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(struct net_device *dev)
1227 const struct net_device_ops *ops = dev->netdev_ops;
1233 * Tell people we are going down, so that they can
1234 * prepare to death, when device is still operating.
1236 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1238 clear_bit(__LINK_STATE_START, &dev->state);
1240 /* Synchronize to scheduled poll. We cannot touch poll list,
1241 * it can be even on different cpu. So just clear netif_running().
1243 * dev->stop() will invoke napi_disable() on all of it's
1244 * napi_struct instances on this device.
1246 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1248 dev_deactivate(dev);
1251 * Call the device specific close. This cannot fail.
1252 * Only if device is UP
1254 * We allow it to be called even after a DETACH hot-plug
1261 * Device is now down.
1264 dev->flags &= ~IFF_UP;
1269 net_dmaengine_put();
1275 * dev_close - shutdown an interface.
1276 * @dev: device to shutdown
1278 * This function moves an active device into down state. A
1279 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1280 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1283 int dev_close(struct net_device *dev)
1285 if (!(dev->flags & IFF_UP))
1291 * Tell people we are down
1293 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1294 call_netdevice_notifiers(NETDEV_DOWN, dev);
1298 EXPORT_SYMBOL(dev_close);
1302 * dev_disable_lro - disable Large Receive Offload on a device
1305 * Disable Large Receive Offload (LRO) on a net device. Must be
1306 * called under RTNL. This is needed if received packets may be
1307 * forwarded to another interface.
1309 void dev_disable_lro(struct net_device *dev)
1311 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1312 dev->ethtool_ops->set_flags) {
1313 u32 flags = dev->ethtool_ops->get_flags(dev);
1314 if (flags & ETH_FLAG_LRO) {
1315 flags &= ~ETH_FLAG_LRO;
1316 dev->ethtool_ops->set_flags(dev, flags);
1319 WARN_ON(dev->features & NETIF_F_LRO);
1321 EXPORT_SYMBOL(dev_disable_lro);
1324 static int dev_boot_phase = 1;
1327 * Device change register/unregister. These are not inline or static
1328 * as we export them to the world.
1332 * register_netdevice_notifier - register a network notifier block
1335 * Register a notifier to be called when network device events occur.
1336 * The notifier passed is linked into the kernel structures and must
1337 * not be reused until it has been unregistered. A negative errno code
1338 * is returned on a failure.
1340 * When registered all registration and up events are replayed
1341 * to the new notifier to allow device to have a race free
1342 * view of the network device list.
1345 int register_netdevice_notifier(struct notifier_block *nb)
1347 struct net_device *dev;
1348 struct net_device *last;
1353 err = raw_notifier_chain_register(&netdev_chain, nb);
1359 for_each_netdev(net, dev) {
1360 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1361 err = notifier_to_errno(err);
1365 if (!(dev->flags & IFF_UP))
1368 nb->notifier_call(nb, NETDEV_UP, dev);
1379 for_each_netdev(net, dev) {
1383 if (dev->flags & IFF_UP) {
1384 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1385 nb->notifier_call(nb, NETDEV_DOWN, dev);
1387 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1388 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1392 raw_notifier_chain_unregister(&netdev_chain, nb);
1395 EXPORT_SYMBOL(register_netdevice_notifier);
1398 * unregister_netdevice_notifier - unregister a network notifier block
1401 * Unregister a notifier previously registered by
1402 * register_netdevice_notifier(). The notifier is unlinked into the
1403 * kernel structures and may then be reused. A negative errno code
1404 * is returned on a failure.
1407 int unregister_netdevice_notifier(struct notifier_block *nb)
1412 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1416 EXPORT_SYMBOL(unregister_netdevice_notifier);
1419 * call_netdevice_notifiers - call all network notifier blocks
1420 * @val: value passed unmodified to notifier function
1421 * @dev: net_device pointer passed unmodified to notifier function
1423 * Call all network notifier blocks. Parameters and return value
1424 * are as for raw_notifier_call_chain().
1427 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1430 return raw_notifier_call_chain(&netdev_chain, val, dev);
1433 /* When > 0 there are consumers of rx skb time stamps */
1434 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1436 void net_enable_timestamp(void)
1438 atomic_inc(&netstamp_needed);
1440 EXPORT_SYMBOL(net_enable_timestamp);
1442 void net_disable_timestamp(void)
1444 atomic_dec(&netstamp_needed);
1446 EXPORT_SYMBOL(net_disable_timestamp);
1448 static inline void net_timestamp_set(struct sk_buff *skb)
1450 if (atomic_read(&netstamp_needed))
1451 __net_timestamp(skb);
1453 skb->tstamp.tv64 = 0;
1456 static inline void net_timestamp_check(struct sk_buff *skb)
1458 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1459 __net_timestamp(skb);
1463 * dev_forward_skb - loopback an skb to another netif
1465 * @dev: destination network device
1466 * @skb: buffer to forward
1469 * NET_RX_SUCCESS (no congestion)
1470 * NET_RX_DROP (packet was dropped, but freed)
1472 * dev_forward_skb can be used for injecting an skb from the
1473 * start_xmit function of one device into the receive queue
1474 * of another device.
1476 * The receiving device may be in another namespace, so
1477 * we have to clear all information in the skb that could
1478 * impact namespace isolation.
1480 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1485 if (unlikely(!(dev->flags & IFF_UP) ||
1486 (skb->len > (dev->mtu + dev->hard_header_len + VLAN_HLEN)))) {
1487 atomic_long_inc(&dev->rx_dropped);
1491 skb_set_dev(skb, dev);
1492 skb->tstamp.tv64 = 0;
1493 skb->pkt_type = PACKET_HOST;
1494 skb->protocol = eth_type_trans(skb, dev);
1495 return netif_rx(skb);
1497 EXPORT_SYMBOL_GPL(dev_forward_skb);
1500 * Support routine. Sends outgoing frames to any network
1501 * taps currently in use.
1504 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1506 struct packet_type *ptype;
1508 #ifdef CONFIG_NET_CLS_ACT
1509 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1510 net_timestamp_set(skb);
1512 net_timestamp_set(skb);
1516 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1517 /* Never send packets back to the socket
1518 * they originated from - MvS (miquels@drinkel.ow.org)
1520 if ((ptype->dev == dev || !ptype->dev) &&
1521 (ptype->af_packet_priv == NULL ||
1522 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1523 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1527 /* skb->nh should be correctly
1528 set by sender, so that the second statement is
1529 just protection against buggy protocols.
1531 skb_reset_mac_header(skb2);
1533 if (skb_network_header(skb2) < skb2->data ||
1534 skb2->network_header > skb2->tail) {
1535 if (net_ratelimit())
1536 printk(KERN_CRIT "protocol %04x is "
1538 ntohs(skb2->protocol),
1540 skb_reset_network_header(skb2);
1543 skb2->transport_header = skb2->network_header;
1544 skb2->pkt_type = PACKET_OUTGOING;
1545 ptype->func(skb2, skb->dev, ptype, skb->dev);
1552 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1553 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1555 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1559 if (txq < 1 || txq > dev->num_tx_queues)
1562 if (dev->reg_state == NETREG_REGISTERED) {
1565 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1570 if (txq < dev->real_num_tx_queues)
1571 qdisc_reset_all_tx_gt(dev, txq);
1574 dev->real_num_tx_queues = txq;
1577 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1581 * netif_set_real_num_rx_queues - set actual number of RX queues used
1582 * @dev: Network device
1583 * @rxq: Actual number of RX queues
1585 * This must be called either with the rtnl_lock held or before
1586 * registration of the net device. Returns 0 on success, or a
1587 * negative error code. If called before registration, it always
1590 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1594 if (rxq < 1 || rxq > dev->num_rx_queues)
1597 if (dev->reg_state == NETREG_REGISTERED) {
1600 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1606 dev->real_num_rx_queues = rxq;
1609 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1612 static inline void __netif_reschedule(struct Qdisc *q)
1614 struct softnet_data *sd;
1615 unsigned long flags;
1617 local_irq_save(flags);
1618 sd = &__get_cpu_var(softnet_data);
1619 q->next_sched = NULL;
1620 *sd->output_queue_tailp = q;
1621 sd->output_queue_tailp = &q->next_sched;
1622 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1623 local_irq_restore(flags);
1626 void __netif_schedule(struct Qdisc *q)
1628 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1629 __netif_reschedule(q);
1631 EXPORT_SYMBOL(__netif_schedule);
1633 void dev_kfree_skb_irq(struct sk_buff *skb)
1635 if (atomic_dec_and_test(&skb->users)) {
1636 struct softnet_data *sd;
1637 unsigned long flags;
1639 local_irq_save(flags);
1640 sd = &__get_cpu_var(softnet_data);
1641 skb->next = sd->completion_queue;
1642 sd->completion_queue = skb;
1643 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1644 local_irq_restore(flags);
1647 EXPORT_SYMBOL(dev_kfree_skb_irq);
1649 void dev_kfree_skb_any(struct sk_buff *skb)
1651 if (in_irq() || irqs_disabled())
1652 dev_kfree_skb_irq(skb);
1656 EXPORT_SYMBOL(dev_kfree_skb_any);
1660 * netif_device_detach - mark device as removed
1661 * @dev: network device
1663 * Mark device as removed from system and therefore no longer available.
1665 void netif_device_detach(struct net_device *dev)
1667 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1668 netif_running(dev)) {
1669 netif_tx_stop_all_queues(dev);
1672 EXPORT_SYMBOL(netif_device_detach);
1675 * netif_device_attach - mark device as attached
1676 * @dev: network device
1678 * Mark device as attached from system and restart if needed.
1680 void netif_device_attach(struct net_device *dev)
1682 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1683 netif_running(dev)) {
1684 netif_tx_wake_all_queues(dev);
1685 __netdev_watchdog_up(dev);
1688 EXPORT_SYMBOL(netif_device_attach);
1690 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1692 return ((features & NETIF_F_NO_CSUM) ||
1693 ((features & NETIF_F_V4_CSUM) &&
1694 protocol == htons(ETH_P_IP)) ||
1695 ((features & NETIF_F_V6_CSUM) &&
1696 protocol == htons(ETH_P_IPV6)) ||
1697 ((features & NETIF_F_FCOE_CRC) &&
1698 protocol == htons(ETH_P_FCOE)));
1701 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1703 __be16 protocol = skb->protocol;
1704 int features = dev->features;
1706 if (vlan_tx_tag_present(skb)) {
1707 features &= dev->vlan_features;
1708 } else if (protocol == htons(ETH_P_8021Q)) {
1709 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1710 protocol = veh->h_vlan_encapsulated_proto;
1711 features &= dev->vlan_features;
1714 return can_checksum_protocol(features, protocol);
1718 * skb_dev_set -- assign a new device to a buffer
1719 * @skb: buffer for the new device
1720 * @dev: network device
1722 * If an skb is owned by a device already, we have to reset
1723 * all data private to the namespace a device belongs to
1724 * before assigning it a new device.
1726 #ifdef CONFIG_NET_NS
1727 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1730 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1733 skb_init_secmark(skb);
1737 skb->ipvs_property = 0;
1738 #ifdef CONFIG_NET_SCHED
1744 EXPORT_SYMBOL(skb_set_dev);
1745 #endif /* CONFIG_NET_NS */
1748 * Invalidate hardware checksum when packet is to be mangled, and
1749 * complete checksum manually on outgoing path.
1751 int skb_checksum_help(struct sk_buff *skb)
1754 int ret = 0, offset;
1756 if (skb->ip_summed == CHECKSUM_COMPLETE)
1757 goto out_set_summed;
1759 if (unlikely(skb_shinfo(skb)->gso_size)) {
1760 /* Let GSO fix up the checksum. */
1761 goto out_set_summed;
1764 offset = skb->csum_start - skb_headroom(skb);
1765 BUG_ON(offset >= skb_headlen(skb));
1766 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1768 offset += skb->csum_offset;
1769 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1771 if (skb_cloned(skb) &&
1772 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1773 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1778 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1780 skb->ip_summed = CHECKSUM_NONE;
1784 EXPORT_SYMBOL(skb_checksum_help);
1787 * skb_gso_segment - Perform segmentation on skb.
1788 * @skb: buffer to segment
1789 * @features: features for the output path (see dev->features)
1791 * This function segments the given skb and returns a list of segments.
1793 * It may return NULL if the skb requires no segmentation. This is
1794 * only possible when GSO is used for verifying header integrity.
1796 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1798 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1799 struct packet_type *ptype;
1800 __be16 type = skb->protocol;
1801 int vlan_depth = ETH_HLEN;
1804 while (type == htons(ETH_P_8021Q)) {
1805 struct vlan_hdr *vh;
1807 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1808 return ERR_PTR(-EINVAL);
1810 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1811 type = vh->h_vlan_encapsulated_proto;
1812 vlan_depth += VLAN_HLEN;
1815 skb_reset_mac_header(skb);
1816 skb->mac_len = skb->network_header - skb->mac_header;
1817 __skb_pull(skb, skb->mac_len);
1819 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1820 struct net_device *dev = skb->dev;
1821 struct ethtool_drvinfo info = {};
1823 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1824 dev->ethtool_ops->get_drvinfo(dev, &info);
1826 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1827 info.driver, dev ? dev->features : 0L,
1828 skb->sk ? skb->sk->sk_route_caps : 0L,
1829 skb->len, skb->data_len, skb->ip_summed);
1831 if (skb_header_cloned(skb) &&
1832 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1833 return ERR_PTR(err);
1837 list_for_each_entry_rcu(ptype,
1838 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1839 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1840 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1841 err = ptype->gso_send_check(skb);
1842 segs = ERR_PTR(err);
1843 if (err || skb_gso_ok(skb, features))
1845 __skb_push(skb, (skb->data -
1846 skb_network_header(skb)));
1848 segs = ptype->gso_segment(skb, features);
1854 __skb_push(skb, skb->data - skb_mac_header(skb));
1858 EXPORT_SYMBOL(skb_gso_segment);
1860 /* Take action when hardware reception checksum errors are detected. */
1862 void netdev_rx_csum_fault(struct net_device *dev)
1864 if (net_ratelimit()) {
1865 printk(KERN_ERR "%s: hw csum failure.\n",
1866 dev ? dev->name : "<unknown>");
1870 EXPORT_SYMBOL(netdev_rx_csum_fault);
1873 /* Actually, we should eliminate this check as soon as we know, that:
1874 * 1. IOMMU is present and allows to map all the memory.
1875 * 2. No high memory really exists on this machine.
1878 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1880 #ifdef CONFIG_HIGHMEM
1882 if (!(dev->features & NETIF_F_HIGHDMA)) {
1883 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1884 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1888 if (PCI_DMA_BUS_IS_PHYS) {
1889 struct device *pdev = dev->dev.parent;
1893 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1894 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1895 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1904 void (*destructor)(struct sk_buff *skb);
1907 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1909 static void dev_gso_skb_destructor(struct sk_buff *skb)
1911 struct dev_gso_cb *cb;
1914 struct sk_buff *nskb = skb->next;
1916 skb->next = nskb->next;
1919 } while (skb->next);
1921 cb = DEV_GSO_CB(skb);
1923 cb->destructor(skb);
1927 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1928 * @skb: buffer to segment
1930 * This function segments the given skb and stores the list of segments
1933 static int dev_gso_segment(struct sk_buff *skb)
1935 struct net_device *dev = skb->dev;
1936 struct sk_buff *segs;
1937 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1940 segs = skb_gso_segment(skb, features);
1942 /* Verifying header integrity only. */
1947 return PTR_ERR(segs);
1950 DEV_GSO_CB(skb)->destructor = skb->destructor;
1951 skb->destructor = dev_gso_skb_destructor;
1957 * Try to orphan skb early, right before transmission by the device.
1958 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1959 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1961 static inline void skb_orphan_try(struct sk_buff *skb)
1963 struct sock *sk = skb->sk;
1965 if (sk && !skb_shinfo(skb)->tx_flags) {
1966 /* skb_tx_hash() wont be able to get sk.
1967 * We copy sk_hash into skb->rxhash
1970 skb->rxhash = sk->sk_hash;
1975 int netif_get_vlan_features(struct sk_buff *skb, struct net_device *dev)
1977 __be16 protocol = skb->protocol;
1979 if (protocol == htons(ETH_P_8021Q)) {
1980 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1981 protocol = veh->h_vlan_encapsulated_proto;
1982 } else if (!skb->vlan_tci)
1983 return dev->features;
1985 if (protocol != htons(ETH_P_8021Q))
1986 return dev->features & dev->vlan_features;
1990 EXPORT_SYMBOL(netif_get_vlan_features);
1993 * Returns true if either:
1994 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1995 * 2. skb is fragmented and the device does not support SG, or if
1996 * at least one of fragments is in highmem and device does not
1997 * support DMA from it.
1999 static inline int skb_needs_linearize(struct sk_buff *skb,
2000 struct net_device *dev)
2002 if (skb_is_nonlinear(skb)) {
2003 int features = dev->features;
2005 if (vlan_tx_tag_present(skb))
2006 features &= dev->vlan_features;
2008 return (skb_has_frag_list(skb) &&
2009 !(features & NETIF_F_FRAGLIST)) ||
2010 (skb_shinfo(skb)->nr_frags &&
2011 (!(features & NETIF_F_SG) ||
2012 illegal_highdma(dev, skb)));
2018 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2019 struct netdev_queue *txq)
2021 const struct net_device_ops *ops = dev->netdev_ops;
2022 int rc = NETDEV_TX_OK;
2024 if (likely(!skb->next)) {
2026 * If device doesnt need skb->dst, release it right now while
2027 * its hot in this cpu cache
2029 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2032 if (!list_empty(&ptype_all))
2033 dev_queue_xmit_nit(skb, dev);
2035 skb_orphan_try(skb);
2037 if (vlan_tx_tag_present(skb) &&
2038 !(dev->features & NETIF_F_HW_VLAN_TX)) {
2039 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2046 if (netif_needs_gso(dev, skb)) {
2047 if (unlikely(dev_gso_segment(skb)))
2052 if (skb_needs_linearize(skb, dev) &&
2053 __skb_linearize(skb))
2056 /* If packet is not checksummed and device does not
2057 * support checksumming for this protocol, complete
2058 * checksumming here.
2060 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2061 skb_set_transport_header(skb, skb->csum_start -
2063 if (!dev_can_checksum(dev, skb) &&
2064 skb_checksum_help(skb))
2069 rc = ops->ndo_start_xmit(skb, dev);
2070 trace_net_dev_xmit(skb, rc);
2071 if (rc == NETDEV_TX_OK)
2072 txq_trans_update(txq);
2078 struct sk_buff *nskb = skb->next;
2080 skb->next = nskb->next;
2084 * If device doesnt need nskb->dst, release it right now while
2085 * its hot in this cpu cache
2087 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2090 rc = ops->ndo_start_xmit(nskb, dev);
2091 trace_net_dev_xmit(nskb, rc);
2092 if (unlikely(rc != NETDEV_TX_OK)) {
2093 if (rc & ~NETDEV_TX_MASK)
2094 goto out_kfree_gso_skb;
2095 nskb->next = skb->next;
2099 txq_trans_update(txq);
2100 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2101 return NETDEV_TX_BUSY;
2102 } while (skb->next);
2105 if (likely(skb->next == NULL))
2106 skb->destructor = DEV_GSO_CB(skb)->destructor;
2113 static u32 hashrnd __read_mostly;
2116 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2117 * to be used as a distribution range.
2119 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2120 unsigned int num_tx_queues)
2124 if (skb_rx_queue_recorded(skb)) {
2125 hash = skb_get_rx_queue(skb);
2126 while (unlikely(hash >= num_tx_queues))
2127 hash -= num_tx_queues;
2131 if (skb->sk && skb->sk->sk_hash)
2132 hash = skb->sk->sk_hash;
2134 hash = (__force u16) skb->protocol ^ skb->rxhash;
2135 hash = jhash_1word(hash, hashrnd);
2137 return (u16) (((u64) hash * num_tx_queues) >> 32);
2139 EXPORT_SYMBOL(__skb_tx_hash);
2141 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2143 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2144 if (net_ratelimit()) {
2145 pr_warning("%s selects TX queue %d, but "
2146 "real number of TX queues is %d\n",
2147 dev->name, queue_index, dev->real_num_tx_queues);
2154 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2157 struct xps_dev_maps *dev_maps;
2158 struct xps_map *map;
2159 int queue_index = -1;
2162 dev_maps = rcu_dereference(dev->xps_maps);
2164 map = rcu_dereference(
2165 dev_maps->cpu_map[raw_smp_processor_id()]);
2168 queue_index = map->queues[0];
2171 if (skb->sk && skb->sk->sk_hash)
2172 hash = skb->sk->sk_hash;
2174 hash = (__force u16) skb->protocol ^
2176 hash = jhash_1word(hash, hashrnd);
2177 queue_index = map->queues[
2178 ((u64)hash * map->len) >> 32];
2180 if (unlikely(queue_index >= dev->real_num_tx_queues))
2192 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2193 struct sk_buff *skb)
2196 const struct net_device_ops *ops = dev->netdev_ops;
2198 if (dev->real_num_tx_queues == 1)
2200 else if (ops->ndo_select_queue) {
2201 queue_index = ops->ndo_select_queue(dev, skb);
2202 queue_index = dev_cap_txqueue(dev, queue_index);
2204 struct sock *sk = skb->sk;
2205 queue_index = sk_tx_queue_get(sk);
2207 if (queue_index < 0 || skb->ooo_okay ||
2208 queue_index >= dev->real_num_tx_queues) {
2209 int old_index = queue_index;
2211 queue_index = get_xps_queue(dev, skb);
2212 if (queue_index < 0)
2213 queue_index = skb_tx_hash(dev, skb);
2215 if (queue_index != old_index && sk) {
2216 struct dst_entry *dst =
2217 rcu_dereference_check(sk->sk_dst_cache, 1);
2219 if (dst && skb_dst(skb) == dst)
2220 sk_tx_queue_set(sk, queue_index);
2225 skb_set_queue_mapping(skb, queue_index);
2226 return netdev_get_tx_queue(dev, queue_index);
2229 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2230 struct net_device *dev,
2231 struct netdev_queue *txq)
2233 spinlock_t *root_lock = qdisc_lock(q);
2234 bool contended = qdisc_is_running(q);
2238 * Heuristic to force contended enqueues to serialize on a
2239 * separate lock before trying to get qdisc main lock.
2240 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2241 * and dequeue packets faster.
2243 if (unlikely(contended))
2244 spin_lock(&q->busylock);
2246 spin_lock(root_lock);
2247 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2250 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2251 qdisc_run_begin(q)) {
2253 * This is a work-conserving queue; there are no old skbs
2254 * waiting to be sent out; and the qdisc is not running -
2255 * xmit the skb directly.
2257 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2259 __qdisc_update_bstats(q, skb->len);
2260 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2261 if (unlikely(contended)) {
2262 spin_unlock(&q->busylock);
2269 rc = NET_XMIT_SUCCESS;
2272 rc = qdisc_enqueue_root(skb, q);
2273 if (qdisc_run_begin(q)) {
2274 if (unlikely(contended)) {
2275 spin_unlock(&q->busylock);
2281 spin_unlock(root_lock);
2282 if (unlikely(contended))
2283 spin_unlock(&q->busylock);
2287 static DEFINE_PER_CPU(int, xmit_recursion);
2288 #define RECURSION_LIMIT 10
2291 * dev_queue_xmit - transmit a buffer
2292 * @skb: buffer to transmit
2294 * Queue a buffer for transmission to a network device. The caller must
2295 * have set the device and priority and built the buffer before calling
2296 * this function. The function can be called from an interrupt.
2298 * A negative errno code is returned on a failure. A success does not
2299 * guarantee the frame will be transmitted as it may be dropped due
2300 * to congestion or traffic shaping.
2302 * -----------------------------------------------------------------------------------
2303 * I notice this method can also return errors from the queue disciplines,
2304 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2307 * Regardless of the return value, the skb is consumed, so it is currently
2308 * difficult to retry a send to this method. (You can bump the ref count
2309 * before sending to hold a reference for retry if you are careful.)
2311 * When calling this method, interrupts MUST be enabled. This is because
2312 * the BH enable code must have IRQs enabled so that it will not deadlock.
2315 int dev_queue_xmit(struct sk_buff *skb)
2317 struct net_device *dev = skb->dev;
2318 struct netdev_queue *txq;
2322 /* Disable soft irqs for various locks below. Also
2323 * stops preemption for RCU.
2327 txq = dev_pick_tx(dev, skb);
2328 q = rcu_dereference_bh(txq->qdisc);
2330 #ifdef CONFIG_NET_CLS_ACT
2331 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2333 trace_net_dev_queue(skb);
2335 rc = __dev_xmit_skb(skb, q, dev, txq);
2339 /* The device has no queue. Common case for software devices:
2340 loopback, all the sorts of tunnels...
2342 Really, it is unlikely that netif_tx_lock protection is necessary
2343 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2345 However, it is possible, that they rely on protection
2348 Check this and shot the lock. It is not prone from deadlocks.
2349 Either shot noqueue qdisc, it is even simpler 8)
2351 if (dev->flags & IFF_UP) {
2352 int cpu = smp_processor_id(); /* ok because BHs are off */
2354 if (txq->xmit_lock_owner != cpu) {
2356 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2357 goto recursion_alert;
2359 HARD_TX_LOCK(dev, txq, cpu);
2361 if (!netif_tx_queue_stopped(txq)) {
2362 __this_cpu_inc(xmit_recursion);
2363 rc = dev_hard_start_xmit(skb, dev, txq);
2364 __this_cpu_dec(xmit_recursion);
2365 if (dev_xmit_complete(rc)) {
2366 HARD_TX_UNLOCK(dev, txq);
2370 HARD_TX_UNLOCK(dev, txq);
2371 if (net_ratelimit())
2372 printk(KERN_CRIT "Virtual device %s asks to "
2373 "queue packet!\n", dev->name);
2375 /* Recursion is detected! It is possible,
2379 if (net_ratelimit())
2380 printk(KERN_CRIT "Dead loop on virtual device "
2381 "%s, fix it urgently!\n", dev->name);
2386 rcu_read_unlock_bh();
2391 rcu_read_unlock_bh();
2394 EXPORT_SYMBOL(dev_queue_xmit);
2397 /*=======================================================================
2399 =======================================================================*/
2401 int netdev_max_backlog __read_mostly = 1000;
2402 int netdev_tstamp_prequeue __read_mostly = 1;
2403 int netdev_budget __read_mostly = 300;
2404 int weight_p __read_mostly = 64; /* old backlog weight */
2406 /* Called with irq disabled */
2407 static inline void ____napi_schedule(struct softnet_data *sd,
2408 struct napi_struct *napi)
2410 list_add_tail(&napi->poll_list, &sd->poll_list);
2411 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2415 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2416 * and src/dst port numbers. Returns a non-zero hash number on success
2419 __u32 __skb_get_rxhash(struct sk_buff *skb)
2421 int nhoff, hash = 0, poff;
2422 struct ipv6hdr *ip6;
2425 u32 addr1, addr2, ihl;
2431 nhoff = skb_network_offset(skb);
2433 switch (skb->protocol) {
2434 case __constant_htons(ETH_P_IP):
2435 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2438 ip = (struct iphdr *) (skb->data + nhoff);
2439 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2442 ip_proto = ip->protocol;
2443 addr1 = (__force u32) ip->saddr;
2444 addr2 = (__force u32) ip->daddr;
2447 case __constant_htons(ETH_P_IPV6):
2448 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2451 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2452 ip_proto = ip6->nexthdr;
2453 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2454 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2462 poff = proto_ports_offset(ip_proto);
2464 nhoff += ihl * 4 + poff;
2465 if (pskb_may_pull(skb, nhoff + 4)) {
2466 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2467 if (ports.v16[1] < ports.v16[0])
2468 swap(ports.v16[0], ports.v16[1]);
2472 /* get a consistent hash (same value on both flow directions) */
2476 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2483 EXPORT_SYMBOL(__skb_get_rxhash);
2487 /* One global table that all flow-based protocols share. */
2488 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2489 EXPORT_SYMBOL(rps_sock_flow_table);
2492 * get_rps_cpu is called from netif_receive_skb and returns the target
2493 * CPU from the RPS map of the receiving queue for a given skb.
2494 * rcu_read_lock must be held on entry.
2496 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2497 struct rps_dev_flow **rflowp)
2499 struct netdev_rx_queue *rxqueue;
2500 struct rps_map *map;
2501 struct rps_dev_flow_table *flow_table;
2502 struct rps_sock_flow_table *sock_flow_table;
2506 if (skb_rx_queue_recorded(skb)) {
2507 u16 index = skb_get_rx_queue(skb);
2508 if (unlikely(index >= dev->real_num_rx_queues)) {
2509 WARN_ONCE(dev->real_num_rx_queues > 1,
2510 "%s received packet on queue %u, but number "
2511 "of RX queues is %u\n",
2512 dev->name, index, dev->real_num_rx_queues);
2515 rxqueue = dev->_rx + index;
2519 map = rcu_dereference(rxqueue->rps_map);
2521 if (map->len == 1) {
2522 tcpu = map->cpus[0];
2523 if (cpu_online(tcpu))
2527 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2531 skb_reset_network_header(skb);
2532 if (!skb_get_rxhash(skb))
2535 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2536 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2537 if (flow_table && sock_flow_table) {
2539 struct rps_dev_flow *rflow;
2541 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2544 next_cpu = sock_flow_table->ents[skb->rxhash &
2545 sock_flow_table->mask];
2548 * If the desired CPU (where last recvmsg was done) is
2549 * different from current CPU (one in the rx-queue flow
2550 * table entry), switch if one of the following holds:
2551 * - Current CPU is unset (equal to RPS_NO_CPU).
2552 * - Current CPU is offline.
2553 * - The current CPU's queue tail has advanced beyond the
2554 * last packet that was enqueued using this table entry.
2555 * This guarantees that all previous packets for the flow
2556 * have been dequeued, thus preserving in order delivery.
2558 if (unlikely(tcpu != next_cpu) &&
2559 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2560 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2561 rflow->last_qtail)) >= 0)) {
2562 tcpu = rflow->cpu = next_cpu;
2563 if (tcpu != RPS_NO_CPU)
2564 rflow->last_qtail = per_cpu(softnet_data,
2565 tcpu).input_queue_head;
2567 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2575 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2577 if (cpu_online(tcpu)) {
2587 /* Called from hardirq (IPI) context */
2588 static void rps_trigger_softirq(void *data)
2590 struct softnet_data *sd = data;
2592 ____napi_schedule(sd, &sd->backlog);
2596 #endif /* CONFIG_RPS */
2599 * Check if this softnet_data structure is another cpu one
2600 * If yes, queue it to our IPI list and return 1
2603 static int rps_ipi_queued(struct softnet_data *sd)
2606 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2609 sd->rps_ipi_next = mysd->rps_ipi_list;
2610 mysd->rps_ipi_list = sd;
2612 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2615 #endif /* CONFIG_RPS */
2620 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2621 * queue (may be a remote CPU queue).
2623 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2624 unsigned int *qtail)
2626 struct softnet_data *sd;
2627 unsigned long flags;
2629 sd = &per_cpu(softnet_data, cpu);
2631 local_irq_save(flags);
2634 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2635 if (skb_queue_len(&sd->input_pkt_queue)) {
2637 __skb_queue_tail(&sd->input_pkt_queue, skb);
2638 input_queue_tail_incr_save(sd, qtail);
2640 local_irq_restore(flags);
2641 return NET_RX_SUCCESS;
2644 /* Schedule NAPI for backlog device
2645 * We can use non atomic operation since we own the queue lock
2647 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2648 if (!rps_ipi_queued(sd))
2649 ____napi_schedule(sd, &sd->backlog);
2657 local_irq_restore(flags);
2659 atomic_long_inc(&skb->dev->rx_dropped);
2665 * netif_rx - post buffer to the network code
2666 * @skb: buffer to post
2668 * This function receives a packet from a device driver and queues it for
2669 * the upper (protocol) levels to process. It always succeeds. The buffer
2670 * may be dropped during processing for congestion control or by the
2674 * NET_RX_SUCCESS (no congestion)
2675 * NET_RX_DROP (packet was dropped)
2679 int netif_rx(struct sk_buff *skb)
2683 /* if netpoll wants it, pretend we never saw it */
2684 if (netpoll_rx(skb))
2687 if (netdev_tstamp_prequeue)
2688 net_timestamp_check(skb);
2690 trace_netif_rx(skb);
2693 struct rps_dev_flow voidflow, *rflow = &voidflow;
2699 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2701 cpu = smp_processor_id();
2703 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2711 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2717 EXPORT_SYMBOL(netif_rx);
2719 int netif_rx_ni(struct sk_buff *skb)
2724 err = netif_rx(skb);
2725 if (local_softirq_pending())
2731 EXPORT_SYMBOL(netif_rx_ni);
2733 static void net_tx_action(struct softirq_action *h)
2735 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2737 if (sd->completion_queue) {
2738 struct sk_buff *clist;
2740 local_irq_disable();
2741 clist = sd->completion_queue;
2742 sd->completion_queue = NULL;
2746 struct sk_buff *skb = clist;
2747 clist = clist->next;
2749 WARN_ON(atomic_read(&skb->users));
2750 trace_kfree_skb(skb, net_tx_action);
2755 if (sd->output_queue) {
2758 local_irq_disable();
2759 head = sd->output_queue;
2760 sd->output_queue = NULL;
2761 sd->output_queue_tailp = &sd->output_queue;
2765 struct Qdisc *q = head;
2766 spinlock_t *root_lock;
2768 head = head->next_sched;
2770 root_lock = qdisc_lock(q);
2771 if (spin_trylock(root_lock)) {
2772 smp_mb__before_clear_bit();
2773 clear_bit(__QDISC_STATE_SCHED,
2776 spin_unlock(root_lock);
2778 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2780 __netif_reschedule(q);
2782 smp_mb__before_clear_bit();
2783 clear_bit(__QDISC_STATE_SCHED,
2791 static inline int deliver_skb(struct sk_buff *skb,
2792 struct packet_type *pt_prev,
2793 struct net_device *orig_dev)
2795 atomic_inc(&skb->users);
2796 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2799 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2800 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2801 /* This hook is defined here for ATM LANE */
2802 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2803 unsigned char *addr) __read_mostly;
2804 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2807 #ifdef CONFIG_NET_CLS_ACT
2808 /* TODO: Maybe we should just force sch_ingress to be compiled in
2809 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2810 * a compare and 2 stores extra right now if we dont have it on
2811 * but have CONFIG_NET_CLS_ACT
2812 * NOTE: This doesnt stop any functionality; if you dont have
2813 * the ingress scheduler, you just cant add policies on ingress.
2816 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2818 struct net_device *dev = skb->dev;
2819 u32 ttl = G_TC_RTTL(skb->tc_verd);
2820 int result = TC_ACT_OK;
2823 if (unlikely(MAX_RED_LOOP < ttl++)) {
2824 if (net_ratelimit())
2825 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2826 skb->skb_iif, dev->ifindex);
2830 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2831 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2834 if (q != &noop_qdisc) {
2835 spin_lock(qdisc_lock(q));
2836 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2837 result = qdisc_enqueue_root(skb, q);
2838 spin_unlock(qdisc_lock(q));
2844 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2845 struct packet_type **pt_prev,
2846 int *ret, struct net_device *orig_dev)
2848 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
2850 if (!rxq || rxq->qdisc == &noop_qdisc)
2854 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2858 switch (ing_filter(skb, rxq)) {
2872 * netdev_rx_handler_register - register receive handler
2873 * @dev: device to register a handler for
2874 * @rx_handler: receive handler to register
2875 * @rx_handler_data: data pointer that is used by rx handler
2877 * Register a receive hander for a device. This handler will then be
2878 * called from __netif_receive_skb. A negative errno code is returned
2881 * The caller must hold the rtnl_mutex.
2883 int netdev_rx_handler_register(struct net_device *dev,
2884 rx_handler_func_t *rx_handler,
2885 void *rx_handler_data)
2889 if (dev->rx_handler)
2892 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2893 rcu_assign_pointer(dev->rx_handler, rx_handler);
2897 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2900 * netdev_rx_handler_unregister - unregister receive handler
2901 * @dev: device to unregister a handler from
2903 * Unregister a receive hander from a device.
2905 * The caller must hold the rtnl_mutex.
2907 void netdev_rx_handler_unregister(struct net_device *dev)
2911 rcu_assign_pointer(dev->rx_handler, NULL);
2912 rcu_assign_pointer(dev->rx_handler_data, NULL);
2914 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2916 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2917 struct net_device *master)
2919 if (skb->pkt_type == PACKET_HOST) {
2920 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2922 memcpy(dest, master->dev_addr, ETH_ALEN);
2926 /* On bonding slaves other than the currently active slave, suppress
2927 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2928 * ARP on active-backup slaves with arp_validate enabled.
2930 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2932 struct net_device *dev = skb->dev;
2934 if (master->priv_flags & IFF_MASTER_ARPMON)
2935 dev->last_rx = jiffies;
2937 if ((master->priv_flags & IFF_MASTER_ALB) &&
2938 (master->priv_flags & IFF_BRIDGE_PORT)) {
2939 /* Do address unmangle. The local destination address
2940 * will be always the one master has. Provides the right
2941 * functionality in a bridge.
2943 skb_bond_set_mac_by_master(skb, master);
2946 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2947 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2948 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2951 if (master->priv_flags & IFF_MASTER_ALB) {
2952 if (skb->pkt_type != PACKET_BROADCAST &&
2953 skb->pkt_type != PACKET_MULTICAST)
2956 if (master->priv_flags & IFF_MASTER_8023AD &&
2957 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2964 EXPORT_SYMBOL(__skb_bond_should_drop);
2966 static int __netif_receive_skb(struct sk_buff *skb)
2968 struct packet_type *ptype, *pt_prev;
2969 rx_handler_func_t *rx_handler;
2970 struct net_device *orig_dev;
2971 struct net_device *master;
2972 struct net_device *null_or_orig;
2973 struct net_device *orig_or_bond;
2974 int ret = NET_RX_DROP;
2977 if (!netdev_tstamp_prequeue)
2978 net_timestamp_check(skb);
2980 trace_netif_receive_skb(skb);
2982 /* if we've gotten here through NAPI, check netpoll */
2983 if (netpoll_receive_skb(skb))
2987 skb->skb_iif = skb->dev->ifindex;
2990 * bonding note: skbs received on inactive slaves should only
2991 * be delivered to pkt handlers that are exact matches. Also
2992 * the deliver_no_wcard flag will be set. If packet handlers
2993 * are sensitive to duplicate packets these skbs will need to
2994 * be dropped at the handler.
2996 null_or_orig = NULL;
2997 orig_dev = skb->dev;
2998 master = ACCESS_ONCE(orig_dev->master);
2999 if (skb->deliver_no_wcard)
3000 null_or_orig = orig_dev;
3002 if (skb_bond_should_drop(skb, master)) {
3003 skb->deliver_no_wcard = 1;
3004 null_or_orig = orig_dev; /* deliver only exact match */
3009 __this_cpu_inc(softnet_data.processed);
3010 skb_reset_network_header(skb);
3011 skb_reset_transport_header(skb);
3012 skb->mac_len = skb->network_header - skb->mac_header;
3018 #ifdef CONFIG_NET_CLS_ACT
3019 if (skb->tc_verd & TC_NCLS) {
3020 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3025 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3026 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
3027 ptype->dev == orig_dev) {
3029 ret = deliver_skb(skb, pt_prev, orig_dev);
3034 #ifdef CONFIG_NET_CLS_ACT
3035 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3041 /* Handle special case of bridge or macvlan */
3042 rx_handler = rcu_dereference(skb->dev->rx_handler);
3045 ret = deliver_skb(skb, pt_prev, orig_dev);
3048 skb = rx_handler(skb);
3053 if (vlan_tx_tag_present(skb)) {
3055 ret = deliver_skb(skb, pt_prev, orig_dev);
3058 if (vlan_hwaccel_do_receive(&skb)) {
3059 ret = __netif_receive_skb(skb);
3061 } else if (unlikely(!skb))
3066 * Make sure frames received on VLAN interfaces stacked on
3067 * bonding interfaces still make their way to any base bonding
3068 * device that may have registered for a specific ptype. The
3069 * handler may have to adjust skb->dev and orig_dev.
3071 orig_or_bond = orig_dev;
3072 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
3073 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
3074 orig_or_bond = vlan_dev_real_dev(skb->dev);
3077 type = skb->protocol;
3078 list_for_each_entry_rcu(ptype,
3079 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3080 if (ptype->type == type && (ptype->dev == null_or_orig ||
3081 ptype->dev == skb->dev || ptype->dev == orig_dev ||
3082 ptype->dev == orig_or_bond)) {
3084 ret = deliver_skb(skb, pt_prev, orig_dev);
3090 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3092 atomic_long_inc(&skb->dev->rx_dropped);
3094 /* Jamal, now you will not able to escape explaining
3095 * me how you were going to use this. :-)
3106 * netif_receive_skb - process receive buffer from network
3107 * @skb: buffer to process
3109 * netif_receive_skb() is the main receive data processing function.
3110 * It always succeeds. The buffer may be dropped during processing
3111 * for congestion control or by the protocol layers.
3113 * This function may only be called from softirq context and interrupts
3114 * should be enabled.
3116 * Return values (usually ignored):
3117 * NET_RX_SUCCESS: no congestion
3118 * NET_RX_DROP: packet was dropped
3120 int netif_receive_skb(struct sk_buff *skb)
3122 if (netdev_tstamp_prequeue)
3123 net_timestamp_check(skb);
3125 if (skb_defer_rx_timestamp(skb))
3126 return NET_RX_SUCCESS;
3130 struct rps_dev_flow voidflow, *rflow = &voidflow;
3135 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3138 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3142 ret = __netif_receive_skb(skb);
3148 return __netif_receive_skb(skb);
3151 EXPORT_SYMBOL(netif_receive_skb);
3153 /* Network device is going away, flush any packets still pending
3154 * Called with irqs disabled.
3156 static void flush_backlog(void *arg)
3158 struct net_device *dev = arg;
3159 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3160 struct sk_buff *skb, *tmp;
3163 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3164 if (skb->dev == dev) {
3165 __skb_unlink(skb, &sd->input_pkt_queue);
3167 input_queue_head_incr(sd);
3172 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3173 if (skb->dev == dev) {
3174 __skb_unlink(skb, &sd->process_queue);
3176 input_queue_head_incr(sd);
3181 static int napi_gro_complete(struct sk_buff *skb)
3183 struct packet_type *ptype;
3184 __be16 type = skb->protocol;
3185 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3188 if (NAPI_GRO_CB(skb)->count == 1) {
3189 skb_shinfo(skb)->gso_size = 0;
3194 list_for_each_entry_rcu(ptype, head, list) {
3195 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3198 err = ptype->gro_complete(skb);
3204 WARN_ON(&ptype->list == head);
3206 return NET_RX_SUCCESS;
3210 return netif_receive_skb(skb);
3213 inline void napi_gro_flush(struct napi_struct *napi)
3215 struct sk_buff *skb, *next;
3217 for (skb = napi->gro_list; skb; skb = next) {
3220 napi_gro_complete(skb);
3223 napi->gro_count = 0;
3224 napi->gro_list = NULL;
3226 EXPORT_SYMBOL(napi_gro_flush);
3228 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3230 struct sk_buff **pp = NULL;
3231 struct packet_type *ptype;
3232 __be16 type = skb->protocol;
3233 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3236 enum gro_result ret;
3238 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3241 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3245 list_for_each_entry_rcu(ptype, head, list) {
3246 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3249 skb_set_network_header(skb, skb_gro_offset(skb));
3250 mac_len = skb->network_header - skb->mac_header;
3251 skb->mac_len = mac_len;
3252 NAPI_GRO_CB(skb)->same_flow = 0;
3253 NAPI_GRO_CB(skb)->flush = 0;
3254 NAPI_GRO_CB(skb)->free = 0;
3256 pp = ptype->gro_receive(&napi->gro_list, skb);
3261 if (&ptype->list == head)
3264 same_flow = NAPI_GRO_CB(skb)->same_flow;
3265 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3268 struct sk_buff *nskb = *pp;
3272 napi_gro_complete(nskb);
3279 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3283 NAPI_GRO_CB(skb)->count = 1;
3284 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3285 skb->next = napi->gro_list;
3286 napi->gro_list = skb;
3290 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3291 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3293 BUG_ON(skb->end - skb->tail < grow);
3295 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3298 skb->data_len -= grow;
3300 skb_shinfo(skb)->frags[0].page_offset += grow;
3301 skb_shinfo(skb)->frags[0].size -= grow;
3303 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3304 put_page(skb_shinfo(skb)->frags[0].page);
3305 memmove(skb_shinfo(skb)->frags,
3306 skb_shinfo(skb)->frags + 1,
3307 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3318 EXPORT_SYMBOL(dev_gro_receive);
3320 static inline gro_result_t
3321 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3325 for (p = napi->gro_list; p; p = p->next) {
3326 unsigned long diffs;
3328 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3329 diffs |= p->vlan_tci ^ skb->vlan_tci;
3330 diffs |= compare_ether_header(skb_mac_header(p),
3331 skb_gro_mac_header(skb));
3332 NAPI_GRO_CB(p)->same_flow = !diffs;
3333 NAPI_GRO_CB(p)->flush = 0;
3336 return dev_gro_receive(napi, skb);
3339 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3343 if (netif_receive_skb(skb))
3348 case GRO_MERGED_FREE:
3359 EXPORT_SYMBOL(napi_skb_finish);
3361 void skb_gro_reset_offset(struct sk_buff *skb)
3363 NAPI_GRO_CB(skb)->data_offset = 0;
3364 NAPI_GRO_CB(skb)->frag0 = NULL;
3365 NAPI_GRO_CB(skb)->frag0_len = 0;
3367 if (skb->mac_header == skb->tail &&
3368 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3369 NAPI_GRO_CB(skb)->frag0 =
3370 page_address(skb_shinfo(skb)->frags[0].page) +
3371 skb_shinfo(skb)->frags[0].page_offset;
3372 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3375 EXPORT_SYMBOL(skb_gro_reset_offset);
3377 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3379 skb_gro_reset_offset(skb);
3381 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3383 EXPORT_SYMBOL(napi_gro_receive);
3385 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3387 __skb_pull(skb, skb_headlen(skb));
3388 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3394 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3396 struct sk_buff *skb = napi->skb;
3399 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3405 EXPORT_SYMBOL(napi_get_frags);
3407 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3413 skb->protocol = eth_type_trans(skb, skb->dev);
3415 if (ret == GRO_HELD)
3416 skb_gro_pull(skb, -ETH_HLEN);
3417 else if (netif_receive_skb(skb))
3422 case GRO_MERGED_FREE:
3423 napi_reuse_skb(napi, skb);
3432 EXPORT_SYMBOL(napi_frags_finish);
3434 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3436 struct sk_buff *skb = napi->skb;
3443 skb_reset_mac_header(skb);
3444 skb_gro_reset_offset(skb);
3446 off = skb_gro_offset(skb);
3447 hlen = off + sizeof(*eth);
3448 eth = skb_gro_header_fast(skb, off);
3449 if (skb_gro_header_hard(skb, hlen)) {
3450 eth = skb_gro_header_slow(skb, hlen, off);
3451 if (unlikely(!eth)) {
3452 napi_reuse_skb(napi, skb);
3458 skb_gro_pull(skb, sizeof(*eth));
3461 * This works because the only protocols we care about don't require
3462 * special handling. We'll fix it up properly at the end.
3464 skb->protocol = eth->h_proto;
3469 EXPORT_SYMBOL(napi_frags_skb);
3471 gro_result_t napi_gro_frags(struct napi_struct *napi)
3473 struct sk_buff *skb = napi_frags_skb(napi);
3478 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3480 EXPORT_SYMBOL(napi_gro_frags);
3483 * net_rps_action sends any pending IPI's for rps.
3484 * Note: called with local irq disabled, but exits with local irq enabled.
3486 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3489 struct softnet_data *remsd = sd->rps_ipi_list;
3492 sd->rps_ipi_list = NULL;
3496 /* Send pending IPI's to kick RPS processing on remote cpus. */
3498 struct softnet_data *next = remsd->rps_ipi_next;
3500 if (cpu_online(remsd->cpu))
3501 __smp_call_function_single(remsd->cpu,
3510 static int process_backlog(struct napi_struct *napi, int quota)
3513 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3516 /* Check if we have pending ipi, its better to send them now,
3517 * not waiting net_rx_action() end.
3519 if (sd->rps_ipi_list) {
3520 local_irq_disable();
3521 net_rps_action_and_irq_enable(sd);
3524 napi->weight = weight_p;
3525 local_irq_disable();
3526 while (work < quota) {
3527 struct sk_buff *skb;
3530 while ((skb = __skb_dequeue(&sd->process_queue))) {
3532 __netif_receive_skb(skb);
3533 local_irq_disable();
3534 input_queue_head_incr(sd);
3535 if (++work >= quota) {
3542 qlen = skb_queue_len(&sd->input_pkt_queue);
3544 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3545 &sd->process_queue);
3547 if (qlen < quota - work) {
3549 * Inline a custom version of __napi_complete().
3550 * only current cpu owns and manipulates this napi,
3551 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3552 * we can use a plain write instead of clear_bit(),
3553 * and we dont need an smp_mb() memory barrier.
3555 list_del(&napi->poll_list);
3558 quota = work + qlen;
3568 * __napi_schedule - schedule for receive
3569 * @n: entry to schedule
3571 * The entry's receive function will be scheduled to run
3573 void __napi_schedule(struct napi_struct *n)
3575 unsigned long flags;
3577 local_irq_save(flags);
3578 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3579 local_irq_restore(flags);
3581 EXPORT_SYMBOL(__napi_schedule);
3583 void __napi_complete(struct napi_struct *n)
3585 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3586 BUG_ON(n->gro_list);
3588 list_del(&n->poll_list);
3589 smp_mb__before_clear_bit();
3590 clear_bit(NAPI_STATE_SCHED, &n->state);
3592 EXPORT_SYMBOL(__napi_complete);
3594 void napi_complete(struct napi_struct *n)
3596 unsigned long flags;
3599 * don't let napi dequeue from the cpu poll list
3600 * just in case its running on a different cpu
3602 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3606 local_irq_save(flags);
3608 local_irq_restore(flags);
3610 EXPORT_SYMBOL(napi_complete);
3612 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3613 int (*poll)(struct napi_struct *, int), int weight)
3615 INIT_LIST_HEAD(&napi->poll_list);
3616 napi->gro_count = 0;
3617 napi->gro_list = NULL;
3620 napi->weight = weight;
3621 list_add(&napi->dev_list, &dev->napi_list);
3623 #ifdef CONFIG_NETPOLL
3624 spin_lock_init(&napi->poll_lock);
3625 napi->poll_owner = -1;
3627 set_bit(NAPI_STATE_SCHED, &napi->state);
3629 EXPORT_SYMBOL(netif_napi_add);
3631 void netif_napi_del(struct napi_struct *napi)
3633 struct sk_buff *skb, *next;
3635 list_del_init(&napi->dev_list);
3636 napi_free_frags(napi);
3638 for (skb = napi->gro_list; skb; skb = next) {
3644 napi->gro_list = NULL;
3645 napi->gro_count = 0;
3647 EXPORT_SYMBOL(netif_napi_del);
3649 static void net_rx_action(struct softirq_action *h)
3651 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3652 unsigned long time_limit = jiffies + 2;
3653 int budget = netdev_budget;
3656 local_irq_disable();
3658 while (!list_empty(&sd->poll_list)) {
3659 struct napi_struct *n;
3662 /* If softirq window is exhuasted then punt.
3663 * Allow this to run for 2 jiffies since which will allow
3664 * an average latency of 1.5/HZ.
3666 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3671 /* Even though interrupts have been re-enabled, this
3672 * access is safe because interrupts can only add new
3673 * entries to the tail of this list, and only ->poll()
3674 * calls can remove this head entry from the list.
3676 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3678 have = netpoll_poll_lock(n);
3682 /* This NAPI_STATE_SCHED test is for avoiding a race
3683 * with netpoll's poll_napi(). Only the entity which
3684 * obtains the lock and sees NAPI_STATE_SCHED set will
3685 * actually make the ->poll() call. Therefore we avoid
3686 * accidently calling ->poll() when NAPI is not scheduled.
3689 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3690 work = n->poll(n, weight);
3694 WARN_ON_ONCE(work > weight);
3698 local_irq_disable();
3700 /* Drivers must not modify the NAPI state if they
3701 * consume the entire weight. In such cases this code
3702 * still "owns" the NAPI instance and therefore can
3703 * move the instance around on the list at-will.
3705 if (unlikely(work == weight)) {
3706 if (unlikely(napi_disable_pending(n))) {
3709 local_irq_disable();
3711 list_move_tail(&n->poll_list, &sd->poll_list);
3714 netpoll_poll_unlock(have);
3717 net_rps_action_and_irq_enable(sd);
3719 #ifdef CONFIG_NET_DMA
3721 * There may not be any more sk_buffs coming right now, so push
3722 * any pending DMA copies to hardware
3724 dma_issue_pending_all();
3731 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3735 static gifconf_func_t *gifconf_list[NPROTO];
3738 * register_gifconf - register a SIOCGIF handler
3739 * @family: Address family
3740 * @gifconf: Function handler
3742 * Register protocol dependent address dumping routines. The handler
3743 * that is passed must not be freed or reused until it has been replaced
3744 * by another handler.
3746 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3748 if (family >= NPROTO)
3750 gifconf_list[family] = gifconf;
3753 EXPORT_SYMBOL(register_gifconf);
3757 * Map an interface index to its name (SIOCGIFNAME)
3761 * We need this ioctl for efficient implementation of the
3762 * if_indextoname() function required by the IPv6 API. Without
3763 * it, we would have to search all the interfaces to find a
3767 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3769 struct net_device *dev;
3773 * Fetch the caller's info block.
3776 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3780 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3786 strcpy(ifr.ifr_name, dev->name);
3789 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3795 * Perform a SIOCGIFCONF call. This structure will change
3796 * size eventually, and there is nothing I can do about it.
3797 * Thus we will need a 'compatibility mode'.
3800 static int dev_ifconf(struct net *net, char __user *arg)
3803 struct net_device *dev;
3810 * Fetch the caller's info block.
3813 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3820 * Loop over the interfaces, and write an info block for each.
3824 for_each_netdev(net, dev) {
3825 for (i = 0; i < NPROTO; i++) {
3826 if (gifconf_list[i]) {
3829 done = gifconf_list[i](dev, NULL, 0);
3831 done = gifconf_list[i](dev, pos + total,
3841 * All done. Write the updated control block back to the caller.
3843 ifc.ifc_len = total;
3846 * Both BSD and Solaris return 0 here, so we do too.
3848 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3851 #ifdef CONFIG_PROC_FS
3853 * This is invoked by the /proc filesystem handler to display a device
3856 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3859 struct net *net = seq_file_net(seq);
3861 struct net_device *dev;
3865 return SEQ_START_TOKEN;
3868 for_each_netdev_rcu(net, dev)
3875 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3877 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3878 first_net_device(seq_file_net(seq)) :
3879 next_net_device((struct net_device *)v);
3882 return rcu_dereference(dev);
3885 void dev_seq_stop(struct seq_file *seq, void *v)
3891 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3893 struct rtnl_link_stats64 temp;
3894 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3896 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3897 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3898 dev->name, stats->rx_bytes, stats->rx_packets,
3900 stats->rx_dropped + stats->rx_missed_errors,
3901 stats->rx_fifo_errors,
3902 stats->rx_length_errors + stats->rx_over_errors +
3903 stats->rx_crc_errors + stats->rx_frame_errors,
3904 stats->rx_compressed, stats->multicast,
3905 stats->tx_bytes, stats->tx_packets,
3906 stats->tx_errors, stats->tx_dropped,
3907 stats->tx_fifo_errors, stats->collisions,
3908 stats->tx_carrier_errors +
3909 stats->tx_aborted_errors +
3910 stats->tx_window_errors +
3911 stats->tx_heartbeat_errors,
3912 stats->tx_compressed);
3916 * Called from the PROCfs module. This now uses the new arbitrary sized
3917 * /proc/net interface to create /proc/net/dev
3919 static int dev_seq_show(struct seq_file *seq, void *v)
3921 if (v == SEQ_START_TOKEN)
3922 seq_puts(seq, "Inter-| Receive "
3924 " face |bytes packets errs drop fifo frame "
3925 "compressed multicast|bytes packets errs "
3926 "drop fifo colls carrier compressed\n");
3928 dev_seq_printf_stats(seq, v);
3932 static struct softnet_data *softnet_get_online(loff_t *pos)
3934 struct softnet_data *sd = NULL;
3936 while (*pos < nr_cpu_ids)
3937 if (cpu_online(*pos)) {
3938 sd = &per_cpu(softnet_data, *pos);
3945 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3947 return softnet_get_online(pos);
3950 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3953 return softnet_get_online(pos);
3956 static void softnet_seq_stop(struct seq_file *seq, void *v)
3960 static int softnet_seq_show(struct seq_file *seq, void *v)
3962 struct softnet_data *sd = v;
3964 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3965 sd->processed, sd->dropped, sd->time_squeeze, 0,
3966 0, 0, 0, 0, /* was fastroute */
3967 sd->cpu_collision, sd->received_rps);
3971 static const struct seq_operations dev_seq_ops = {
3972 .start = dev_seq_start,
3973 .next = dev_seq_next,
3974 .stop = dev_seq_stop,
3975 .show = dev_seq_show,
3978 static int dev_seq_open(struct inode *inode, struct file *file)
3980 return seq_open_net(inode, file, &dev_seq_ops,
3981 sizeof(struct seq_net_private));
3984 static const struct file_operations dev_seq_fops = {
3985 .owner = THIS_MODULE,
3986 .open = dev_seq_open,
3988 .llseek = seq_lseek,
3989 .release = seq_release_net,
3992 static const struct seq_operations softnet_seq_ops = {
3993 .start = softnet_seq_start,
3994 .next = softnet_seq_next,
3995 .stop = softnet_seq_stop,
3996 .show = softnet_seq_show,
3999 static int softnet_seq_open(struct inode *inode, struct file *file)
4001 return seq_open(file, &softnet_seq_ops);
4004 static const struct file_operations softnet_seq_fops = {
4005 .owner = THIS_MODULE,
4006 .open = softnet_seq_open,
4008 .llseek = seq_lseek,
4009 .release = seq_release,
4012 static void *ptype_get_idx(loff_t pos)
4014 struct packet_type *pt = NULL;
4018 list_for_each_entry_rcu(pt, &ptype_all, list) {
4024 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4025 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4034 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4038 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4041 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4043 struct packet_type *pt;
4044 struct list_head *nxt;
4048 if (v == SEQ_START_TOKEN)
4049 return ptype_get_idx(0);
4052 nxt = pt->list.next;
4053 if (pt->type == htons(ETH_P_ALL)) {
4054 if (nxt != &ptype_all)
4057 nxt = ptype_base[0].next;
4059 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4061 while (nxt == &ptype_base[hash]) {
4062 if (++hash >= PTYPE_HASH_SIZE)
4064 nxt = ptype_base[hash].next;
4067 return list_entry(nxt, struct packet_type, list);
4070 static void ptype_seq_stop(struct seq_file *seq, void *v)
4076 static int ptype_seq_show(struct seq_file *seq, void *v)
4078 struct packet_type *pt = v;
4080 if (v == SEQ_START_TOKEN)
4081 seq_puts(seq, "Type Device Function\n");
4082 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4083 if (pt->type == htons(ETH_P_ALL))
4084 seq_puts(seq, "ALL ");
4086 seq_printf(seq, "%04x", ntohs(pt->type));
4088 seq_printf(seq, " %-8s %pF\n",
4089 pt->dev ? pt->dev->name : "", pt->func);
4095 static const struct seq_operations ptype_seq_ops = {
4096 .start = ptype_seq_start,
4097 .next = ptype_seq_next,
4098 .stop = ptype_seq_stop,
4099 .show = ptype_seq_show,
4102 static int ptype_seq_open(struct inode *inode, struct file *file)
4104 return seq_open_net(inode, file, &ptype_seq_ops,
4105 sizeof(struct seq_net_private));
4108 static const struct file_operations ptype_seq_fops = {
4109 .owner = THIS_MODULE,
4110 .open = ptype_seq_open,
4112 .llseek = seq_lseek,
4113 .release = seq_release_net,
4117 static int __net_init dev_proc_net_init(struct net *net)
4121 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4123 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4125 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4128 if (wext_proc_init(net))
4134 proc_net_remove(net, "ptype");
4136 proc_net_remove(net, "softnet_stat");
4138 proc_net_remove(net, "dev");
4142 static void __net_exit dev_proc_net_exit(struct net *net)
4144 wext_proc_exit(net);
4146 proc_net_remove(net, "ptype");
4147 proc_net_remove(net, "softnet_stat");
4148 proc_net_remove(net, "dev");
4151 static struct pernet_operations __net_initdata dev_proc_ops = {
4152 .init = dev_proc_net_init,
4153 .exit = dev_proc_net_exit,
4156 static int __init dev_proc_init(void)
4158 return register_pernet_subsys(&dev_proc_ops);
4161 #define dev_proc_init() 0
4162 #endif /* CONFIG_PROC_FS */
4166 * netdev_set_master - set up master/slave pair
4167 * @slave: slave device
4168 * @master: new master device
4170 * Changes the master device of the slave. Pass %NULL to break the
4171 * bonding. The caller must hold the RTNL semaphore. On a failure
4172 * a negative errno code is returned. On success the reference counts
4173 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4174 * function returns zero.
4176 int netdev_set_master(struct net_device *slave, struct net_device *master)
4178 struct net_device *old = slave->master;
4188 slave->master = master;
4195 slave->flags |= IFF_SLAVE;
4197 slave->flags &= ~IFF_SLAVE;
4199 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4202 EXPORT_SYMBOL(netdev_set_master);
4204 static void dev_change_rx_flags(struct net_device *dev, int flags)
4206 const struct net_device_ops *ops = dev->netdev_ops;
4208 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4209 ops->ndo_change_rx_flags(dev, flags);
4212 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4214 unsigned short old_flags = dev->flags;
4220 dev->flags |= IFF_PROMISC;
4221 dev->promiscuity += inc;
4222 if (dev->promiscuity == 0) {
4225 * If inc causes overflow, untouch promisc and return error.
4228 dev->flags &= ~IFF_PROMISC;
4230 dev->promiscuity -= inc;
4231 printk(KERN_WARNING "%s: promiscuity touches roof, "
4232 "set promiscuity failed, promiscuity feature "
4233 "of device might be broken.\n", dev->name);
4237 if (dev->flags != old_flags) {
4238 printk(KERN_INFO "device %s %s promiscuous mode\n",
4239 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4241 if (audit_enabled) {
4242 current_uid_gid(&uid, &gid);
4243 audit_log(current->audit_context, GFP_ATOMIC,
4244 AUDIT_ANOM_PROMISCUOUS,
4245 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4246 dev->name, (dev->flags & IFF_PROMISC),
4247 (old_flags & IFF_PROMISC),
4248 audit_get_loginuid(current),
4250 audit_get_sessionid(current));
4253 dev_change_rx_flags(dev, IFF_PROMISC);
4259 * dev_set_promiscuity - update promiscuity count on a device
4263 * Add or remove promiscuity from a device. While the count in the device
4264 * remains above zero the interface remains promiscuous. Once it hits zero
4265 * the device reverts back to normal filtering operation. A negative inc
4266 * value is used to drop promiscuity on the device.
4267 * Return 0 if successful or a negative errno code on error.
4269 int dev_set_promiscuity(struct net_device *dev, int inc)
4271 unsigned short old_flags = dev->flags;
4274 err = __dev_set_promiscuity(dev, inc);
4277 if (dev->flags != old_flags)
4278 dev_set_rx_mode(dev);
4281 EXPORT_SYMBOL(dev_set_promiscuity);
4284 * dev_set_allmulti - update allmulti count on a device
4288 * Add or remove reception of all multicast frames to a device. While the
4289 * count in the device remains above zero the interface remains listening
4290 * to all interfaces. Once it hits zero the device reverts back to normal
4291 * filtering operation. A negative @inc value is used to drop the counter
4292 * when releasing a resource needing all multicasts.
4293 * Return 0 if successful or a negative errno code on error.
4296 int dev_set_allmulti(struct net_device *dev, int inc)
4298 unsigned short old_flags = dev->flags;
4302 dev->flags |= IFF_ALLMULTI;
4303 dev->allmulti += inc;
4304 if (dev->allmulti == 0) {
4307 * If inc causes overflow, untouch allmulti and return error.
4310 dev->flags &= ~IFF_ALLMULTI;
4312 dev->allmulti -= inc;
4313 printk(KERN_WARNING "%s: allmulti touches roof, "
4314 "set allmulti failed, allmulti feature of "
4315 "device might be broken.\n", dev->name);
4319 if (dev->flags ^ old_flags) {
4320 dev_change_rx_flags(dev, IFF_ALLMULTI);
4321 dev_set_rx_mode(dev);
4325 EXPORT_SYMBOL(dev_set_allmulti);
4328 * Upload unicast and multicast address lists to device and
4329 * configure RX filtering. When the device doesn't support unicast
4330 * filtering it is put in promiscuous mode while unicast addresses
4333 void __dev_set_rx_mode(struct net_device *dev)
4335 const struct net_device_ops *ops = dev->netdev_ops;
4337 /* dev_open will call this function so the list will stay sane. */
4338 if (!(dev->flags&IFF_UP))
4341 if (!netif_device_present(dev))
4344 if (ops->ndo_set_rx_mode)
4345 ops->ndo_set_rx_mode(dev);
4347 /* Unicast addresses changes may only happen under the rtnl,
4348 * therefore calling __dev_set_promiscuity here is safe.
4350 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4351 __dev_set_promiscuity(dev, 1);
4352 dev->uc_promisc = 1;
4353 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4354 __dev_set_promiscuity(dev, -1);
4355 dev->uc_promisc = 0;
4358 if (ops->ndo_set_multicast_list)
4359 ops->ndo_set_multicast_list(dev);
4363 void dev_set_rx_mode(struct net_device *dev)
4365 netif_addr_lock_bh(dev);
4366 __dev_set_rx_mode(dev);
4367 netif_addr_unlock_bh(dev);
4371 * dev_get_flags - get flags reported to userspace
4374 * Get the combination of flag bits exported through APIs to userspace.
4376 unsigned dev_get_flags(const struct net_device *dev)
4380 flags = (dev->flags & ~(IFF_PROMISC |
4385 (dev->gflags & (IFF_PROMISC |
4388 if (netif_running(dev)) {
4389 if (netif_oper_up(dev))
4390 flags |= IFF_RUNNING;
4391 if (netif_carrier_ok(dev))
4392 flags |= IFF_LOWER_UP;
4393 if (netif_dormant(dev))
4394 flags |= IFF_DORMANT;
4399 EXPORT_SYMBOL(dev_get_flags);
4401 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4403 int old_flags = dev->flags;
4409 * Set the flags on our device.
4412 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4413 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4415 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4419 * Load in the correct multicast list now the flags have changed.
4422 if ((old_flags ^ flags) & IFF_MULTICAST)
4423 dev_change_rx_flags(dev, IFF_MULTICAST);
4425 dev_set_rx_mode(dev);
4428 * Have we downed the interface. We handle IFF_UP ourselves
4429 * according to user attempts to set it, rather than blindly
4434 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4435 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4438 dev_set_rx_mode(dev);
4441 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4442 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4444 dev->gflags ^= IFF_PROMISC;
4445 dev_set_promiscuity(dev, inc);
4448 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4449 is important. Some (broken) drivers set IFF_PROMISC, when
4450 IFF_ALLMULTI is requested not asking us and not reporting.
4452 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4453 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4455 dev->gflags ^= IFF_ALLMULTI;
4456 dev_set_allmulti(dev, inc);
4462 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4464 unsigned int changes = dev->flags ^ old_flags;
4466 if (changes & IFF_UP) {
4467 if (dev->flags & IFF_UP)
4468 call_netdevice_notifiers(NETDEV_UP, dev);
4470 call_netdevice_notifiers(NETDEV_DOWN, dev);
4473 if (dev->flags & IFF_UP &&
4474 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4475 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4479 * dev_change_flags - change device settings
4481 * @flags: device state flags
4483 * Change settings on device based state flags. The flags are
4484 * in the userspace exported format.
4486 int dev_change_flags(struct net_device *dev, unsigned flags)
4489 int old_flags = dev->flags;
4491 ret = __dev_change_flags(dev, flags);
4495 changes = old_flags ^ dev->flags;
4497 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4499 __dev_notify_flags(dev, old_flags);
4502 EXPORT_SYMBOL(dev_change_flags);
4505 * dev_set_mtu - Change maximum transfer unit
4507 * @new_mtu: new transfer unit
4509 * Change the maximum transfer size of the network device.
4511 int dev_set_mtu(struct net_device *dev, int new_mtu)
4513 const struct net_device_ops *ops = dev->netdev_ops;
4516 if (new_mtu == dev->mtu)
4519 /* MTU must be positive. */
4523 if (!netif_device_present(dev))
4527 if (ops->ndo_change_mtu)
4528 err = ops->ndo_change_mtu(dev, new_mtu);
4532 if (!err && dev->flags & IFF_UP)
4533 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4536 EXPORT_SYMBOL(dev_set_mtu);
4539 * dev_set_mac_address - Change Media Access Control Address
4543 * Change the hardware (MAC) address of the device
4545 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4547 const struct net_device_ops *ops = dev->netdev_ops;
4550 if (!ops->ndo_set_mac_address)
4552 if (sa->sa_family != dev->type)
4554 if (!netif_device_present(dev))
4556 err = ops->ndo_set_mac_address(dev, sa);
4558 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4561 EXPORT_SYMBOL(dev_set_mac_address);
4564 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4566 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4569 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4575 case SIOCGIFFLAGS: /* Get interface flags */
4576 ifr->ifr_flags = (short) dev_get_flags(dev);
4579 case SIOCGIFMETRIC: /* Get the metric on the interface
4580 (currently unused) */
4581 ifr->ifr_metric = 0;
4584 case SIOCGIFMTU: /* Get the MTU of a device */
4585 ifr->ifr_mtu = dev->mtu;
4590 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4592 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4593 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4594 ifr->ifr_hwaddr.sa_family = dev->type;
4602 ifr->ifr_map.mem_start = dev->mem_start;
4603 ifr->ifr_map.mem_end = dev->mem_end;
4604 ifr->ifr_map.base_addr = dev->base_addr;
4605 ifr->ifr_map.irq = dev->irq;
4606 ifr->ifr_map.dma = dev->dma;
4607 ifr->ifr_map.port = dev->if_port;
4611 ifr->ifr_ifindex = dev->ifindex;
4615 ifr->ifr_qlen = dev->tx_queue_len;
4619 /* dev_ioctl() should ensure this case
4631 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4633 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4636 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4637 const struct net_device_ops *ops;
4642 ops = dev->netdev_ops;
4645 case SIOCSIFFLAGS: /* Set interface flags */
4646 return dev_change_flags(dev, ifr->ifr_flags);
4648 case SIOCSIFMETRIC: /* Set the metric on the interface
4649 (currently unused) */
4652 case SIOCSIFMTU: /* Set the MTU of a device */
4653 return dev_set_mtu(dev, ifr->ifr_mtu);
4656 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4658 case SIOCSIFHWBROADCAST:
4659 if (ifr->ifr_hwaddr.sa_family != dev->type)
4661 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4662 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4663 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4667 if (ops->ndo_set_config) {
4668 if (!netif_device_present(dev))
4670 return ops->ndo_set_config(dev, &ifr->ifr_map);
4675 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4676 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4678 if (!netif_device_present(dev))
4680 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4683 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4684 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4686 if (!netif_device_present(dev))
4688 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4691 if (ifr->ifr_qlen < 0)
4693 dev->tx_queue_len = ifr->ifr_qlen;
4697 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4698 return dev_change_name(dev, ifr->ifr_newname);
4701 * Unknown or private ioctl
4704 if ((cmd >= SIOCDEVPRIVATE &&
4705 cmd <= SIOCDEVPRIVATE + 15) ||
4706 cmd == SIOCBONDENSLAVE ||
4707 cmd == SIOCBONDRELEASE ||
4708 cmd == SIOCBONDSETHWADDR ||
4709 cmd == SIOCBONDSLAVEINFOQUERY ||
4710 cmd == SIOCBONDINFOQUERY ||
4711 cmd == SIOCBONDCHANGEACTIVE ||
4712 cmd == SIOCGMIIPHY ||
4713 cmd == SIOCGMIIREG ||
4714 cmd == SIOCSMIIREG ||
4715 cmd == SIOCBRADDIF ||
4716 cmd == SIOCBRDELIF ||
4717 cmd == SIOCSHWTSTAMP ||
4718 cmd == SIOCWANDEV) {
4720 if (ops->ndo_do_ioctl) {
4721 if (netif_device_present(dev))
4722 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4734 * This function handles all "interface"-type I/O control requests. The actual
4735 * 'doing' part of this is dev_ifsioc above.
4739 * dev_ioctl - network device ioctl
4740 * @net: the applicable net namespace
4741 * @cmd: command to issue
4742 * @arg: pointer to a struct ifreq in user space
4744 * Issue ioctl functions to devices. This is normally called by the
4745 * user space syscall interfaces but can sometimes be useful for
4746 * other purposes. The return value is the return from the syscall if
4747 * positive or a negative errno code on error.
4750 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4756 /* One special case: SIOCGIFCONF takes ifconf argument
4757 and requires shared lock, because it sleeps writing
4761 if (cmd == SIOCGIFCONF) {
4763 ret = dev_ifconf(net, (char __user *) arg);
4767 if (cmd == SIOCGIFNAME)
4768 return dev_ifname(net, (struct ifreq __user *)arg);
4770 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4773 ifr.ifr_name[IFNAMSIZ-1] = 0;
4775 colon = strchr(ifr.ifr_name, ':');
4780 * See which interface the caller is talking about.
4785 * These ioctl calls:
4786 * - can be done by all.
4787 * - atomic and do not require locking.
4798 dev_load(net, ifr.ifr_name);
4800 ret = dev_ifsioc_locked(net, &ifr, cmd);
4805 if (copy_to_user(arg, &ifr,
4806 sizeof(struct ifreq)))
4812 dev_load(net, ifr.ifr_name);
4814 ret = dev_ethtool(net, &ifr);
4819 if (copy_to_user(arg, &ifr,
4820 sizeof(struct ifreq)))
4826 * These ioctl calls:
4827 * - require superuser power.
4828 * - require strict serialization.
4834 if (!capable(CAP_NET_ADMIN))
4836 dev_load(net, ifr.ifr_name);
4838 ret = dev_ifsioc(net, &ifr, cmd);
4843 if (copy_to_user(arg, &ifr,
4844 sizeof(struct ifreq)))
4850 * These ioctl calls:
4851 * - require superuser power.
4852 * - require strict serialization.
4853 * - do not return a value
4863 case SIOCSIFHWBROADCAST:
4866 case SIOCBONDENSLAVE:
4867 case SIOCBONDRELEASE:
4868 case SIOCBONDSETHWADDR:
4869 case SIOCBONDCHANGEACTIVE:
4873 if (!capable(CAP_NET_ADMIN))
4876 case SIOCBONDSLAVEINFOQUERY:
4877 case SIOCBONDINFOQUERY:
4878 dev_load(net, ifr.ifr_name);
4880 ret = dev_ifsioc(net, &ifr, cmd);
4885 /* Get the per device memory space. We can add this but
4886 * currently do not support it */
4888 /* Set the per device memory buffer space.
4889 * Not applicable in our case */
4894 * Unknown or private ioctl.
4897 if (cmd == SIOCWANDEV ||
4898 (cmd >= SIOCDEVPRIVATE &&
4899 cmd <= SIOCDEVPRIVATE + 15)) {
4900 dev_load(net, ifr.ifr_name);
4902 ret = dev_ifsioc(net, &ifr, cmd);
4904 if (!ret && copy_to_user(arg, &ifr,
4905 sizeof(struct ifreq)))
4909 /* Take care of Wireless Extensions */
4910 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4911 return wext_handle_ioctl(net, &ifr, cmd, arg);
4918 * dev_new_index - allocate an ifindex
4919 * @net: the applicable net namespace
4921 * Returns a suitable unique value for a new device interface
4922 * number. The caller must hold the rtnl semaphore or the
4923 * dev_base_lock to be sure it remains unique.
4925 static int dev_new_index(struct net *net)
4931 if (!__dev_get_by_index(net, ifindex))
4936 /* Delayed registration/unregisteration */
4937 static LIST_HEAD(net_todo_list);
4939 static void net_set_todo(struct net_device *dev)
4941 list_add_tail(&dev->todo_list, &net_todo_list);
4944 static void rollback_registered_many(struct list_head *head)
4946 struct net_device *dev, *tmp;
4948 BUG_ON(dev_boot_phase);
4951 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4952 /* Some devices call without registering
4953 * for initialization unwind. Remove those
4954 * devices and proceed with the remaining.
4956 if (dev->reg_state == NETREG_UNINITIALIZED) {
4957 pr_debug("unregister_netdevice: device %s/%p never "
4958 "was registered\n", dev->name, dev);
4961 list_del(&dev->unreg_list);
4965 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4967 /* If device is running, close it first. */
4970 /* And unlink it from device chain. */
4971 unlist_netdevice(dev);
4973 dev->reg_state = NETREG_UNREGISTERING;
4978 list_for_each_entry(dev, head, unreg_list) {
4979 /* Shutdown queueing discipline. */
4983 /* Notify protocols, that we are about to destroy
4984 this device. They should clean all the things.
4986 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4988 if (!dev->rtnl_link_ops ||
4989 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4990 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4993 * Flush the unicast and multicast chains
4998 if (dev->netdev_ops->ndo_uninit)
4999 dev->netdev_ops->ndo_uninit(dev);
5001 /* Notifier chain MUST detach us from master device. */
5002 WARN_ON(dev->master);
5004 /* Remove entries from kobject tree */
5005 netdev_unregister_kobject(dev);
5008 /* Process any work delayed until the end of the batch */
5009 dev = list_first_entry(head, struct net_device, unreg_list);
5010 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5014 list_for_each_entry(dev, head, unreg_list)
5018 static void rollback_registered(struct net_device *dev)
5022 list_add(&dev->unreg_list, &single);
5023 rollback_registered_many(&single);
5026 unsigned long netdev_fix_features(unsigned long features, const char *name)
5028 /* Fix illegal SG+CSUM combinations. */
5029 if ((features & NETIF_F_SG) &&
5030 !(features & NETIF_F_ALL_CSUM)) {
5032 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
5033 "checksum feature.\n", name);
5034 features &= ~NETIF_F_SG;
5037 /* TSO requires that SG is present as well. */
5038 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
5040 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
5041 "SG feature.\n", name);
5042 features &= ~NETIF_F_TSO;
5045 if (features & NETIF_F_UFO) {
5046 /* maybe split UFO into V4 and V6? */
5047 if (!((features & NETIF_F_GEN_CSUM) ||
5048 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5049 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5051 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5052 "since no checksum offload features.\n",
5054 features &= ~NETIF_F_UFO;
5057 if (!(features & NETIF_F_SG)) {
5059 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5060 "since no NETIF_F_SG feature.\n", name);
5061 features &= ~NETIF_F_UFO;
5067 EXPORT_SYMBOL(netdev_fix_features);
5070 * netif_stacked_transfer_operstate - transfer operstate
5071 * @rootdev: the root or lower level device to transfer state from
5072 * @dev: the device to transfer operstate to
5074 * Transfer operational state from root to device. This is normally
5075 * called when a stacking relationship exists between the root
5076 * device and the device(a leaf device).
5078 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5079 struct net_device *dev)
5081 if (rootdev->operstate == IF_OPER_DORMANT)
5082 netif_dormant_on(dev);
5084 netif_dormant_off(dev);
5086 if (netif_carrier_ok(rootdev)) {
5087 if (!netif_carrier_ok(dev))
5088 netif_carrier_on(dev);
5090 if (netif_carrier_ok(dev))
5091 netif_carrier_off(dev);
5094 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5097 static int netif_alloc_rx_queues(struct net_device *dev)
5099 unsigned int i, count = dev->num_rx_queues;
5100 struct netdev_rx_queue *rx;
5104 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5106 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5111 for (i = 0; i < count; i++)
5117 static void netdev_init_one_queue(struct net_device *dev,
5118 struct netdev_queue *queue, void *_unused)
5120 /* Initialize queue lock */
5121 spin_lock_init(&queue->_xmit_lock);
5122 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5123 queue->xmit_lock_owner = -1;
5124 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5128 static int netif_alloc_netdev_queues(struct net_device *dev)
5130 unsigned int count = dev->num_tx_queues;
5131 struct netdev_queue *tx;
5135 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5137 pr_err("netdev: Unable to allocate %u tx queues.\n",
5143 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5144 spin_lock_init(&dev->tx_global_lock);
5150 * register_netdevice - register a network device
5151 * @dev: device to register
5153 * Take a completed network device structure and add it to the kernel
5154 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5155 * chain. 0 is returned on success. A negative errno code is returned
5156 * on a failure to set up the device, or if the name is a duplicate.
5158 * Callers must hold the rtnl semaphore. You may want
5159 * register_netdev() instead of this.
5162 * The locking appears insufficient to guarantee two parallel registers
5163 * will not get the same name.
5166 int register_netdevice(struct net_device *dev)
5169 struct net *net = dev_net(dev);
5171 BUG_ON(dev_boot_phase);
5176 /* When net_device's are persistent, this will be fatal. */
5177 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5180 spin_lock_init(&dev->addr_list_lock);
5181 netdev_set_addr_lockdep_class(dev);
5185 /* Init, if this function is available */
5186 if (dev->netdev_ops->ndo_init) {
5187 ret = dev->netdev_ops->ndo_init(dev);
5195 ret = dev_get_valid_name(dev, dev->name, 0);
5199 dev->ifindex = dev_new_index(net);
5200 if (dev->iflink == -1)
5201 dev->iflink = dev->ifindex;
5203 /* Fix illegal checksum combinations */
5204 if ((dev->features & NETIF_F_HW_CSUM) &&
5205 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5206 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5208 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5211 if ((dev->features & NETIF_F_NO_CSUM) &&
5212 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5213 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5215 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5218 dev->features = netdev_fix_features(dev->features, dev->name);
5220 /* Enable software GSO if SG is supported. */
5221 if (dev->features & NETIF_F_SG)
5222 dev->features |= NETIF_F_GSO;
5224 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5225 * vlan_dev_init() will do the dev->features check, so these features
5226 * are enabled only if supported by underlying device.
5228 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5230 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5231 ret = notifier_to_errno(ret);
5235 ret = netdev_register_kobject(dev);
5238 dev->reg_state = NETREG_REGISTERED;
5241 * Default initial state at registry is that the
5242 * device is present.
5245 set_bit(__LINK_STATE_PRESENT, &dev->state);
5247 dev_init_scheduler(dev);
5249 list_netdevice(dev);
5251 /* Notify protocols, that a new device appeared. */
5252 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5253 ret = notifier_to_errno(ret);
5255 rollback_registered(dev);
5256 dev->reg_state = NETREG_UNREGISTERED;
5259 * Prevent userspace races by waiting until the network
5260 * device is fully setup before sending notifications.
5262 if (!dev->rtnl_link_ops ||
5263 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5264 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5270 if (dev->netdev_ops->ndo_uninit)
5271 dev->netdev_ops->ndo_uninit(dev);
5274 EXPORT_SYMBOL(register_netdevice);
5277 * init_dummy_netdev - init a dummy network device for NAPI
5278 * @dev: device to init
5280 * This takes a network device structure and initialize the minimum
5281 * amount of fields so it can be used to schedule NAPI polls without
5282 * registering a full blown interface. This is to be used by drivers
5283 * that need to tie several hardware interfaces to a single NAPI
5284 * poll scheduler due to HW limitations.
5286 int init_dummy_netdev(struct net_device *dev)
5288 /* Clear everything. Note we don't initialize spinlocks
5289 * are they aren't supposed to be taken by any of the
5290 * NAPI code and this dummy netdev is supposed to be
5291 * only ever used for NAPI polls
5293 memset(dev, 0, sizeof(struct net_device));
5295 /* make sure we BUG if trying to hit standard
5296 * register/unregister code path
5298 dev->reg_state = NETREG_DUMMY;
5300 /* NAPI wants this */
5301 INIT_LIST_HEAD(&dev->napi_list);
5303 /* a dummy interface is started by default */
5304 set_bit(__LINK_STATE_PRESENT, &dev->state);
5305 set_bit(__LINK_STATE_START, &dev->state);
5307 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5308 * because users of this 'device' dont need to change
5314 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5318 * register_netdev - register a network device
5319 * @dev: device to register
5321 * Take a completed network device structure and add it to the kernel
5322 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5323 * chain. 0 is returned on success. A negative errno code is returned
5324 * on a failure to set up the device, or if the name is a duplicate.
5326 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5327 * and expands the device name if you passed a format string to
5330 int register_netdev(struct net_device *dev)
5337 * If the name is a format string the caller wants us to do a
5340 if (strchr(dev->name, '%')) {
5341 err = dev_alloc_name(dev, dev->name);
5346 err = register_netdevice(dev);
5351 EXPORT_SYMBOL(register_netdev);
5353 int netdev_refcnt_read(const struct net_device *dev)
5357 for_each_possible_cpu(i)
5358 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5361 EXPORT_SYMBOL(netdev_refcnt_read);
5364 * netdev_wait_allrefs - wait until all references are gone.
5366 * This is called when unregistering network devices.
5368 * Any protocol or device that holds a reference should register
5369 * for netdevice notification, and cleanup and put back the
5370 * reference if they receive an UNREGISTER event.
5371 * We can get stuck here if buggy protocols don't correctly
5374 static void netdev_wait_allrefs(struct net_device *dev)
5376 unsigned long rebroadcast_time, warning_time;
5379 linkwatch_forget_dev(dev);
5381 rebroadcast_time = warning_time = jiffies;
5382 refcnt = netdev_refcnt_read(dev);
5384 while (refcnt != 0) {
5385 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5388 /* Rebroadcast unregister notification */
5389 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5390 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5391 * should have already handle it the first time */
5393 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5395 /* We must not have linkwatch events
5396 * pending on unregister. If this
5397 * happens, we simply run the queue
5398 * unscheduled, resulting in a noop
5401 linkwatch_run_queue();
5406 rebroadcast_time = jiffies;
5411 refcnt = netdev_refcnt_read(dev);
5413 if (time_after(jiffies, warning_time + 10 * HZ)) {
5414 printk(KERN_EMERG "unregister_netdevice: "
5415 "waiting for %s to become free. Usage "
5418 warning_time = jiffies;
5427 * register_netdevice(x1);
5428 * register_netdevice(x2);
5430 * unregister_netdevice(y1);
5431 * unregister_netdevice(y2);
5437 * We are invoked by rtnl_unlock().
5438 * This allows us to deal with problems:
5439 * 1) We can delete sysfs objects which invoke hotplug
5440 * without deadlocking with linkwatch via keventd.
5441 * 2) Since we run with the RTNL semaphore not held, we can sleep
5442 * safely in order to wait for the netdev refcnt to drop to zero.
5444 * We must not return until all unregister events added during
5445 * the interval the lock was held have been completed.
5447 void netdev_run_todo(void)
5449 struct list_head list;
5451 /* Snapshot list, allow later requests */
5452 list_replace_init(&net_todo_list, &list);
5456 while (!list_empty(&list)) {
5457 struct net_device *dev
5458 = list_first_entry(&list, struct net_device, todo_list);
5459 list_del(&dev->todo_list);
5461 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5462 printk(KERN_ERR "network todo '%s' but state %d\n",
5463 dev->name, dev->reg_state);
5468 dev->reg_state = NETREG_UNREGISTERED;
5470 on_each_cpu(flush_backlog, dev, 1);
5472 netdev_wait_allrefs(dev);
5475 BUG_ON(netdev_refcnt_read(dev));
5476 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5477 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5478 WARN_ON(dev->dn_ptr);
5480 if (dev->destructor)
5481 dev->destructor(dev);
5483 /* Free network device */
5484 kobject_put(&dev->dev.kobj);
5489 * dev_txq_stats_fold - fold tx_queues stats
5490 * @dev: device to get statistics from
5491 * @stats: struct rtnl_link_stats64 to hold results
5493 void dev_txq_stats_fold(const struct net_device *dev,
5494 struct rtnl_link_stats64 *stats)
5496 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5498 struct netdev_queue *txq;
5500 for (i = 0; i < dev->num_tx_queues; i++) {
5501 txq = netdev_get_tx_queue(dev, i);
5502 spin_lock_bh(&txq->_xmit_lock);
5503 tx_bytes += txq->tx_bytes;
5504 tx_packets += txq->tx_packets;
5505 tx_dropped += txq->tx_dropped;
5506 spin_unlock_bh(&txq->_xmit_lock);
5508 if (tx_bytes || tx_packets || tx_dropped) {
5509 stats->tx_bytes = tx_bytes;
5510 stats->tx_packets = tx_packets;
5511 stats->tx_dropped = tx_dropped;
5514 EXPORT_SYMBOL(dev_txq_stats_fold);
5516 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5517 * fields in the same order, with only the type differing.
5519 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5520 const struct net_device_stats *netdev_stats)
5522 #if BITS_PER_LONG == 64
5523 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5524 memcpy(stats64, netdev_stats, sizeof(*stats64));
5526 size_t i, n = sizeof(*stats64) / sizeof(u64);
5527 const unsigned long *src = (const unsigned long *)netdev_stats;
5528 u64 *dst = (u64 *)stats64;
5530 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5531 sizeof(*stats64) / sizeof(u64));
5532 for (i = 0; i < n; i++)
5538 * dev_get_stats - get network device statistics
5539 * @dev: device to get statistics from
5540 * @storage: place to store stats
5542 * Get network statistics from device. Return @storage.
5543 * The device driver may provide its own method by setting
5544 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5545 * otherwise the internal statistics structure is used.
5547 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5548 struct rtnl_link_stats64 *storage)
5550 const struct net_device_ops *ops = dev->netdev_ops;
5552 if (ops->ndo_get_stats64) {
5553 memset(storage, 0, sizeof(*storage));
5554 ops->ndo_get_stats64(dev, storage);
5555 } else if (ops->ndo_get_stats) {
5556 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5558 netdev_stats_to_stats64(storage, &dev->stats);
5559 dev_txq_stats_fold(dev, storage);
5561 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5564 EXPORT_SYMBOL(dev_get_stats);
5566 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5568 struct netdev_queue *queue = dev_ingress_queue(dev);
5570 #ifdef CONFIG_NET_CLS_ACT
5573 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5576 netdev_init_one_queue(dev, queue, NULL);
5577 queue->qdisc = &noop_qdisc;
5578 queue->qdisc_sleeping = &noop_qdisc;
5579 rcu_assign_pointer(dev->ingress_queue, queue);
5585 * alloc_netdev_mq - allocate network device
5586 * @sizeof_priv: size of private data to allocate space for
5587 * @name: device name format string
5588 * @setup: callback to initialize device
5589 * @queue_count: the number of subqueues to allocate
5591 * Allocates a struct net_device with private data area for driver use
5592 * and performs basic initialization. Also allocates subquue structs
5593 * for each queue on the device at the end of the netdevice.
5595 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5596 void (*setup)(struct net_device *), unsigned int queue_count)
5598 struct net_device *dev;
5600 struct net_device *p;
5602 BUG_ON(strlen(name) >= sizeof(dev->name));
5604 if (queue_count < 1) {
5605 pr_err("alloc_netdev: Unable to allocate device "
5606 "with zero queues.\n");
5610 alloc_size = sizeof(struct net_device);
5612 /* ensure 32-byte alignment of private area */
5613 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5614 alloc_size += sizeof_priv;
5616 /* ensure 32-byte alignment of whole construct */
5617 alloc_size += NETDEV_ALIGN - 1;
5619 p = kzalloc(alloc_size, GFP_KERNEL);
5621 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5625 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5626 dev->padded = (char *)dev - (char *)p;
5628 dev->pcpu_refcnt = alloc_percpu(int);
5629 if (!dev->pcpu_refcnt)
5632 if (dev_addr_init(dev))
5638 dev_net_set(dev, &init_net);
5640 dev->num_tx_queues = queue_count;
5641 dev->real_num_tx_queues = queue_count;
5642 if (netif_alloc_netdev_queues(dev))
5646 dev->num_rx_queues = queue_count;
5647 dev->real_num_rx_queues = queue_count;
5648 if (netif_alloc_rx_queues(dev))
5652 dev->gso_max_size = GSO_MAX_SIZE;
5654 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5655 dev->ethtool_ntuple_list.count = 0;
5656 INIT_LIST_HEAD(&dev->napi_list);
5657 INIT_LIST_HEAD(&dev->unreg_list);
5658 INIT_LIST_HEAD(&dev->link_watch_list);
5659 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5661 strcpy(dev->name, name);
5665 free_percpu(dev->pcpu_refcnt);
5675 EXPORT_SYMBOL(alloc_netdev_mq);
5678 * free_netdev - free network device
5681 * This function does the last stage of destroying an allocated device
5682 * interface. The reference to the device object is released.
5683 * If this is the last reference then it will be freed.
5685 void free_netdev(struct net_device *dev)
5687 struct napi_struct *p, *n;
5689 release_net(dev_net(dev));
5696 kfree(rcu_dereference_raw(dev->ingress_queue));
5698 /* Flush device addresses */
5699 dev_addr_flush(dev);
5701 /* Clear ethtool n-tuple list */
5702 ethtool_ntuple_flush(dev);
5704 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5707 free_percpu(dev->pcpu_refcnt);
5708 dev->pcpu_refcnt = NULL;
5710 /* Compatibility with error handling in drivers */
5711 if (dev->reg_state == NETREG_UNINITIALIZED) {
5712 kfree((char *)dev - dev->padded);
5716 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5717 dev->reg_state = NETREG_RELEASED;
5719 /* will free via device release */
5720 put_device(&dev->dev);
5722 EXPORT_SYMBOL(free_netdev);
5725 * synchronize_net - Synchronize with packet receive processing
5727 * Wait for packets currently being received to be done.
5728 * Does not block later packets from starting.
5730 void synchronize_net(void)
5735 EXPORT_SYMBOL(synchronize_net);
5738 * unregister_netdevice_queue - remove device from the kernel
5742 * This function shuts down a device interface and removes it
5743 * from the kernel tables.
5744 * If head not NULL, device is queued to be unregistered later.
5746 * Callers must hold the rtnl semaphore. You may want
5747 * unregister_netdev() instead of this.
5750 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5755 list_move_tail(&dev->unreg_list, head);
5757 rollback_registered(dev);
5758 /* Finish processing unregister after unlock */
5762 EXPORT_SYMBOL(unregister_netdevice_queue);
5765 * unregister_netdevice_many - unregister many devices
5766 * @head: list of devices
5768 void unregister_netdevice_many(struct list_head *head)
5770 struct net_device *dev;
5772 if (!list_empty(head)) {
5773 rollback_registered_many(head);
5774 list_for_each_entry(dev, head, unreg_list)
5778 EXPORT_SYMBOL(unregister_netdevice_many);
5781 * unregister_netdev - remove device from the kernel
5784 * This function shuts down a device interface and removes it
5785 * from the kernel tables.
5787 * This is just a wrapper for unregister_netdevice that takes
5788 * the rtnl semaphore. In general you want to use this and not
5789 * unregister_netdevice.
5791 void unregister_netdev(struct net_device *dev)
5794 unregister_netdevice(dev);
5797 EXPORT_SYMBOL(unregister_netdev);
5800 * dev_change_net_namespace - move device to different nethost namespace
5802 * @net: network namespace
5803 * @pat: If not NULL name pattern to try if the current device name
5804 * is already taken in the destination network namespace.
5806 * This function shuts down a device interface and moves it
5807 * to a new network namespace. On success 0 is returned, on
5808 * a failure a netagive errno code is returned.
5810 * Callers must hold the rtnl semaphore.
5813 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5819 /* Don't allow namespace local devices to be moved. */
5821 if (dev->features & NETIF_F_NETNS_LOCAL)
5824 /* Ensure the device has been registrered */
5826 if (dev->reg_state != NETREG_REGISTERED)
5829 /* Get out if there is nothing todo */
5831 if (net_eq(dev_net(dev), net))
5834 /* Pick the destination device name, and ensure
5835 * we can use it in the destination network namespace.
5838 if (__dev_get_by_name(net, dev->name)) {
5839 /* We get here if we can't use the current device name */
5842 if (dev_get_valid_name(dev, pat, 1))
5847 * And now a mini version of register_netdevice unregister_netdevice.
5850 /* If device is running close it first. */
5853 /* And unlink it from device chain */
5855 unlist_netdevice(dev);
5859 /* Shutdown queueing discipline. */
5862 /* Notify protocols, that we are about to destroy
5863 this device. They should clean all the things.
5865 Note that dev->reg_state stays at NETREG_REGISTERED.
5866 This is wanted because this way 8021q and macvlan know
5867 the device is just moving and can keep their slaves up.
5869 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5870 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5873 * Flush the unicast and multicast chains
5878 /* Actually switch the network namespace */
5879 dev_net_set(dev, net);
5881 /* If there is an ifindex conflict assign a new one */
5882 if (__dev_get_by_index(net, dev->ifindex)) {
5883 int iflink = (dev->iflink == dev->ifindex);
5884 dev->ifindex = dev_new_index(net);
5886 dev->iflink = dev->ifindex;
5889 /* Fixup kobjects */
5890 err = device_rename(&dev->dev, dev->name);
5893 /* Add the device back in the hashes */
5894 list_netdevice(dev);
5896 /* Notify protocols, that a new device appeared. */
5897 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5900 * Prevent userspace races by waiting until the network
5901 * device is fully setup before sending notifications.
5903 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5910 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5912 static int dev_cpu_callback(struct notifier_block *nfb,
5913 unsigned long action,
5916 struct sk_buff **list_skb;
5917 struct sk_buff *skb;
5918 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5919 struct softnet_data *sd, *oldsd;
5921 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5924 local_irq_disable();
5925 cpu = smp_processor_id();
5926 sd = &per_cpu(softnet_data, cpu);
5927 oldsd = &per_cpu(softnet_data, oldcpu);
5929 /* Find end of our completion_queue. */
5930 list_skb = &sd->completion_queue;
5932 list_skb = &(*list_skb)->next;
5933 /* Append completion queue from offline CPU. */
5934 *list_skb = oldsd->completion_queue;
5935 oldsd->completion_queue = NULL;
5937 /* Append output queue from offline CPU. */
5938 if (oldsd->output_queue) {
5939 *sd->output_queue_tailp = oldsd->output_queue;
5940 sd->output_queue_tailp = oldsd->output_queue_tailp;
5941 oldsd->output_queue = NULL;
5942 oldsd->output_queue_tailp = &oldsd->output_queue;
5945 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5948 /* Process offline CPU's input_pkt_queue */
5949 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5951 input_queue_head_incr(oldsd);
5953 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5955 input_queue_head_incr(oldsd);
5963 * netdev_increment_features - increment feature set by one
5964 * @all: current feature set
5965 * @one: new feature set
5966 * @mask: mask feature set
5968 * Computes a new feature set after adding a device with feature set
5969 * @one to the master device with current feature set @all. Will not
5970 * enable anything that is off in @mask. Returns the new feature set.
5972 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5975 /* If device needs checksumming, downgrade to it. */
5976 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5977 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5978 else if (mask & NETIF_F_ALL_CSUM) {
5979 /* If one device supports v4/v6 checksumming, set for all. */
5980 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5981 !(all & NETIF_F_GEN_CSUM)) {
5982 all &= ~NETIF_F_ALL_CSUM;
5983 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5986 /* If one device supports hw checksumming, set for all. */
5987 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5988 all &= ~NETIF_F_ALL_CSUM;
5989 all |= NETIF_F_HW_CSUM;
5993 one |= NETIF_F_ALL_CSUM;
5995 one |= all & NETIF_F_ONE_FOR_ALL;
5996 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5997 all |= one & mask & NETIF_F_ONE_FOR_ALL;
6001 EXPORT_SYMBOL(netdev_increment_features);
6003 static struct hlist_head *netdev_create_hash(void)
6006 struct hlist_head *hash;
6008 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6010 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6011 INIT_HLIST_HEAD(&hash[i]);
6016 /* Initialize per network namespace state */
6017 static int __net_init netdev_init(struct net *net)
6019 INIT_LIST_HEAD(&net->dev_base_head);
6021 net->dev_name_head = netdev_create_hash();
6022 if (net->dev_name_head == NULL)
6025 net->dev_index_head = netdev_create_hash();
6026 if (net->dev_index_head == NULL)
6032 kfree(net->dev_name_head);
6038 * netdev_drivername - network driver for the device
6039 * @dev: network device
6040 * @buffer: buffer for resulting name
6041 * @len: size of buffer
6043 * Determine network driver for device.
6045 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6047 const struct device_driver *driver;
6048 const struct device *parent;
6050 if (len <= 0 || !buffer)
6054 parent = dev->dev.parent;
6059 driver = parent->driver;
6060 if (driver && driver->name)
6061 strlcpy(buffer, driver->name, len);
6065 static int __netdev_printk(const char *level, const struct net_device *dev,
6066 struct va_format *vaf)
6070 if (dev && dev->dev.parent)
6071 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6072 netdev_name(dev), vaf);
6074 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6076 r = printk("%s(NULL net_device): %pV", level, vaf);
6081 int netdev_printk(const char *level, const struct net_device *dev,
6082 const char *format, ...)
6084 struct va_format vaf;
6088 va_start(args, format);
6093 r = __netdev_printk(level, dev, &vaf);
6098 EXPORT_SYMBOL(netdev_printk);
6100 #define define_netdev_printk_level(func, level) \
6101 int func(const struct net_device *dev, const char *fmt, ...) \
6104 struct va_format vaf; \
6107 va_start(args, fmt); \
6112 r = __netdev_printk(level, dev, &vaf); \
6117 EXPORT_SYMBOL(func);
6119 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6120 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6121 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6122 define_netdev_printk_level(netdev_err, KERN_ERR);
6123 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6124 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6125 define_netdev_printk_level(netdev_info, KERN_INFO);
6127 static void __net_exit netdev_exit(struct net *net)
6129 kfree(net->dev_name_head);
6130 kfree(net->dev_index_head);
6133 static struct pernet_operations __net_initdata netdev_net_ops = {
6134 .init = netdev_init,
6135 .exit = netdev_exit,
6138 static void __net_exit default_device_exit(struct net *net)
6140 struct net_device *dev, *aux;
6142 * Push all migratable network devices back to the
6143 * initial network namespace
6146 for_each_netdev_safe(net, dev, aux) {
6148 char fb_name[IFNAMSIZ];
6150 /* Ignore unmoveable devices (i.e. loopback) */
6151 if (dev->features & NETIF_F_NETNS_LOCAL)
6154 /* Leave virtual devices for the generic cleanup */
6155 if (dev->rtnl_link_ops)
6158 /* Push remaing network devices to init_net */
6159 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6160 err = dev_change_net_namespace(dev, &init_net, fb_name);
6162 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6163 __func__, dev->name, err);
6170 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6172 /* At exit all network devices most be removed from a network
6173 * namespace. Do this in the reverse order of registeration.
6174 * Do this across as many network namespaces as possible to
6175 * improve batching efficiency.
6177 struct net_device *dev;
6179 LIST_HEAD(dev_kill_list);
6182 list_for_each_entry(net, net_list, exit_list) {
6183 for_each_netdev_reverse(net, dev) {
6184 if (dev->rtnl_link_ops)
6185 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6187 unregister_netdevice_queue(dev, &dev_kill_list);
6190 unregister_netdevice_many(&dev_kill_list);
6194 static struct pernet_operations __net_initdata default_device_ops = {
6195 .exit = default_device_exit,
6196 .exit_batch = default_device_exit_batch,
6200 * Initialize the DEV module. At boot time this walks the device list and
6201 * unhooks any devices that fail to initialise (normally hardware not
6202 * present) and leaves us with a valid list of present and active devices.
6207 * This is called single threaded during boot, so no need
6208 * to take the rtnl semaphore.
6210 static int __init net_dev_init(void)
6212 int i, rc = -ENOMEM;
6214 BUG_ON(!dev_boot_phase);
6216 if (dev_proc_init())
6219 if (netdev_kobject_init())
6222 INIT_LIST_HEAD(&ptype_all);
6223 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6224 INIT_LIST_HEAD(&ptype_base[i]);
6226 if (register_pernet_subsys(&netdev_net_ops))
6230 * Initialise the packet receive queues.
6233 for_each_possible_cpu(i) {
6234 struct softnet_data *sd = &per_cpu(softnet_data, i);
6236 memset(sd, 0, sizeof(*sd));
6237 skb_queue_head_init(&sd->input_pkt_queue);
6238 skb_queue_head_init(&sd->process_queue);
6239 sd->completion_queue = NULL;
6240 INIT_LIST_HEAD(&sd->poll_list);
6241 sd->output_queue = NULL;
6242 sd->output_queue_tailp = &sd->output_queue;
6244 sd->csd.func = rps_trigger_softirq;
6250 sd->backlog.poll = process_backlog;
6251 sd->backlog.weight = weight_p;
6252 sd->backlog.gro_list = NULL;
6253 sd->backlog.gro_count = 0;
6258 /* The loopback device is special if any other network devices
6259 * is present in a network namespace the loopback device must
6260 * be present. Since we now dynamically allocate and free the
6261 * loopback device ensure this invariant is maintained by
6262 * keeping the loopback device as the first device on the
6263 * list of network devices. Ensuring the loopback devices
6264 * is the first device that appears and the last network device
6267 if (register_pernet_device(&loopback_net_ops))
6270 if (register_pernet_device(&default_device_ops))
6273 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6274 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6276 hotcpu_notifier(dev_cpu_callback, 0);
6284 subsys_initcall(net_dev_init);
6286 static int __init initialize_hashrnd(void)
6288 get_random_bytes(&hashrnd, sizeof(hashrnd));
6292 late_initcall_sync(initialize_hashrnd);