *n;
5751 release_net(dev_net(dev));
5758 kfree(rcu_dereference_raw(dev->ingress_queue));
5760 /* Flush device addresses */
5761 dev_addr_flush(dev);
5763 /* Clear ethtool n-tuple list */
5764 ethtool_ntuple_flush(dev);
5766 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5769 free_percpu(dev->pcpu_refcnt);
5770 dev->pcpu_refcnt = NULL;
5772 /* Compatibility with error handling in drivers */
5773 if (dev->reg_state == NETREG_UNINITIALIZED) {
5774 kfree((char *)dev - dev->padded);
5778 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5779 dev->reg_state = NETREG_RELEASED;
5781 /* will free via device release */
5782 put_device(&dev->dev);
5784 EXPORT_SYMBOL(free_netdev);
5787 * synchronize_net - Synchronize with packet receive processing
5789 * Wait for packets currently being received to be done.
5790 * Does not block later packets from starting.
5792 void synchronize_net(void)
5797 EXPORT_SYMBOL(synchronize_net);
5800 * unregister_netdevice_queue - remove device from the kernel
5804 * This function shuts down a device interface and removes it
5805 * from the kernel tables.
5806 * If head not NULL, device is queued to be unregistered later.
5808 * Callers must hold the rtnl semaphore. You may want
5809 * unregister_netdev() instead of this.
5812 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5817 list_move_tail(&dev->unreg_list, head);
5819 rollback_registered(dev);
5820 /* Finish processing unregister after unlock */
5824 EXPORT_SYMBOL(unregister_netdevice_queue);
5827 * unregister_netdevice_many - unregister many devices
5828 * @head: list of devices
5830 void unregister_netdevice_many(struct list_head *head)
5832 struct net_device *dev;
5834 if (!list_empty(head)) {
5835 rollback_registered_many(head);
5836 list_for_each_entry(dev, head, unreg_list)
5840 EXPORT_SYMBOL(unregister_netdevice_many);
5843 * unregister_netdev - remove device from the kernel
5846 * This function shuts down a device interface and removes it
5847 * from the kernel tables.
5849 * This is just a wrapper for unregister_netdevice that takes
5850 * the rtnl semaphore. In general you want to use this and not
5851 * unregister_netdevice.
5853 void unregister_netdev(struct net_device *dev)
5856 unregister_netdevice(dev);
5859 EXPORT_SYMBOL(unregister_netdev);
5862 * dev_change_net_namespace - move device to different nethost namespace
5864 * @net: network namespace
5865 * @pat: If not NULL name pattern to try if the current device name
5866 * is already taken in the destination network namespace.
5868 * This function shuts down a device interface and moves it
5869 * to a new network namespace. On success 0 is returned, on
5870 * a failure a netagive errno code is returned.
5872 * Callers must hold the rtnl semaphore.
5875 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5881 /* Don't allow namespace local devices to be moved. */
5883 if (dev->features & NETIF_F_NETNS_LOCAL)
5886 /* Ensure the device has been registrered */
5888 if (dev->reg_state != NETREG_REGISTERED)
5891 /* Get out if there is nothing todo */
5893 if (net_eq(dev_net(dev), net))
5896 /* Pick the destination device name, and ensure
5897 * we can use it in the destination network namespace.
5900 if (__dev_get_by_name(net, dev->name)) {
5901 /* We get here if we can't use the current device name */
5904 if (dev_get_valid_name(dev, pat, 1))
5909 * And now a mini version of register_netdevice unregister_netdevice.
5912 /* If device is running close it first. */
5915 /* And unlink it from device chain */
5917 unlist_netdevice(dev);
5921 /* Shutdown queueing discipline. */
5924 /* Notify protocols, that we are about to destroy
5925 this device. They should clean all the things.
5927 Note that dev->reg_state stays at NETREG_REGISTERED.
5928 This is wanted because this way 8021q and macvlan know
5929 the device is just moving and can keep their slaves up.
5931 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5932 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5935 * Flush the unicast and multicast chains
5940 /* Actually switch the network namespace */
5941 dev_net_set(dev, net);
5943 /* If there is an ifindex conflict assign a new one */
5944 if (__dev_get_by_index(net, dev->ifindex)) {
5945 int iflink = (dev->iflink == dev->ifindex);
5946 dev->ifindex = dev_new_index(net);
5948 dev->iflink = dev->ifindex;
5951 /* Fixup kobjects */
5952 err = device_rename(&dev->dev, dev->name);
5955 /* Add the device back in the hashes */
5956 list_netdevice(dev);
5958 /* Notify protocols, that a new device appeared. */
5959 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5962 * Prevent userspace races by waiting until the network
5963 * device is fully setup before sending notifications.
5965 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5972 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5974 static int dev_cpu_callback(struct notifier_block *nfb,
5975 unsigned long action,
5978 struct sk_buff **list_skb;
5979 struct sk_buff *skb;
5980 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5981 struct softnet_data *sd, *oldsd;
5983 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5986 local_irq_disable();
5987 cpu = smp_processor_id();
5988 sd = &per_cpu(softnet_data, cpu);
5989 oldsd = &per_cpu(softnet_data, oldcpu);
5991 /* Find end of our completion_queue. */
5992 list_skb = &sd->completion_queue;
5994 list_skb = &(*list_skb)->next;
5995 /* Append completion queue from offline CPU. */
5996 *list_skb = oldsd->completion_queue;
5997 oldsd->completion_queue = NULL;
5999 /* Append output queue from offline CPU. */
6000 if (oldsd->output_queue) {
6001 *sd->output_queue_tailp = oldsd->output_queue;
6002 sd->output_queue_tailp = oldsd->output_queue_tailp;
6003 oldsd->output_queue = NULL;
6004 oldsd->output_queue_tailp = &oldsd->output_queue;
6007 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6010 /* Process offline CPU's input_pkt_queue */
6011 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6013 input_queue_head_incr(oldsd);
6015 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6017 input_queue_head_incr(oldsd);
6025 * netdev_increment_features - increment feature set by one
6026 * @all: current feature set
6027 * @one: new feature set
6028 * @mask: mask feature set
6030 * Computes a new feature set after adding a device with feature set
6031 * @one to the master device with current feature set @all. Will not
6032 * enable anything that is off in @mask. Returns the new feature set.
6034 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
6037 /* If device needs checksumming, downgrade to it. */
6038 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
6039 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
6040 else if (mask & NETIF_F_ALL_CSUM) {
6041 /* If one device supports v4/v6 checksumming, set for all. */
6042 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
6043 !(all & NETIF_F_GEN_CSUM)) {
6044 all &= ~NETIF_F_ALL_CSUM;
6045 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
6048 /* If one device supports hw checksumming, set for all. */
6049 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
6050 all &= ~NETIF_F_ALL_CSUM;
6051 all |= NETIF_F_HW_CSUM;
6055 one |= NETIF_F_ALL_CSUM;
6057 one |= all & NETIF_F_ONE_FOR_ALL;
6058 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
6059 all |= one & mask & NETIF_F_ONE_FOR_ALL;
6063 EXPORT_SYMBOL(netdev_increment_features);
6065 static struct hlist_head *netdev_create_hash(void)
6068 struct hlist_head *hash;
6070 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6072 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6073 INIT_HLIST_HEAD(&hash[i]);
6078 /* Initialize per network namespace state */
6079 static int __net_init netdev_init(struct net *net)
6081 INIT_LIST_HEAD(&net->dev_base_head);
6083 net->dev_name_head = netdev_create_hash();
6084 if (net->dev_name_head == NULL)
6087 net->dev_index_head = netdev_create_hash();
6088 if (net->dev_index_head == NULL)
6094 kfree(net->dev_name_head);
6100 * netdev_drivername - network driver for the device
6101 * @dev: network device
6102 * @buffer: buffer for resulting name
6103 * @len: size of buffer
6105 * Determine network driver for device.
6107 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6109 const struct device_driver *driver;
6110 const struct device *parent;
6112 if (len <= 0 || !buffer)
6116 parent = dev->dev.parent;
6121 driver = parent->driver;
6122 if (driver && driver->name)
6123 strlcpy(buffer, driver->name, len);
6127 static int __netdev_printk(const char *level, const struct net_device *dev,
6128 struct va_format *vaf)
6132 if (dev && dev->dev.parent)
6133 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6134 netdev_name(dev), vaf);
6136 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6138 r = printk("%s(NULL net_device): %pV", level, vaf);
6143 int netdev_printk(const char *level, const struct net_device *dev,
6144 const char *format, ...)
6146 struct va_format vaf;
6150 va_start(args, format);
6155 r = __netdev_printk(level, dev, &vaf);
6160 EXPORT_SYMBOL(netdev_printk);
6162 #define define_netdev_printk_level(func, level) \
6163 int func(const struct net_device *dev, const char *fmt, ...) \
6166 struct va_format vaf; \
6169 va_start(args, fmt); \
6174 r = __netdev_printk(level, dev, &vaf); \
6179 EXPORT_SYMBOL(func);
6181 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6182 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6183 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6184 define_netdev_printk_level(netdev_err, KERN_ERR);
6185 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6186 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6187 define_netdev_printk_level(netdev_info, KERN_INFO);
6189 static void __net_exit netdev_exit(struct net *net)
6191 kfree(net->dev_name_head);
6192 kfree(net->dev_index_head);
6195 static struct pernet_operations __net_initdata netdev_net_ops = {
6196 .init = netdev_init,
6197 .exit = netdev_exit,
6200 static void __net_exit default_device_exit(struct net *net)
6202 struct net_device *dev, *aux;
6204 * Push all migratable network devices back to the
6205 * initial network namespace
6208 for_each_netdev_safe(net, dev, aux) {
6210 char fb_name[IFNAMSIZ];
6212 /* Ignore unmoveable devices (i.e. loopback) */
6213 if (dev->features & NETIF_F_NETNS_LOCAL)
6216 /* Leave virtual devices for the generic cleanup */
6217 if (dev->rtnl_link_ops)
6220 /* Push remaing network devices to init_net */
6221 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6222 err = dev_change_net_namespace(dev, &init_net, fb_name);
6224 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6225 __func__, dev->name, err);
6232 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6234 /* At exit all network devices most be removed from a network
6235 * namespace. Do this in the reverse order of registeration.
6236 * Do this across as many network namespaces as possible to
6237 * improve batching efficiency.
6239 struct net_device *dev;
6241 LIST_HEAD(dev_kill_list);
6244 list_for_each_entry(net, net_list, exit_list) {
6245 for_each_netdev_reverse(net, dev) {
6246 if (dev->rtnl_link_ops)
6247 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6249 unregister_netdevice_queue(dev, &dev_kill_list);
6252 unregister_netdevice_many(&dev_kill_list);
6256 static struct pernet_operations __net_initdata default_device_ops = {
6257 .exit = default_device_exit,
6258 .exit_batch = default_device_exit_batch,
6262 * Initialize the DEV module. At boot time this walks the device list and
6263 * unhooks any devices that fail to initialise (normally hardware not
6264 * present) and leaves us with a valid list of present and active devices.
6269 * This is called single threaded during boot, so no need
6270 * to take the rtnl semaphore.
6272 static int __init net_dev_init(void)
6274 int i, rc = -ENOMEM;
6276 BUG_ON(!dev_boot_phase);
6278 if (dev_proc_init())
6281 if (netdev_kobject_init())
6284 INIT_LIST_HEAD(&ptype_all);
6285 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6286 INIT_LIST_HEAD(&ptype_base[i]);
6288 if (register_pernet_subsys(&netdev_net_ops))
6292 * Initialise the packet receive queues.
6295 for_each_possible_cpu(i) {
6296 struct softnet_data *sd = &per_cpu(softnet_data, i);
6298 memset(sd, 0, sizeof(*sd));
6299 skb_queue_head_init(&sd->input_pkt_queue);
6300 skb_queue_head_init(&sd->process_queue);
6301 sd->completion_queue = NULL;
6302 INIT_LIST_HEAD(&sd->poll_list);
6303 sd->output_queue = NULL;
6304 sd->output_queue_tailp = &sd->output_queue;
6306 sd->csd.func = rps_trigger_softirq;
6312 sd->backlog.poll = process_backlog;
6313 sd->backlog.weight = weight_p;
6314 sd->backlog.gro_list = NULL;
6315 sd->backlog.gro_count = 0;
6320 /* The loopback device is special if any other network devices
6321 * is present in a network namespace the loopback device must
6322 * be present. Since we now dynamically allocate and free the
6323 * loopback device ensure this invariant is maintained by
6324 * keeping the loopback device as the first device on the
6325 * list of network devices. Ensuring the loopback devices
6326 * is the first device that appears and the last network device
6329 if (register_pernet_device(&loopback_net_ops))
6332 if (register_pernet_device(&default_device_ops))
6335 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6336 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6338 hotcpu_notifier(dev_cpu_callback, 0);
6346 subsys_initcall(net_dev_init);
6348 static int __init initialize_hashrnd(void)
6350 get_random_bytes(&hashrnd, sizeof(hashrnd));
6354 late_initcall_sync(initialize_hashrnd);