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
1974 * @features: device features as applicable to this skb
1976 * This function segments the given skb and stores the list of segments
1979 static int dev_gso_segment(struct sk_buff *skb, int features)
1981 struct sk_buff *segs;
1983 segs = skb_gso_segment(skb, features);
1985 /* Verifying header integrity only. */
1990 return PTR_ERR(segs);
1993 DEV_GSO_CB(skb)->destructor = skb->destructor;
1994 skb->destructor = dev_gso_skb_destructor;
2000 * Try to orphan skb early, right before transmission by the device.
2001 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2002 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2004 static inline void skb_orphan_try(struct sk_buff *skb)
2006 struct sock *sk = skb->sk;
2008 if (sk && !skb_shinfo(skb)->tx_flags) {
2009 /* skb_tx_hash() wont be able to get sk.
2010 * We copy sk_hash into skb->rxhash
2013 skb->rxhash = sk->sk_hash;
2018 static int harmonize_features(struct sk_buff *skb, __be16 protocol, int features)
2020 if (!can_checksum_protocol(protocol, features)) {
2021 features &= ~NETIF_F_ALL_CSUM;
2022 features &= ~NETIF_F_SG;
2023 } else if (illegal_highdma(skb->dev, skb)) {
2024 features &= ~NETIF_F_SG;
2030 int netif_skb_features(struct sk_buff *skb)
2032 __be16 protocol = skb->protocol;
2033 int features = skb->dev->features;
2035 if (protocol == htons(ETH_P_8021Q)) {
2036 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2037 protocol = veh->h_vlan_encapsulated_proto;
2038 } else if (!vlan_tx_tag_present(skb)) {
2039 return harmonize_features(skb, protocol, features);
2042 features &= skb->dev->vlan_features;
2044 if (protocol != htons(ETH_P_8021Q)) {
2045 return harmonize_features(skb, protocol, features);
2047 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2049 return harmonize_features(skb, protocol, features);
2052 EXPORT_SYMBOL(netif_skb_features);
2055 * Returns true if either:
2056 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2057 * 2. skb is fragmented and the device does not support SG, or if
2058 * at least one of fragments is in highmem and device does not
2059 * support DMA from it.
2061 static inline int skb_needs_linearize(struct sk_buff *skb,
2064 return skb_is_nonlinear(skb) &&
2065 ((skb_has_frag_list(skb) &&
2066 !(features & NETIF_F_FRAGLIST)) ||
2067 (skb_shinfo(skb)->nr_frags &&
2068 !(features & NETIF_F_SG)));
2071 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2072 struct netdev_queue *txq)
2074 const struct net_device_ops *ops = dev->netdev_ops;
2075 int rc = NETDEV_TX_OK;
2077 if (likely(!skb->next)) {
2081 * If device doesnt need skb->dst, release it right now while
2082 * its hot in this cpu cache
2084 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2087 if (!list_empty(&ptype_all))
2088 dev_queue_xmit_nit(skb, dev);
2090 skb_orphan_try(skb);
2092 features = netif_skb_features(skb);
2094 if (vlan_tx_tag_present(skb) &&
2095 !(features & NETIF_F_HW_VLAN_TX)) {
2096 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2103 if (netif_needs_gso(skb, features)) {
2104 if (unlikely(dev_gso_segment(skb, features)))
2109 if (skb_needs_linearize(skb, features) &&
2110 __skb_linearize(skb))
2113 /* If packet is not checksummed and device does not
2114 * support checksumming for this protocol, complete
2115 * checksumming here.
2117 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2118 skb_set_transport_header(skb,
2119 skb_checksum_start_offset(skb));
2120 if (!dev_can_checksum(dev, skb) &&
2121 skb_checksum_help(skb))
2126 rc = ops->ndo_start_xmit(skb, dev);
2127 trace_net_dev_xmit(skb, rc);
2128 if (rc == NETDEV_TX_OK)
2129 txq_trans_update(txq);
2135 struct sk_buff *nskb = skb->next;
2137 skb->next = nskb->next;
2141 * If device doesnt need nskb->dst, release it right now while
2142 * its hot in this cpu cache
2144 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2147 rc = ops->ndo_start_xmit(nskb, dev);
2148 trace_net_dev_xmit(nskb, rc);
2149 if (unlikely(rc != NETDEV_TX_OK)) {
2150 if (rc & ~NETDEV_TX_MASK)
2151 goto out_kfree_gso_skb;
2152 nskb->next = skb->next;
2156 txq_trans_update(txq);
2157 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2158 return NETDEV_TX_BUSY;
2159 } while (skb->next);
2162 if (likely(skb->next == NULL))
2163 skb->destructor = DEV_GSO_CB(skb)->destructor;
2170 static u32 hashrnd __read_mostly;
2173 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2174 * to be used as a distribution range.
2176 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2177 unsigned int num_tx_queues)
2181 if (skb_rx_queue_recorded(skb)) {
2182 hash = skb_get_rx_queue(skb);
2183 while (unlikely(hash >= num_tx_queues))
2184 hash -= num_tx_queues;
2188 if (skb->sk && skb->sk->sk_hash)
2189 hash = skb->sk->sk_hash;
2191 hash = (__force u16) skb->protocol ^ skb->rxhash;
2192 hash = jhash_1word(hash, hashrnd);
2194 return (u16) (((u64) hash * num_tx_queues) >> 32);
2196 EXPORT_SYMBOL(__skb_tx_hash);
2198 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2200 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2201 if (net_ratelimit()) {
2202 pr_warning("%s selects TX queue %d, but "
2203 "real number of TX queues is %d\n",
2204 dev->name, queue_index, dev->real_num_tx_queues);
2211 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2214 struct xps_dev_maps *dev_maps;
2215 struct xps_map *map;
2216 int queue_index = -1;
2219 dev_maps = rcu_dereference(dev->xps_maps);
2221 map = rcu_dereference(
2222 dev_maps->cpu_map[raw_smp_processor_id()]);
2225 queue_index = map->queues[0];
2228 if (skb->sk && skb->sk->sk_hash)
2229 hash = skb->sk->sk_hash;
2231 hash = (__force u16) skb->protocol ^
2233 hash = jhash_1word(hash, hashrnd);
2234 queue_index = map->queues[
2235 ((u64)hash * map->len) >> 32];
2237 if (unlikely(queue_index >= dev->real_num_tx_queues))
2249 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2250 struct sk_buff *skb)
2253 const struct net_device_ops *ops = dev->netdev_ops;
2255 if (dev->real_num_tx_queues == 1)
2257 else if (ops->ndo_select_queue) {
2258 queue_index = ops->ndo_select_queue(dev, skb);
2259 queue_index = dev_cap_txqueue(dev, queue_index);
2261 struct sock *sk = skb->sk;
2262 queue_index = sk_tx_queue_get(sk);
2264 if (queue_index < 0 || skb->ooo_okay ||
2265 queue_index >= dev->real_num_tx_queues) {
2266 int old_index = queue_index;
2268 queue_index = get_xps_queue(dev, skb);
2269 if (queue_index < 0)
2270 queue_index = skb_tx_hash(dev, skb);
2272 if (queue_index != old_index && sk) {
2273 struct dst_entry *dst =
2274 rcu_dereference_check(sk->sk_dst_cache, 1);
2276 if (dst && skb_dst(skb) == dst)
2277 sk_tx_queue_set(sk, queue_index);
2282 skb_set_queue_mapping(skb, queue_index);
2283 return netdev_get_tx_queue(dev, queue_index);
2286 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2287 struct net_device *dev,
2288 struct netdev_queue *txq)
2290 spinlock_t *root_lock = qdisc_lock(q);
2291 bool contended = qdisc_is_running(q);
2295 * Heuristic to force contended enqueues to serialize on a
2296 * separate lock before trying to get qdisc main lock.
2297 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2298 * and dequeue packets faster.
2300 if (unlikely(contended))
2301 spin_lock(&q->busylock);
2303 spin_lock(root_lock);
2304 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2307 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2308 qdisc_run_begin(q)) {
2310 * This is a work-conserving queue; there are no old skbs
2311 * waiting to be sent out; and the qdisc is not running -
2312 * xmit the skb directly.
2314 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2316 __qdisc_update_bstats(q, skb->len);
2317 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2318 if (unlikely(contended)) {
2319 spin_unlock(&q->busylock);
2326 rc = NET_XMIT_SUCCESS;
2329 rc = qdisc_enqueue_root(skb, q);
2330 if (qdisc_run_begin(q)) {
2331 if (unlikely(contended)) {
2332 spin_unlock(&q->busylock);
2338 spin_unlock(root_lock);
2339 if (unlikely(contended))
2340 spin_unlock(&q->busylock);
2344 static DEFINE_PER_CPU(int, xmit_recursion);
2345 #define RECURSION_LIMIT 10
2348 * dev_queue_xmit - transmit a buffer
2349 * @skb: buffer to transmit
2351 * Queue a buffer for transmission to a network device. The caller must
2352 * have set the device and priority and built the buffer before calling
2353 * this function. The function can be called from an interrupt.
2355 * A negative errno code is returned on a failure. A success does not
2356 * guarantee the frame will be transmitted as it may be dropped due
2357 * to congestion or traffic shaping.
2359 * -----------------------------------------------------------------------------------
2360 * I notice this method can also return errors from the queue disciplines,
2361 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2364 * Regardless of the return value, the skb is consumed, so it is currently
2365 * difficult to retry a send to this method. (You can bump the ref count
2366 * before sending to hold a reference for retry if you are careful.)
2368 * When calling this method, interrupts MUST be enabled. This is because
2369 * the BH enable code must have IRQs enabled so that it will not deadlock.
2372 int dev_queue_xmit(struct sk_buff *skb)
2374 struct net_device *dev = skb->dev;
2375 struct netdev_queue *txq;
2379 /* Disable soft irqs for various locks below. Also
2380 * stops preemption for RCU.
2384 txq = dev_pick_tx(dev, skb);
2385 q = rcu_dereference_bh(txq->qdisc);
2387 #ifdef CONFIG_NET_CLS_ACT
2388 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2390 trace_net_dev_queue(skb);
2392 rc = __dev_xmit_skb(skb, q, dev, txq);
2396 /* The device has no queue. Common case for software devices:
2397 loopback, all the sorts of tunnels...
2399 Really, it is unlikely that netif_tx_lock protection is necessary
2400 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2402 However, it is possible, that they rely on protection
2405 Check this and shot the lock. It is not prone from deadlocks.
2406 Either shot noqueue qdisc, it is even simpler 8)
2408 if (dev->flags & IFF_UP) {
2409 int cpu = smp_processor_id(); /* ok because BHs are off */
2411 if (txq->xmit_lock_owner != cpu) {
2413 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2414 goto recursion_alert;
2416 HARD_TX_LOCK(dev, txq, cpu);
2418 if (!netif_tx_queue_stopped(txq)) {
2419 __this_cpu_inc(xmit_recursion);
2420 rc = dev_hard_start_xmit(skb, dev, txq);
2421 __this_cpu_dec(xmit_recursion);
2422 if (dev_xmit_complete(rc)) {
2423 HARD_TX_UNLOCK(dev, txq);
2427 HARD_TX_UNLOCK(dev, txq);
2428 if (net_ratelimit())
2429 printk(KERN_CRIT "Virtual device %s asks to "
2430 "queue packet!\n", dev->name);
2432 /* Recursion is detected! It is possible,
2436 if (net_ratelimit())
2437 printk(KERN_CRIT "Dead loop on virtual device "
2438 "%s, fix it urgently!\n", dev->name);
2443 rcu_read_unlock_bh();
2448 rcu_read_unlock_bh();
2451 EXPORT_SYMBOL(dev_queue_xmit);
2454 /*=======================================================================
2456 =======================================================================*/
2458 int netdev_max_backlog __read_mostly = 1000;
2459 int netdev_tstamp_prequeue __read_mostly = 1;
2460 int netdev_budget __read_mostly = 300;
2461 int weight_p __read_mostly = 64; /* old backlog weight */
2463 /* Called with irq disabled */
2464 static inline void ____napi_schedule(struct softnet_data *sd,
2465 struct napi_struct *napi)
2467 list_add_tail(&napi->poll_list, &sd->poll_list);
2468 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2472 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2473 * and src/dst port numbers. Returns a non-zero hash number on success
2476 __u32 __skb_get_rxhash(struct sk_buff *skb)
2478 int nhoff, hash = 0, poff;
2479 struct ipv6hdr *ip6;
2482 u32 addr1, addr2, ihl;
2488 nhoff = skb_network_offset(skb);
2490 switch (skb->protocol) {
2491 case __constant_htons(ETH_P_IP):
2492 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2495 ip = (struct iphdr *) (skb->data + nhoff);
2496 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2499 ip_proto = ip->protocol;
2500 addr1 = (__force u32) ip->saddr;
2501 addr2 = (__force u32) ip->daddr;
2504 case __constant_htons(ETH_P_IPV6):
2505 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2508 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2509 ip_proto = ip6->nexthdr;
2510 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2511 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2519 poff = proto_ports_offset(ip_proto);
2521 nhoff += ihl * 4 + poff;
2522 if (pskb_may_pull(skb, nhoff + 4)) {
2523 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2524 if (ports.v16[1] < ports.v16[0])
2525 swap(ports.v16[0], ports.v16[1]);
2529 /* get a consistent hash (same value on both flow directions) */
2533 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2540 EXPORT_SYMBOL(__skb_get_rxhash);
2544 /* One global table that all flow-based protocols share. */
2545 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2546 EXPORT_SYMBOL(rps_sock_flow_table);
2549 * get_rps_cpu is called from netif_receive_skb and returns the target
2550 * CPU from the RPS map of the receiving queue for a given skb.
2551 * rcu_read_lock must be held on entry.
2553 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2554 struct rps_dev_flow **rflowp)
2556 struct netdev_rx_queue *rxqueue;
2557 struct rps_map *map;
2558 struct rps_dev_flow_table *flow_table;
2559 struct rps_sock_flow_table *sock_flow_table;
2563 if (skb_rx_queue_recorded(skb)) {
2564 u16 index = skb_get_rx_queue(skb);
2565 if (unlikely(index >= dev->real_num_rx_queues)) {
2566 WARN_ONCE(dev->real_num_rx_queues > 1,
2567 "%s received packet on queue %u, but number "
2568 "of RX queues is %u\n",
2569 dev->name, index, dev->real_num_rx_queues);
2572 rxqueue = dev->_rx + index;
2576 map = rcu_dereference(rxqueue->rps_map);
2578 if (map->len == 1) {
2579 tcpu = map->cpus[0];
2580 if (cpu_online(tcpu))
2584 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2588 skb_reset_network_header(skb);
2589 if (!skb_get_rxhash(skb))
2592 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2593 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2594 if (flow_table && sock_flow_table) {
2596 struct rps_dev_flow *rflow;
2598 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2601 next_cpu = sock_flow_table->ents[skb->rxhash &
2602 sock_flow_table->mask];
2605 * If the desired CPU (where last recvmsg was done) is
2606 * different from current CPU (one in the rx-queue flow
2607 * table entry), switch if one of the following holds:
2608 * - Current CPU is unset (equal to RPS_NO_CPU).
2609 * - Current CPU is offline.
2610 * - The current CPU's queue tail has advanced beyond the
2611 * last packet that was enqueued using this table entry.
2612 * This guarantees that all previous packets for the flow
2613 * have been dequeued, thus preserving in order delivery.
2615 if (unlikely(tcpu != next_cpu) &&
2616 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2617 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2618 rflow->last_qtail)) >= 0)) {
2619 tcpu = rflow->cpu = next_cpu;
2620 if (tcpu != RPS_NO_CPU)
2621 rflow->last_qtail = per_cpu(softnet_data,
2622 tcpu).input_queue_head;
2624 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2632 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2634 if (cpu_online(tcpu)) {
2644 /* Called from hardirq (IPI) context */
2645 static void rps_trigger_softirq(void *data)
2647 struct softnet_data *sd = data;
2649 ____napi_schedule(sd, &sd->backlog);
2653 #endif /* CONFIG_RPS */
2656 * Check if this softnet_data structure is another cpu one
2657 * If yes, queue it to our IPI list and return 1
2660 static int rps_ipi_queued(struct softnet_data *sd)
2663 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2666 sd->rps_ipi_next = mysd->rps_ipi_list;
2667 mysd->rps_ipi_list = sd;
2669 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2672 #endif /* CONFIG_RPS */
2677 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2678 * queue (may be a remote CPU queue).
2680 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2681 unsigned int *qtail)
2683 struct softnet_data *sd;
2684 unsigned long flags;
2686 sd = &per_cpu(softnet_data, cpu);
2688 local_irq_save(flags);
2691 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2692 if (skb_queue_len(&sd->input_pkt_queue)) {
2694 __skb_queue_tail(&sd->input_pkt_queue, skb);
2695 input_queue_tail_incr_save(sd, qtail);
2697 local_irq_restore(flags);
2698 return NET_RX_SUCCESS;
2701 /* Schedule NAPI for backlog device
2702 * We can use non atomic operation since we own the queue lock
2704 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2705 if (!rps_ipi_queued(sd))
2706 ____napi_schedule(sd, &sd->backlog);
2714 local_irq_restore(flags);
2716 atomic_long_inc(&skb->dev->rx_dropped);
2722 * netif_rx - post buffer to the network code
2723 * @skb: buffer to post
2725 * This function receives a packet from a device driver and queues it for
2726 * the upper (protocol) levels to process. It always succeeds. The buffer
2727 * may be dropped during processing for congestion control or by the
2731 * NET_RX_SUCCESS (no congestion)
2732 * NET_RX_DROP (packet was dropped)
2736 int netif_rx(struct sk_buff *skb)
2740 /* if netpoll wants it, pretend we never saw it */
2741 if (netpoll_rx(skb))
2744 if (netdev_tstamp_prequeue)
2745 net_timestamp_check(skb);
2747 trace_netif_rx(skb);
2750 struct rps_dev_flow voidflow, *rflow = &voidflow;
2756 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2758 cpu = smp_processor_id();
2760 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2768 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2774 EXPORT_SYMBOL(netif_rx);
2776 int netif_rx_ni(struct sk_buff *skb)
2781 err = netif_rx(skb);
2782 if (local_softirq_pending())
2788 EXPORT_SYMBOL(netif_rx_ni);
2790 static void net_tx_action(struct softirq_action *h)
2792 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2794 if (sd->completion_queue) {
2795 struct sk_buff *clist;
2797 local_irq_disable();
2798 clist = sd->completion_queue;
2799 sd->completion_queue = NULL;
2803 struct sk_buff *skb = clist;
2804 clist = clist->next;
2806 WARN_ON(atomic_read(&skb->users));
2807 trace_kfree_skb(skb, net_tx_action);
2812 if (sd->output_queue) {
2815 local_irq_disable();
2816 head = sd->output_queue;
2817 sd->output_queue = NULL;
2818 sd->output_queue_tailp = &sd->output_queue;
2822 struct Qdisc *q = head;
2823 spinlock_t *root_lock;
2825 head = head->next_sched;
2827 root_lock = qdisc_lock(q);
2828 if (spin_trylock(root_lock)) {
2829 smp_mb__before_clear_bit();
2830 clear_bit(__QDISC_STATE_SCHED,
2833 spin_unlock(root_lock);
2835 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2837 __netif_reschedule(q);
2839 smp_mb__before_clear_bit();
2840 clear_bit(__QDISC_STATE_SCHED,
2848 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2849 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2850 /* This hook is defined here for ATM LANE */
2851 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2852 unsigned char *addr) __read_mostly;
2853 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2856 #ifdef CONFIG_NET_CLS_ACT
2857 /* TODO: Maybe we should just force sch_ingress to be compiled in
2858 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2859 * a compare and 2 stores extra right now if we dont have it on
2860 * but have CONFIG_NET_CLS_ACT
2861 * NOTE: This doesnt stop any functionality; if you dont have
2862 * the ingress scheduler, you just cant add policies on ingress.
2865 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2867 struct net_device *dev = skb->dev;
2868 u32 ttl = G_TC_RTTL(skb->tc_verd);
2869 int result = TC_ACT_OK;
2872 if (unlikely(MAX_RED_LOOP < ttl++)) {
2873 if (net_ratelimit())
2874 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2875 skb->skb_iif, dev->ifindex);
2879 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2880 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2883 if (q != &noop_qdisc) {
2884 spin_lock(qdisc_lock(q));
2885 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2886 result = qdisc_enqueue_root(skb, q);
2887 spin_unlock(qdisc_lock(q));
2893 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2894 struct packet_type **pt_prev,
2895 int *ret, struct net_device *orig_dev)
2897 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
2899 if (!rxq || rxq->qdisc == &noop_qdisc)
2903 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2907 switch (ing_filter(skb, rxq)) {
2921 * netdev_rx_handler_register - register receive handler
2922 * @dev: device to register a handler for
2923 * @rx_handler: receive handler to register
2924 * @rx_handler_data: data pointer that is used by rx handler
2926 * Register a receive hander for a device. This handler will then be
2927 * called from __netif_receive_skb. A negative errno code is returned
2930 * The caller must hold the rtnl_mutex.
2932 int netdev_rx_handler_register(struct net_device *dev,
2933 rx_handler_func_t *rx_handler,
2934 void *rx_handler_data)
2938 if (dev->rx_handler)
2941 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2942 rcu_assign_pointer(dev->rx_handler, rx_handler);
2946 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2949 * netdev_rx_handler_unregister - unregister receive handler
2950 * @dev: device to unregister a handler from
2952 * Unregister a receive hander from a device.
2954 * The caller must hold the rtnl_mutex.
2956 void netdev_rx_handler_unregister(struct net_device *dev)
2960 rcu_assign_pointer(dev->rx_handler, NULL);
2961 rcu_assign_pointer(dev->rx_handler_data, NULL);
2963 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2965 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2966 struct net_device *master)
2968 if (skb->pkt_type == PACKET_HOST) {
2969 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2971 memcpy(dest, master->dev_addr, ETH_ALEN);
2975 /* On bonding slaves other than the currently active slave, suppress
2976 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2977 * ARP on active-backup slaves with arp_validate enabled.
2979 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2981 struct net_device *dev = skb->dev;
2983 if (master->priv_flags & IFF_MASTER_ARPMON)
2984 dev->last_rx = jiffies;
2986 if ((master->priv_flags & IFF_MASTER_ALB) &&
2987 (master->priv_flags & IFF_BRIDGE_PORT)) {
2988 /* Do address unmangle. The local destination address
2989 * will be always the one master has. Provides the right
2990 * functionality in a bridge.
2992 skb_bond_set_mac_by_master(skb, master);
2995 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2996 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2997 skb->protocol == __cpu_to_be16(ETH_P_ARP))
3000 if (master->priv_flags & IFF_MASTER_ALB) {
3001 if (skb->pkt_type != PACKET_BROADCAST &&
3002 skb->pkt_type != PACKET_MULTICAST)
3005 if (master->priv_flags & IFF_MASTER_8023AD &&
3006 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
3013 EXPORT_SYMBOL(__skb_bond_should_drop);
3015 static int __netif_receive_skb(struct sk_buff *skb)
3017 struct packet_type *ptype, *pt_prev;
3018 rx_handler_func_t *rx_handler;
3019 struct net_device *orig_dev;
3020 struct net_device *master;
3021 struct net_device *null_or_orig;
3022 struct net_device *orig_or_bond;
3023 int ret = NET_RX_DROP;
3026 if (!netdev_tstamp_prequeue)
3027 net_timestamp_check(skb);
3029 trace_netif_receive_skb(skb);
3031 /* if we've gotten here through NAPI, check netpoll */
3032 if (netpoll_receive_skb(skb))
3036 skb->skb_iif = skb->dev->ifindex;
3039 * bonding note: skbs received on inactive slaves should only
3040 * be delivered to pkt handlers that are exact matches. Also
3041 * the deliver_no_wcard flag will be set. If packet handlers
3042 * are sensitive to duplicate packets these skbs will need to
3043 * be dropped at the handler.
3045 null_or_orig = NULL;
3046 orig_dev = skb->dev;
3047 master = ACCESS_ONCE(orig_dev->master);
3048 if (skb->deliver_no_wcard)
3049 null_or_orig = orig_dev;
3051 if (skb_bond_should_drop(skb, master)) {
3052 skb->deliver_no_wcard = 1;
3053 null_or_orig = orig_dev; /* deliver only exact match */
3058 __this_cpu_inc(softnet_data.processed);
3059 skb_reset_network_header(skb);
3060 skb_reset_transport_header(skb);
3061 skb->mac_len = skb->network_header - skb->mac_header;
3067 #ifdef CONFIG_NET_CLS_ACT
3068 if (skb->tc_verd & TC_NCLS) {
3069 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3074 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3075 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
3076 ptype->dev == orig_dev) {
3078 ret = deliver_skb(skb, pt_prev, orig_dev);
3083 #ifdef CONFIG_NET_CLS_ACT
3084 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3090 /* Handle special case of bridge or macvlan */
3091 rx_handler = rcu_dereference(skb->dev->rx_handler);
3094 ret = deliver_skb(skb, pt_prev, orig_dev);
3097 skb = rx_handler(skb);
3102 if (vlan_tx_tag_present(skb)) {
3104 ret = deliver_skb(skb, pt_prev, orig_dev);
3107 if (vlan_hwaccel_do_receive(&skb)) {
3108 ret = __netif_receive_skb(skb);
3110 } else if (unlikely(!skb))
3115 * Make sure frames received on VLAN interfaces stacked on
3116 * bonding interfaces still make their way to any base bonding
3117 * device that may have registered for a specific ptype. The
3118 * handler may have to adjust skb->dev and orig_dev.
3120 orig_or_bond = orig_dev;
3121 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
3122 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
3123 orig_or_bond = vlan_dev_real_dev(skb->dev);
3126 type = skb->protocol;
3127 list_for_each_entry_rcu(ptype,
3128 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3129 if (ptype->type == type && (ptype->dev == null_or_orig ||
3130 ptype->dev == skb->dev || ptype->dev == orig_dev ||
3131 ptype->dev == orig_or_bond)) {
3133 ret = deliver_skb(skb, pt_prev, orig_dev);
3139 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3141 atomic_long_inc(&skb->dev->rx_dropped);
3143 /* Jamal, now you will not able to escape explaining
3144 * me how you were going to use this. :-)
3155 * netif_receive_skb - process receive buffer from network
3156 * @skb: buffer to process
3158 * netif_receive_skb() is the main receive data processing function.
3159 * It always succeeds. The buffer may be dropped during processing
3160 * for congestion control or by the protocol layers.
3162 * This function may only be called from softirq context and interrupts
3163 * should be enabled.
3165 * Return values (usually ignored):
3166 * NET_RX_SUCCESS: no congestion
3167 * NET_RX_DROP: packet was dropped
3169 int netif_receive_skb(struct sk_buff *skb)
3171 if (netdev_tstamp_prequeue)
3172 net_timestamp_check(skb);
3174 if (skb_defer_rx_timestamp(skb))
3175 return NET_RX_SUCCESS;
3179 struct rps_dev_flow voidflow, *rflow = &voidflow;
3184 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3187 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3191 ret = __netif_receive_skb(skb);
3197 return __netif_receive_skb(skb);
3200 EXPORT_SYMBOL(netif_receive_skb);
3202 /* Network device is going away, flush any packets still pending
3203 * Called with irqs disabled.
3205 static void flush_backlog(void *arg)
3207 struct net_device *dev = arg;
3208 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3209 struct sk_buff *skb, *tmp;
3212 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3213 if (skb->dev == dev) {
3214 __skb_unlink(skb, &sd->input_pkt_queue);
3216 input_queue_head_incr(sd);
3221 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3222 if (skb->dev == dev) {
3223 __skb_unlink(skb, &sd->process_queue);
3225 input_queue_head_incr(sd);
3230 static int napi_gro_complete(struct sk_buff *skb)
3232 struct packet_type *ptype;
3233 __be16 type = skb->protocol;
3234 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3237 if (NAPI_GRO_CB(skb)->count == 1) {
3238 skb_shinfo(skb)->gso_size = 0;
3243 list_for_each_entry_rcu(ptype, head, list) {
3244 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3247 err = ptype->gro_complete(skb);
3253 WARN_ON(&ptype->list == head);
3255 return NET_RX_SUCCESS;
3259 return netif_receive_skb(skb);
3262 inline void napi_gro_flush(struct napi_struct *napi)
3264 struct sk_buff *skb, *next;
3266 for (skb = napi->gro_list; skb; skb = next) {
3269 napi_gro_complete(skb);
3272 napi->gro_count = 0;
3273 napi->gro_list = NULL;
3275 EXPORT_SYMBOL(napi_gro_flush);
3277 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3279 struct sk_buff **pp = NULL;
3280 struct packet_type *ptype;
3281 __be16 type = skb->protocol;
3282 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3285 enum gro_result ret;
3287 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3290 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3294 list_for_each_entry_rcu(ptype, head, list) {
3295 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3298 skb_set_network_header(skb, skb_gro_offset(skb));
3299 mac_len = skb->network_header - skb->mac_header;
3300 skb->mac_len = mac_len;
3301 NAPI_GRO_CB(skb)->same_flow = 0;
3302 NAPI_GRO_CB(skb)->flush = 0;
3303 NAPI_GRO_CB(skb)->free = 0;
3305 pp = ptype->gro_receive(&napi->gro_list, skb);
3310 if (&ptype->list == head)
3313 same_flow = NAPI_GRO_CB(skb)->same_flow;
3314 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3317 struct sk_buff *nskb = *pp;
3321 napi_gro_complete(nskb);
3328 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3332 NAPI_GRO_CB(skb)->count = 1;
3333 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3334 skb->next = napi->gro_list;
3335 napi->gro_list = skb;
3339 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3340 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3342 BUG_ON(skb->end - skb->tail < grow);
3344 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3347 skb->data_len -= grow;
3349 skb_shinfo(skb)->frags[0].page_offset += grow;
3350 skb_shinfo(skb)->frags[0].size -= grow;
3352 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3353 put_page(skb_shinfo(skb)->frags[0].page);
3354 memmove(skb_shinfo(skb)->frags,
3355 skb_shinfo(skb)->frags + 1,
3356 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3367 EXPORT_SYMBOL(dev_gro_receive);
3369 static inline gro_result_t
3370 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3374 for (p = napi->gro_list; p; p = p->next) {
3375 unsigned long diffs;
3377 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3378 diffs |= p->vlan_tci ^ skb->vlan_tci;
3379 diffs |= compare_ether_header(skb_mac_header(p),
3380 skb_gro_mac_header(skb));
3381 NAPI_GRO_CB(p)->same_flow = !diffs;
3382 NAPI_GRO_CB(p)->flush = 0;
3385 return dev_gro_receive(napi, skb);
3388 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3392 if (netif_receive_skb(skb))
3397 case GRO_MERGED_FREE:
3408 EXPORT_SYMBOL(napi_skb_finish);
3410 void skb_gro_reset_offset(struct sk_buff *skb)
3412 NAPI_GRO_CB(skb)->data_offset = 0;
3413 NAPI_GRO_CB(skb)->frag0 = NULL;
3414 NAPI_GRO_CB(skb)->frag0_len = 0;
3416 if (skb->mac_header == skb->tail &&
3417 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3418 NAPI_GRO_CB(skb)->frag0 =
3419 page_address(skb_shinfo(skb)->frags[0].page) +
3420 skb_shinfo(skb)->frags[0].page_offset;
3421 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3424 EXPORT_SYMBOL(skb_gro_reset_offset);
3426 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3428 skb_gro_reset_offset(skb);
3430 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3432 EXPORT_SYMBOL(napi_gro_receive);
3434 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3436 __skb_pull(skb, skb_headlen(skb));
3437 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3443 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3445 struct sk_buff *skb = napi->skb;
3448 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3454 EXPORT_SYMBOL(napi_get_frags);
3456 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3462 skb->protocol = eth_type_trans(skb, skb->dev);
3464 if (ret == GRO_HELD)
3465 skb_gro_pull(skb, -ETH_HLEN);
3466 else if (netif_receive_skb(skb))
3471 case GRO_MERGED_FREE:
3472 napi_reuse_skb(napi, skb);
3481 EXPORT_SYMBOL(napi_frags_finish);
3483 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3485 struct sk_buff *skb = napi->skb;
3492 skb_reset_mac_header(skb);
3493 skb_gro_reset_offset(skb);
3495 off = skb_gro_offset(skb);
3496 hlen = off + sizeof(*eth);
3497 eth = skb_gro_header_fast(skb, off);
3498 if (skb_gro_header_hard(skb, hlen)) {
3499 eth = skb_gro_header_slow(skb, hlen, off);
3500 if (unlikely(!eth)) {
3501 napi_reuse_skb(napi, skb);
3507 skb_gro_pull(skb, sizeof(*eth));
3510 * This works because the only protocols we care about don't require
3511 * special handling. We'll fix it up properly at the end.
3513 skb->protocol = eth->h_proto;
3518 EXPORT_SYMBOL(napi_frags_skb);
3520 gro_result_t napi_gro_frags(struct napi_struct *napi)
3522 struct sk_buff *skb = napi_frags_skb(napi);
3527 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3529 EXPORT_SYMBOL(napi_gro_frags);
3532 * net_rps_action sends any pending IPI's for rps.
3533 * Note: called with local irq disabled, but exits with local irq enabled.
3535 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3538 struct softnet_data *remsd = sd->rps_ipi_list;
3541 sd->rps_ipi_list = NULL;
3545 /* Send pending IPI's to kick RPS processing on remote cpus. */
3547 struct softnet_data *next = remsd->rps_ipi_next;
3549 if (cpu_online(remsd->cpu))
3550 __smp_call_function_single(remsd->cpu,
3559 static int process_backlog(struct napi_struct *napi, int quota)
3562 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3565 /* Check if we have pending ipi, its better to send them now,
3566 * not waiting net_rx_action() end.
3568 if (sd->rps_ipi_list) {
3569 local_irq_disable();
3570 net_rps_action_and_irq_enable(sd);
3573 napi->weight = weight_p;
3574 local_irq_disable();
3575 while (work < quota) {
3576 struct sk_buff *skb;
3579 while ((skb = __skb_dequeue(&sd->process_queue))) {
3581 __netif_receive_skb(skb);
3582 local_irq_disable();
3583 input_queue_head_incr(sd);
3584 if (++work >= quota) {
3591 qlen = skb_queue_len(&sd->input_pkt_queue);
3593 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3594 &sd->process_queue);
3596 if (qlen < quota - work) {
3598 * Inline a custom version of __napi_complete().
3599 * only current cpu owns and manipulates this napi,
3600 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3601 * we can use a plain write instead of clear_bit(),
3602 * and we dont need an smp_mb() memory barrier.
3604 list_del(&napi->poll_list);
3607 quota = work + qlen;
3617 * __napi_schedule - schedule for receive
3618 * @n: entry to schedule
3620 * The entry's receive function will be scheduled to run
3622 void __napi_schedule(struct napi_struct *n)
3624 unsigned long flags;
3626 local_irq_save(flags);
3627 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3628 local_irq_restore(flags);
3630 EXPORT_SYMBOL(__napi_schedule);
3632 void __napi_complete(struct napi_struct *n)
3634 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3635 BUG_ON(n->gro_list);
3637 list_del(&n->poll_list);
3638 smp_mb__before_clear_bit();
3639 clear_bit(NAPI_STATE_SCHED, &n->state);
3641 EXPORT_SYMBOL(__napi_complete);
3643 void napi_complete(struct napi_struct *n)
3645 unsigned long flags;
3648 * don't let napi dequeue from the cpu poll list
3649 * just in case its running on a different cpu
3651 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3655 local_irq_save(flags);
3657 local_irq_restore(flags);
3659 EXPORT_SYMBOL(napi_complete);
3661 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3662 int (*poll)(struct napi_struct *, int), int weight)
3664 INIT_LIST_HEAD(&napi->poll_list);
3665 napi->gro_count = 0;
3666 napi->gro_list = NULL;
3669 napi->weight = weight;
3670 list_add(&napi->dev_list, &dev->napi_list);
3672 #ifdef CONFIG_NETPOLL
3673 spin_lock_init(&napi->poll_lock);
3674 napi->poll_owner = -1;
3676 set_bit(NAPI_STATE_SCHED, &napi->state);
3678 EXPORT_SYMBOL(netif_napi_add);
3680 void netif_napi_del(struct napi_struct *napi)
3682 struct sk_buff *skb, *next;
3684 list_del_init(&napi->dev_list);
3685 napi_free_frags(napi);
3687 for (skb = napi->gro_list; skb; skb = next) {
3693 napi->gro_list = NULL;
3694 napi->gro_count = 0;
3696 EXPORT_SYMBOL(netif_napi_del);
3698 static void net_rx_action(struct softirq_action *h)
3700 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3701 unsigned long time_limit = jiffies + 2;
3702 int budget = netdev_budget;
3705 local_irq_disable();
3707 while (!list_empty(&sd->poll_list)) {
3708 struct napi_struct *n;
3711 /* If softirq window is exhuasted then punt.
3712 * Allow this to run for 2 jiffies since which will allow
3713 * an average latency of 1.5/HZ.
3715 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3720 /* Even though interrupts have been re-enabled, this
3721 * access is safe because interrupts can only add new
3722 * entries to the tail of this list, and only ->poll()
3723 * calls can remove this head entry from the list.
3725 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3727 have = netpoll_poll_lock(n);
3731 /* This NAPI_STATE_SCHED test is for avoiding a race
3732 * with netpoll's poll_napi(). Only the entity which
3733 * obtains the lock and sees NAPI_STATE_SCHED set will
3734 * actually make the ->poll() call. Therefore we avoid
3735 * accidently calling ->poll() when NAPI is not scheduled.
3738 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3739 work = n->poll(n, weight);
3743 WARN_ON_ONCE(work > weight);
3747 local_irq_disable();
3749 /* Drivers must not modify the NAPI state if they
3750 * consume the entire weight. In such cases this code
3751 * still "owns" the NAPI instance and therefore can
3752 * move the instance around on the list at-will.
3754 if (unlikely(work == weight)) {
3755 if (unlikely(napi_disable_pending(n))) {
3758 local_irq_disable();
3760 list_move_tail(&n->poll_list, &sd->poll_list);
3763 netpoll_poll_unlock(have);
3766 net_rps_action_and_irq_enable(sd);
3768 #ifdef CONFIG_NET_DMA
3770 * There may not be any more sk_buffs coming right now, so push
3771 * any pending DMA copies to hardware
3773 dma_issue_pending_all();
3780 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3784 static gifconf_func_t *gifconf_list[NPROTO];
3787 * register_gifconf - register a SIOCGIF handler
3788 * @family: Address family
3789 * @gifconf: Function handler
3791 * Register protocol dependent address dumping routines. The handler
3792 * that is passed must not be freed or reused until it has been replaced
3793 * by another handler.
3795 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3797 if (family >= NPROTO)
3799 gifconf_list[family] = gifconf;
3802 EXPORT_SYMBOL(register_gifconf);
3806 * Map an interface index to its name (SIOCGIFNAME)
3810 * We need this ioctl for efficient implementation of the
3811 * if_indextoname() function required by the IPv6 API. Without
3812 * it, we would have to search all the interfaces to find a
3816 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3818 struct net_device *dev;
3822 * Fetch the caller's info block.
3825 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3829 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3835 strcpy(ifr.ifr_name, dev->name);
3838 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3844 * Perform a SIOCGIFCONF call. This structure will change
3845 * size eventually, and there is nothing I can do about it.
3846 * Thus we will need a 'compatibility mode'.
3849 static int dev_ifconf(struct net *net, char __user *arg)
3852 struct net_device *dev;
3859 * Fetch the caller's info block.
3862 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3869 * Loop over the interfaces, and write an info block for each.
3873 for_each_netdev(net, dev) {
3874 for (i = 0; i < NPROTO; i++) {
3875 if (gifconf_list[i]) {
3878 done = gifconf_list[i](dev, NULL, 0);
3880 done = gifconf_list[i](dev, pos + total,
3890 * All done. Write the updated control block back to the caller.
3892 ifc.ifc_len = total;
3895 * Both BSD and Solaris return 0 here, so we do too.
3897 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3900 #ifdef CONFIG_PROC_FS
3902 * This is invoked by the /proc filesystem handler to display a device
3905 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3908 struct net *net = seq_file_net(seq);
3910 struct net_device *dev;
3914 return SEQ_START_TOKEN;
3917 for_each_netdev_rcu(net, dev)
3924 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3926 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3927 first_net_device(seq_file_net(seq)) :
3928 next_net_device((struct net_device *)v);
3931 return rcu_dereference(dev);
3934 void dev_seq_stop(struct seq_file *seq, void *v)
3940 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3942 struct rtnl_link_stats64 temp;
3943 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3945 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3946 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3947 dev->name, stats->rx_bytes, stats->rx_packets,
3949 stats->rx_dropped + stats->rx_missed_errors,
3950 stats->rx_fifo_errors,
3951 stats->rx_length_errors + stats->rx_over_errors +
3952 stats->rx_crc_errors + stats->rx_frame_errors,
3953 stats->rx_compressed, stats->multicast,
3954 stats->tx_bytes, stats->tx_packets,
3955 stats->tx_errors, stats->tx_dropped,
3956 stats->tx_fifo_errors, stats->collisions,
3957 stats->tx_carrier_errors +
3958 stats->tx_aborted_errors +
3959 stats->tx_window_errors +
3960 stats->tx_heartbeat_errors,
3961 stats->tx_compressed);
3965 * Called from the PROCfs module. This now uses the new arbitrary sized
3966 * /proc/net interface to create /proc/net/dev
3968 static int dev_seq_show(struct seq_file *seq, void *v)
3970 if (v == SEQ_START_TOKEN)
3971 seq_puts(seq, "Inter-| Receive "
3973 " face |bytes packets errs drop fifo frame "
3974 "compressed multicast|bytes packets errs "
3975 "drop fifo colls carrier compressed\n");
3977 dev_seq_printf_stats(seq, v);
3981 static struct softnet_data *softnet_get_online(loff_t *pos)
3983 struct softnet_data *sd = NULL;
3985 while (*pos < nr_cpu_ids)
3986 if (cpu_online(*pos)) {
3987 sd = &per_cpu(softnet_data, *pos);
3994 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3996 return softnet_get_online(pos);
3999 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4002 return softnet_get_online(pos);
4005 static void softnet_seq_stop(struct seq_file *seq, void *v)
4009 static int softnet_seq_show(struct seq_file *seq, void *v)
4011 struct softnet_data *sd = v;
4013 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4014 sd->processed, sd->dropped, sd->time_squeeze, 0,
4015 0, 0, 0, 0, /* was fastroute */
4016 sd->cpu_collision, sd->received_rps);
4020 static const struct seq_operations dev_seq_ops = {
4021 .start = dev_seq_start,
4022 .next = dev_seq_next,
4023 .stop = dev_seq_stop,
4024 .show = dev_seq_show,
4027 static int dev_seq_open(struct inode *inode, struct file *file)
4029 return seq_open_net(inode, file, &dev_seq_ops,
4030 sizeof(struct seq_net_private));
4033 static const struct file_operations dev_seq_fops = {
4034 .owner = THIS_MODULE,
4035 .open = dev_seq_open,
4037 .llseek = seq_lseek,
4038 .release = seq_release_net,
4041 static const struct seq_operations softnet_seq_ops = {
4042 .start = softnet_seq_start,
4043 .next = softnet_seq_next,
4044 .stop = softnet_seq_stop,
4045 .show = softnet_seq_show,
4048 static int softnet_seq_open(struct inode *inode, struct file *file)
4050 return seq_open(file, &softnet_seq_ops);
4053 static const struct file_operations softnet_seq_fops = {
4054 .owner = THIS_MODULE,
4055 .open = softnet_seq_open,
4057 .llseek = seq_lseek,
4058 .release = seq_release,
4061 static void *ptype_get_idx(loff_t pos)
4063 struct packet_type *pt = NULL;
4067 list_for_each_entry_rcu(pt, &ptype_all, list) {
4073 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4074 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4083 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4087 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4090 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4092 struct packet_type *pt;
4093 struct list_head *nxt;
4097 if (v == SEQ_START_TOKEN)
4098 return ptype_get_idx(0);
4101 nxt = pt->list.next;
4102 if (pt->type == htons(ETH_P_ALL)) {
4103 if (nxt != &ptype_all)
4106 nxt = ptype_base[0].next;
4108 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4110 while (nxt == &ptype_base[hash]) {
4111 if (++hash >= PTYPE_HASH_SIZE)
4113 nxt = ptype_base[hash].next;
4116 return list_entry(nxt, struct packet_type, list);
4119 static void ptype_seq_stop(struct seq_file *seq, void *v)
4125 static int ptype_seq_show(struct seq_file *seq, void *v)
4127 struct packet_type *pt = v;
4129 if (v == SEQ_START_TOKEN)
4130 seq_puts(seq, "Type Device Function\n");
4131 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4132 if (pt->type == htons(ETH_P_ALL))
4133 seq_puts(seq, "ALL ");
4135 seq_printf(seq, "%04x", ntohs(pt->type));
4137 seq_printf(seq, " %-8s %pF\n",
4138 pt->dev ? pt->dev->name : "", pt->func);
4144 static const struct seq_operations ptype_seq_ops = {
4145 .start = ptype_seq_start,
4146 .next = ptype_seq_next,
4147 .stop = ptype_seq_stop,
4148 .show = ptype_seq_show,
4151 static int ptype_seq_open(struct inode *inode, struct file *file)
4153 return seq_open_net(inode, file, &ptype_seq_ops,
4154 sizeof(struct seq_net_private));
4157 static const struct file_operations ptype_seq_fops = {
4158 .owner = THIS_MODULE,
4159 .open = ptype_seq_open,
4161 .llseek = seq_lseek,
4162 .release = seq_release_net,
4166 static int __net_init dev_proc_net_init(struct net *net)
4170 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4172 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4174 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4177 if (wext_proc_init(net))
4183 proc_net_remove(net, "ptype");
4185 proc_net_remove(net, "softnet_stat");
4187 proc_net_remove(net, "dev");
4191 static void __net_exit dev_proc_net_exit(struct net *net)
4193 wext_proc_exit(net);
4195 proc_net_remove(net, "ptype");
4196 proc_net_remove(net, "softnet_stat");
4197 proc_net_remove(net, "dev");
4200 static struct pernet_operations __net_initdata dev_proc_ops = {
4201 .init = dev_proc_net_init,
4202 .exit = dev_proc_net_exit,
4205 static int __init dev_proc_init(void)
4207 return register_pernet_subsys(&dev_proc_ops);
4210 #define dev_proc_init() 0
4211 #endif /* CONFIG_PROC_FS */
4215 * netdev_set_master - set up master/slave pair
4216 * @slave: slave device
4217 * @master: new master device
4219 * Changes the master device of the slave. Pass %NULL to break the
4220 * bonding. The caller must hold the RTNL semaphore. On a failure
4221 * a negative errno code is returned. On success the reference counts
4222 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4223 * function returns zero.
4225 int netdev_set_master(struct net_device *slave, struct net_device *master)
4227 struct net_device *old = slave->master;
4237 slave->master = master;
4244 slave->flags |= IFF_SLAVE;
4246 slave->flags &= ~IFF_SLAVE;
4248 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4251 EXPORT_SYMBOL(netdev_set_master);
4253 static void dev_change_rx_flags(struct net_device *dev, int flags)
4255 const struct net_device_ops *ops = dev->netdev_ops;
4257 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4258 ops->ndo_change_rx_flags(dev, flags);
4261 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4263 unsigned short old_flags = dev->flags;
4269 dev->flags |= IFF_PROMISC;
4270 dev->promiscuity += inc;
4271 if (dev->promiscuity == 0) {
4274 * If inc causes overflow, untouch promisc and return error.
4277 dev->flags &= ~IFF_PROMISC;
4279 dev->promiscuity -= inc;
4280 printk(KERN_WARNING "%s: promiscuity touches roof, "
4281 "set promiscuity failed, promiscuity feature "
4282 "of device might be broken.\n", dev->name);
4286 if (dev->flags != old_flags) {
4287 printk(KERN_INFO "device %s %s promiscuous mode\n",
4288 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4290 if (audit_enabled) {
4291 current_uid_gid(&uid, &gid);
4292 audit_log(current->audit_context, GFP_ATOMIC,
4293 AUDIT_ANOM_PROMISCUOUS,
4294 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4295 dev->name, (dev->flags & IFF_PROMISC),
4296 (old_flags & IFF_PROMISC),
4297 audit_get_loginuid(current),
4299 audit_get_sessionid(current));
4302 dev_change_rx_flags(dev, IFF_PROMISC);
4308 * dev_set_promiscuity - update promiscuity count on a device
4312 * Add or remove promiscuity from a device. While the count in the device
4313 * remains above zero the interface remains promiscuous. Once it hits zero
4314 * the device reverts back to normal filtering operation. A negative inc
4315 * value is used to drop promiscuity on the device.
4316 * Return 0 if successful or a negative errno code on error.
4318 int dev_set_promiscuity(struct net_device *dev, int inc)
4320 unsigned short old_flags = dev->flags;
4323 err = __dev_set_promiscuity(dev, inc);
4326 if (dev->flags != old_flags)
4327 dev_set_rx_mode(dev);
4330 EXPORT_SYMBOL(dev_set_promiscuity);
4333 * dev_set_allmulti - update allmulti count on a device
4337 * Add or remove reception of all multicast frames to a device. While the
4338 * count in the device remains above zero the interface remains listening
4339 * to all interfaces. Once it hits zero the device reverts back to normal
4340 * filtering operation. A negative @inc value is used to drop the counter
4341 * when releasing a resource needing all multicasts.
4342 * Return 0 if successful or a negative errno code on error.
4345 int dev_set_allmulti(struct net_device *dev, int inc)
4347 unsigned short old_flags = dev->flags;
4351 dev->flags |= IFF_ALLMULTI;
4352 dev->allmulti += inc;
4353 if (dev->allmulti == 0) {
4356 * If inc causes overflow, untouch allmulti and return error.
4359 dev->flags &= ~IFF_ALLMULTI;
4361 dev->allmulti -= inc;
4362 printk(KERN_WARNING "%s: allmulti touches roof, "
4363 "set allmulti failed, allmulti feature of "
4364 "device might be broken.\n", dev->name);
4368 if (dev->flags ^ old_flags) {
4369 dev_change_rx_flags(dev, IFF_ALLMULTI);
4370 dev_set_rx_mode(dev);
4374 EXPORT_SYMBOL(dev_set_allmulti);
4377 * Upload unicast and multicast address lists to device and
4378 * configure RX filtering. When the device doesn't support unicast
4379 * filtering it is put in promiscuous mode while unicast addresses
4382 void __dev_set_rx_mode(struct net_device *dev)
4384 const struct net_device_ops *ops = dev->netdev_ops;
4386 /* dev_open will call this function so the list will stay sane. */
4387 if (!(dev->flags&IFF_UP))
4390 if (!netif_device_present(dev))
4393 if (ops->ndo_set_rx_mode)
4394 ops->ndo_set_rx_mode(dev);
4396 /* Unicast addresses changes may only happen under the rtnl,
4397 * therefore calling __dev_set_promiscuity here is safe.
4399 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4400 __dev_set_promiscuity(dev, 1);
4401 dev->uc_promisc = 1;
4402 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4403 __dev_set_promiscuity(dev, -1);
4404 dev->uc_promisc = 0;
4407 if (ops->ndo_set_multicast_list)
4408 ops->ndo_set_multicast_list(dev);
4412 void dev_set_rx_mode(struct net_device *dev)
4414 netif_addr_lock_bh(dev);
4415 __dev_set_rx_mode(dev);
4416 netif_addr_unlock_bh(dev);
4420 * dev_get_flags - get flags reported to userspace
4423 * Get the combination of flag bits exported through APIs to userspace.
4425 unsigned dev_get_flags(const struct net_device *dev)
4429 flags = (dev->flags & ~(IFF_PROMISC |
4434 (dev->gflags & (IFF_PROMISC |
4437 if (netif_running(dev)) {
4438 if (netif_oper_up(dev))
4439 flags |= IFF_RUNNING;
4440 if (netif_carrier_ok(dev))
4441 flags |= IFF_LOWER_UP;
4442 if (netif_dormant(dev))
4443 flags |= IFF_DORMANT;
4448 EXPORT_SYMBOL(dev_get_flags);
4450 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4452 int old_flags = dev->flags;
4458 * Set the flags on our device.
4461 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4462 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4464 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4468 * Load in the correct multicast list now the flags have changed.
4471 if ((old_flags ^ flags) & IFF_MULTICAST)
4472 dev_change_rx_flags(dev, IFF_MULTICAST);
4474 dev_set_rx_mode(dev);
4477 * Have we downed the interface. We handle IFF_UP ourselves
4478 * according to user attempts to set it, rather than blindly
4483 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4484 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4487 dev_set_rx_mode(dev);
4490 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4491 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4493 dev->gflags ^= IFF_PROMISC;
4494 dev_set_promiscuity(dev, inc);
4497 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4498 is important. Some (broken) drivers set IFF_PROMISC, when
4499 IFF_ALLMULTI is requested not asking us and not reporting.
4501 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4502 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4504 dev->gflags ^= IFF_ALLMULTI;
4505 dev_set_allmulti(dev, inc);
4511 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4513 unsigned int changes = dev->flags ^ old_flags;
4515 if (changes & IFF_UP) {
4516 if (dev->flags & IFF_UP)
4517 call_netdevice_notifiers(NETDEV_UP, dev);
4519 call_netdevice_notifiers(NETDEV_DOWN, dev);
4522 if (dev->flags & IFF_UP &&
4523 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4524 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4528 * dev_change_flags - change device settings
4530 * @flags: device state flags
4532 * Change settings on device based state flags. The flags are
4533 * in the userspace exported format.
4535 int dev_change_flags(struct net_device *dev, unsigned flags)
4538 int old_flags = dev->flags;
4540 ret = __dev_change_flags(dev, flags);
4544 changes = old_flags ^ dev->flags;
4546 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4548 __dev_notify_flags(dev, old_flags);
4551 EXPORT_SYMBOL(dev_change_flags);
4554 * dev_set_mtu - Change maximum transfer unit
4556 * @new_mtu: new transfer unit
4558 * Change the maximum transfer size of the network device.
4560 int dev_set_mtu(struct net_device *dev, int new_mtu)
4562 const struct net_device_ops *ops = dev->netdev_ops;
4565 if (new_mtu == dev->mtu)
4568 /* MTU must be positive. */
4572 if (!netif_device_present(dev))
4576 if (ops->ndo_change_mtu)
4577 err = ops->ndo_change_mtu(dev, new_mtu);
4581 if (!err && dev->flags & IFF_UP)
4582 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4585 EXPORT_SYMBOL(dev_set_mtu);
4588 * dev_set_mac_address - Change Media Access Control Address
4592 * Change the hardware (MAC) address of the device
4594 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4596 const struct net_device_ops *ops = dev->netdev_ops;
4599 if (!ops->ndo_set_mac_address)
4601 if (sa->sa_family != dev->type)
4603 if (!netif_device_present(dev))
4605 err = ops->ndo_set_mac_address(dev, sa);
4607 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4610 EXPORT_SYMBOL(dev_set_mac_address);
4613 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4615 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4618 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4624 case SIOCGIFFLAGS: /* Get interface flags */
4625 ifr->ifr_flags = (short) dev_get_flags(dev);
4628 case SIOCGIFMETRIC: /* Get the metric on the interface
4629 (currently unused) */
4630 ifr->ifr_metric = 0;
4633 case SIOCGIFMTU: /* Get the MTU of a device */
4634 ifr->ifr_mtu = dev->mtu;
4639 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4641 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4642 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4643 ifr->ifr_hwaddr.sa_family = dev->type;
4651 ifr->ifr_map.mem_start = dev->mem_start;
4652 ifr->ifr_map.mem_end = dev->mem_end;
4653 ifr->ifr_map.base_addr = dev->base_addr;
4654 ifr->ifr_map.irq = dev->irq;
4655 ifr->ifr_map.dma = dev->dma;
4656 ifr->ifr_map.port = dev->if_port;
4660 ifr->ifr_ifindex = dev->ifindex;
4664 ifr->ifr_qlen = dev->tx_queue_len;
4668 /* dev_ioctl() should ensure this case
4680 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4682 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4685 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4686 const struct net_device_ops *ops;
4691 ops = dev->netdev_ops;
4694 case SIOCSIFFLAGS: /* Set interface flags */
4695 return dev_change_flags(dev, ifr->ifr_flags);
4697 case SIOCSIFMETRIC: /* Set the metric on the interface
4698 (currently unused) */
4701 case SIOCSIFMTU: /* Set the MTU of a device */
4702 return dev_set_mtu(dev, ifr->ifr_mtu);
4705 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4707 case SIOCSIFHWBROADCAST:
4708 if (ifr->ifr_hwaddr.sa_family != dev->type)
4710 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4711 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4712 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4716 if (ops->ndo_set_config) {
4717 if (!netif_device_present(dev))
4719 return ops->ndo_set_config(dev, &ifr->ifr_map);
4724 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4725 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4727 if (!netif_device_present(dev))
4729 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4732 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4733 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4735 if (!netif_device_present(dev))
4737 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4740 if (ifr->ifr_qlen < 0)
4742 dev->tx_queue_len = ifr->ifr_qlen;
4746 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4747 return dev_change_name(dev, ifr->ifr_newname);
4750 * Unknown or private ioctl
4753 if ((cmd >= SIOCDEVPRIVATE &&
4754 cmd <= SIOCDEVPRIVATE + 15) ||
4755 cmd == SIOCBONDENSLAVE ||
4756 cmd == SIOCBONDRELEASE ||
4757 cmd == SIOCBONDSETHWADDR ||
4758 cmd == SIOCBONDSLAVEINFOQUERY ||
4759 cmd == SIOCBONDINFOQUERY ||
4760 cmd == SIOCBONDCHANGEACTIVE ||
4761 cmd == SIOCGMIIPHY ||
4762 cmd == SIOCGMIIREG ||
4763 cmd == SIOCSMIIREG ||
4764 cmd == SIOCBRADDIF ||
4765 cmd == SIOCBRDELIF ||
4766 cmd == SIOCSHWTSTAMP ||
4767 cmd == SIOCWANDEV) {
4769 if (ops->ndo_do_ioctl) {
4770 if (netif_device_present(dev))
4771 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4783 * This function handles all "interface"-type I/O control requests. The actual
4784 * 'doing' part of this is dev_ifsioc above.
4788 * dev_ioctl - network device ioctl
4789 * @net: the applicable net namespace
4790 * @cmd: command to issue
4791 * @arg: pointer to a struct ifreq in user space
4793 * Issue ioctl functions to devices. This is normally called by the
4794 * user space syscall interfaces but can sometimes be useful for
4795 * other purposes. The return value is the return from the syscall if
4796 * positive or a negative errno code on error.
4799 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4805 /* One special case: SIOCGIFCONF takes ifconf argument
4806 and requires shared lock, because it sleeps writing
4810 if (cmd == SIOCGIFCONF) {
4812 ret = dev_ifconf(net, (char __user *) arg);
4816 if (cmd == SIOCGIFNAME)
4817 return dev_ifname(net, (struct ifreq __user *)arg);
4819 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4822 ifr.ifr_name[IFNAMSIZ-1] = 0;
4824 colon = strchr(ifr.ifr_name, ':');
4829 * See which interface the caller is talking about.
4834 * These ioctl calls:
4835 * - can be done by all.
4836 * - atomic and do not require locking.
4847 dev_load(net, ifr.ifr_name);
4849 ret = dev_ifsioc_locked(net, &ifr, cmd);
4854 if (copy_to_user(arg, &ifr,
4855 sizeof(struct ifreq)))
4861 dev_load(net, ifr.ifr_name);
4863 ret = dev_ethtool(net, &ifr);
4868 if (copy_to_user(arg, &ifr,
4869 sizeof(struct ifreq)))
4875 * These ioctl calls:
4876 * - require superuser power.
4877 * - require strict serialization.
4883 if (!capable(CAP_NET_ADMIN))
4885 dev_load(net, ifr.ifr_name);
4887 ret = dev_ifsioc(net, &ifr, cmd);
4892 if (copy_to_user(arg, &ifr,
4893 sizeof(struct ifreq)))
4899 * These ioctl calls:
4900 * - require superuser power.
4901 * - require strict serialization.
4902 * - do not return a value
4912 case SIOCSIFHWBROADCAST:
4915 case SIOCBONDENSLAVE:
4916 case SIOCBONDRELEASE:
4917 case SIOCBONDSETHWADDR:
4918 case SIOCBONDCHANGEACTIVE:
4922 if (!capable(CAP_NET_ADMIN))
4925 case SIOCBONDSLAVEINFOQUERY:
4926 case SIOCBONDINFOQUERY:
4927 dev_load(net, ifr.ifr_name);
4929 ret = dev_ifsioc(net, &ifr, cmd);
4934 /* Get the per device memory space. We can add this but
4935 * currently do not support it */
4937 /* Set the per device memory buffer space.
4938 * Not applicable in our case */
4943 * Unknown or private ioctl.
4946 if (cmd == SIOCWANDEV ||
4947 (cmd >= SIOCDEVPRIVATE &&
4948 cmd <= SIOCDEVPRIVATE + 15)) {
4949 dev_load(net, ifr.ifr_name);
4951 ret = dev_ifsioc(net, &ifr, cmd);
4953 if (!ret && copy_to_user(arg, &ifr,
4954 sizeof(struct ifreq)))
4958 /* Take care of Wireless Extensions */
4959 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4960 return wext_handle_ioctl(net, &ifr, cmd, arg);
4967 * dev_new_index - allocate an ifindex
4968 * @net: the applicable net namespace
4970 * Returns a suitable unique value for a new device interface
4971 * number. The caller must hold the rtnl semaphore or the
4972 * dev_base_lock to be sure it remains unique.
4974 static int dev_new_index(struct net *net)
4980 if (!__dev_get_by_index(net, ifindex))
4985 /* Delayed registration/unregisteration */
4986 static LIST_HEAD(net_todo_list);
4988 static void net_set_todo(struct net_device *dev)
4990 list_add_tail(&dev->todo_list, &net_todo_list);
4993 static void rollback_registered_many(struct list_head *head)
4995 struct net_device *dev, *tmp;
4997 BUG_ON(dev_boot_phase);
5000 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5001 /* Some devices call without registering
5002 * for initialization unwind. Remove those
5003 * devices and proceed with the remaining.
5005 if (dev->reg_state == NETREG_UNINITIALIZED) {
5006 pr_debug("unregister_netdevice: device %s/%p never "
5007 "was registered\n", dev->name, dev);
5010 list_del(&dev->unreg_list);
5014 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5017 /* If device is running, close it first. */
5018 dev_close_many(head);
5020 list_for_each_entry(dev, head, unreg_list) {
5021 /* And unlink it from device chain. */
5022 unlist_netdevice(dev);
5024 dev->reg_state = NETREG_UNREGISTERING;
5029 list_for_each_entry(dev, head, unreg_list) {
5030 /* Shutdown queueing discipline. */
5034 /* Notify protocols, that we are about to destroy
5035 this device. They should clean all the things.
5037 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5039 if (!dev->rtnl_link_ops ||
5040 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5041 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5044 * Flush the unicast and multicast chains
5049 if (dev->netdev_ops->ndo_uninit)
5050 dev->netdev_ops->ndo_uninit(dev);
5052 /* Notifier chain MUST detach us from master device. */
5053 WARN_ON(dev->master);
5055 /* Remove entries from kobject tree */
5056 netdev_unregister_kobject(dev);
5059 /* Process any work delayed until the end of the batch */
5060 dev = list_first_entry(head, struct net_device, unreg_list);
5061 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5065 list_for_each_entry(dev, head, unreg_list)
5069 static void rollback_registered(struct net_device *dev)
5073 list_add(&dev->unreg_list, &single);
5074 rollback_registered_many(&single);
5077 unsigned long netdev_fix_features(unsigned long features, const char *name)
5079 /* Fix illegal SG+CSUM combinations. */
5080 if ((features & NETIF_F_SG) &&
5081 !(features & NETIF_F_ALL_CSUM)) {
5083 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
5084 "checksum feature.\n", name);
5085 features &= ~NETIF_F_SG;
5088 /* TSO requires that SG is present as well. */
5089 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
5091 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
5092 "SG feature.\n", name);
5093 features &= ~NETIF_F_TSO;
5096 if (features & NETIF_F_UFO) {
5097 /* maybe split UFO into V4 and V6? */
5098 if (!((features & NETIF_F_GEN_CSUM) ||
5099 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5100 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5102 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5103 "since no checksum offload features.\n",
5105 features &= ~NETIF_F_UFO;
5108 if (!(features & NETIF_F_SG)) {
5110 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5111 "since no NETIF_F_SG feature.\n", name);
5112 features &= ~NETIF_F_UFO;
5118 EXPORT_SYMBOL(netdev_fix_features);
5121 * netif_stacked_transfer_operstate - transfer operstate
5122 * @rootdev: the root or lower level device to transfer state from
5123 * @dev: the device to transfer operstate to
5125 * Transfer operational state from root to device. This is normally
5126 * called when a stacking relationship exists between the root
5127 * device and the device(a leaf device).
5129 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5130 struct net_device *dev)
5132 if (rootdev->operstate == IF_OPER_DORMANT)
5133 netif_dormant_on(dev);
5135 netif_dormant_off(dev);
5137 if (netif_carrier_ok(rootdev)) {
5138 if (!netif_carrier_ok(dev))
5139 netif_carrier_on(dev);
5141 if (netif_carrier_ok(dev))
5142 netif_carrier_off(dev);
5145 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5148 static int netif_alloc_rx_queues(struct net_device *dev)
5150 unsigned int i, count = dev->num_rx_queues;
5151 struct netdev_rx_queue *rx;
5155 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5157 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5162 for (i = 0; i < count; i++)
5168 static void netdev_init_one_queue(struct net_device *dev,
5169 struct netdev_queue *queue, void *_unused)
5171 /* Initialize queue lock */
5172 spin_lock_init(&queue->_xmit_lock);
5173 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5174 queue->xmit_lock_owner = -1;
5175 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5179 static int netif_alloc_netdev_queues(struct net_device *dev)
5181 unsigned int count = dev->num_tx_queues;
5182 struct netdev_queue *tx;
5186 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5188 pr_err("netdev: Unable to allocate %u tx queues.\n",
5194 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5195 spin_lock_init(&dev->tx_global_lock);
5201 * register_netdevice - register a network device
5202 * @dev: device to register
5204 * Take a completed network device structure and add it to the kernel
5205 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5206 * chain. 0 is returned on success. A negative errno code is returned
5207 * on a failure to set up the device, or if the name is a duplicate.
5209 * Callers must hold the rtnl semaphore. You may want
5210 * register_netdev() instead of this.
5213 * The locking appears insufficient to guarantee two parallel registers
5214 * will not get the same name.
5217 int register_netdevice(struct net_device *dev)
5220 struct net *net = dev_net(dev);
5222 BUG_ON(dev_boot_phase);
5227 /* When net_device's are persistent, this will be fatal. */
5228 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5231 spin_lock_init(&dev->addr_list_lock);
5232 netdev_set_addr_lockdep_class(dev);
5236 /* Init, if this function is available */
5237 if (dev->netdev_ops->ndo_init) {
5238 ret = dev->netdev_ops->ndo_init(dev);
5246 ret = dev_get_valid_name(dev, dev->name, 0);
5250 dev->ifindex = dev_new_index(net);
5251 if (dev->iflink == -1)
5252 dev->iflink = dev->ifindex;
5254 /* Fix illegal checksum combinations */
5255 if ((dev->features & NETIF_F_HW_CSUM) &&
5256 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5257 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5259 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5262 if ((dev->features & NETIF_F_NO_CSUM) &&
5263 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5264 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5266 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5269 dev->features = netdev_fix_features(dev->features, dev->name);
5271 /* Enable software GSO if SG is supported. */
5272 if (dev->features & NETIF_F_SG)
5273 dev->features |= NETIF_F_GSO;
5275 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5276 * vlan_dev_init() will do the dev->features check, so these features
5277 * are enabled only if supported by underlying device.
5279 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5281 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5282 ret = notifier_to_errno(ret);
5286 ret = netdev_register_kobject(dev);
5289 dev->reg_state = NETREG_REGISTERED;
5292 * Default initial state at registry is that the
5293 * device is present.
5296 set_bit(__LINK_STATE_PRESENT, &dev->state);
5298 dev_init_scheduler(dev);
5300 list_netdevice(dev);
5302 /* Notify protocols, that a new device appeared. */
5303 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5304 ret = notifier_to_errno(ret);
5306 rollback_registered(dev);
5307 dev->reg_state = NETREG_UNREGISTERED;
5310 * Prevent userspace races by waiting until the network
5311 * device is fully setup before sending notifications.
5313 if (!dev->rtnl_link_ops ||
5314 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5315 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5321 if (dev->netdev_ops->ndo_uninit)
5322 dev->netdev_ops->ndo_uninit(dev);
5325 EXPORT_SYMBOL(register_netdevice);
5328 * init_dummy_netdev - init a dummy network device for NAPI
5329 * @dev: device to init
5331 * This takes a network device structure and initialize the minimum
5332 * amount of fields so it can be used to schedule NAPI polls without
5333 * registering a full blown interface. This is to be used by drivers
5334 * that need to tie several hardware interfaces to a single NAPI
5335 * poll scheduler due to HW limitations.
5337 int init_dummy_netdev(struct net_device *dev)
5339 /* Clear everything. Note we don't initialize spinlocks
5340 * are they aren't supposed to be taken by any of the
5341 * NAPI code and this dummy netdev is supposed to be
5342 * only ever used for NAPI polls
5344 memset(dev, 0, sizeof(struct net_device));
5346 /* make sure we BUG if trying to hit standard
5347 * register/unregister code path
5349 dev->reg_state = NETREG_DUMMY;
5351 /* NAPI wants this */
5352 INIT_LIST_HEAD(&dev->napi_list);
5354 /* a dummy interface is started by default */
5355 set_bit(__LINK_STATE_PRESENT, &dev->state);
5356 set_bit(__LINK_STATE_START, &dev->state);
5358 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5359 * because users of this 'device' dont need to change
5365 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5369 * register_netdev - register a network device
5370 * @dev: device to register
5372 * Take a completed network device structure and add it to the kernel
5373 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5374 * chain. 0 is returned on success. A negative errno code is returned
5375 * on a failure to set up the device, or if the name is a duplicate.
5377 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5378 * and expands the device name if you passed a format string to
5381 int register_netdev(struct net_device *dev)
5388 * If the name is a format string the caller wants us to do a
5391 if (strchr(dev->name, '%')) {
5392 err = dev_alloc_name(dev, dev->name);
5397 err = register_netdevice(dev);
5402 EXPORT_SYMBOL(register_netdev);
5404 int netdev_refcnt_read(const struct net_device *dev)
5408 for_each_possible_cpu(i)
5409 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5412 EXPORT_SYMBOL(netdev_refcnt_read);
5415 * netdev_wait_allrefs - wait until all references are gone.
5417 * This is called when unregistering network devices.
5419 * Any protocol or device that holds a reference should register
5420 * for netdevice notification, and cleanup and put back the
5421 * reference if they receive an UNREGISTER event.
5422 * We can get stuck here if buggy protocols don't correctly
5425 static void netdev_wait_allrefs(struct net_device *dev)
5427 unsigned long rebroadcast_time, warning_time;
5430 linkwatch_forget_dev(dev);
5432 rebroadcast_time = warning_time = jiffies;
5433 refcnt = netdev_refcnt_read(dev);
5435 while (refcnt != 0) {
5436 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5439 /* Rebroadcast unregister notification */
5440 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5441 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5442 * should have already handle it the first time */
5444 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5446 /* We must not have linkwatch events
5447 * pending on unregister. If this
5448 * happens, we simply run the queue
5449 * unscheduled, resulting in a noop
5452 linkwatch_run_queue();
5457 rebroadcast_time = jiffies;
5462 refcnt = netdev_refcnt_read(dev);
5464 if (time_after(jiffies, warning_time + 10 * HZ)) {
5465 printk(KERN_EMERG "unregister_netdevice: "
5466 "waiting for %s to become free. Usage "
5469 warning_time = jiffies;
5478 * register_netdevice(x1);
5479 * register_netdevice(x2);
5481 * unregister_netdevice(y1);
5482 * unregister_netdevice(y2);
5488 * We are invoked by rtnl_unlock().
5489 * This allows us to deal with problems:
5490 * 1) We can delete sysfs objects which invoke hotplug
5491 * without deadlocking with linkwatch via keventd.
5492 * 2) Since we run with the RTNL semaphore not held, we can sleep
5493 * safely in order to wait for the netdev refcnt to drop to zero.
5495 * We must not return until all unregister events added during
5496 * the interval the lock was held have been completed.
5498 void netdev_run_todo(void)
5500 struct list_head list;
5502 /* Snapshot list, allow later requests */
5503 list_replace_init(&net_todo_list, &list);
5507 while (!list_empty(&list)) {
5508 struct net_device *dev
5509 = list_first_entry(&list, struct net_device, todo_list);
5510 list_del(&dev->todo_list);
5512 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5513 printk(KERN_ERR "network todo '%s' but state %d\n",
5514 dev->name, dev->reg_state);
5519 dev->reg_state = NETREG_UNREGISTERED;
5521 on_each_cpu(flush_backlog, dev, 1);
5523 netdev_wait_allrefs(dev);
5526 BUG_ON(netdev_refcnt_read(dev));
5527 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5528 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5529 WARN_ON(dev->dn_ptr);
5531 if (dev->destructor)
5532 dev->destructor(dev);
5534 /* Free network device */
5535 kobject_put(&dev->dev.kobj);
5540 * dev_txq_stats_fold - fold tx_queues stats
5541 * @dev: device to get statistics from
5542 * @stats: struct rtnl_link_stats64 to hold results
5544 void dev_txq_stats_fold(const struct net_device *dev,
5545 struct rtnl_link_stats64 *stats)
5547 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5549 struct netdev_queue *txq;
5551 for (i = 0; i < dev->num_tx_queues; i++) {
5552 txq = netdev_get_tx_queue(dev, i);
5553 spin_lock_bh(&txq->_xmit_lock);
5554 tx_bytes += txq->tx_bytes;
5555 tx_packets += txq->tx_packets;
5556 tx_dropped += txq->tx_dropped;
5557 spin_unlock_bh(&txq->_xmit_lock);
5559 if (tx_bytes || tx_packets || tx_dropped) {
5560 stats->tx_bytes = tx_bytes;
5561 stats->tx_packets = tx_packets;
5562 stats->tx_dropped = tx_dropped;
5565 EXPORT_SYMBOL(dev_txq_stats_fold);
5567 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5568 * fields in the same order, with only the type differing.
5570 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5571 const struct net_device_stats *netdev_stats)
5573 #if BITS_PER_LONG == 64
5574 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5575 memcpy(stats64, netdev_stats, sizeof(*stats64));
5577 size_t i, n = sizeof(*stats64) / sizeof(u64);
5578 const unsigned long *src = (const unsigned long *)netdev_stats;
5579 u64 *dst = (u64 *)stats64;
5581 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5582 sizeof(*stats64) / sizeof(u64));
5583 for (i = 0; i < n; i++)
5589 * dev_get_stats - get network device statistics
5590 * @dev: device to get statistics from
5591 * @storage: place to store stats
5593 * Get network statistics from device. Return @storage.
5594 * The device driver may provide its own method by setting
5595 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5596 * otherwise the internal statistics structure is used.
5598 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5599 struct rtnl_link_stats64 *storage)
5601 const struct net_device_ops *ops = dev->netdev_ops;
5603 if (ops->ndo_get_stats64) {
5604 memset(storage, 0, sizeof(*storage));
5605 ops->ndo_get_stats64(dev, storage);
5606 } else if (ops->ndo_get_stats) {
5607 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5609 netdev_stats_to_stats64(storage, &dev->stats);
5610 dev_txq_stats_fold(dev, storage);
5612 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5615 EXPORT_SYMBOL(dev_get_stats);
5617 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5619 struct netdev_queue *queue = dev_ingress_queue(dev);
5621 #ifdef CONFIG_NET_CLS_ACT
5624 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5627 netdev_init_one_queue(dev, queue, NULL);
5628 queue->qdisc = &noop_qdisc;
5629 queue->qdisc_sleeping = &noop_qdisc;
5630 rcu_assign_pointer(dev->ingress_queue, queue);
5636 * alloc_netdev_mq - allocate network device
5637 * @sizeof_priv: size of private data to allocate space for
5638 * @name: device name format string
5639 * @setup: callback to initialize device
5640 * @queue_count: the number of subqueues to allocate
5642 * Allocates a struct net_device with private data area for driver use
5643 * and performs basic initialization. Also allocates subquue structs
5644 * for each queue on the device at the end of the netdevice.
5646 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5647 void (*setup)(struct net_device *), unsigned int queue_count)
5649 struct net_device *dev;
5651 struct net_device *p;
5653 BUG_ON(strlen(name) >= sizeof(dev->name));
5655 if (queue_count < 1) {
5656 pr_err("alloc_netdev: Unable to allocate device "
5657 "with zero queues.\n");
5661 alloc_size = sizeof(struct net_device);
5663 /* ensure 32-byte alignment of private area */
5664 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5665 alloc_size += sizeof_priv;
5667 /* ensure 32-byte alignment of whole construct */
5668 alloc_size += NETDEV_ALIGN - 1;
5670 p = kzalloc(alloc_size, GFP_KERNEL);
5672 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5676 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5677 dev->padded = (char *)dev - (char *)p;
5679 dev->pcpu_refcnt = alloc_percpu(int);
5680 if (!dev->pcpu_refcnt)
5683 if (dev_addr_init(dev))
5689 dev_net_set(dev, &init_net);
5691 dev->num_tx_queues = queue_count;
5692 dev->real_num_tx_queues = queue_count;
5693 if (netif_alloc_netdev_queues(dev))
5697 dev->num_rx_queues = queue_count;
5698 dev->real_num_rx_queues = queue_count;
5699 if (netif_alloc_rx_queues(dev))
5703 dev->gso_max_size = GSO_MAX_SIZE;
5705 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5706 dev->ethtool_ntuple_list.count = 0;
5707 INIT_LIST_HEAD(&dev->napi_list);
5708 INIT_LIST_HEAD(&dev->unreg_list);
5709 INIT_LIST_HEAD(&dev->link_watch_list);
5710 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5712 strcpy(dev->name, name);
5716 free_percpu(dev->pcpu_refcnt);
5726 EXPORT_SYMBOL(alloc_netdev_mq);
5729 * free_netdev - free network device
5732 * This function does the last stage of destroying an allocated device
5733 * interface. The reference to the device object is released.
5734 * If this is the last reference then it will be freed.
5736 void free_netdev(struct net_device *dev)
5738 struct napi_struct *p, *n;
5740 release_net(dev_net(dev));
5747 kfree(rcu_dereference_raw(dev->ingress_queue));
5749 /* Flush device addresses */
5750 dev_addr_flush(dev);
5752 /* Clear ethtool n-tuple list */
5753 ethtool_ntuple_flush(dev);
5755 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5758 free_percpu(dev->pcpu_refcnt);
5759 dev->pcpu_refcnt = NULL;
5761 /* Compatibility with error handling in drivers */
5762 if (dev->reg_state == NETREG_UNINITIALIZED) {
5763 kfree((char *)dev - dev->padded);
5767 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5768 dev->reg_state = NETREG_RELEASED;
5770 /* will free via device release */
5771 put_device(&dev->dev);
5773 EXPORT_SYMBOL(free_netdev);
5776 * synchronize_net - Synchronize with packet receive processing
5778 * Wait for packets currently being received to be done.
5779 * Does not block later packets from starting.
5781 void synchronize_net(void)
5786 EXPORT_SYMBOL(synchronize_net);
5789 * unregister_netdevice_queue - remove device from the kernel
5793 * This function shuts down a device interface and removes it
5794 * from the kernel tables.
5795 * If head not NULL, device is queued to be unregistered later.
5797 * Callers must hold the rtnl semaphore. You may want
5798 * unregister_netdev() instead of this.
5801 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5806 list_move_tail(&dev->unreg_list, head);
5808 rollback_registered(dev);
5809 /* Finish processing unregister after unlock */
5813 EXPORT_SYMBOL(unregister_netdevice_queue);
5816 * unregister_netdevice_many - unregister many devices
5817 * @head: list of devices
5819 void unregister_netdevice_many(struct list_head *head)
5821 struct net_device *dev;
5823 if (!list_empty(head)) {
5824 rollback_registered_many(head);
5825 list_for_each_entry(dev, head, unreg_list)
5829 EXPORT_SYMBOL(unregister_netdevice_many);
5832 * unregister_netdev - remove device from the kernel
5835 * This function shuts down a device interface and removes it
5836 * from the kernel tables.
5838 * This is just a wrapper for unregister_netdevice that takes
5839 * the rtnl semaphore. In general you want to use this and not
5840 * unregister_netdevice.
5842 void unregister_netdev(struct net_device *dev)
5845 unregister_netdevice(dev);
5848 EXPORT_SYMBOL(unregister_netdev);
5851 * dev_change_net_namespace - move device to different nethost namespace
5853 * @net: network namespace
5854 * @pat: If not NULL name pattern to try if the current device name
5855 * is already taken in the destination network namespace.
5857 * This function shuts down a device interface and moves it
5858 * to a new network namespace. On success 0 is returned, on
5859 * a failure a netagive errno code is returned.
5861 * Callers must hold the rtnl semaphore.
5864 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5870 /* Don't allow namespace local devices to be moved. */
5872 if (dev->features & NETIF_F_NETNS_LOCAL)
5875 /* Ensure the device has been registrered */
5877 if (dev->reg_state != NETREG_REGISTERED)
5880 /* Get out if there is nothing todo */
5882 if (net_eq(dev_net(dev), net))
5885 /* Pick the destination device name, and ensure
5886 * we can use it in the destination network namespace.
5889 if (__dev_get_by_name(net, dev->name)) {
5890 /* We get here if we can't use the current device name */
5893 if (dev_get_valid_name(dev, pat, 1))
5898 * And now a mini version of register_netdevice unregister_netdevice.
5901 /* If device is running close it first. */
5904 /* And unlink it from device chain */
5906 unlist_netdevice(dev);
5910 /* Shutdown queueing discipline. */
5913 /* Notify protocols, that we are about to destroy
5914 this device. They should clean all the things.
5916 Note that dev->reg_state stays at NETREG_REGISTERED.
5917 This is wanted because this way 8021q and macvlan know
5918 the device is just moving and can keep their slaves up.
5920 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5921 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5924 * Flush the unicast and multicast chains
5929 /* Actually switch the network namespace */
5930 dev_net_set(dev, net);
5932 /* If there is an ifindex conflict assign a new one */
5933 if (__dev_get_by_index(net, dev->ifindex)) {
5934 int iflink = (dev->iflink == dev->ifindex);
5935 dev->ifindex = dev_new_index(net);
5937 dev->iflink = dev->ifindex;
5940 /* Fixup kobjects */
5941 err = device_rename(&dev->dev, dev->name);
5944 /* Add the device back in the hashes */
5945 list_netdevice(dev);
5947 /* Notify protocols, that a new device appeared. */
5948 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5951 * Prevent userspace races by waiting until the network
5952 * device is fully setup before sending notifications.
5954 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5961 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5963 static int dev_cpu_callback(struct notifier_block *nfb,
5964 unsigned long action,
5967 struct sk_buff **list_skb;
5968 struct sk_buff *skb;
5969 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5970 struct softnet_data *sd, *oldsd;
5972 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5975 local_irq_disable();
5976 cpu = smp_processor_id();
5977 sd = &per_cpu(softnet_data, cpu);
5978 oldsd = &per_cpu(softnet_data, oldcpu);
5980 /* Find end of our completion_queue. */
5981 list_skb = &sd->completion_queue;
5983 list_skb = &(*list_skb)->next;
5984 /* Append completion queue from offline CPU. */
5985 *list_skb = oldsd->completion_queue;
5986 oldsd->completion_queue = NULL;
5988 /* Append output queue from offline CPU. */
5989 if (oldsd->output_queue) {
5990 *sd->output_queue_tailp = oldsd->output_queue;
5991 sd->output_queue_tailp = oldsd->output_queue_tailp;
5992 oldsd->output_queue = NULL;
5993 oldsd->output_queue_tailp = &oldsd->output_queue;
5996 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5999 /* Process offline CPU's input_pkt_queue */
6000 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6002 input_queue_head_incr(oldsd);
6004 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6006 input_queue_head_incr(oldsd);
6014 * netdev_increment_features - increment feature set by one
6015 * @all: current feature set
6016 * @one: new feature set
6017 * @mask: mask feature set
6019 * Computes a new feature set after adding a device with feature set
6020 * @one to the master device with current feature set @all. Will not
6021 * enable anything that is off in @mask. Returns the new feature set.
6023 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
6026 /* If device needs checksumming, downgrade to it. */
6027 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
6028 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
6029 else if (mask & NETIF_F_ALL_CSUM) {
6030 /* If one device supports v4/v6 checksumming, set for all. */
6031 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
6032 !(all & NETIF_F_GEN_CSUM)) {
6033 all &= ~NETIF_F_ALL_CSUM;
6034 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
6037 /* If one device supports hw checksumming, set for all. */
6038 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
6039 all &= ~NETIF_F_ALL_CSUM;
6040 all |= NETIF_F_HW_CSUM;
6044 one |= NETIF_F_ALL_CSUM;
6046 one |= all & NETIF_F_ONE_FOR_ALL;
6047 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
6048 all |= one & mask & NETIF_F_ONE_FOR_ALL;
6052 EXPORT_SYMBOL(netdev_increment_features);
6054 static struct hlist_head *netdev_create_hash(void)
6057 struct hlist_head *hash;
6059 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6061 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6062 INIT_HLIST_HEAD(&hash[i]);
6067 /* Initialize per network namespace state */
6068 static int __net_init netdev_init(struct net *net)
6070 INIT_LIST_HEAD(&net->dev_base_head);
6072 net->dev_name_head = netdev_create_hash();
6073 if (net->dev_name_head == NULL)
6076 net->dev_index_head = netdev_create_hash();
6077 if (net->dev_index_head == NULL)
6083 kfree(net->dev_name_head);
6089 * netdev_drivername - network driver for the device
6090 * @dev: network device
6091 * @buffer: buffer for resulting name
6092 * @len: size of buffer
6094 * Determine network driver for device.
6096 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6098 const struct device_driver *driver;
6099 const struct device *parent;
6101 if (len <= 0 || !buffer)
6105 parent = dev->dev.parent;
6110 driver = parent->driver;
6111 if (driver && driver->name)
6112 strlcpy(buffer, driver->name, len);
6116 static int __netdev_printk(const char *level, const struct net_device *dev,
6117 struct va_format *vaf)
6121 if (dev && dev->dev.parent)
6122 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6123 netdev_name(dev), vaf);
6125 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6127 r = printk("%s(NULL net_device): %pV", level, vaf);
6132 int netdev_printk(const char *level, const struct net_device *dev,
6133 const char *format, ...)
6135 struct va_format vaf;
6139 va_start(args, format);
6144 r = __netdev_printk(level, dev, &vaf);
6149 EXPORT_SYMBOL(netdev_printk);
6151 #define define_netdev_printk_level(func, level) \
6152 int func(const struct net_device *dev, const char *fmt, ...) \
6155 struct va_format vaf; \
6158 va_start(args, fmt); \
6163 r = __netdev_printk(level, dev, &vaf); \
6168 EXPORT_SYMBOL(func);
6170 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6171 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6172 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6173 define_netdev_printk_level(netdev_err, KERN_ERR);
6174 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6175 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6176 define_netdev_printk_level(netdev_info, KERN_INFO);
6178 static void __net_exit netdev_exit(struct net *net)
6180 kfree(net->dev_name_head);
6181 kfree(net->dev_index_head);
6184 static struct pernet_operations __net_initdata netdev_net_ops = {
6185 .init = netdev_init,
6186 .exit = netdev_exit,
6189 static void __net_exit default_device_exit(struct net *net)
6191 struct net_device *dev, *aux;
6193 * Push all migratable network devices back to the
6194 * initial network namespace
6197 for_each_netdev_safe(net, dev, aux) {
6199 char fb_name[IFNAMSIZ];
6201 /* Ignore unmoveable devices (i.e. loopback) */
6202 if (dev->features & NETIF_F_NETNS_LOCAL)
6205 /* Leave virtual devices for the generic cleanup */
6206 if (dev->rtnl_link_ops)
6209 /* Push remaing network devices to init_net */
6210 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6211 err = dev_change_net_namespace(dev, &init_net, fb_name);
6213 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6214 __func__, dev->name, err);
6221 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6223 /* At exit all network devices most be removed from a network
6224 * namespace. Do this in the reverse order of registeration.
6225 * Do this across as many network namespaces as possible to
6226 * improve batching efficiency.
6228 struct net_device *dev;
6230 LIST_HEAD(dev_kill_list);
6233 list_for_each_entry(net, net_list, exit_list) {
6234 for_each_netdev_reverse(net, dev) {
6235 if (dev->rtnl_link_ops)
6236 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6238 unregister_netdevice_queue(dev, &dev_kill_list);
6241 unregister_netdevice_many(&dev_kill_list);
6245 static struct pernet_operations __net_initdata default_device_ops = {
6246 .exit = default_device_exit,
6247 .exit_batch = default_device_exit_batch,
6251 * Initialize the DEV module. At boot time this walks the device list and
6252 * unhooks any devices that fail to initialise (normally hardware not
6253 * present) and leaves us with a valid list of present and active devices.
6258 * This is called single threaded during boot, so no need
6259 * to take the rtnl semaphore.
6261 static int __init net_dev_init(void)
6263 int i, rc = -ENOMEM;
6265 BUG_ON(!dev_boot_phase);
6267 if (dev_proc_init())
6270 if (netdev_kobject_init())
6273 INIT_LIST_HEAD(&ptype_all);
6274 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6275 INIT_LIST_HEAD(&ptype_base[i]);
6277 if (register_pernet_subsys(&netdev_net_ops))
6281 * Initialise the packet receive queues.
6284 for_each_possible_cpu(i) {
6285 struct softnet_data *sd = &per_cpu(softnet_data, i);
6287 memset(sd, 0, sizeof(*sd));
6288 skb_queue_head_init(&sd->input_pkt_queue);
6289 skb_queue_head_init(&sd->process_queue);
6290 sd->completion_queue = NULL;
6291 INIT_LIST_HEAD(&sd->poll_list);
6292 sd->output_queue = NULL;
6293 sd->output_queue_tailp = &sd->output_queue;
6295 sd->csd.func = rps_trigger_softirq;
6301 sd->backlog.poll = process_backlog;
6302 sd->backlog.weight = weight_p;
6303 sd->backlog.gro_list = NULL;
6304 sd->backlog.gro_count = 0;
6309 /* The loopback device is special if any other network devices
6310 * is present in a network namespace the loopback device must
6311 * be present. Since we now dynamically allocate and free the
6312 * loopback device ensure this invariant is maintained by
6313 * keeping the loopback device as the first device on the
6314 * list of network devices. Ensuring the loopback devices
6315 * is the first device that appears and the last network device
6318 if (register_pernet_device(&loopback_net_ops))
6321 if (register_pernet_device(&default_device_ops))
6324 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6325 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6327 hotcpu_notifier(dev_cpu_callback, 0);
6335 subsys_initcall(net_dev_init);
6337 static int __init initialize_hashrnd(void)
6339 get_random_bytes(&hashrnd, sizeof(hashrnd));
6343 late_initcall_sync(initialize_hashrnd);