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 static 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);
1596 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1597 * @dev: Network device
1598 * @txq: number of queues available
1600 * If real_num_tx_queues is changed the tc mappings may no longer be
1601 * valid. To resolve this verify the tc mapping remains valid and if
1602 * not NULL the mapping. With no priorities mapping to this
1603 * offset/count pair it will no longer be used. In the worst case TC0
1604 * is invalid nothing can be done so disable priority mappings. If is
1605 * expected that drivers will fix this mapping if they can before
1606 * calling netif_set_real_num_tx_queues.
1608 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1611 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1613 /* If TC0 is invalidated disable TC mapping */
1614 if (tc->offset + tc->count > txq) {
1615 pr_warning("Number of in use tx queues changed "
1616 "invalidating tc mappings. Priority "
1617 "traffic classification disabled!\n");
1622 /* Invalidated prio to tc mappings set to TC0 */
1623 for (i = 1; i < TC_BITMASK + 1; i++) {
1624 int q = netdev_get_prio_tc_map(dev, i);
1626 tc = &dev->tc_to_txq[q];
1627 if (tc->offset + tc->count > txq) {
1628 pr_warning("Number of in use tx queues "
1629 "changed. Priority %i to tc "
1630 "mapping %i is no longer valid "
1631 "setting map to 0\n",
1633 netdev_set_prio_tc_map(dev, i, 0);
1639 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1640 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1642 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1646 if (txq < 1 || txq > dev->num_tx_queues)
1649 if (dev->reg_state == NETREG_REGISTERED) {
1652 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1658 netif_setup_tc(dev, txq);
1660 if (txq < dev->real_num_tx_queues)
1661 qdisc_reset_all_tx_gt(dev, txq);
1664 dev->real_num_tx_queues = txq;
1667 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1671 * netif_set_real_num_rx_queues - set actual number of RX queues used
1672 * @dev: Network device
1673 * @rxq: Actual number of RX queues
1675 * This must be called either with the rtnl_lock held or before
1676 * registration of the net device. Returns 0 on success, or a
1677 * negative error code. If called before registration, it always
1680 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1684 if (rxq < 1 || rxq > dev->num_rx_queues)
1687 if (dev->reg_state == NETREG_REGISTERED) {
1690 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1696 dev->real_num_rx_queues = rxq;
1699 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1702 static inline void __netif_reschedule(struct Qdisc *q)
1704 struct softnet_data *sd;
1705 unsigned long flags;
1707 local_irq_save(flags);
1708 sd = &__get_cpu_var(softnet_data);
1709 q->next_sched = NULL;
1710 *sd->output_queue_tailp = q;
1711 sd->output_queue_tailp = &q->next_sched;
1712 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1713 local_irq_restore(flags);
1716 void __netif_schedule(struct Qdisc *q)
1718 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1719 __netif_reschedule(q);
1721 EXPORT_SYMBOL(__netif_schedule);
1723 void dev_kfree_skb_irq(struct sk_buff *skb)
1725 if (atomic_dec_and_test(&skb->users)) {
1726 struct softnet_data *sd;
1727 unsigned long flags;
1729 local_irq_save(flags);
1730 sd = &__get_cpu_var(softnet_data);
1731 skb->next = sd->completion_queue;
1732 sd->completion_queue = skb;
1733 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1734 local_irq_restore(flags);
1737 EXPORT_SYMBOL(dev_kfree_skb_irq);
1739 void dev_kfree_skb_any(struct sk_buff *skb)
1741 if (in_irq() || irqs_disabled())
1742 dev_kfree_skb_irq(skb);
1746 EXPORT_SYMBOL(dev_kfree_skb_any);
1750 * netif_device_detach - mark device as removed
1751 * @dev: network device
1753 * Mark device as removed from system and therefore no longer available.
1755 void netif_device_detach(struct net_device *dev)
1757 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1758 netif_running(dev)) {
1759 netif_tx_stop_all_queues(dev);
1762 EXPORT_SYMBOL(netif_device_detach);
1765 * netif_device_attach - mark device as attached
1766 * @dev: network device
1768 * Mark device as attached from system and restart if needed.
1770 void netif_device_attach(struct net_device *dev)
1772 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1773 netif_running(dev)) {
1774 netif_tx_wake_all_queues(dev);
1775 __netdev_watchdog_up(dev);
1778 EXPORT_SYMBOL(netif_device_attach);
1781 * skb_dev_set -- assign a new device to a buffer
1782 * @skb: buffer for the new device
1783 * @dev: network device
1785 * If an skb is owned by a device already, we have to reset
1786 * all data private to the namespace a device belongs to
1787 * before assigning it a new device.
1789 #ifdef CONFIG_NET_NS
1790 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1793 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1796 skb_init_secmark(skb);
1800 skb->ipvs_property = 0;
1801 #ifdef CONFIG_NET_SCHED
1807 EXPORT_SYMBOL(skb_set_dev);
1808 #endif /* CONFIG_NET_NS */
1811 * Invalidate hardware checksum when packet is to be mangled, and
1812 * complete checksum manually on outgoing path.
1814 int skb_checksum_help(struct sk_buff *skb)
1817 int ret = 0, offset;
1819 if (skb->ip_summed == CHECKSUM_COMPLETE)
1820 goto out_set_summed;
1822 if (unlikely(skb_shinfo(skb)->gso_size)) {
1823 /* Let GSO fix up the checksum. */
1824 goto out_set_summed;
1827 offset = skb_checksum_start_offset(skb);
1828 BUG_ON(offset >= skb_headlen(skb));
1829 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1831 offset += skb->csum_offset;
1832 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1834 if (skb_cloned(skb) &&
1835 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1836 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1841 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1843 skb->ip_summed = CHECKSUM_NONE;
1847 EXPORT_SYMBOL(skb_checksum_help);
1850 * skb_gso_segment - Perform segmentation on skb.
1851 * @skb: buffer to segment
1852 * @features: features for the output path (see dev->features)
1854 * This function segments the given skb and returns a list of segments.
1856 * It may return NULL if the skb requires no segmentation. This is
1857 * only possible when GSO is used for verifying header integrity.
1859 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1861 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1862 struct packet_type *ptype;
1863 __be16 type = skb->protocol;
1864 int vlan_depth = ETH_HLEN;
1867 while (type == htons(ETH_P_8021Q)) {
1868 struct vlan_hdr *vh;
1870 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1871 return ERR_PTR(-EINVAL);
1873 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1874 type = vh->h_vlan_encapsulated_proto;
1875 vlan_depth += VLAN_HLEN;
1878 skb_reset_mac_header(skb);
1879 skb->mac_len = skb->network_header - skb->mac_header;
1880 __skb_pull(skb, skb->mac_len);
1882 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1883 struct net_device *dev = skb->dev;
1884 struct ethtool_drvinfo info = {};
1886 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1887 dev->ethtool_ops->get_drvinfo(dev, &info);
1889 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1890 info.driver, dev ? dev->features : 0L,
1891 skb->sk ? skb->sk->sk_route_caps : 0L,
1892 skb->len, skb->data_len, skb->ip_summed);
1894 if (skb_header_cloned(skb) &&
1895 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1896 return ERR_PTR(err);
1900 list_for_each_entry_rcu(ptype,
1901 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1902 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1903 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1904 err = ptype->gso_send_check(skb);
1905 segs = ERR_PTR(err);
1906 if (err || skb_gso_ok(skb, features))
1908 __skb_push(skb, (skb->data -
1909 skb_network_header(skb)));
1911 segs = ptype->gso_segment(skb, features);
1917 __skb_push(skb, skb->data - skb_mac_header(skb));
1921 EXPORT_SYMBOL(skb_gso_segment);
1923 /* Take action when hardware reception checksum errors are detected. */
1925 void netdev_rx_csum_fault(struct net_device *dev)
1927 if (net_ratelimit()) {
1928 printk(KERN_ERR "%s: hw csum failure.\n",
1929 dev ? dev->name : "<unknown>");
1933 EXPORT_SYMBOL(netdev_rx_csum_fault);
1936 /* Actually, we should eliminate this check as soon as we know, that:
1937 * 1. IOMMU is present and allows to map all the memory.
1938 * 2. No high memory really exists on this machine.
1941 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1943 #ifdef CONFIG_HIGHMEM
1945 if (!(dev->features & NETIF_F_HIGHDMA)) {
1946 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1947 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1951 if (PCI_DMA_BUS_IS_PHYS) {
1952 struct device *pdev = dev->dev.parent;
1956 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1957 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1958 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1967 void (*destructor)(struct sk_buff *skb);
1970 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1972 static void dev_gso_skb_destructor(struct sk_buff *skb)
1974 struct dev_gso_cb *cb;
1977 struct sk_buff *nskb = skb->next;
1979 skb->next = nskb->next;
1982 } while (skb->next);
1984 cb = DEV_GSO_CB(skb);
1986 cb->destructor(skb);
1990 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1991 * @skb: buffer to segment
1992 * @features: device features as applicable to this skb
1994 * This function segments the given skb and stores the list of segments
1997 static int dev_gso_segment(struct sk_buff *skb, int features)
1999 struct sk_buff *segs;
2001 segs = skb_gso_segment(skb, features);
2003 /* Verifying header integrity only. */
2008 return PTR_ERR(segs);
2011 DEV_GSO_CB(skb)->destructor = skb->destructor;
2012 skb->destructor = dev_gso_skb_destructor;
2018 * Try to orphan skb early, right before transmission by the device.
2019 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2020 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2022 static inline void skb_orphan_try(struct sk_buff *skb)
2024 struct sock *sk = skb->sk;
2026 if (sk && !skb_shinfo(skb)->tx_flags) {
2027 /* skb_tx_hash() wont be able to get sk.
2028 * We copy sk_hash into skb->rxhash
2031 skb->rxhash = sk->sk_hash;
2036 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2038 return ((features & NETIF_F_GEN_CSUM) ||
2039 ((features & NETIF_F_V4_CSUM) &&
2040 protocol == htons(ETH_P_IP)) ||
2041 ((features & NETIF_F_V6_CSUM) &&
2042 protocol == htons(ETH_P_IPV6)) ||
2043 ((features & NETIF_F_FCOE_CRC) &&
2044 protocol == htons(ETH_P_FCOE)));
2047 static int harmonize_features(struct sk_buff *skb, __be16 protocol, int features)
2049 if (!can_checksum_protocol(protocol, features)) {
2050 features &= ~NETIF_F_ALL_CSUM;
2051 features &= ~NETIF_F_SG;
2052 } else if (illegal_highdma(skb->dev, skb)) {
2053 features &= ~NETIF_F_SG;
2059 int netif_skb_features(struct sk_buff *skb)
2061 __be16 protocol = skb->protocol;
2062 int features = skb->dev->features;
2064 if (protocol == htons(ETH_P_8021Q)) {
2065 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2066 protocol = veh->h_vlan_encapsulated_proto;
2067 } else if (!vlan_tx_tag_present(skb)) {
2068 return harmonize_features(skb, protocol, features);
2071 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2073 if (protocol != htons(ETH_P_8021Q)) {
2074 return harmonize_features(skb, protocol, features);
2076 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2077 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2078 return harmonize_features(skb, protocol, features);
2081 EXPORT_SYMBOL(netif_skb_features);
2084 * Returns true if either:
2085 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2086 * 2. skb is fragmented and the device does not support SG, or if
2087 * at least one of fragments is in highmem and device does not
2088 * support DMA from it.
2090 static inline int skb_needs_linearize(struct sk_buff *skb,
2093 return skb_is_nonlinear(skb) &&
2094 ((skb_has_frag_list(skb) &&
2095 !(features & NETIF_F_FRAGLIST)) ||
2096 (skb_shinfo(skb)->nr_frags &&
2097 !(features & NETIF_F_SG)));
2100 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2101 struct netdev_queue *txq)
2103 const struct net_device_ops *ops = dev->netdev_ops;
2104 int rc = NETDEV_TX_OK;
2106 if (likely(!skb->next)) {
2110 * If device doesnt need skb->dst, release it right now while
2111 * its hot in this cpu cache
2113 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2116 if (!list_empty(&ptype_all))
2117 dev_queue_xmit_nit(skb, dev);
2119 skb_orphan_try(skb);
2121 features = netif_skb_features(skb);
2123 if (vlan_tx_tag_present(skb) &&
2124 !(features & NETIF_F_HW_VLAN_TX)) {
2125 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2132 if (netif_needs_gso(skb, features)) {
2133 if (unlikely(dev_gso_segment(skb, features)))
2138 if (skb_needs_linearize(skb, features) &&
2139 __skb_linearize(skb))
2142 /* If packet is not checksummed and device does not
2143 * support checksumming for this protocol, complete
2144 * checksumming here.
2146 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2147 skb_set_transport_header(skb,
2148 skb_checksum_start_offset(skb));
2149 if (!(features & NETIF_F_ALL_CSUM) &&
2150 skb_checksum_help(skb))
2155 rc = ops->ndo_start_xmit(skb, dev);
2156 trace_net_dev_xmit(skb, rc);
2157 if (rc == NETDEV_TX_OK)
2158 txq_trans_update(txq);
2164 struct sk_buff *nskb = skb->next;
2166 skb->next = nskb->next;
2170 * If device doesnt need nskb->dst, release it right now while
2171 * its hot in this cpu cache
2173 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2176 rc = ops->ndo_start_xmit(nskb, dev);
2177 trace_net_dev_xmit(nskb, rc);
2178 if (unlikely(rc != NETDEV_TX_OK)) {
2179 if (rc & ~NETDEV_TX_MASK)
2180 goto out_kfree_gso_skb;
2181 nskb->next = skb->next;
2185 txq_trans_update(txq);
2186 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2187 return NETDEV_TX_BUSY;
2188 } while (skb->next);
2191 if (likely(skb->next == NULL))
2192 skb->destructor = DEV_GSO_CB(skb)->destructor;
2199 static u32 hashrnd __read_mostly;
2202 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2203 * to be used as a distribution range.
2205 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2206 unsigned int num_tx_queues)
2210 u16 qcount = num_tx_queues;
2212 if (skb_rx_queue_recorded(skb)) {
2213 hash = skb_get_rx_queue(skb);
2214 while (unlikely(hash >= num_tx_queues))
2215 hash -= num_tx_queues;
2220 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2221 qoffset = dev->tc_to_txq[tc].offset;
2222 qcount = dev->tc_to_txq[tc].count;
2225 if (skb->sk && skb->sk->sk_hash)
2226 hash = skb->sk->sk_hash;
2228 hash = (__force u16) skb->protocol ^ skb->rxhash;
2229 hash = jhash_1word(hash, hashrnd);
2231 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2233 EXPORT_SYMBOL(__skb_tx_hash);
2235 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2237 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2238 if (net_ratelimit()) {
2239 pr_warning("%s selects TX queue %d, but "
2240 "real number of TX queues is %d\n",
2241 dev->name, queue_index, dev->real_num_tx_queues);
2248 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2251 struct xps_dev_maps *dev_maps;
2252 struct xps_map *map;
2253 int queue_index = -1;
2256 dev_maps = rcu_dereference(dev->xps_maps);
2258 map = rcu_dereference(
2259 dev_maps->cpu_map[raw_smp_processor_id()]);
2262 queue_index = map->queues[0];
2265 if (skb->sk && skb->sk->sk_hash)
2266 hash = skb->sk->sk_hash;
2268 hash = (__force u16) skb->protocol ^
2270 hash = jhash_1word(hash, hashrnd);
2271 queue_index = map->queues[
2272 ((u64)hash * map->len) >> 32];
2274 if (unlikely(queue_index >= dev->real_num_tx_queues))
2286 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2287 struct sk_buff *skb)
2290 const struct net_device_ops *ops = dev->netdev_ops;
2292 if (dev->real_num_tx_queues == 1)
2294 else if (ops->ndo_select_queue) {
2295 queue_index = ops->ndo_select_queue(dev, skb);
2296 queue_index = dev_cap_txqueue(dev, queue_index);
2298 struct sock *sk = skb->sk;
2299 queue_index = sk_tx_queue_get(sk);
2301 if (queue_index < 0 || skb->ooo_okay ||
2302 queue_index >= dev->real_num_tx_queues) {
2303 int old_index = queue_index;
2305 queue_index = get_xps_queue(dev, skb);
2306 if (queue_index < 0)
2307 queue_index = skb_tx_hash(dev, skb);
2309 if (queue_index != old_index && sk) {
2310 struct dst_entry *dst =
2311 rcu_dereference_check(sk->sk_dst_cache, 1);
2313 if (dst && skb_dst(skb) == dst)
2314 sk_tx_queue_set(sk, queue_index);
2319 skb_set_queue_mapping(skb, queue_index);
2320 return netdev_get_tx_queue(dev, queue_index);
2323 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2324 struct net_device *dev,
2325 struct netdev_queue *txq)
2327 spinlock_t *root_lock = qdisc_lock(q);
2331 qdisc_skb_cb(skb)->pkt_len = skb->len;
2332 qdisc_calculate_pkt_len(skb, q);
2334 * Heuristic to force contended enqueues to serialize on a
2335 * separate lock before trying to get qdisc main lock.
2336 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2337 * and dequeue packets faster.
2339 contended = qdisc_is_running(q);
2340 if (unlikely(contended))
2341 spin_lock(&q->busylock);
2343 spin_lock(root_lock);
2344 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2347 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2348 qdisc_run_begin(q)) {
2350 * This is a work-conserving queue; there are no old skbs
2351 * waiting to be sent out; and the qdisc is not running -
2352 * xmit the skb directly.
2354 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2357 qdisc_bstats_update(q, skb);
2359 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2360 if (unlikely(contended)) {
2361 spin_unlock(&q->busylock);
2368 rc = NET_XMIT_SUCCESS;
2371 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2372 if (qdisc_run_begin(q)) {
2373 if (unlikely(contended)) {
2374 spin_unlock(&q->busylock);
2380 spin_unlock(root_lock);
2381 if (unlikely(contended))
2382 spin_unlock(&q->busylock);
2386 static DEFINE_PER_CPU(int, xmit_recursion);
2387 #define RECURSION_LIMIT 10
2390 * dev_queue_xmit - transmit a buffer
2391 * @skb: buffer to transmit
2393 * Queue a buffer for transmission to a network device. The caller must
2394 * have set the device and priority and built the buffer before calling
2395 * this function. The function can be called from an interrupt.
2397 * A negative errno code is returned on a failure. A success does not
2398 * guarantee the frame will be transmitted as it may be dropped due
2399 * to congestion or traffic shaping.
2401 * -----------------------------------------------------------------------------------
2402 * I notice this method can also return errors from the queue disciplines,
2403 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2406 * Regardless of the return value, the skb is consumed, so it is currently
2407 * difficult to retry a send to this method. (You can bump the ref count
2408 * before sending to hold a reference for retry if you are careful.)
2410 * When calling this method, interrupts MUST be enabled. This is because
2411 * the BH enable code must have IRQs enabled so that it will not deadlock.
2414 int dev_queue_xmit(struct sk_buff *skb)
2416 struct net_device *dev = skb->dev;
2417 struct netdev_queue *txq;
2421 /* Disable soft irqs for various locks below. Also
2422 * stops preemption for RCU.
2426 txq = dev_pick_tx(dev, skb);
2427 q = rcu_dereference_bh(txq->qdisc);
2429 #ifdef CONFIG_NET_CLS_ACT
2430 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2432 trace_net_dev_queue(skb);
2434 rc = __dev_xmit_skb(skb, q, dev, txq);
2438 /* The device has no queue. Common case for software devices:
2439 loopback, all the sorts of tunnels...
2441 Really, it is unlikely that netif_tx_lock protection is necessary
2442 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2444 However, it is possible, that they rely on protection
2447 Check this and shot the lock. It is not prone from deadlocks.
2448 Either shot noqueue qdisc, it is even simpler 8)
2450 if (dev->flags & IFF_UP) {
2451 int cpu = smp_processor_id(); /* ok because BHs are off */
2453 if (txq->xmit_lock_owner != cpu) {
2455 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2456 goto recursion_alert;
2458 HARD_TX_LOCK(dev, txq, cpu);
2460 if (!netif_tx_queue_stopped(txq)) {
2461 __this_cpu_inc(xmit_recursion);
2462 rc = dev_hard_start_xmit(skb, dev, txq);
2463 __this_cpu_dec(xmit_recursion);
2464 if (dev_xmit_complete(rc)) {
2465 HARD_TX_UNLOCK(dev, txq);
2469 HARD_TX_UNLOCK(dev, txq);
2470 if (net_ratelimit())
2471 printk(KERN_CRIT "Virtual device %s asks to "
2472 "queue packet!\n", dev->name);
2474 /* Recursion is detected! It is possible,
2478 if (net_ratelimit())
2479 printk(KERN_CRIT "Dead loop on virtual device "
2480 "%s, fix it urgently!\n", dev->name);
2485 rcu_read_unlock_bh();
2490 rcu_read_unlock_bh();
2493 EXPORT_SYMBOL(dev_queue_xmit);
2496 /*=======================================================================
2498 =======================================================================*/
2500 int netdev_max_backlog __read_mostly = 1000;
2501 int netdev_tstamp_prequeue __read_mostly = 1;
2502 int netdev_budget __read_mostly = 300;
2503 int weight_p __read_mostly = 64; /* old backlog weight */
2505 /* Called with irq disabled */
2506 static inline void ____napi_schedule(struct softnet_data *sd,
2507 struct napi_struct *napi)
2509 list_add_tail(&napi->poll_list, &sd->poll_list);
2510 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2514 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2515 * and src/dst port numbers. Returns a non-zero hash number on success
2518 __u32 __skb_get_rxhash(struct sk_buff *skb)
2520 int nhoff, hash = 0, poff;
2521 struct ipv6hdr *ip6;
2524 u32 addr1, addr2, ihl;
2530 nhoff = skb_network_offset(skb);
2532 switch (skb->protocol) {
2533 case __constant_htons(ETH_P_IP):
2534 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2537 ip = (struct iphdr *) (skb->data + nhoff);
2538 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2541 ip_proto = ip->protocol;
2542 addr1 = (__force u32) ip->saddr;
2543 addr2 = (__force u32) ip->daddr;
2546 case __constant_htons(ETH_P_IPV6):
2547 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2550 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2551 ip_proto = ip6->nexthdr;
2552 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2553 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2561 poff = proto_ports_offset(ip_proto);
2563 nhoff += ihl * 4 + poff;
2564 if (pskb_may_pull(skb, nhoff + 4)) {
2565 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2566 if (ports.v16[1] < ports.v16[0])
2567 swap(ports.v16[0], ports.v16[1]);
2571 /* get a consistent hash (same value on both flow directions) */
2575 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2582 EXPORT_SYMBOL(__skb_get_rxhash);
2586 /* One global table that all flow-based protocols share. */
2587 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2588 EXPORT_SYMBOL(rps_sock_flow_table);
2591 * get_rps_cpu is called from netif_receive_skb and returns the target
2592 * CPU from the RPS map of the receiving queue for a given skb.
2593 * rcu_read_lock must be held on entry.
2595 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2596 struct rps_dev_flow **rflowp)
2598 struct netdev_rx_queue *rxqueue;
2599 struct rps_map *map;
2600 struct rps_dev_flow_table *flow_table;
2601 struct rps_sock_flow_table *sock_flow_table;
2605 if (skb_rx_queue_recorded(skb)) {
2606 u16 index = skb_get_rx_queue(skb);
2607 if (unlikely(index >= dev->real_num_rx_queues)) {
2608 WARN_ONCE(dev->real_num_rx_queues > 1,
2609 "%s received packet on queue %u, but number "
2610 "of RX queues is %u\n",
2611 dev->name, index, dev->real_num_rx_queues);
2614 rxqueue = dev->_rx + index;
2618 map = rcu_dereference(rxqueue->rps_map);
2620 if (map->len == 1) {
2621 tcpu = map->cpus[0];
2622 if (cpu_online(tcpu))
2626 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2630 skb_reset_network_header(skb);
2631 if (!skb_get_rxhash(skb))
2634 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2635 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2636 if (flow_table && sock_flow_table) {
2638 struct rps_dev_flow *rflow;
2640 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2643 next_cpu = sock_flow_table->ents[skb->rxhash &
2644 sock_flow_table->mask];
2647 * If the desired CPU (where last recvmsg was done) is
2648 * different from current CPU (one in the rx-queue flow
2649 * table entry), switch if one of the following holds:
2650 * - Current CPU is unset (equal to RPS_NO_CPU).
2651 * - Current CPU is offline.
2652 * - The current CPU's queue tail has advanced beyond the
2653 * last packet that was enqueued using this table entry.
2654 * This guarantees that all previous packets for the flow
2655 * have been dequeued, thus preserving in order delivery.
2657 if (unlikely(tcpu != next_cpu) &&
2658 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2659 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2660 rflow->last_qtail)) >= 0)) {
2661 tcpu = rflow->cpu = next_cpu;
2662 if (tcpu != RPS_NO_CPU)
2663 rflow->last_qtail = per_cpu(softnet_data,
2664 tcpu).input_queue_head;
2666 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2674 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2676 if (cpu_online(tcpu)) {
2686 /* Called from hardirq (IPI) context */
2687 static void rps_trigger_softirq(void *data)
2689 struct softnet_data *sd = data;
2691 ____napi_schedule(sd, &sd->backlog);
2695 #endif /* CONFIG_RPS */
2698 * Check if this softnet_data structure is another cpu one
2699 * If yes, queue it to our IPI list and return 1
2702 static int rps_ipi_queued(struct softnet_data *sd)
2705 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2708 sd->rps_ipi_next = mysd->rps_ipi_list;
2709 mysd->rps_ipi_list = sd;
2711 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2714 #endif /* CONFIG_RPS */
2719 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2720 * queue (may be a remote CPU queue).
2722 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2723 unsigned int *qtail)
2725 struct softnet_data *sd;
2726 unsigned long flags;
2728 sd = &per_cpu(softnet_data, cpu);
2730 local_irq_save(flags);
2733 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2734 if (skb_queue_len(&sd->input_pkt_queue)) {
2736 __skb_queue_tail(&sd->input_pkt_queue, skb);
2737 input_queue_tail_incr_save(sd, qtail);
2739 local_irq_restore(flags);
2740 return NET_RX_SUCCESS;
2743 /* Schedule NAPI for backlog device
2744 * We can use non atomic operation since we own the queue lock
2746 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2747 if (!rps_ipi_queued(sd))
2748 ____napi_schedule(sd, &sd->backlog);
2756 local_irq_restore(flags);
2758 atomic_long_inc(&skb->dev->rx_dropped);
2764 * netif_rx - post buffer to the network code
2765 * @skb: buffer to post
2767 * This function receives a packet from a device driver and queues it for
2768 * the upper (protocol) levels to process. It always succeeds. The buffer
2769 * may be dropped during processing for congestion control or by the
2773 * NET_RX_SUCCESS (no congestion)
2774 * NET_RX_DROP (packet was dropped)
2778 int netif_rx(struct sk_buff *skb)
2782 /* if netpoll wants it, pretend we never saw it */
2783 if (netpoll_rx(skb))
2786 if (netdev_tstamp_prequeue)
2787 net_timestamp_check(skb);
2789 trace_netif_rx(skb);
2792 struct rps_dev_flow voidflow, *rflow = &voidflow;
2798 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2800 cpu = smp_processor_id();
2802 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2810 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2816 EXPORT_SYMBOL(netif_rx);
2818 int netif_rx_ni(struct sk_buff *skb)
2823 err = netif_rx(skb);
2824 if (local_softirq_pending())
2830 EXPORT_SYMBOL(netif_rx_ni);
2832 static void net_tx_action(struct softirq_action *h)
2834 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2836 if (sd->completion_queue) {
2837 struct sk_buff *clist;
2839 local_irq_disable();
2840 clist = sd->completion_queue;
2841 sd->completion_queue = NULL;
2845 struct sk_buff *skb = clist;
2846 clist = clist->next;
2848 WARN_ON(atomic_read(&skb->users));
2849 trace_kfree_skb(skb, net_tx_action);
2854 if (sd->output_queue) {
2857 local_irq_disable();
2858 head = sd->output_queue;
2859 sd->output_queue = NULL;
2860 sd->output_queue_tailp = &sd->output_queue;
2864 struct Qdisc *q = head;
2865 spinlock_t *root_lock;
2867 head = head->next_sched;
2869 root_lock = qdisc_lock(q);
2870 if (spin_trylock(root_lock)) {
2871 smp_mb__before_clear_bit();
2872 clear_bit(__QDISC_STATE_SCHED,
2875 spin_unlock(root_lock);
2877 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2879 __netif_reschedule(q);
2881 smp_mb__before_clear_bit();
2882 clear_bit(__QDISC_STATE_SCHED,
2890 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2891 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2892 /* This hook is defined here for ATM LANE */
2893 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2894 unsigned char *addr) __read_mostly;
2895 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2898 #ifdef CONFIG_NET_CLS_ACT
2899 /* TODO: Maybe we should just force sch_ingress to be compiled in
2900 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2901 * a compare and 2 stores extra right now if we dont have it on
2902 * but have CONFIG_NET_CLS_ACT
2903 * NOTE: This doesnt stop any functionality; if you dont have
2904 * the ingress scheduler, you just cant add policies on ingress.
2907 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2909 struct net_device *dev = skb->dev;
2910 u32 ttl = G_TC_RTTL(skb->tc_verd);
2911 int result = TC_ACT_OK;
2914 if (unlikely(MAX_RED_LOOP < ttl++)) {
2915 if (net_ratelimit())
2916 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2917 skb->skb_iif, dev->ifindex);
2921 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2922 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2925 if (q != &noop_qdisc) {
2926 spin_lock(qdisc_lock(q));
2927 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2928 result = qdisc_enqueue_root(skb, q);
2929 spin_unlock(qdisc_lock(q));
2935 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2936 struct packet_type **pt_prev,
2937 int *ret, struct net_device *orig_dev)
2939 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
2941 if (!rxq || rxq->qdisc == &noop_qdisc)
2945 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2949 switch (ing_filter(skb, rxq)) {
2963 * netdev_rx_handler_register - register receive handler
2964 * @dev: device to register a handler for
2965 * @rx_handler: receive handler to register
2966 * @rx_handler_data: data pointer that is used by rx handler
2968 * Register a receive hander for a device. This handler will then be
2969 * called from __netif_receive_skb. A negative errno code is returned
2972 * The caller must hold the rtnl_mutex.
2974 int netdev_rx_handler_register(struct net_device *dev,
2975 rx_handler_func_t *rx_handler,
2976 void *rx_handler_data)
2980 if (dev->rx_handler)
2983 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2984 rcu_assign_pointer(dev->rx_handler, rx_handler);
2988 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2991 * netdev_rx_handler_unregister - unregister receive handler
2992 * @dev: device to unregister a handler from
2994 * Unregister a receive hander from a device.
2996 * The caller must hold the rtnl_mutex.
2998 void netdev_rx_handler_unregister(struct net_device *dev)
3002 rcu_assign_pointer(dev->rx_handler, NULL);
3003 rcu_assign_pointer(dev->rx_handler_data, NULL);
3005 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3007 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
3008 struct net_device *master)
3010 if (skb->pkt_type == PACKET_HOST) {
3011 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
3013 memcpy(dest, master->dev_addr, ETH_ALEN);
3017 /* On bonding slaves other than the currently active slave, suppress
3018 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
3019 * ARP on active-backup slaves with arp_validate enabled.
3021 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
3023 struct net_device *dev = skb->dev;
3025 if (master->priv_flags & IFF_MASTER_ARPMON)
3026 dev->last_rx = jiffies;
3028 if ((master->priv_flags & IFF_MASTER_ALB) &&
3029 (master->priv_flags & IFF_BRIDGE_PORT)) {
3030 /* Do address unmangle. The local destination address
3031 * will be always the one master has. Provides the right
3032 * functionality in a bridge.
3034 skb_bond_set_mac_by_master(skb, master);
3037 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
3038 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
3039 skb->protocol == __cpu_to_be16(ETH_P_ARP))
3042 if (master->priv_flags & IFF_MASTER_ALB) {
3043 if (skb->pkt_type != PACKET_BROADCAST &&
3044 skb->pkt_type != PACKET_MULTICAST)
3047 if (master->priv_flags & IFF_MASTER_8023AD &&
3048 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
3055 EXPORT_SYMBOL(__skb_bond_should_drop);
3057 static int __netif_receive_skb(struct sk_buff *skb)
3059 struct packet_type *ptype, *pt_prev;
3060 rx_handler_func_t *rx_handler;
3061 struct net_device *orig_dev;
3062 struct net_device *master;
3063 struct net_device *null_or_orig;
3064 struct net_device *orig_or_bond;
3065 int ret = NET_RX_DROP;
3068 if (!netdev_tstamp_prequeue)
3069 net_timestamp_check(skb);
3071 trace_netif_receive_skb(skb);
3073 /* if we've gotten here through NAPI, check netpoll */
3074 if (netpoll_receive_skb(skb))
3078 skb->skb_iif = skb->dev->ifindex;
3081 * bonding note: skbs received on inactive slaves should only
3082 * be delivered to pkt handlers that are exact matches. Also
3083 * the deliver_no_wcard flag will be set. If packet handlers
3084 * are sensitive to duplicate packets these skbs will need to
3085 * be dropped at the handler.
3087 null_or_orig = NULL;
3088 orig_dev = skb->dev;
3089 master = ACCESS_ONCE(orig_dev->master);
3090 if (skb->deliver_no_wcard)
3091 null_or_orig = orig_dev;
3093 if (skb_bond_should_drop(skb, master)) {
3094 skb->deliver_no_wcard = 1;
3095 null_or_orig = orig_dev; /* deliver only exact match */
3100 __this_cpu_inc(softnet_data.processed);
3101 skb_reset_network_header(skb);
3102 skb_reset_transport_header(skb);
3103 skb->mac_len = skb->network_header - skb->mac_header;
3109 #ifdef CONFIG_NET_CLS_ACT
3110 if (skb->tc_verd & TC_NCLS) {
3111 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3116 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3117 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
3118 ptype->dev == orig_dev) {
3120 ret = deliver_skb(skb, pt_prev, orig_dev);
3125 #ifdef CONFIG_NET_CLS_ACT
3126 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3132 /* Handle special case of bridge or macvlan */
3133 rx_handler = rcu_dereference(skb->dev->rx_handler);
3136 ret = deliver_skb(skb, pt_prev, orig_dev);
3139 skb = rx_handler(skb);
3144 if (vlan_tx_tag_present(skb)) {
3146 ret = deliver_skb(skb, pt_prev, orig_dev);
3149 if (vlan_hwaccel_do_receive(&skb)) {
3150 ret = __netif_receive_skb(skb);
3152 } else if (unlikely(!skb))
3157 * Make sure frames received on VLAN interfaces stacked on
3158 * bonding interfaces still make their way to any base bonding
3159 * device that may have registered for a specific ptype. The
3160 * handler may have to adjust skb->dev and orig_dev.
3162 orig_or_bond = orig_dev;
3163 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
3164 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
3165 orig_or_bond = vlan_dev_real_dev(skb->dev);
3168 type = skb->protocol;
3169 list_for_each_entry_rcu(ptype,
3170 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3171 if (ptype->type == type && (ptype->dev == null_or_orig ||
3172 ptype->dev == skb->dev || ptype->dev == orig_dev ||
3173 ptype->dev == orig_or_bond)) {
3175 ret = deliver_skb(skb, pt_prev, orig_dev);
3181 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3183 atomic_long_inc(&skb->dev->rx_dropped);
3185 /* Jamal, now you will not able to escape explaining
3186 * me how you were going to use this. :-)
3197 * netif_receive_skb - process receive buffer from network
3198 * @skb: buffer to process
3200 * netif_receive_skb() is the main receive data processing function.
3201 * It always succeeds. The buffer may be dropped during processing
3202 * for congestion control or by the protocol layers.
3204 * This function may only be called from softirq context and interrupts
3205 * should be enabled.
3207 * Return values (usually ignored):
3208 * NET_RX_SUCCESS: no congestion
3209 * NET_RX_DROP: packet was dropped
3211 int netif_receive_skb(struct sk_buff *skb)
3213 if (netdev_tstamp_prequeue)
3214 net_timestamp_check(skb);
3216 if (skb_defer_rx_timestamp(skb))
3217 return NET_RX_SUCCESS;
3221 struct rps_dev_flow voidflow, *rflow = &voidflow;
3226 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3229 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3233 ret = __netif_receive_skb(skb);
3239 return __netif_receive_skb(skb);
3242 EXPORT_SYMBOL(netif_receive_skb);
3244 /* Network device is going away, flush any packets still pending
3245 * Called with irqs disabled.
3247 static void flush_backlog(void *arg)
3249 struct net_device *dev = arg;
3250 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3251 struct sk_buff *skb, *tmp;
3254 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3255 if (skb->dev == dev) {
3256 __skb_unlink(skb, &sd->input_pkt_queue);
3258 input_queue_head_incr(sd);
3263 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3264 if (skb->dev == dev) {
3265 __skb_unlink(skb, &sd->process_queue);
3267 input_queue_head_incr(sd);
3272 static int napi_gro_complete(struct sk_buff *skb)
3274 struct packet_type *ptype;
3275 __be16 type = skb->protocol;
3276 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3279 if (NAPI_GRO_CB(skb)->count == 1) {
3280 skb_shinfo(skb)->gso_size = 0;
3285 list_for_each_entry_rcu(ptype, head, list) {
3286 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3289 err = ptype->gro_complete(skb);
3295 WARN_ON(&ptype->list == head);
3297 return NET_RX_SUCCESS;
3301 return netif_receive_skb(skb);
3304 inline void napi_gro_flush(struct napi_struct *napi)
3306 struct sk_buff *skb, *next;
3308 for (skb = napi->gro_list; skb; skb = next) {
3311 napi_gro_complete(skb);
3314 napi->gro_count = 0;
3315 napi->gro_list = NULL;
3317 EXPORT_SYMBOL(napi_gro_flush);
3319 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3321 struct sk_buff **pp = NULL;
3322 struct packet_type *ptype;
3323 __be16 type = skb->protocol;
3324 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3327 enum gro_result ret;
3329 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3332 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3336 list_for_each_entry_rcu(ptype, head, list) {
3337 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3340 skb_set_network_header(skb, skb_gro_offset(skb));
3341 mac_len = skb->network_header - skb->mac_header;
3342 skb->mac_len = mac_len;
3343 NAPI_GRO_CB(skb)->same_flow = 0;
3344 NAPI_GRO_CB(skb)->flush = 0;
3345 NAPI_GRO_CB(skb)->free = 0;
3347 pp = ptype->gro_receive(&napi->gro_list, skb);
3352 if (&ptype->list == head)
3355 same_flow = NAPI_GRO_CB(skb)->same_flow;
3356 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3359 struct sk_buff *nskb = *pp;
3363 napi_gro_complete(nskb);
3370 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3374 NAPI_GRO_CB(skb)->count = 1;
3375 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3376 skb->next = napi->gro_list;
3377 napi->gro_list = skb;
3381 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3382 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3384 BUG_ON(skb->end - skb->tail < grow);
3386 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3389 skb->data_len -= grow;
3391 skb_shinfo(skb)->frags[0].page_offset += grow;
3392 skb_shinfo(skb)->frags[0].size -= grow;
3394 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3395 put_page(skb_shinfo(skb)->frags[0].page);
3396 memmove(skb_shinfo(skb)->frags,
3397 skb_shinfo(skb)->frags + 1,
3398 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3409 EXPORT_SYMBOL(dev_gro_receive);
3411 static inline gro_result_t
3412 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3416 for (p = napi->gro_list; p; p = p->next) {
3417 unsigned long diffs;
3419 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3420 diffs |= p->vlan_tci ^ skb->vlan_tci;
3421 diffs |= compare_ether_header(skb_mac_header(p),
3422 skb_gro_mac_header(skb));
3423 NAPI_GRO_CB(p)->same_flow = !diffs;
3424 NAPI_GRO_CB(p)->flush = 0;
3427 return dev_gro_receive(napi, skb);
3430 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3434 if (netif_receive_skb(skb))
3439 case GRO_MERGED_FREE:
3450 EXPORT_SYMBOL(napi_skb_finish);
3452 void skb_gro_reset_offset(struct sk_buff *skb)
3454 NAPI_GRO_CB(skb)->data_offset = 0;
3455 NAPI_GRO_CB(skb)->frag0 = NULL;
3456 NAPI_GRO_CB(skb)->frag0_len = 0;
3458 if (skb->mac_header == skb->tail &&
3459 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3460 NAPI_GRO_CB(skb)->frag0 =
3461 page_address(skb_shinfo(skb)->frags[0].page) +
3462 skb_shinfo(skb)->frags[0].page_offset;
3463 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3466 EXPORT_SYMBOL(skb_gro_reset_offset);
3468 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3470 skb_gro_reset_offset(skb);
3472 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3474 EXPORT_SYMBOL(napi_gro_receive);
3476 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3478 __skb_pull(skb, skb_headlen(skb));
3479 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3485 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3487 struct sk_buff *skb = napi->skb;
3490 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3496 EXPORT_SYMBOL(napi_get_frags);
3498 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3504 skb->protocol = eth_type_trans(skb, skb->dev);
3506 if (ret == GRO_HELD)
3507 skb_gro_pull(skb, -ETH_HLEN);
3508 else if (netif_receive_skb(skb))
3513 case GRO_MERGED_FREE:
3514 napi_reuse_skb(napi, skb);
3523 EXPORT_SYMBOL(napi_frags_finish);
3525 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3527 struct sk_buff *skb = napi->skb;
3534 skb_reset_mac_header(skb);
3535 skb_gro_reset_offset(skb);
3537 off = skb_gro_offset(skb);
3538 hlen = off + sizeof(*eth);
3539 eth = skb_gro_header_fast(skb, off);
3540 if (skb_gro_header_hard(skb, hlen)) {
3541 eth = skb_gro_header_slow(skb, hlen, off);
3542 if (unlikely(!eth)) {
3543 napi_reuse_skb(napi, skb);
3549 skb_gro_pull(skb, sizeof(*eth));
3552 * This works because the only protocols we care about don't require
3553 * special handling. We'll fix it up properly at the end.
3555 skb->protocol = eth->h_proto;
3560 EXPORT_SYMBOL(napi_frags_skb);
3562 gro_result_t napi_gro_frags(struct napi_struct *napi)
3564 struct sk_buff *skb = napi_frags_skb(napi);
3569 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3571 EXPORT_SYMBOL(napi_gro_frags);
3574 * net_rps_action sends any pending IPI's for rps.
3575 * Note: called with local irq disabled, but exits with local irq enabled.
3577 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3580 struct softnet_data *remsd = sd->rps_ipi_list;
3583 sd->rps_ipi_list = NULL;
3587 /* Send pending IPI's to kick RPS processing on remote cpus. */
3589 struct softnet_data *next = remsd->rps_ipi_next;
3591 if (cpu_online(remsd->cpu))
3592 __smp_call_function_single(remsd->cpu,
3601 static int process_backlog(struct napi_struct *napi, int quota)
3604 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3607 /* Check if we have pending ipi, its better to send them now,
3608 * not waiting net_rx_action() end.
3610 if (sd->rps_ipi_list) {
3611 local_irq_disable();
3612 net_rps_action_and_irq_enable(sd);
3615 napi->weight = weight_p;
3616 local_irq_disable();
3617 while (work < quota) {
3618 struct sk_buff *skb;
3621 while ((skb = __skb_dequeue(&sd->process_queue))) {
3623 __netif_receive_skb(skb);
3624 local_irq_disable();
3625 input_queue_head_incr(sd);
3626 if (++work >= quota) {
3633 qlen = skb_queue_len(&sd->input_pkt_queue);
3635 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3636 &sd->process_queue);
3638 if (qlen < quota - work) {
3640 * Inline a custom version of __napi_complete().
3641 * only current cpu owns and manipulates this napi,
3642 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3643 * we can use a plain write instead of clear_bit(),
3644 * and we dont need an smp_mb() memory barrier.
3646 list_del(&napi->poll_list);
3649 quota = work + qlen;
3659 * __napi_schedule - schedule for receive
3660 * @n: entry to schedule
3662 * The entry's receive function will be scheduled to run
3664 void __napi_schedule(struct napi_struct *n)
3666 unsigned long flags;
3668 local_irq_save(flags);
3669 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3670 local_irq_restore(flags);
3672 EXPORT_SYMBOL(__napi_schedule);
3674 void __napi_complete(struct napi_struct *n)
3676 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3677 BUG_ON(n->gro_list);
3679 list_del(&n->poll_list);
3680 smp_mb__before_clear_bit();
3681 clear_bit(NAPI_STATE_SCHED, &n->state);
3683 EXPORT_SYMBOL(__napi_complete);
3685 void napi_complete(struct napi_struct *n)
3687 unsigned long flags;
3690 * don't let napi dequeue from the cpu poll list
3691 * just in case its running on a different cpu
3693 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3697 local_irq_save(flags);
3699 local_irq_restore(flags);
3701 EXPORT_SYMBOL(napi_complete);
3703 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3704 int (*poll)(struct napi_struct *, int), int weight)
3706 INIT_LIST_HEAD(&napi->poll_list);
3707 napi->gro_count = 0;
3708 napi->gro_list = NULL;
3711 napi->weight = weight;
3712 list_add(&napi->dev_list, &dev->napi_list);
3714 #ifdef CONFIG_NETPOLL
3715 spin_lock_init(&napi->poll_lock);
3716 napi->poll_owner = -1;
3718 set_bit(NAPI_STATE_SCHED, &napi->state);
3720 EXPORT_SYMBOL(netif_napi_add);
3722 void netif_napi_del(struct napi_struct *napi)
3724 struct sk_buff *skb, *next;
3726 list_del_init(&napi->dev_list);
3727 napi_free_frags(napi);
3729 for (skb = napi->gro_list; skb; skb = next) {
3735 napi->gro_list = NULL;
3736 napi->gro_count = 0;
3738 EXPORT_SYMBOL(netif_napi_del);
3740 static void net_rx_action(struct softirq_action *h)
3742 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3743 unsigned long time_limit = jiffies + 2;
3744 int budget = netdev_budget;
3747 local_irq_disable();
3749 while (!list_empty(&sd->poll_list)) {
3750 struct napi_struct *n;
3753 /* If softirq window is exhuasted then punt.
3754 * Allow this to run for 2 jiffies since which will allow
3755 * an average latency of 1.5/HZ.
3757 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3762 /* Even though interrupts have been re-enabled, this
3763 * access is safe because interrupts can only add new
3764 * entries to the tail of this list, and only ->poll()
3765 * calls can remove this head entry from the list.
3767 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3769 have = netpoll_poll_lock(n);
3773 /* This NAPI_STATE_SCHED test is for avoiding a race
3774 * with netpoll's poll_napi(). Only the entity which
3775 * obtains the lock and sees NAPI_STATE_SCHED set will
3776 * actually make the ->poll() call. Therefore we avoid
3777 * accidently calling ->poll() when NAPI is not scheduled.
3780 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3781 work = n->poll(n, weight);
3785 WARN_ON_ONCE(work > weight);
3789 local_irq_disable();
3791 /* Drivers must not modify the NAPI state if they
3792 * consume the entire weight. In such cases this code
3793 * still "owns" the NAPI instance and therefore can
3794 * move the instance around on the list at-will.
3796 if (unlikely(work == weight)) {
3797 if (unlikely(napi_disable_pending(n))) {
3800 local_irq_disable();
3802 list_move_tail(&n->poll_list, &sd->poll_list);
3805 netpoll_poll_unlock(have);
3808 net_rps_action_and_irq_enable(sd);
3810 #ifdef CONFIG_NET_DMA
3812 * There may not be any more sk_buffs coming right now, so push
3813 * any pending DMA copies to hardware
3815 dma_issue_pending_all();
3822 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3826 static gifconf_func_t *gifconf_list[NPROTO];
3829 * register_gifconf - register a SIOCGIF handler
3830 * @family: Address family
3831 * @gifconf: Function handler
3833 * Register protocol dependent address dumping routines. The handler
3834 * that is passed must not be freed or reused until it has been replaced
3835 * by another handler.
3837 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3839 if (family >= NPROTO)
3841 gifconf_list[family] = gifconf;
3844 EXPORT_SYMBOL(register_gifconf);
3848 * Map an interface index to its name (SIOCGIFNAME)
3852 * We need this ioctl for efficient implementation of the
3853 * if_indextoname() function required by the IPv6 API. Without
3854 * it, we would have to search all the interfaces to find a
3858 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3860 struct net_device *dev;
3864 * Fetch the caller's info block.
3867 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3871 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3877 strcpy(ifr.ifr_name, dev->name);
3880 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3886 * Perform a SIOCGIFCONF call. This structure will change
3887 * size eventually, and there is nothing I can do about it.
3888 * Thus we will need a 'compatibility mode'.
3891 static int dev_ifconf(struct net *net, char __user *arg)
3894 struct net_device *dev;
3901 * Fetch the caller's info block.
3904 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3911 * Loop over the interfaces, and write an info block for each.
3915 for_each_netdev(net, dev) {
3916 for (i = 0; i < NPROTO; i++) {
3917 if (gifconf_list[i]) {
3920 done = gifconf_list[i](dev, NULL, 0);
3922 done = gifconf_list[i](dev, pos + total,
3932 * All done. Write the updated control block back to the caller.
3934 ifc.ifc_len = total;
3937 * Both BSD and Solaris return 0 here, so we do too.
3939 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3942 #ifdef CONFIG_PROC_FS
3944 * This is invoked by the /proc filesystem handler to display a device
3947 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3950 struct net *net = seq_file_net(seq);
3952 struct net_device *dev;
3956 return SEQ_START_TOKEN;
3959 for_each_netdev_rcu(net, dev)
3966 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3968 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3969 first_net_device(seq_file_net(seq)) :
3970 next_net_device((struct net_device *)v);
3973 return rcu_dereference(dev);
3976 void dev_seq_stop(struct seq_file *seq, void *v)
3982 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3984 struct rtnl_link_stats64 temp;
3985 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3987 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3988 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3989 dev->name, stats->rx_bytes, stats->rx_packets,
3991 stats->rx_dropped + stats->rx_missed_errors,
3992 stats->rx_fifo_errors,
3993 stats->rx_length_errors + stats->rx_over_errors +
3994 stats->rx_crc_errors + stats->rx_frame_errors,
3995 stats->rx_compressed, stats->multicast,
3996 stats->tx_bytes, stats->tx_packets,
3997 stats->tx_errors, stats->tx_dropped,
3998 stats->tx_fifo_errors, stats->collisions,
3999 stats->tx_carrier_errors +
4000 stats->tx_aborted_errors +
4001 stats->tx_window_errors +
4002 stats->tx_heartbeat_errors,
4003 stats->tx_compressed);
4007 * Called from the PROCfs module. This now uses the new arbitrary sized
4008 * /proc/net interface to create /proc/net/dev
4010 static int dev_seq_show(struct seq_file *seq, void *v)
4012 if (v == SEQ_START_TOKEN)
4013 seq_puts(seq, "Inter-| Receive "
4015 " face |bytes packets errs drop fifo frame "
4016 "compressed multicast|bytes packets errs "
4017 "drop fifo colls carrier compressed\n");
4019 dev_seq_printf_stats(seq, v);
4023 static struct softnet_data *softnet_get_online(loff_t *pos)
4025 struct softnet_data *sd = NULL;
4027 while (*pos < nr_cpu_ids)
4028 if (cpu_online(*pos)) {
4029 sd = &per_cpu(softnet_data, *pos);
4036 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4038 return softnet_get_online(pos);
4041 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4044 return softnet_get_online(pos);
4047 static void softnet_seq_stop(struct seq_file *seq, void *v)
4051 static int softnet_seq_show(struct seq_file *seq, void *v)
4053 struct softnet_data *sd = v;
4055 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4056 sd->processed, sd->dropped, sd->time_squeeze, 0,
4057 0, 0, 0, 0, /* was fastroute */
4058 sd->cpu_collision, sd->received_rps);
4062 static const struct seq_operations dev_seq_ops = {
4063 .start = dev_seq_start,
4064 .next = dev_seq_next,
4065 .stop = dev_seq_stop,
4066 .show = dev_seq_show,
4069 static int dev_seq_open(struct inode *inode, struct file *file)
4071 return seq_open_net(inode, file, &dev_seq_ops,
4072 sizeof(struct seq_net_private));
4075 static const struct file_operations dev_seq_fops = {
4076 .owner = THIS_MODULE,
4077 .open = dev_seq_open,
4079 .llseek = seq_lseek,
4080 .release = seq_release_net,
4083 static const struct seq_operations softnet_seq_ops = {
4084 .start = softnet_seq_start,
4085 .next = softnet_seq_next,
4086 .stop = softnet_seq_stop,
4087 .show = softnet_seq_show,
4090 static int softnet_seq_open(struct inode *inode, struct file *file)
4092 return seq_open(file, &softnet_seq_ops);
4095 static const struct file_operations softnet_seq_fops = {
4096 .owner = THIS_MODULE,
4097 .open = softnet_seq_open,
4099 .llseek = seq_lseek,
4100 .release = seq_release,
4103 static void *ptype_get_idx(loff_t pos)
4105 struct packet_type *pt = NULL;
4109 list_for_each_entry_rcu(pt, &ptype_all, list) {
4115 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4116 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4125 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4129 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4132 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4134 struct packet_type *pt;
4135 struct list_head *nxt;
4139 if (v == SEQ_START_TOKEN)
4140 return ptype_get_idx(0);
4143 nxt = pt->list.next;
4144 if (pt->type == htons(ETH_P_ALL)) {
4145 if (nxt != &ptype_all)
4148 nxt = ptype_base[0].next;
4150 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4152 while (nxt == &ptype_base[hash]) {
4153 if (++hash >= PTYPE_HASH_SIZE)
4155 nxt = ptype_base[hash].next;
4158 return list_entry(nxt, struct packet_type, list);
4161 static void ptype_seq_stop(struct seq_file *seq, void *v)
4167 static int ptype_seq_show(struct seq_file *seq, void *v)
4169 struct packet_type *pt = v;
4171 if (v == SEQ_START_TOKEN)
4172 seq_puts(seq, "Type Device Function\n");
4173 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4174 if (pt->type == htons(ETH_P_ALL))
4175 seq_puts(seq, "ALL ");
4177 seq_printf(seq, "%04x", ntohs(pt->type));
4179 seq_printf(seq, " %-8s %pF\n",
4180 pt->dev ? pt->dev->name : "", pt->func);
4186 static const struct seq_operations ptype_seq_ops = {
4187 .start = ptype_seq_start,
4188 .next = ptype_seq_next,
4189 .stop = ptype_seq_stop,
4190 .show = ptype_seq_show,
4193 static int ptype_seq_open(struct inode *inode, struct file *file)
4195 return seq_open_net(inode, file, &ptype_seq_ops,
4196 sizeof(struct seq_net_private));
4199 static const struct file_operations ptype_seq_fops = {
4200 .owner = THIS_MODULE,
4201 .open = ptype_seq_open,
4203 .llseek = seq_lseek,
4204 .release = seq_release_net,
4208 static int __net_init dev_proc_net_init(struct net *net)
4212 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4214 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4216 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4219 if (wext_proc_init(net))
4225 proc_net_remove(net, "ptype");
4227 proc_net_remove(net, "softnet_stat");
4229 proc_net_remove(net, "dev");
4233 static void __net_exit dev_proc_net_exit(struct net *net)
4235 wext_proc_exit(net);
4237 proc_net_remove(net, "ptype");
4238 proc_net_remove(net, "softnet_stat");
4239 proc_net_remove(net, "dev");
4242 static struct pernet_operations __net_initdata dev_proc_ops = {
4243 .init = dev_proc_net_init,
4244 .exit = dev_proc_net_exit,
4247 static int __init dev_proc_init(void)
4249 return register_pernet_subsys(&dev_proc_ops);
4252 #define dev_proc_init() 0
4253 #endif /* CONFIG_PROC_FS */
4257 * netdev_set_master - set up master/slave pair
4258 * @slave: slave device
4259 * @master: new master device
4261 * Changes the master device of the slave. Pass %NULL to break the
4262 * bonding. The caller must hold the RTNL semaphore. On a failure
4263 * a negative errno code is returned. On success the reference counts
4264 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4265 * function returns zero.
4267 int netdev_set_master(struct net_device *slave, struct net_device *master)
4269 struct net_device *old = slave->master;
4279 slave->master = master;
4286 slave->flags |= IFF_SLAVE;
4288 slave->flags &= ~IFF_SLAVE;
4290 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4293 EXPORT_SYMBOL(netdev_set_master);
4295 static void dev_change_rx_flags(struct net_device *dev, int flags)
4297 const struct net_device_ops *ops = dev->netdev_ops;
4299 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4300 ops->ndo_change_rx_flags(dev, flags);
4303 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4305 unsigned short old_flags = dev->flags;
4311 dev->flags |= IFF_PROMISC;
4312 dev->promiscuity += inc;
4313 if (dev->promiscuity == 0) {
4316 * If inc causes overflow, untouch promisc and return error.
4319 dev->flags &= ~IFF_PROMISC;
4321 dev->promiscuity -= inc;
4322 printk(KERN_WARNING "%s: promiscuity touches roof, "
4323 "set promiscuity failed, promiscuity feature "
4324 "of device might be broken.\n", dev->name);
4328 if (dev->flags != old_flags) {
4329 printk(KERN_INFO "device %s %s promiscuous mode\n",
4330 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4332 if (audit_enabled) {
4333 current_uid_gid(&uid, &gid);
4334 audit_log(current->audit_context, GFP_ATOMIC,
4335 AUDIT_ANOM_PROMISCUOUS,
4336 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4337 dev->name, (dev->flags & IFF_PROMISC),
4338 (old_flags & IFF_PROMISC),
4339 audit_get_loginuid(current),
4341 audit_get_sessionid(current));
4344 dev_change_rx_flags(dev, IFF_PROMISC);
4350 * dev_set_promiscuity - update promiscuity count on a device
4354 * Add or remove promiscuity from a device. While the count in the device
4355 * remains above zero the interface remains promiscuous. Once it hits zero
4356 * the device reverts back to normal filtering operation. A negative inc
4357 * value is used to drop promiscuity on the device.
4358 * Return 0 if successful or a negative errno code on error.
4360 int dev_set_promiscuity(struct net_device *dev, int inc)
4362 unsigned short old_flags = dev->flags;
4365 err = __dev_set_promiscuity(dev, inc);
4368 if (dev->flags != old_flags)
4369 dev_set_rx_mode(dev);
4372 EXPORT_SYMBOL(dev_set_promiscuity);
4375 * dev_set_allmulti - update allmulti count on a device
4379 * Add or remove reception of all multicast frames to a device. While the
4380 * count in the device remains above zero the interface remains listening
4381 * to all interfaces. Once it hits zero the device reverts back to normal
4382 * filtering operation. A negative @inc value is used to drop the counter
4383 * when releasing a resource needing all multicasts.
4384 * Return 0 if successful or a negative errno code on error.
4387 int dev_set_allmulti(struct net_device *dev, int inc)
4389 unsigned short old_flags = dev->flags;
4393 dev->flags |= IFF_ALLMULTI;
4394 dev->allmulti += inc;
4395 if (dev->allmulti == 0) {
4398 * If inc causes overflow, untouch allmulti and return error.
4401 dev->flags &= ~IFF_ALLMULTI;
4403 dev->allmulti -= inc;
4404 printk(KERN_WARNING "%s: allmulti touches roof, "
4405 "set allmulti failed, allmulti feature of "
4406 "device might be broken.\n", dev->name);
4410 if (dev->flags ^ old_flags) {
4411 dev_change_rx_flags(dev, IFF_ALLMULTI);
4412 dev_set_rx_mode(dev);
4416 EXPORT_SYMBOL(dev_set_allmulti);
4419 * Upload unicast and multicast address lists to device and
4420 * configure RX filtering. When the device doesn't support unicast
4421 * filtering it is put in promiscuous mode while unicast addresses
4424 void __dev_set_rx_mode(struct net_device *dev)
4426 const struct net_device_ops *ops = dev->netdev_ops;
4428 /* dev_open will call this function so the list will stay sane. */
4429 if (!(dev->flags&IFF_UP))
4432 if (!netif_device_present(dev))
4435 if (ops->ndo_set_rx_mode)
4436 ops->ndo_set_rx_mode(dev);
4438 /* Unicast addresses changes may only happen under the rtnl,
4439 * therefore calling __dev_set_promiscuity here is safe.
4441 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4442 __dev_set_promiscuity(dev, 1);
4443 dev->uc_promisc = 1;
4444 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4445 __dev_set_promiscuity(dev, -1);
4446 dev->uc_promisc = 0;
4449 if (ops->ndo_set_multicast_list)
4450 ops->ndo_set_multicast_list(dev);
4454 void dev_set_rx_mode(struct net_device *dev)
4456 netif_addr_lock_bh(dev);
4457 __dev_set_rx_mode(dev);
4458 netif_addr_unlock_bh(dev);
4462 * dev_get_flags - get flags reported to userspace
4465 * Get the combination of flag bits exported through APIs to userspace.
4467 unsigned dev_get_flags(const struct net_device *dev)
4471 flags = (dev->flags & ~(IFF_PROMISC |
4476 (dev->gflags & (IFF_PROMISC |
4479 if (netif_running(dev)) {
4480 if (netif_oper_up(dev))
4481 flags |= IFF_RUNNING;
4482 if (netif_carrier_ok(dev))
4483 flags |= IFF_LOWER_UP;
4484 if (netif_dormant(dev))
4485 flags |= IFF_DORMANT;
4490 EXPORT_SYMBOL(dev_get_flags);
4492 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4494 int old_flags = dev->flags;
4500 * Set the flags on our device.
4503 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4504 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4506 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4510 * Load in the correct multicast list now the flags have changed.
4513 if ((old_flags ^ flags) & IFF_MULTICAST)
4514 dev_change_rx_flags(dev, IFF_MULTICAST);
4516 dev_set_rx_mode(dev);
4519 * Have we downed the interface. We handle IFF_UP ourselves
4520 * according to user attempts to set it, rather than blindly
4525 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4526 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4529 dev_set_rx_mode(dev);
4532 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4533 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4535 dev->gflags ^= IFF_PROMISC;
4536 dev_set_promiscuity(dev, inc);
4539 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4540 is important. Some (broken) drivers set IFF_PROMISC, when
4541 IFF_ALLMULTI is requested not asking us and not reporting.
4543 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4544 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4546 dev->gflags ^= IFF_ALLMULTI;
4547 dev_set_allmulti(dev, inc);
4553 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4555 unsigned int changes = dev->flags ^ old_flags;
4557 if (changes & IFF_UP) {
4558 if (dev->flags & IFF_UP)
4559 call_netdevice_notifiers(NETDEV_UP, dev);
4561 call_netdevice_notifiers(NETDEV_DOWN, dev);
4564 if (dev->flags & IFF_UP &&
4565 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4566 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4570 * dev_change_flags - change device settings
4572 * @flags: device state flags
4574 * Change settings on device based state flags. The flags are
4575 * in the userspace exported format.
4577 int dev_change_flags(struct net_device *dev, unsigned flags)
4580 int old_flags = dev->flags;
4582 ret = __dev_change_flags(dev, flags);
4586 changes = old_flags ^ dev->flags;
4588 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4590 __dev_notify_flags(dev, old_flags);
4593 EXPORT_SYMBOL(dev_change_flags);
4596 * dev_set_mtu - Change maximum transfer unit
4598 * @new_mtu: new transfer unit
4600 * Change the maximum transfer size of the network device.
4602 int dev_set_mtu(struct net_device *dev, int new_mtu)
4604 const struct net_device_ops *ops = dev->netdev_ops;
4607 if (new_mtu == dev->mtu)
4610 /* MTU must be positive. */
4614 if (!netif_device_present(dev))
4618 if (ops->ndo_change_mtu)
4619 err = ops->ndo_change_mtu(dev, new_mtu);
4623 if (!err && dev->flags & IFF_UP)
4624 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4627 EXPORT_SYMBOL(dev_set_mtu);
4630 * dev_set_group - Change group this device belongs to
4632 * @new_group: group this device should belong to
4634 void dev_set_group(struct net_device *dev, int new_group)
4636 dev->group = new_group;
4638 EXPORT_SYMBOL(dev_set_group);
4641 * dev_set_mac_address - Change Media Access Control Address
4645 * Change the hardware (MAC) address of the device
4647 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4649 const struct net_device_ops *ops = dev->netdev_ops;
4652 if (!ops->ndo_set_mac_address)
4654 if (sa->sa_family != dev->type)
4656 if (!netif_device_present(dev))
4658 err = ops->ndo_set_mac_address(dev, sa);
4660 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4663 EXPORT_SYMBOL(dev_set_mac_address);
4666 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4668 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4671 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4677 case SIOCGIFFLAGS: /* Get interface flags */
4678 ifr->ifr_flags = (short) dev_get_flags(dev);
4681 case SIOCGIFMETRIC: /* Get the metric on the interface
4682 (currently unused) */
4683 ifr->ifr_metric = 0;
4686 case SIOCGIFMTU: /* Get the MTU of a device */
4687 ifr->ifr_mtu = dev->mtu;
4692 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4694 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4695 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4696 ifr->ifr_hwaddr.sa_family = dev->type;
4704 ifr->ifr_map.mem_start = dev->mem_start;
4705 ifr->ifr_map.mem_end = dev->mem_end;
4706 ifr->ifr_map.base_addr = dev->base_addr;
4707 ifr->ifr_map.irq = dev->irq;
4708 ifr->ifr_map.dma = dev->dma;
4709 ifr->ifr_map.port = dev->if_port;
4713 ifr->ifr_ifindex = dev->ifindex;
4717 ifr->ifr_qlen = dev->tx_queue_len;
4721 /* dev_ioctl() should ensure this case
4733 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4735 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4738 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4739 const struct net_device_ops *ops;
4744 ops = dev->netdev_ops;
4747 case SIOCSIFFLAGS: /* Set interface flags */
4748 return dev_change_flags(dev, ifr->ifr_flags);
4750 case SIOCSIFMETRIC: /* Set the metric on the interface
4751 (currently unused) */
4754 case SIOCSIFMTU: /* Set the MTU of a device */
4755 return dev_set_mtu(dev, ifr->ifr_mtu);
4758 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4760 case SIOCSIFHWBROADCAST:
4761 if (ifr->ifr_hwaddr.sa_family != dev->type)
4763 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4764 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4765 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4769 if (ops->ndo_set_config) {
4770 if (!netif_device_present(dev))
4772 return ops->ndo_set_config(dev, &ifr->ifr_map);
4777 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4778 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4780 if (!netif_device_present(dev))
4782 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4785 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4786 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4788 if (!netif_device_present(dev))
4790 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4793 if (ifr->ifr_qlen < 0)
4795 dev->tx_queue_len = ifr->ifr_qlen;
4799 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4800 return dev_change_name(dev, ifr->ifr_newname);
4803 * Unknown or private ioctl
4806 if ((cmd >= SIOCDEVPRIVATE &&
4807 cmd <= SIOCDEVPRIVATE + 15) ||
4808 cmd == SIOCBONDENSLAVE ||
4809 cmd == SIOCBONDRELEASE ||
4810 cmd == SIOCBONDSETHWADDR ||
4811 cmd == SIOCBONDSLAVEINFOQUERY ||
4812 cmd == SIOCBONDINFOQUERY ||
4813 cmd == SIOCBONDCHANGEACTIVE ||
4814 cmd == SIOCGMIIPHY ||
4815 cmd == SIOCGMIIREG ||
4816 cmd == SIOCSMIIREG ||
4817 cmd == SIOCBRADDIF ||
4818 cmd == SIOCBRDELIF ||
4819 cmd == SIOCSHWTSTAMP ||
4820 cmd == SIOCWANDEV) {
4822 if (ops->ndo_do_ioctl) {
4823 if (netif_device_present(dev))
4824 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4836 * This function handles all "interface"-type I/O control requests. The actual
4837 * 'doing' part of this is dev_ifsioc above.
4841 * dev_ioctl - network device ioctl
4842 * @net: the applicable net namespace
4843 * @cmd: command to issue
4844 * @arg: pointer to a struct ifreq in user space
4846 * Issue ioctl functions to devices. This is normally called by the
4847 * user space syscall interfaces but can sometimes be useful for
4848 * other purposes. The return value is the return from the syscall if
4849 * positive or a negative errno code on error.
4852 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4858 /* One special case: SIOCGIFCONF takes ifconf argument
4859 and requires shared lock, because it sleeps writing
4863 if (cmd == SIOCGIFCONF) {
4865 ret = dev_ifconf(net, (char __user *) arg);
4869 if (cmd == SIOCGIFNAME)
4870 return dev_ifname(net, (struct ifreq __user *)arg);
4872 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4875 ifr.ifr_name[IFNAMSIZ-1] = 0;
4877 colon = strchr(ifr.ifr_name, ':');
4882 * See which interface the caller is talking about.
4887 * These ioctl calls:
4888 * - can be done by all.
4889 * - atomic and do not require locking.
4900 dev_load(net, ifr.ifr_name);
4902 ret = dev_ifsioc_locked(net, &ifr, cmd);
4907 if (copy_to_user(arg, &ifr,
4908 sizeof(struct ifreq)))
4914 dev_load(net, ifr.ifr_name);
4916 ret = dev_ethtool(net, &ifr);
4921 if (copy_to_user(arg, &ifr,
4922 sizeof(struct ifreq)))
4928 * These ioctl calls:
4929 * - require superuser power.
4930 * - require strict serialization.
4936 if (!capable(CAP_NET_ADMIN))
4938 dev_load(net, ifr.ifr_name);
4940 ret = dev_ifsioc(net, &ifr, cmd);
4945 if (copy_to_user(arg, &ifr,
4946 sizeof(struct ifreq)))
4952 * These ioctl calls:
4953 * - require superuser power.
4954 * - require strict serialization.
4955 * - do not return a value
4965 case SIOCSIFHWBROADCAST:
4968 case SIOCBONDENSLAVE:
4969 case SIOCBONDRELEASE:
4970 case SIOCBONDSETHWADDR:
4971 case SIOCBONDCHANGEACTIVE:
4975 if (!capable(CAP_NET_ADMIN))
4978 case SIOCBONDSLAVEINFOQUERY:
4979 case SIOCBONDINFOQUERY:
4980 dev_load(net, ifr.ifr_name);
4982 ret = dev_ifsioc(net, &ifr, cmd);
4987 /* Get the per device memory space. We can add this but
4988 * currently do not support it */
4990 /* Set the per device memory buffer space.
4991 * Not applicable in our case */
4996 * Unknown or private ioctl.
4999 if (cmd == SIOCWANDEV ||
5000 (cmd >= SIOCDEVPRIVATE &&
5001 cmd <= SIOCDEVPRIVATE + 15)) {
5002 dev_load(net, ifr.ifr_name);
5004 ret = dev_ifsioc(net, &ifr, cmd);
5006 if (!ret && copy_to_user(arg, &ifr,
5007 sizeof(struct ifreq)))
5011 /* Take care of Wireless Extensions */
5012 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5013 return wext_handle_ioctl(net, &ifr, cmd, arg);
5020 * dev_new_index - allocate an ifindex
5021 * @net: the applicable net namespace
5023 * Returns a suitable unique value for a new device interface
5024 * number. The caller must hold the rtnl semaphore or the
5025 * dev_base_lock to be sure it remains unique.
5027 static int dev_new_index(struct net *net)
5033 if (!__dev_get_by_index(net, ifindex))
5038 /* Delayed registration/unregisteration */
5039 static LIST_HEAD(net_todo_list);
5041 static void net_set_todo(struct net_device *dev)
5043 list_add_tail(&dev->todo_list, &net_todo_list);
5046 static void rollback_registered_many(struct list_head *head)
5048 struct net_device *dev, *tmp;
5050 BUG_ON(dev_boot_phase);
5053 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5054 /* Some devices call without registering
5055 * for initialization unwind. Remove those
5056 * devices and proceed with the remaining.
5058 if (dev->reg_state == NETREG_UNINITIALIZED) {
5059 pr_debug("unregister_netdevice: device %s/%p never "
5060 "was registered\n", dev->name, dev);
5063 list_del(&dev->unreg_list);
5067 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5070 /* If device is running, close it first. */
5071 dev_close_many(head);
5073 list_for_each_entry(dev, head, unreg_list) {
5074 /* And unlink it from device chain. */
5075 unlist_netdevice(dev);
5077 dev->reg_state = NETREG_UNREGISTERING;
5082 list_for_each_entry(dev, head, unreg_list) {
5083 /* Shutdown queueing discipline. */
5087 /* Notify protocols, that we are about to destroy
5088 this device. They should clean all the things.
5090 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5092 if (!dev->rtnl_link_ops ||
5093 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5094 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5097 * Flush the unicast and multicast chains
5102 if (dev->netdev_ops->ndo_uninit)
5103 dev->netdev_ops->ndo_uninit(dev);
5105 /* Notifier chain MUST detach us from master device. */
5106 WARN_ON(dev->master);
5108 /* Remove entries from kobject tree */
5109 netdev_unregister_kobject(dev);
5112 /* Process any work delayed until the end of the batch */
5113 dev = list_first_entry(head, struct net_device, unreg_list);
5114 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5118 list_for_each_entry(dev, head, unreg_list)
5122 static void rollback_registered(struct net_device *dev)
5126 list_add(&dev->unreg_list, &single);
5127 rollback_registered_many(&single);
5130 unsigned long netdev_fix_features(unsigned long features, const char *name)
5132 /* Fix illegal SG+CSUM combinations. */
5133 if ((features & NETIF_F_SG) &&
5134 !(features & NETIF_F_ALL_CSUM)) {
5136 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
5137 "checksum feature.\n", name);
5138 features &= ~NETIF_F_SG;
5141 /* TSO requires that SG is present as well. */
5142 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
5144 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
5145 "SG feature.\n", name);
5146 features &= ~NETIF_F_TSO;
5149 if (features & NETIF_F_UFO) {
5150 /* maybe split UFO into V4 and V6? */
5151 if (!((features & NETIF_F_GEN_CSUM) ||
5152 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5153 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5155 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5156 "since no checksum offload features.\n",
5158 features &= ~NETIF_F_UFO;
5161 if (!(features & NETIF_F_SG)) {
5163 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5164 "since no NETIF_F_SG feature.\n", name);
5165 features &= ~NETIF_F_UFO;
5171 EXPORT_SYMBOL(netdev_fix_features);
5174 * netif_stacked_transfer_operstate - transfer operstate
5175 * @rootdev: the root or lower level device to transfer state from
5176 * @dev: the device to transfer operstate to
5178 * Transfer operational state from root to device. This is normally
5179 * called when a stacking relationship exists between the root
5180 * device and the device(a leaf device).
5182 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5183 struct net_device *dev)
5185 if (rootdev->operstate == IF_OPER_DORMANT)
5186 netif_dormant_on(dev);
5188 netif_dormant_off(dev);
5190 if (netif_carrier_ok(rootdev)) {
5191 if (!netif_carrier_ok(dev))
5192 netif_carrier_on(dev);
5194 if (netif_carrier_ok(dev))
5195 netif_carrier_off(dev);
5198 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5201 static int netif_alloc_rx_queues(struct net_device *dev)
5203 unsigned int i, count = dev->num_rx_queues;
5204 struct netdev_rx_queue *rx;
5208 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5210 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5215 for (i = 0; i < count; i++)
5221 static void netdev_init_one_queue(struct net_device *dev,
5222 struct netdev_queue *queue, void *_unused)
5224 /* Initialize queue lock */
5225 spin_lock_init(&queue->_xmit_lock);
5226 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5227 queue->xmit_lock_owner = -1;
5228 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5232 static int netif_alloc_netdev_queues(struct net_device *dev)
5234 unsigned int count = dev->num_tx_queues;
5235 struct netdev_queue *tx;
5239 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5241 pr_err("netdev: Unable to allocate %u tx queues.\n",
5247 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5248 spin_lock_init(&dev->tx_global_lock);
5254 * register_netdevice - register a network device
5255 * @dev: device to register
5257 * Take a completed network device structure and add it to the kernel
5258 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5259 * chain. 0 is returned on success. A negative errno code is returned
5260 * on a failure to set up the device, or if the name is a duplicate.
5262 * Callers must hold the rtnl semaphore. You may want
5263 * register_netdev() instead of this.
5266 * The locking appears insufficient to guarantee two parallel registers
5267 * will not get the same name.
5270 int register_netdevice(struct net_device *dev)
5273 struct net *net = dev_net(dev);
5275 BUG_ON(dev_boot_phase);
5280 /* When net_device's are persistent, this will be fatal. */
5281 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5284 spin_lock_init(&dev->addr_list_lock);
5285 netdev_set_addr_lockdep_class(dev);
5289 /* Init, if this function is available */
5290 if (dev->netdev_ops->ndo_init) {
5291 ret = dev->netdev_ops->ndo_init(dev);
5299 ret = dev_get_valid_name(dev, dev->name, 0);
5303 dev->ifindex = dev_new_index(net);
5304 if (dev->iflink == -1)
5305 dev->iflink = dev->ifindex;
5307 /* Fix illegal checksum combinations */
5308 if ((dev->features & NETIF_F_HW_CSUM) &&
5309 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5310 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5312 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5315 if ((dev->features & NETIF_F_NO_CSUM) &&
5316 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5317 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5319 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5322 dev->features = netdev_fix_features(dev->features, dev->name);
5324 /* Enable software GSO if SG is supported. */
5325 if (dev->features & NETIF_F_SG)
5326 dev->features |= NETIF_F_GSO;
5328 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5329 * vlan_dev_init() will do the dev->features check, so these features
5330 * are enabled only if supported by underlying device.
5332 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5334 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5335 ret = notifier_to_errno(ret);
5339 ret = netdev_register_kobject(dev);
5342 dev->reg_state = NETREG_REGISTERED;
5345 * Default initial state at registry is that the
5346 * device is present.
5349 set_bit(__LINK_STATE_PRESENT, &dev->state);
5351 dev_init_scheduler(dev);
5353 list_netdevice(dev);
5355 /* Notify protocols, that a new device appeared. */
5356 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5357 ret = notifier_to_errno(ret);
5359 rollback_registered(dev);
5360 dev->reg_state = NETREG_UNREGISTERED;
5363 * Prevent userspace races by waiting until the network
5364 * device is fully setup before sending notifications.
5366 if (!dev->rtnl_link_ops ||
5367 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5368 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5374 if (dev->netdev_ops->ndo_uninit)
5375 dev->netdev_ops->ndo_uninit(dev);
5378 EXPORT_SYMBOL(register_netdevice);
5381 * init_dummy_netdev - init a dummy network device for NAPI
5382 * @dev: device to init
5384 * This takes a network device structure and initialize the minimum
5385 * amount of fields so it can be used to schedule NAPI polls without
5386 * registering a full blown interface. This is to be used by drivers
5387 * that need to tie several hardware interfaces to a single NAPI
5388 * poll scheduler due to HW limitations.
5390 int init_dummy_netdev(struct net_device *dev)
5392 /* Clear everything. Note we don't initialize spinlocks
5393 * are they aren't supposed to be taken by any of the
5394 * NAPI code and this dummy netdev is supposed to be
5395 * only ever used for NAPI polls
5397 memset(dev, 0, sizeof(struct net_device));
5399 /* make sure we BUG if trying to hit standard
5400 * register/unregister code path
5402 dev->reg_state = NETREG_DUMMY;
5404 /* NAPI wants this */
5405 INIT_LIST_HEAD(&dev->napi_list);
5407 /* a dummy interface is started by default */
5408 set_bit(__LINK_STATE_PRESENT, &dev->state);
5409 set_bit(__LINK_STATE_START, &dev->state);
5411 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5412 * because users of this 'device' dont need to change
5418 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5422 * register_netdev - register a network device
5423 * @dev: device to register
5425 * Take a completed network device structure and add it to the kernel
5426 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5427 * chain. 0 is returned on success. A negative errno code is returned
5428 * on a failure to set up the device, or if the name is a duplicate.
5430 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5431 * and expands the device name if you passed a format string to
5434 int register_netdev(struct net_device *dev)
5441 * If the name is a format string the caller wants us to do a
5444 if (strchr(dev->name, '%')) {
5445 err = dev_alloc_name(dev, dev->name);
5450 err = register_netdevice(dev);
5455 EXPORT_SYMBOL(register_netdev);
5457 int netdev_refcnt_read(const struct net_device *dev)
5461 for_each_possible_cpu(i)
5462 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5465 EXPORT_SYMBOL(netdev_refcnt_read);
5468 * netdev_wait_allrefs - wait until all references are gone.
5470 * This is called when unregistering network devices.
5472 * Any protocol or device that holds a reference should register
5473 * for netdevice notification, and cleanup and put back the
5474 * reference if they receive an UNREGISTER event.
5475 * We can get stuck here if buggy protocols don't correctly
5478 static void netdev_wait_allrefs(struct net_device *dev)
5480 unsigned long rebroadcast_time, warning_time;
5483 linkwatch_forget_dev(dev);
5485 rebroadcast_time = warning_time = jiffies;
5486 refcnt = netdev_refcnt_read(dev);
5488 while (refcnt != 0) {
5489 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5492 /* Rebroadcast unregister notification */
5493 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5494 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5495 * should have already handle it the first time */
5497 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5499 /* We must not have linkwatch events
5500 * pending on unregister. If this
5501 * happens, we simply run the queue
5502 * unscheduled, resulting in a noop
5505 linkwatch_run_queue();
5510 rebroadcast_time = jiffies;
5515 refcnt = netdev_refcnt_read(dev);
5517 if (time_after(jiffies, warning_time + 10 * HZ)) {
5518 printk(KERN_EMERG "unregister_netdevice: "
5519 "waiting for %s to become free. Usage "
5522 warning_time = jiffies;
5531 * register_netdevice(x1);
5532 * register_netdevice(x2);
5534 * unregister_netdevice(y1);
5535 * unregister_netdevice(y2);
5541 * We are invoked by rtnl_unlock().
5542 * This allows us to deal with problems:
5543 * 1) We can delete sysfs objects which invoke hotplug
5544 * without deadlocking with linkwatch via keventd.
5545 * 2) Since we run with the RTNL semaphore not held, we can sleep
5546 * safely in order to wait for the netdev refcnt to drop to zero.
5548 * We must not return until all unregister events added during
5549 * the interval the lock was held have been completed.
5551 void netdev_run_todo(void)
5553 struct list_head list;
5555 /* Snapshot list, allow later requests */
5556 list_replace_init(&net_todo_list, &list);
5560 while (!list_empty(&list)) {
5561 struct net_device *dev
5562 = list_first_entry(&list, struct net_device, todo_list);
5563 list_del(&dev->todo_list);
5565 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5566 printk(KERN_ERR "network todo '%s' but state %d\n",
5567 dev->name, dev->reg_state);
5572 dev->reg_state = NETREG_UNREGISTERED;
5574 on_each_cpu(flush_backlog, dev, 1);
5576 netdev_wait_allrefs(dev);
5579 BUG_ON(netdev_refcnt_read(dev));
5580 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5581 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5582 WARN_ON(dev->dn_ptr);
5584 if (dev->destructor)
5585 dev->destructor(dev);
5587 /* Free network device */
5588 kobject_put(&dev->dev.kobj);
5592 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5593 * fields in the same order, with only the type differing.
5595 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5596 const struct net_device_stats *netdev_stats)
5598 #if BITS_PER_LONG == 64
5599 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5600 memcpy(stats64, netdev_stats, sizeof(*stats64));
5602 size_t i, n = sizeof(*stats64) / sizeof(u64);
5603 const unsigned long *src = (const unsigned long *)netdev_stats;
5604 u64 *dst = (u64 *)stats64;
5606 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5607 sizeof(*stats64) / sizeof(u64));
5608 for (i = 0; i < n; i++)
5614 * dev_get_stats - get network device statistics
5615 * @dev: device to get statistics from
5616 * @storage: place to store stats
5618 * Get network statistics from device. Return @storage.
5619 * The device driver may provide its own method by setting
5620 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5621 * otherwise the internal statistics structure is used.
5623 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5624 struct rtnl_link_stats64 *storage)
5626 const struct net_device_ops *ops = dev->netdev_ops;
5628 if (ops->ndo_get_stats64) {
5629 memset(storage, 0, sizeof(*storage));
5630 ops->ndo_get_stats64(dev, storage);
5631 } else if (ops->ndo_get_stats) {
5632 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5634 netdev_stats_to_stats64(storage, &dev->stats);
5636 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5639 EXPORT_SYMBOL(dev_get_stats);
5641 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5643 struct netdev_queue *queue = dev_ingress_queue(dev);
5645 #ifdef CONFIG_NET_CLS_ACT
5648 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5651 netdev_init_one_queue(dev, queue, NULL);
5652 queue->qdisc = &noop_qdisc;
5653 queue->qdisc_sleeping = &noop_qdisc;
5654 rcu_assign_pointer(dev->ingress_queue, queue);
5660 * alloc_netdev_mqs - allocate network device
5661 * @sizeof_priv: size of private data to allocate space for
5662 * @name: device name format string
5663 * @setup: callback to initialize device
5664 * @txqs: the number of TX subqueues to allocate
5665 * @rxqs: the number of RX subqueues to allocate
5667 * Allocates a struct net_device with private data area for driver use
5668 * and performs basic initialization. Also allocates subquue structs
5669 * for each queue on the device.
5671 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5672 void (*setup)(struct net_device *),
5673 unsigned int txqs, unsigned int rxqs)
5675 struct net_device *dev;
5677 struct net_device *p;
5679 BUG_ON(strlen(name) >= sizeof(dev->name));
5682 pr_err("alloc_netdev: Unable to allocate device "
5683 "with zero queues.\n");
5689 pr_err("alloc_netdev: Unable to allocate device "
5690 "with zero RX queues.\n");
5695 alloc_size = sizeof(struct net_device);
5697 /* ensure 32-byte alignment of private area */
5698 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5699 alloc_size += sizeof_priv;
5701 /* ensure 32-byte alignment of whole construct */
5702 alloc_size += NETDEV_ALIGN - 1;
5704 p = kzalloc(alloc_size, GFP_KERNEL);
5706 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5710 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5711 dev->padded = (char *)dev - (char *)p;
5713 dev->pcpu_refcnt = alloc_percpu(int);
5714 if (!dev->pcpu_refcnt)
5717 if (dev_addr_init(dev))
5723 dev_net_set(dev, &init_net);
5725 dev->num_tx_queues = txqs;
5726 dev->real_num_tx_queues = txqs;
5727 if (netif_alloc_netdev_queues(dev))
5731 dev->num_rx_queues = rxqs;
5732 dev->real_num_rx_queues = rxqs;
5733 if (netif_alloc_rx_queues(dev))
5737 dev->gso_max_size = GSO_MAX_SIZE;
5739 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5740 dev->ethtool_ntuple_list.count = 0;
5741 INIT_LIST_HEAD(&dev->napi_list);
5742 INIT_LIST_HEAD(&dev->unreg_list);
5743 INIT_LIST_HEAD(&dev->link_watch_list);
5744 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5746 strcpy(dev->name, name);
5747 dev->group = INIT_NETDEV_GROUP;
5751 free_percpu(dev->pcpu_refcnt);
5761 EXPORT_SYMBOL(alloc_netdev_mqs);
5764 * free_netdev - free network device
5767 * This function does the last stage of destroying an allocated device
5768 * interface. The reference to the device object is released.
5769 * If this is the last reference then it will be freed.
5771 void free_netdev(struct net_device *dev)
5773 struct napi_struct *p, *n;
5775 release_net(dev_net(dev));
5782 kfree(rcu_dereference_raw(dev->ingress_queue));
5784 /* Flush device addresses */
5785 dev_addr_flush(dev);
5787 /* Clear ethtool n-tuple list */
5788 ethtool_ntuple_flush(dev);
5790 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5793 free_percpu(dev->pcpu_refcnt);
5794 dev->pcpu_refcnt = NULL;
5796 /* Compatibility with error handling in drivers */
5797 if (dev->reg_state == NETREG_UNINITIALIZED) {
5798 kfree((char *)dev - dev->padded);
5802 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5803 dev->reg_state = NETREG_RELEASED;
5805 /* will free via device release */
5806 put_device(&dev->dev);
5808 EXPORT_SYMBOL(free_netdev);
5811 * synchronize_net - Synchronize with packet receive processing
5813 * Wait for packets currently being received to be done.
5814 * Does not block later packets from starting.
5816 void synchronize_net(void)
5821 EXPORT_SYMBOL(synchronize_net);
5824 * unregister_netdevice_queue - remove device from the kernel
5828 * This function shuts down a device interface and removes it
5829 * from the kernel tables.
5830 * If head not NULL, device is queued to be unregistered later.
5832 * Callers must hold the rtnl semaphore. You may want
5833 * unregister_netdev() instead of this.
5836 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5841 list_move_tail(&dev->unreg_list, head);
5843 rollback_registered(dev);
5844 /* Finish processing unregister after unlock */
5848 EXPORT_SYMBOL(unregister_netdevice_queue);
5851 * unregister_netdevice_many - unregister many devices
5852 * @head: list of devices
5854 void unregister_netdevice_many(struct list_head *head)
5856 struct net_device *dev;
5858 if (!list_empty(head)) {
5859 rollback_registered_many(head);
5860 list_for_each_entry(dev, head, unreg_list)
5864 EXPORT_SYMBOL(unregister_netdevice_many);
5867 * unregister_netdev - remove device from the kernel
5870 * This function shuts down a device interface and removes it
5871 * from the kernel tables.
5873 * This is just a wrapper for unregister_netdevice that takes
5874 * the rtnl semaphore. In general you want to use this and not
5875 * unregister_netdevice.
5877 void unregister_netdev(struct net_device *dev)
5880 unregister_netdevice(dev);
5883 EXPORT_SYMBOL(unregister_netdev);
5886 * dev_change_net_namespace - move device to different nethost namespace
5888 * @net: network namespace
5889 * @pat: If not NULL name pattern to try if the current device name
5890 * is already taken in the destination network namespace.
5892 * This function shuts down a device interface and moves it
5893 * to a new network namespace. On success 0 is returned, on
5894 * a failure a netagive errno code is returned.
5896 * Callers must hold the rtnl semaphore.
5899 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5905 /* Don't allow namespace local devices to be moved. */
5907 if (dev->features & NETIF_F_NETNS_LOCAL)
5910 /* Ensure the device has been registrered */
5912 if (dev->reg_state != NETREG_REGISTERED)
5915 /* Get out if there is nothing todo */
5917 if (net_eq(dev_net(dev), net))
5920 /* Pick the destination device name, and ensure
5921 * we can use it in the destination network namespace.
5924 if (__dev_get_by_name(net, dev->name)) {
5925 /* We get here if we can't use the current device name */
5928 if (dev_get_valid_name(dev, pat, 1))
5933 * And now a mini version of register_netdevice unregister_netdevice.
5936 /* If device is running close it first. */
5939 /* And unlink it from device chain */
5941 unlist_netdevice(dev);
5945 /* Shutdown queueing discipline. */
5948 /* Notify protocols, that we are about to destroy
5949 this device. They should clean all the things.
5951 Note that dev->reg_state stays at NETREG_REGISTERED.
5952 This is wanted because this way 8021q and macvlan know
5953 the device is just moving and can keep their slaves up.
5955 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5956 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5959 * Flush the unicast and multicast chains
5964 /* Actually switch the network namespace */
5965 dev_net_set(dev, net);
5967 /* If there is an ifindex conflict assign a new one */
5968 if (__dev_get_by_index(net, dev->ifindex)) {
5969 int iflink = (dev->iflink == dev->ifindex);
5970 dev->ifindex = dev_new_index(net);
5972 dev->iflink = dev->ifindex;
5975 /* Fixup kobjects */
5976 err = device_rename(&dev->dev, dev->name);
5979 /* Add the device back in the hashes */
5980 list_netdevice(dev);
5982 /* Notify protocols, that a new device appeared. */
5983 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5986 * Prevent userspace races by waiting until the network
5987 * device is fully setup before sending notifications.
5989 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5996 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5998 static int dev_cpu_callback(struct notifier_block *nfb,
5999 unsigned long action,
6002 struct sk_buff **list_skb;
6003 struct sk_buff *skb;
6004 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6005 struct softnet_data *sd, *oldsd;
6007 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6010 local_irq_disable();
6011 cpu = smp_processor_id();
6012 sd = &per_cpu(softnet_data, cpu);
6013 oldsd = &per_cpu(softnet_data, oldcpu);
6015 /* Find end of our completion_queue. */
6016 list_skb = &sd->completion_queue;
6018 list_skb = &(*list_skb)->next;
6019 /* Append completion queue from offline CPU. */
6020 *list_skb = oldsd->completion_queue;
6021 oldsd->completion_queue = NULL;
6023 /* Append output queue from offline CPU. */
6024 if (oldsd->output_queue) {
6025 *sd->output_queue_tailp = oldsd->output_queue;
6026 sd->output_queue_tailp = oldsd->output_queue_tailp;
6027 oldsd->output_queue = NULL;
6028 oldsd->output_queue_tailp = &oldsd->output_queue;
6031 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6034 /* Process offline CPU's input_pkt_queue */
6035 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6037 input_queue_head_incr(oldsd);
6039 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6041 input_queue_head_incr(oldsd);
6049 * netdev_increment_features - increment feature set by one
6050 * @all: current feature set
6051 * @one: new feature set
6052 * @mask: mask feature set
6054 * Computes a new feature set after adding a device with feature set
6055 * @one to the master device with current feature set @all. Will not
6056 * enable anything that is off in @mask. Returns the new feature set.
6058 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
6061 /* If device needs checksumming, downgrade to it. */
6062 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
6063 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
6064 else if (mask & NETIF_F_ALL_CSUM) {
6065 /* If one device supports v4/v6 checksumming, set for all. */
6066 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
6067 !(all & NETIF_F_GEN_CSUM)) {
6068 all &= ~NETIF_F_ALL_CSUM;
6069 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
6072 /* If one device supports hw checksumming, set for all. */
6073 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
6074 all &= ~NETIF_F_ALL_CSUM;
6075 all |= NETIF_F_HW_CSUM;
6079 one |= NETIF_F_ALL_CSUM;
6081 one |= all & NETIF_F_ONE_FOR_ALL;
6082 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
6083 all |= one & mask & NETIF_F_ONE_FOR_ALL;
6087 EXPORT_SYMBOL(netdev_increment_features);
6089 static struct hlist_head *netdev_create_hash(void)
6092 struct hlist_head *hash;
6094 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6096 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6097 INIT_HLIST_HEAD(&hash[i]);
6102 /* Initialize per network namespace state */
6103 static int __net_init netdev_init(struct net *net)
6105 INIT_LIST_HEAD(&net->dev_base_head);
6107 net->dev_name_head = netdev_create_hash();
6108 if (net->dev_name_head == NULL)
6111 net->dev_index_head = netdev_create_hash();
6112 if (net->dev_index_head == NULL)
6118 kfree(net->dev_name_head);
6124 * netdev_drivername - network driver for the device
6125 * @dev: network device
6126 * @buffer: buffer for resulting name
6127 * @len: size of buffer
6129 * Determine network driver for device.
6131 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6133 const struct device_driver *driver;
6134 const struct device *parent;
6136 if (len <= 0 || !buffer)
6140 parent = dev->dev.parent;
6145 driver = parent->driver;
6146 if (driver && driver->name)
6147 strlcpy(buffer, driver->name, len);
6151 static int __netdev_printk(const char *level, const struct net_device *dev,
6152 struct va_format *vaf)
6156 if (dev && dev->dev.parent)
6157 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6158 netdev_name(dev), vaf);
6160 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6162 r = printk("%s(NULL net_device): %pV", level, vaf);
6167 int netdev_printk(const char *level, const struct net_device *dev,
6168 const char *format, ...)
6170 struct va_format vaf;
6174 va_start(args, format);
6179 r = __netdev_printk(level, dev, &vaf);
6184 EXPORT_SYMBOL(netdev_printk);
6186 #define define_netdev_printk_level(func, level) \
6187 int func(const struct net_device *dev, const char *fmt, ...) \
6190 struct va_format vaf; \
6193 va_start(args, fmt); \
6198 r = __netdev_printk(level, dev, &vaf); \
6203 EXPORT_SYMBOL(func);
6205 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6206 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6207 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6208 define_netdev_printk_level(netdev_err, KERN_ERR);
6209 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6210 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6211 define_netdev_printk_level(netdev_info, KERN_INFO);
6213 static void __net_exit netdev_exit(struct net *net)
6215 kfree(net->dev_name_head);
6216 kfree(net->dev_index_head);
6219 static struct pernet_operations __net_initdata netdev_net_ops = {
6220 .init = netdev_init,
6221 .exit = netdev_exit,
6224 static void __net_exit default_device_exit(struct net *net)
6226 struct net_device *dev, *aux;
6228 * Push all migratable network devices back to the
6229 * initial network namespace
6232 for_each_netdev_safe(net, dev, aux) {
6234 char fb_name[IFNAMSIZ];
6236 /* Ignore unmoveable devices (i.e. loopback) */
6237 if (dev->features & NETIF_F_NETNS_LOCAL)
6240 /* Leave virtual devices for the generic cleanup */
6241 if (dev->rtnl_link_ops)
6244 /* Push remaing network devices to init_net */
6245 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6246 err = dev_change_net_namespace(dev, &init_net, fb_name);
6248 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6249 __func__, dev->name, err);
6256 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6258 /* At exit all network devices most be removed from a network
6259 * namespace. Do this in the reverse order of registration.
6260 * Do this across as many network namespaces as possible to
6261 * improve batching efficiency.
6263 struct net_device *dev;
6265 LIST_HEAD(dev_kill_list);
6268 list_for_each_entry(net, net_list, exit_list) {
6269 for_each_netdev_reverse(net, dev) {
6270 if (dev->rtnl_link_ops)
6271 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6273 unregister_netdevice_queue(dev, &dev_kill_list);
6276 unregister_netdevice_many(&dev_kill_list);
6280 static struct pernet_operations __net_initdata default_device_ops = {
6281 .exit = default_device_exit,
6282 .exit_batch = default_device_exit_batch,
6286 * Initialize the DEV module. At boot time this walks the device list and
6287 * unhooks any devices that fail to initialise (normally hardware not
6288 * present) and leaves us with a valid list of present and active devices.
6293 * This is called single threaded during boot, so no need
6294 * to take the rtnl semaphore.
6296 static int __init net_dev_init(void)
6298 int i, rc = -ENOMEM;
6300 BUG_ON(!dev_boot_phase);
6302 if (dev_proc_init())
6305 if (netdev_kobject_init())
6308 INIT_LIST_HEAD(&ptype_all);
6309 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6310 INIT_LIST_HEAD(&ptype_base[i]);
6312 if (register_pernet_subsys(&netdev_net_ops))
6316 * Initialise the packet receive queues.
6319 for_each_possible_cpu(i) {
6320 struct softnet_data *sd = &per_cpu(softnet_data, i);
6322 memset(sd, 0, sizeof(*sd));
6323 skb_queue_head_init(&sd->input_pkt_queue);
6324 skb_queue_head_init(&sd->process_queue);
6325 sd->completion_queue = NULL;
6326 INIT_LIST_HEAD(&sd->poll_list);
6327 sd->output_queue = NULL;
6328 sd->output_queue_tailp = &sd->output_queue;
6330 sd->csd.func = rps_trigger_softirq;
6336 sd->backlog.poll = process_backlog;
6337 sd->backlog.weight = weight_p;
6338 sd->backlog.gro_list = NULL;
6339 sd->backlog.gro_count = 0;
6344 /* The loopback device is special if any other network devices
6345 * is present in a network namespace the loopback device must
6346 * be present. Since we now dynamically allocate and free the
6347 * loopback device ensure this invariant is maintained by
6348 * keeping the loopback device as the first device on the
6349 * list of network devices. Ensuring the loopback devices
6350 * is the first device that appears and the last network device
6353 if (register_pernet_device(&loopback_net_ops))
6356 if (register_pernet_device(&default_device_ops))
6359 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6360 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6362 hotcpu_notifier(dev_cpu_callback, 0);
6370 subsys_initcall(net_dev_init);
6372 static int __init initialize_hashrnd(void)
6374 get_random_bytes(&hashrnd, sizeof(hashrnd));
6378 late_initcall_sync(initialize_hashrnd);