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 - 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 the
753 * rtnl semaphore. The returned device has not had its ref count increased
754 * and the caller must therefore be careful about locking
757 * If the API was consistent this would be __dev_get_by_hwaddr
760 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
762 struct net_device *dev;
766 for_each_netdev(net, dev)
767 if (dev->type == type &&
768 !memcmp(dev->dev_addr, ha, dev->addr_len))
773 EXPORT_SYMBOL(dev_getbyhwaddr);
775 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
777 struct net_device *dev;
780 for_each_netdev(net, dev)
781 if (dev->type == type)
786 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
788 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
790 struct net_device *dev, *ret = NULL;
793 for_each_netdev_rcu(net, dev)
794 if (dev->type == type) {
802 EXPORT_SYMBOL(dev_getfirstbyhwtype);
805 * dev_get_by_flags_rcu - find any device with given flags
806 * @net: the applicable net namespace
807 * @if_flags: IFF_* values
808 * @mask: bitmask of bits in if_flags to check
810 * Search for any interface with the given flags. Returns NULL if a device
811 * is not found or a pointer to the device. Must be called inside
812 * rcu_read_lock(), and result refcount is unchanged.
815 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
818 struct net_device *dev, *ret;
821 for_each_netdev_rcu(net, dev) {
822 if (((dev->flags ^ if_flags) & mask) == 0) {
829 EXPORT_SYMBOL(dev_get_by_flags_rcu);
832 * dev_valid_name - check if name is okay for network device
835 * Network device names need to be valid file names to
836 * to allow sysfs to work. We also disallow any kind of
839 int dev_valid_name(const char *name)
843 if (strlen(name) >= IFNAMSIZ)
845 if (!strcmp(name, ".") || !strcmp(name, ".."))
849 if (*name == '/' || isspace(*name))
855 EXPORT_SYMBOL(dev_valid_name);
858 * __dev_alloc_name - allocate a name for a device
859 * @net: network namespace to allocate the device name in
860 * @name: name format string
861 * @buf: scratch buffer and result name string
863 * Passed a format string - eg "lt%d" it will try and find a suitable
864 * id. It scans list of devices to build up a free map, then chooses
865 * the first empty slot. The caller must hold the dev_base or rtnl lock
866 * while allocating the name and adding the device in order to avoid
868 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
869 * Returns the number of the unit assigned or a negative errno code.
872 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
876 const int max_netdevices = 8*PAGE_SIZE;
877 unsigned long *inuse;
878 struct net_device *d;
880 p = strnchr(name, IFNAMSIZ-1, '%');
883 * Verify the string as this thing may have come from
884 * the user. There must be either one "%d" and no other "%"
887 if (p[1] != 'd' || strchr(p + 2, '%'))
890 /* Use one page as a bit array of possible slots */
891 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
895 for_each_netdev(net, d) {
896 if (!sscanf(d->name, name, &i))
898 if (i < 0 || i >= max_netdevices)
901 /* avoid cases where sscanf is not exact inverse of printf */
902 snprintf(buf, IFNAMSIZ, name, i);
903 if (!strncmp(buf, d->name, IFNAMSIZ))
907 i = find_first_zero_bit(inuse, max_netdevices);
908 free_page((unsigned long) inuse);
912 snprintf(buf, IFNAMSIZ, name, i);
913 if (!__dev_get_by_name(net, buf))
916 /* It is possible to run out of possible slots
917 * when the name is long and there isn't enough space left
918 * for the digits, or if all bits are used.
924 * dev_alloc_name - allocate a name for a device
926 * @name: name format string
928 * Passed a format string - eg "lt%d" it will try and find a suitable
929 * id. It scans list of devices to build up a free map, then chooses
930 * the first empty slot. The caller must hold the dev_base or rtnl lock
931 * while allocating the name and adding the device in order to avoid
933 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
934 * Returns the number of the unit assigned or a negative errno code.
937 int dev_alloc_name(struct net_device *dev, const char *name)
943 BUG_ON(!dev_net(dev));
945 ret = __dev_alloc_name(net, name, buf);
947 strlcpy(dev->name, buf, IFNAMSIZ);
950 EXPORT_SYMBOL(dev_alloc_name);
952 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
956 BUG_ON(!dev_net(dev));
959 if (!dev_valid_name(name))
962 if (fmt && strchr(name, '%'))
963 return dev_alloc_name(dev, name);
964 else if (__dev_get_by_name(net, name))
966 else if (dev->name != name)
967 strlcpy(dev->name, name, IFNAMSIZ);
973 * dev_change_name - change name of a device
975 * @newname: name (or format string) must be at least IFNAMSIZ
977 * Change name of a device, can pass format strings "eth%d".
980 int dev_change_name(struct net_device *dev, const char *newname)
982 char oldname[IFNAMSIZ];
988 BUG_ON(!dev_net(dev));
991 if (dev->flags & IFF_UP)
994 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
997 memcpy(oldname, dev->name, IFNAMSIZ);
999 err = dev_get_valid_name(dev, newname, 1);
1004 ret = device_rename(&dev->dev, dev->name);
1006 memcpy(dev->name, oldname, IFNAMSIZ);
1010 write_lock_bh(&dev_base_lock);
1011 hlist_del(&dev->name_hlist);
1012 write_unlock_bh(&dev_base_lock);
1016 write_lock_bh(&dev_base_lock);
1017 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1018 write_unlock_bh(&dev_base_lock);
1020 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1021 ret = notifier_to_errno(ret);
1024 /* err >= 0 after dev_alloc_name() or stores the first errno */
1027 memcpy(dev->name, oldname, IFNAMSIZ);
1031 "%s: name change rollback failed: %d.\n",
1040 * dev_set_alias - change ifalias of a device
1042 * @alias: name up to IFALIASZ
1043 * @len: limit of bytes to copy from info
1045 * Set ifalias for a device,
1047 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1051 if (len >= IFALIASZ)
1056 kfree(dev->ifalias);
1057 dev->ifalias = NULL;
1062 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1066 strlcpy(dev->ifalias, alias, len+1);
1072 * netdev_features_change - device changes features
1073 * @dev: device to cause notification
1075 * Called to indicate a device has changed features.
1077 void netdev_features_change(struct net_device *dev)
1079 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1081 EXPORT_SYMBOL(netdev_features_change);
1084 * netdev_state_change - device changes state
1085 * @dev: device to cause notification
1087 * Called to indicate a device has changed state. This function calls
1088 * the notifier chains for netdev_chain and sends a NEWLINK message
1089 * to the routing socket.
1091 void netdev_state_change(struct net_device *dev)
1093 if (dev->flags & IFF_UP) {
1094 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1095 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1098 EXPORT_SYMBOL(netdev_state_change);
1100 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1102 return call_netdevice_notifiers(event, dev);
1104 EXPORT_SYMBOL(netdev_bonding_change);
1107 * dev_load - load a network module
1108 * @net: the applicable net namespace
1109 * @name: name of interface
1111 * If a network interface is not present and the process has suitable
1112 * privileges this function loads the module. If module loading is not
1113 * available in this kernel then it becomes a nop.
1116 void dev_load(struct net *net, const char *name)
1118 struct net_device *dev;
1121 dev = dev_get_by_name_rcu(net, name);
1124 if (!dev && capable(CAP_NET_ADMIN))
1125 request_module("%s", name);
1127 EXPORT_SYMBOL(dev_load);
1129 static int __dev_open(struct net_device *dev)
1131 const struct net_device_ops *ops = dev->netdev_ops;
1137 * Is it even present?
1139 if (!netif_device_present(dev))
1142 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1143 ret = notifier_to_errno(ret);
1148 * Call device private open method
1150 set_bit(__LINK_STATE_START, &dev->state);
1152 if (ops->ndo_validate_addr)
1153 ret = ops->ndo_validate_addr(dev);
1155 if (!ret && ops->ndo_open)
1156 ret = ops->ndo_open(dev);
1159 * If it went open OK then:
1163 clear_bit(__LINK_STATE_START, &dev->state);
1168 dev->flags |= IFF_UP;
1173 net_dmaengine_get();
1176 * Initialize multicasting status
1178 dev_set_rx_mode(dev);
1181 * Wakeup transmit queue engine
1190 * dev_open - prepare an interface for use.
1191 * @dev: device to open
1193 * Takes a device from down to up state. The device's private open
1194 * function is invoked and then the multicast lists are loaded. Finally
1195 * the device is moved into the up state and a %NETDEV_UP message is
1196 * sent to the netdev notifier chain.
1198 * Calling this function on an active interface is a nop. On a failure
1199 * a negative errno code is returned.
1201 int dev_open(struct net_device *dev)
1208 if (dev->flags & IFF_UP)
1214 ret = __dev_open(dev);
1219 * ... and announce new interface.
1221 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1222 call_netdevice_notifiers(NETDEV_UP, dev);
1226 EXPORT_SYMBOL(dev_open);
1228 static int __dev_close(struct net_device *dev)
1230 const struct net_device_ops *ops = dev->netdev_ops;
1236 * Tell people we are going down, so that they can
1237 * prepare to death, when device is still operating.
1239 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1241 clear_bit(__LINK_STATE_START, &dev->state);
1243 /* Synchronize to scheduled poll. We cannot touch poll list,
1244 * it can be even on different cpu. So just clear netif_running().
1246 * dev->stop() will invoke napi_disable() on all of it's
1247 * napi_struct instances on this device.
1249 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1251 dev_deactivate(dev);
1254 * Call the device specific close. This cannot fail.
1255 * Only if device is UP
1257 * We allow it to be called even after a DETACH hot-plug
1264 * Device is now down.
1267 dev->flags &= ~IFF_UP;
1272 net_dmaengine_put();
1278 * dev_close - shutdown an interface.
1279 * @dev: device to shutdown
1281 * This function moves an active device into down state. A
1282 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1283 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1286 int dev_close(struct net_device *dev)
1288 if (!(dev->flags & IFF_UP))
1294 * Tell people we are down
1296 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1297 call_netdevice_notifiers(NETDEV_DOWN, dev);
1301 EXPORT_SYMBOL(dev_close);
1305 * dev_disable_lro - disable Large Receive Offload on a device
1308 * Disable Large Receive Offload (LRO) on a net device. Must be
1309 * called under RTNL. This is needed if received packets may be
1310 * forwarded to another interface.
1312 void dev_disable_lro(struct net_device *dev)
1314 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1315 dev->ethtool_ops->set_flags) {
1316 u32 flags = dev->ethtool_ops->get_flags(dev);
1317 if (flags & ETH_FLAG_LRO) {
1318 flags &= ~ETH_FLAG_LRO;
1319 dev->ethtool_ops->set_flags(dev, flags);
1322 WARN_ON(dev->features & NETIF_F_LRO);
1324 EXPORT_SYMBOL(dev_disable_lro);
1327 static int dev_boot_phase = 1;
1330 * Device change register/unregister. These are not inline or static
1331 * as we export them to the world.
1335 * register_netdevice_notifier - register a network notifier block
1338 * Register a notifier to be called when network device events occur.
1339 * The notifier passed is linked into the kernel structures and must
1340 * not be reused until it has been unregistered. A negative errno code
1341 * is returned on a failure.
1343 * When registered all registration and up events are replayed
1344 * to the new notifier to allow device to have a race free
1345 * view of the network device list.
1348 int register_netdevice_notifier(struct notifier_block *nb)
1350 struct net_device *dev;
1351 struct net_device *last;
1356 err = raw_notifier_chain_register(&netdev_chain, nb);
1362 for_each_netdev(net, dev) {
1363 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1364 err = notifier_to_errno(err);
1368 if (!(dev->flags & IFF_UP))
1371 nb->notifier_call(nb, NETDEV_UP, dev);
1382 for_each_netdev(net, dev) {
1386 if (dev->flags & IFF_UP) {
1387 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1388 nb->notifier_call(nb, NETDEV_DOWN, dev);
1390 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1391 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1395 raw_notifier_chain_unregister(&netdev_chain, nb);
1398 EXPORT_SYMBOL(register_netdevice_notifier);
1401 * unregister_netdevice_notifier - unregister a network notifier block
1404 * Unregister a notifier previously registered by
1405 * register_netdevice_notifier(). The notifier is unlinked into the
1406 * kernel structures and may then be reused. A negative errno code
1407 * is returned on a failure.
1410 int unregister_netdevice_notifier(struct notifier_block *nb)
1415 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1419 EXPORT_SYMBOL(unregister_netdevice_notifier);
1422 * call_netdevice_notifiers - call all network notifier blocks
1423 * @val: value passed unmodified to notifier function
1424 * @dev: net_device pointer passed unmodified to notifier function
1426 * Call all network notifier blocks. Parameters and return value
1427 * are as for raw_notifier_call_chain().
1430 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1433 return raw_notifier_call_chain(&netdev_chain, val, dev);
1436 /* When > 0 there are consumers of rx skb time stamps */
1437 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1439 void net_enable_timestamp(void)
1441 atomic_inc(&netstamp_needed);
1443 EXPORT_SYMBOL(net_enable_timestamp);
1445 void net_disable_timestamp(void)
1447 atomic_dec(&netstamp_needed);
1449 EXPORT_SYMBOL(net_disable_timestamp);
1451 static inline void net_timestamp_set(struct sk_buff *skb)
1453 if (atomic_read(&netstamp_needed))
1454 __net_timestamp(skb);
1456 skb->tstamp.tv64 = 0;
1459 static inline void net_timestamp_check(struct sk_buff *skb)
1461 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1462 __net_timestamp(skb);
1466 * dev_forward_skb - loopback an skb to another netif
1468 * @dev: destination network device
1469 * @skb: buffer to forward
1472 * NET_RX_SUCCESS (no congestion)
1473 * NET_RX_DROP (packet was dropped, but freed)
1475 * dev_forward_skb can be used for injecting an skb from the
1476 * start_xmit function of one device into the receive queue
1477 * of another device.
1479 * The receiving device may be in another namespace, so
1480 * we have to clear all information in the skb that could
1481 * impact namespace isolation.
1483 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1488 if (unlikely(!(dev->flags & IFF_UP) ||
1489 (skb->len > (dev->mtu + dev->hard_header_len + VLAN_HLEN)))) {
1490 atomic_long_inc(&dev->rx_dropped);
1494 skb_set_dev(skb, dev);
1495 skb->tstamp.tv64 = 0;
1496 skb->pkt_type = PACKET_HOST;
1497 skb->protocol = eth_type_trans(skb, dev);
1498 return netif_rx(skb);
1500 EXPORT_SYMBOL_GPL(dev_forward_skb);
1503 * Support routine. Sends outgoing frames to any network
1504 * taps currently in use.
1507 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1509 struct packet_type *ptype;
1511 #ifdef CONFIG_NET_CLS_ACT
1512 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1513 net_timestamp_set(skb);
1515 net_timestamp_set(skb);
1519 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1520 /* Never send packets back to the socket
1521 * they originated from - MvS (miquels@drinkel.ow.org)
1523 if ((ptype->dev == dev || !ptype->dev) &&
1524 (ptype->af_packet_priv == NULL ||
1525 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1526 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1530 /* skb->nh should be correctly
1531 set by sender, so that the second statement is
1532 just protection against buggy protocols.
1534 skb_reset_mac_header(skb2);
1536 if (skb_network_header(skb2) < skb2->data ||
1537 skb2->network_header > skb2->tail) {
1538 if (net_ratelimit())
1539 printk(KERN_CRIT "protocol %04x is "
1541 ntohs(skb2->protocol),
1543 skb_reset_network_header(skb2);
1546 skb2->transport_header = skb2->network_header;
1547 skb2->pkt_type = PACKET_OUTGOING;
1548 ptype->func(skb2, skb->dev, ptype, skb->dev);
1555 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1556 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1558 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1560 if (txq < 1 || txq > dev->num_tx_queues)
1563 if (dev->reg_state == NETREG_REGISTERED) {
1566 if (txq < dev->real_num_tx_queues)
1567 qdisc_reset_all_tx_gt(dev, txq);
1570 dev->real_num_tx_queues = txq;
1573 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1577 * netif_set_real_num_rx_queues - set actual number of RX queues used
1578 * @dev: Network device
1579 * @rxq: Actual number of RX queues
1581 * This must be called either with the rtnl_lock held or before
1582 * registration of the net device. Returns 0 on success, or a
1583 * negative error code. If called before registration, it always
1586 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1590 if (rxq < 1 || rxq > dev->num_rx_queues)
1593 if (dev->reg_state == NETREG_REGISTERED) {
1596 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1602 dev->real_num_rx_queues = rxq;
1605 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1608 static inline void __netif_reschedule(struct Qdisc *q)
1610 struct softnet_data *sd;
1611 unsigned long flags;
1613 local_irq_save(flags);
1614 sd = &__get_cpu_var(softnet_data);
1615 q->next_sched = NULL;
1616 *sd->output_queue_tailp = q;
1617 sd->output_queue_tailp = &q->next_sched;
1618 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1619 local_irq_restore(flags);
1622 void __netif_schedule(struct Qdisc *q)
1624 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1625 __netif_reschedule(q);
1627 EXPORT_SYMBOL(__netif_schedule);
1629 void dev_kfree_skb_irq(struct sk_buff *skb)
1631 if (atomic_dec_and_test(&skb->users)) {
1632 struct softnet_data *sd;
1633 unsigned long flags;
1635 local_irq_save(flags);
1636 sd = &__get_cpu_var(softnet_data);
1637 skb->next = sd->completion_queue;
1638 sd->completion_queue = skb;
1639 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1640 local_irq_restore(flags);
1643 EXPORT_SYMBOL(dev_kfree_skb_irq);
1645 void dev_kfree_skb_any(struct sk_buff *skb)
1647 if (in_irq() || irqs_disabled())
1648 dev_kfree_skb_irq(skb);
1652 EXPORT_SYMBOL(dev_kfree_skb_any);
1656 * netif_device_detach - mark device as removed
1657 * @dev: network device
1659 * Mark device as removed from system and therefore no longer available.
1661 void netif_device_detach(struct net_device *dev)
1663 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1664 netif_running(dev)) {
1665 netif_tx_stop_all_queues(dev);
1668 EXPORT_SYMBOL(netif_device_detach);
1671 * netif_device_attach - mark device as attached
1672 * @dev: network device
1674 * Mark device as attached from system and restart if needed.
1676 void netif_device_attach(struct net_device *dev)
1678 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1679 netif_running(dev)) {
1680 netif_tx_wake_all_queues(dev);
1681 __netdev_watchdog_up(dev);
1684 EXPORT_SYMBOL(netif_device_attach);
1686 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1688 return ((features & NETIF_F_NO_CSUM) ||
1689 ((features & NETIF_F_V4_CSUM) &&
1690 protocol == htons(ETH_P_IP)) ||
1691 ((features & NETIF_F_V6_CSUM) &&
1692 protocol == htons(ETH_P_IPV6)) ||
1693 ((features & NETIF_F_FCOE_CRC) &&
1694 protocol == htons(ETH_P_FCOE)));
1697 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1699 __be16 protocol = skb->protocol;
1700 int features = dev->features;
1702 if (vlan_tx_tag_present(skb)) {
1703 features &= dev->vlan_features;
1704 } else if (protocol == htons(ETH_P_8021Q)) {
1705 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1706 protocol = veh->h_vlan_encapsulated_proto;
1707 features &= dev->vlan_features;
1710 return can_checksum_protocol(features, protocol);
1714 * skb_dev_set -- assign a new device to a buffer
1715 * @skb: buffer for the new device
1716 * @dev: network device
1718 * If an skb is owned by a device already, we have to reset
1719 * all data private to the namespace a device belongs to
1720 * before assigning it a new device.
1722 #ifdef CONFIG_NET_NS
1723 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1726 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1729 skb_init_secmark(skb);
1733 skb->ipvs_property = 0;
1734 #ifdef CONFIG_NET_SCHED
1740 EXPORT_SYMBOL(skb_set_dev);
1741 #endif /* CONFIG_NET_NS */
1744 * Invalidate hardware checksum when packet is to be mangled, and
1745 * complete checksum manually on outgoing path.
1747 int skb_checksum_help(struct sk_buff *skb)
1750 int ret = 0, offset;
1752 if (skb->ip_summed == CHECKSUM_COMPLETE)
1753 goto out_set_summed;
1755 if (unlikely(skb_shinfo(skb)->gso_size)) {
1756 /* Let GSO fix up the checksum. */
1757 goto out_set_summed;
1760 offset = skb->csum_start - skb_headroom(skb);
1761 BUG_ON(offset >= skb_headlen(skb));
1762 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1764 offset += skb->csum_offset;
1765 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1767 if (skb_cloned(skb) &&
1768 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1769 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1774 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1776 skb->ip_summed = CHECKSUM_NONE;
1780 EXPORT_SYMBOL(skb_checksum_help);
1783 * skb_gso_segment - Perform segmentation on skb.
1784 * @skb: buffer to segment
1785 * @features: features for the output path (see dev->features)
1787 * This function segments the given skb and returns a list of segments.
1789 * It may return NULL if the skb requires no segmentation. This is
1790 * only possible when GSO is used for verifying header integrity.
1792 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1794 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1795 struct packet_type *ptype;
1796 __be16 type = skb->protocol;
1797 int vlan_depth = ETH_HLEN;
1800 while (type == htons(ETH_P_8021Q)) {
1801 struct vlan_hdr *vh;
1803 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1804 return ERR_PTR(-EINVAL);
1806 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1807 type = vh->h_vlan_encapsulated_proto;
1808 vlan_depth += VLAN_HLEN;
1811 skb_reset_mac_header(skb);
1812 skb->mac_len = skb->network_header - skb->mac_header;
1813 __skb_pull(skb, skb->mac_len);
1815 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1816 struct net_device *dev = skb->dev;
1817 struct ethtool_drvinfo info = {};
1819 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1820 dev->ethtool_ops->get_drvinfo(dev, &info);
1822 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1823 info.driver, dev ? dev->features : 0L,
1824 skb->sk ? skb->sk->sk_route_caps : 0L,
1825 skb->len, skb->data_len, skb->ip_summed);
1827 if (skb_header_cloned(skb) &&
1828 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1829 return ERR_PTR(err);
1833 list_for_each_entry_rcu(ptype,
1834 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1835 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1836 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1837 err = ptype->gso_send_check(skb);
1838 segs = ERR_PTR(err);
1839 if (err || skb_gso_ok(skb, features))
1841 __skb_push(skb, (skb->data -
1842 skb_network_header(skb)));
1844 segs = ptype->gso_segment(skb, features);
1850 __skb_push(skb, skb->data - skb_mac_header(skb));
1854 EXPORT_SYMBOL(skb_gso_segment);
1856 /* Take action when hardware reception checksum errors are detected. */
1858 void netdev_rx_csum_fault(struct net_device *dev)
1860 if (net_ratelimit()) {
1861 printk(KERN_ERR "%s: hw csum failure.\n",
1862 dev ? dev->name : "<unknown>");
1866 EXPORT_SYMBOL(netdev_rx_csum_fault);
1869 /* Actually, we should eliminate this check as soon as we know, that:
1870 * 1. IOMMU is present and allows to map all the memory.
1871 * 2. No high memory really exists on this machine.
1874 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1876 #ifdef CONFIG_HIGHMEM
1878 if (!(dev->features & NETIF_F_HIGHDMA)) {
1879 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1880 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1884 if (PCI_DMA_BUS_IS_PHYS) {
1885 struct device *pdev = dev->dev.parent;
1889 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1890 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1891 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1900 void (*destructor)(struct sk_buff *skb);
1903 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1905 static void dev_gso_skb_destructor(struct sk_buff *skb)
1907 struct dev_gso_cb *cb;
1910 struct sk_buff *nskb = skb->next;
1912 skb->next = nskb->next;
1915 } while (skb->next);
1917 cb = DEV_GSO_CB(skb);
1919 cb->destructor(skb);
1923 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1924 * @skb: buffer to segment
1926 * This function segments the given skb and stores the list of segments
1929 static int dev_gso_segment(struct sk_buff *skb)
1931 struct net_device *dev = skb->dev;
1932 struct sk_buff *segs;
1933 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1936 segs = skb_gso_segment(skb, features);
1938 /* Verifying header integrity only. */
1943 return PTR_ERR(segs);
1946 DEV_GSO_CB(skb)->destructor = skb->destructor;
1947 skb->destructor = dev_gso_skb_destructor;
1953 * Try to orphan skb early, right before transmission by the device.
1954 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1955 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1957 static inline void skb_orphan_try(struct sk_buff *skb)
1959 struct sock *sk = skb->sk;
1961 if (sk && !skb_shinfo(skb)->tx_flags) {
1962 /* skb_tx_hash() wont be able to get sk.
1963 * We copy sk_hash into skb->rxhash
1966 skb->rxhash = sk->sk_hash;
1971 int netif_get_vlan_features(struct sk_buff *skb, struct net_device *dev)
1973 __be16 protocol = skb->protocol;
1975 if (protocol == htons(ETH_P_8021Q)) {
1976 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1977 protocol = veh->h_vlan_encapsulated_proto;
1978 } else if (!skb->vlan_tci)
1979 return dev->features;
1981 if (protocol != htons(ETH_P_8021Q))
1982 return dev->features & dev->vlan_features;
1988 * Returns true if either:
1989 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1990 * 2. skb is fragmented and the device does not support SG, or if
1991 * at least one of fragments is in highmem and device does not
1992 * support DMA from it.
1994 static inline int skb_needs_linearize(struct sk_buff *skb,
1995 struct net_device *dev)
1997 if (skb_is_nonlinear(skb)) {
1998 int features = dev->features;
2000 if (vlan_tx_tag_present(skb))
2001 features &= dev->vlan_features;
2003 return (skb_has_frag_list(skb) &&
2004 !(features & NETIF_F_FRAGLIST)) ||
2005 (skb_shinfo(skb)->nr_frags &&
2006 (!(features & NETIF_F_SG) ||
2007 illegal_highdma(dev, skb)));
2013 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2014 struct netdev_queue *txq)
2016 const struct net_device_ops *ops = dev->netdev_ops;
2017 int rc = NETDEV_TX_OK;
2019 if (likely(!skb->next)) {
2020 if (!list_empty(&ptype_all))
2021 dev_queue_xmit_nit(skb, dev);
2024 * If device doesnt need skb->dst, release it right now while
2025 * its hot in this cpu cache
2027 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2030 skb_orphan_try(skb);
2032 if (vlan_tx_tag_present(skb) &&
2033 !(dev->features & NETIF_F_HW_VLAN_TX)) {
2034 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2041 if (netif_needs_gso(dev, skb)) {
2042 if (unlikely(dev_gso_segment(skb)))
2047 if (skb_needs_linearize(skb, dev) &&
2048 __skb_linearize(skb))
2051 /* If packet is not checksummed and device does not
2052 * support checksumming for this protocol, complete
2053 * checksumming here.
2055 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2056 skb_set_transport_header(skb, skb->csum_start -
2058 if (!dev_can_checksum(dev, skb) &&
2059 skb_checksum_help(skb))
2064 rc = ops->ndo_start_xmit(skb, dev);
2065 trace_net_dev_xmit(skb, rc);
2066 if (rc == NETDEV_TX_OK)
2067 txq_trans_update(txq);
2073 struct sk_buff *nskb = skb->next;
2075 skb->next = nskb->next;
2079 * If device doesnt need nskb->dst, release it right now while
2080 * its hot in this cpu cache
2082 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2085 rc = ops->ndo_start_xmit(nskb, dev);
2086 trace_net_dev_xmit(nskb, rc);
2087 if (unlikely(rc != NETDEV_TX_OK)) {
2088 if (rc & ~NETDEV_TX_MASK)
2089 goto out_kfree_gso_skb;
2090 nskb->next = skb->next;
2094 txq_trans_update(txq);
2095 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2096 return NETDEV_TX_BUSY;
2097 } while (skb->next);
2100 if (likely(skb->next == NULL))
2101 skb->destructor = DEV_GSO_CB(skb)->destructor;
2108 static u32 hashrnd __read_mostly;
2110 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
2114 if (skb_rx_queue_recorded(skb)) {
2115 hash = skb_get_rx_queue(skb);
2116 while (unlikely(hash >= dev->real_num_tx_queues))
2117 hash -= dev->real_num_tx_queues;
2121 if (skb->sk && skb->sk->sk_hash)
2122 hash = skb->sk->sk_hash;
2124 hash = (__force u16) skb->protocol ^ skb->rxhash;
2125 hash = jhash_1word(hash, hashrnd);
2127 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2129 EXPORT_SYMBOL(skb_tx_hash);
2131 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2133 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2134 if (net_ratelimit()) {
2135 pr_warning("%s selects TX queue %d, but "
2136 "real number of TX queues is %d\n",
2137 dev->name, queue_index, dev->real_num_tx_queues);
2144 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2145 struct sk_buff *skb)
2148 const struct net_device_ops *ops = dev->netdev_ops;
2150 if (ops->ndo_select_queue) {
2151 queue_index = ops->ndo_select_queue(dev, skb);
2152 queue_index = dev_cap_txqueue(dev, queue_index);
2154 struct sock *sk = skb->sk;
2155 queue_index = sk_tx_queue_get(sk);
2156 if (queue_index < 0 || queue_index >= dev->real_num_tx_queues) {
2159 if (dev->real_num_tx_queues > 1)
2160 queue_index = skb_tx_hash(dev, skb);
2163 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2165 if (dst && skb_dst(skb) == dst)
2166 sk_tx_queue_set(sk, queue_index);
2171 skb_set_queue_mapping(skb, queue_index);
2172 return netdev_get_tx_queue(dev, queue_index);
2175 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2176 struct net_device *dev,
2177 struct netdev_queue *txq)
2179 spinlock_t *root_lock = qdisc_lock(q);
2180 bool contended = qdisc_is_running(q);
2184 * Heuristic to force contended enqueues to serialize on a
2185 * separate lock before trying to get qdisc main lock.
2186 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2187 * and dequeue packets faster.
2189 if (unlikely(contended))
2190 spin_lock(&q->busylock);
2192 spin_lock(root_lock);
2193 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2196 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2197 qdisc_run_begin(q)) {
2199 * This is a work-conserving queue; there are no old skbs
2200 * waiting to be sent out; and the qdisc is not running -
2201 * xmit the skb directly.
2203 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2205 __qdisc_update_bstats(q, skb->len);
2206 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2207 if (unlikely(contended)) {
2208 spin_unlock(&q->busylock);
2215 rc = NET_XMIT_SUCCESS;
2218 rc = qdisc_enqueue_root(skb, q);
2219 if (qdisc_run_begin(q)) {
2220 if (unlikely(contended)) {
2221 spin_unlock(&q->busylock);
2227 spin_unlock(root_lock);
2228 if (unlikely(contended))
2229 spin_unlock(&q->busylock);
2233 static DEFINE_PER_CPU(int, xmit_recursion);
2234 #define RECURSION_LIMIT 10
2237 * dev_queue_xmit - transmit a buffer
2238 * @skb: buffer to transmit
2240 * Queue a buffer for transmission to a network device. The caller must
2241 * have set the device and priority and built the buffer before calling
2242 * this function. The function can be called from an interrupt.
2244 * A negative errno code is returned on a failure. A success does not
2245 * guarantee the frame will be transmitted as it may be dropped due
2246 * to congestion or traffic shaping.
2248 * -----------------------------------------------------------------------------------
2249 * I notice this method can also return errors from the queue disciplines,
2250 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2253 * Regardless of the return value, the skb is consumed, so it is currently
2254 * difficult to retry a send to this method. (You can bump the ref count
2255 * before sending to hold a reference for retry if you are careful.)
2257 * When calling this method, interrupts MUST be enabled. This is because
2258 * the BH enable code must have IRQs enabled so that it will not deadlock.
2261 int dev_queue_xmit(struct sk_buff *skb)
2263 struct net_device *dev = skb->dev;
2264 struct netdev_queue *txq;
2268 /* Disable soft irqs for various locks below. Also
2269 * stops preemption for RCU.
2273 txq = dev_pick_tx(dev, skb);
2274 q = rcu_dereference_bh(txq->qdisc);
2276 #ifdef CONFIG_NET_CLS_ACT
2277 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2279 trace_net_dev_queue(skb);
2281 rc = __dev_xmit_skb(skb, q, dev, txq);
2285 /* The device has no queue. Common case for software devices:
2286 loopback, all the sorts of tunnels...
2288 Really, it is unlikely that netif_tx_lock protection is necessary
2289 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2291 However, it is possible, that they rely on protection
2294 Check this and shot the lock. It is not prone from deadlocks.
2295 Either shot noqueue qdisc, it is even simpler 8)
2297 if (dev->flags & IFF_UP) {
2298 int cpu = smp_processor_id(); /* ok because BHs are off */
2300 if (txq->xmit_lock_owner != cpu) {
2302 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2303 goto recursion_alert;
2305 HARD_TX_LOCK(dev, txq, cpu);
2307 if (!netif_tx_queue_stopped(txq)) {
2308 __this_cpu_inc(xmit_recursion);
2309 rc = dev_hard_start_xmit(skb, dev, txq);
2310 __this_cpu_dec(xmit_recursion);
2311 if (dev_xmit_complete(rc)) {
2312 HARD_TX_UNLOCK(dev, txq);
2316 HARD_TX_UNLOCK(dev, txq);
2317 if (net_ratelimit())
2318 printk(KERN_CRIT "Virtual device %s asks to "
2319 "queue packet!\n", dev->name);
2321 /* Recursion is detected! It is possible,
2325 if (net_ratelimit())
2326 printk(KERN_CRIT "Dead loop on virtual device "
2327 "%s, fix it urgently!\n", dev->name);
2332 rcu_read_unlock_bh();
2337 rcu_read_unlock_bh();
2340 EXPORT_SYMBOL(dev_queue_xmit);
2343 /*=======================================================================
2345 =======================================================================*/
2347 int netdev_max_backlog __read_mostly = 1000;
2348 int netdev_tstamp_prequeue __read_mostly = 1;
2349 int netdev_budget __read_mostly = 300;
2350 int weight_p __read_mostly = 64; /* old backlog weight */
2352 /* Called with irq disabled */
2353 static inline void ____napi_schedule(struct softnet_data *sd,
2354 struct napi_struct *napi)
2356 list_add_tail(&napi->poll_list, &sd->poll_list);
2357 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2361 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2362 * and src/dst port numbers. Returns a non-zero hash number on success
2365 __u32 __skb_get_rxhash(struct sk_buff *skb)
2367 int nhoff, hash = 0, poff;
2368 struct ipv6hdr *ip6;
2371 u32 addr1, addr2, ihl;
2377 nhoff = skb_network_offset(skb);
2379 switch (skb->protocol) {
2380 case __constant_htons(ETH_P_IP):
2381 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2384 ip = (struct iphdr *) (skb->data + nhoff);
2385 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2388 ip_proto = ip->protocol;
2389 addr1 = (__force u32) ip->saddr;
2390 addr2 = (__force u32) ip->daddr;
2393 case __constant_htons(ETH_P_IPV6):
2394 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2397 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2398 ip_proto = ip6->nexthdr;
2399 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2400 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2408 poff = proto_ports_offset(ip_proto);
2410 nhoff += ihl * 4 + poff;
2411 if (pskb_may_pull(skb, nhoff + 4)) {
2412 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2413 if (ports.v16[1] < ports.v16[0])
2414 swap(ports.v16[0], ports.v16[1]);
2418 /* get a consistent hash (same value on both flow directions) */
2422 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2429 EXPORT_SYMBOL(__skb_get_rxhash);
2433 /* One global table that all flow-based protocols share. */
2434 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2435 EXPORT_SYMBOL(rps_sock_flow_table);
2438 * get_rps_cpu is called from netif_receive_skb and returns the target
2439 * CPU from the RPS map of the receiving queue for a given skb.
2440 * rcu_read_lock must be held on entry.
2442 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2443 struct rps_dev_flow **rflowp)
2445 struct netdev_rx_queue *rxqueue;
2446 struct rps_map *map;
2447 struct rps_dev_flow_table *flow_table;
2448 struct rps_sock_flow_table *sock_flow_table;
2452 if (skb_rx_queue_recorded(skb)) {
2453 u16 index = skb_get_rx_queue(skb);
2454 if (unlikely(index >= dev->real_num_rx_queues)) {
2455 WARN_ONCE(dev->real_num_rx_queues > 1,
2456 "%s received packet on queue %u, but number "
2457 "of RX queues is %u\n",
2458 dev->name, index, dev->real_num_rx_queues);
2461 rxqueue = dev->_rx + index;
2465 map = rcu_dereference(rxqueue->rps_map);
2467 if (map->len == 1) {
2468 tcpu = map->cpus[0];
2469 if (cpu_online(tcpu))
2473 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2477 skb_reset_network_header(skb);
2478 if (!skb_get_rxhash(skb))
2481 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2482 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2483 if (flow_table && sock_flow_table) {
2485 struct rps_dev_flow *rflow;
2487 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2490 next_cpu = sock_flow_table->ents[skb->rxhash &
2491 sock_flow_table->mask];
2494 * If the desired CPU (where last recvmsg was done) is
2495 * different from current CPU (one in the rx-queue flow
2496 * table entry), switch if one of the following holds:
2497 * - Current CPU is unset (equal to RPS_NO_CPU).
2498 * - Current CPU is offline.
2499 * - The current CPU's queue tail has advanced beyond the
2500 * last packet that was enqueued using this table entry.
2501 * This guarantees that all previous packets for the flow
2502 * have been dequeued, thus preserving in order delivery.
2504 if (unlikely(tcpu != next_cpu) &&
2505 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2506 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2507 rflow->last_qtail)) >= 0)) {
2508 tcpu = rflow->cpu = next_cpu;
2509 if (tcpu != RPS_NO_CPU)
2510 rflow->last_qtail = per_cpu(softnet_data,
2511 tcpu).input_queue_head;
2513 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2521 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2523 if (cpu_online(tcpu)) {
2533 /* Called from hardirq (IPI) context */
2534 static void rps_trigger_softirq(void *data)
2536 struct softnet_data *sd = data;
2538 ____napi_schedule(sd, &sd->backlog);
2542 #endif /* CONFIG_RPS */
2545 * Check if this softnet_data structure is another cpu one
2546 * If yes, queue it to our IPI list and return 1
2549 static int rps_ipi_queued(struct softnet_data *sd)
2552 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2555 sd->rps_ipi_next = mysd->rps_ipi_list;
2556 mysd->rps_ipi_list = sd;
2558 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2561 #endif /* CONFIG_RPS */
2566 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2567 * queue (may be a remote CPU queue).
2569 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2570 unsigned int *qtail)
2572 struct softnet_data *sd;
2573 unsigned long flags;
2575 sd = &per_cpu(softnet_data, cpu);
2577 local_irq_save(flags);
2580 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2581 if (skb_queue_len(&sd->input_pkt_queue)) {
2583 __skb_queue_tail(&sd->input_pkt_queue, skb);
2584 input_queue_tail_incr_save(sd, qtail);
2586 local_irq_restore(flags);
2587 return NET_RX_SUCCESS;
2590 /* Schedule NAPI for backlog device
2591 * We can use non atomic operation since we own the queue lock
2593 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2594 if (!rps_ipi_queued(sd))
2595 ____napi_schedule(sd, &sd->backlog);
2603 local_irq_restore(flags);
2605 atomic_long_inc(&skb->dev->rx_dropped);
2611 * netif_rx - post buffer to the network code
2612 * @skb: buffer to post
2614 * This function receives a packet from a device driver and queues it for
2615 * the upper (protocol) levels to process. It always succeeds. The buffer
2616 * may be dropped during processing for congestion control or by the
2620 * NET_RX_SUCCESS (no congestion)
2621 * NET_RX_DROP (packet was dropped)
2625 int netif_rx(struct sk_buff *skb)
2629 /* if netpoll wants it, pretend we never saw it */
2630 if (netpoll_rx(skb))
2633 if (netdev_tstamp_prequeue)
2634 net_timestamp_check(skb);
2636 trace_netif_rx(skb);
2639 struct rps_dev_flow voidflow, *rflow = &voidflow;
2645 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2647 cpu = smp_processor_id();
2649 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2657 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2663 EXPORT_SYMBOL(netif_rx);
2665 int netif_rx_ni(struct sk_buff *skb)
2670 err = netif_rx(skb);
2671 if (local_softirq_pending())
2677 EXPORT_SYMBOL(netif_rx_ni);
2679 static void net_tx_action(struct softirq_action *h)
2681 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2683 if (sd->completion_queue) {
2684 struct sk_buff *clist;
2686 local_irq_disable();
2687 clist = sd->completion_queue;
2688 sd->completion_queue = NULL;
2692 struct sk_buff *skb = clist;
2693 clist = clist->next;
2695 WARN_ON(atomic_read(&skb->users));
2696 trace_kfree_skb(skb, net_tx_action);
2701 if (sd->output_queue) {
2704 local_irq_disable();
2705 head = sd->output_queue;
2706 sd->output_queue = NULL;
2707 sd->output_queue_tailp = &sd->output_queue;
2711 struct Qdisc *q = head;
2712 spinlock_t *root_lock;
2714 head = head->next_sched;
2716 root_lock = qdisc_lock(q);
2717 if (spin_trylock(root_lock)) {
2718 smp_mb__before_clear_bit();
2719 clear_bit(__QDISC_STATE_SCHED,
2722 spin_unlock(root_lock);
2724 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2726 __netif_reschedule(q);
2728 smp_mb__before_clear_bit();
2729 clear_bit(__QDISC_STATE_SCHED,
2737 static inline int deliver_skb(struct sk_buff *skb,
2738 struct packet_type *pt_prev,
2739 struct net_device *orig_dev)
2741 atomic_inc(&skb->users);
2742 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2745 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2746 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2747 /* This hook is defined here for ATM LANE */
2748 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2749 unsigned char *addr) __read_mostly;
2750 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2753 #ifdef CONFIG_NET_CLS_ACT
2754 /* TODO: Maybe we should just force sch_ingress to be compiled in
2755 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2756 * a compare and 2 stores extra right now if we dont have it on
2757 * but have CONFIG_NET_CLS_ACT
2758 * NOTE: This doesnt stop any functionality; if you dont have
2759 * the ingress scheduler, you just cant add policies on ingress.
2762 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2764 struct net_device *dev = skb->dev;
2765 u32 ttl = G_TC_RTTL(skb->tc_verd);
2766 int result = TC_ACT_OK;
2769 if (unlikely(MAX_RED_LOOP < ttl++)) {
2770 if (net_ratelimit())
2771 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2772 skb->skb_iif, dev->ifindex);
2776 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2777 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2780 if (q != &noop_qdisc) {
2781 spin_lock(qdisc_lock(q));
2782 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2783 result = qdisc_enqueue_root(skb, q);
2784 spin_unlock(qdisc_lock(q));
2790 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2791 struct packet_type **pt_prev,
2792 int *ret, struct net_device *orig_dev)
2794 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
2796 if (!rxq || rxq->qdisc == &noop_qdisc)
2800 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2804 switch (ing_filter(skb, rxq)) {
2818 * netdev_rx_handler_register - register receive handler
2819 * @dev: device to register a handler for
2820 * @rx_handler: receive handler to register
2821 * @rx_handler_data: data pointer that is used by rx handler
2823 * Register a receive hander for a device. This handler will then be
2824 * called from __netif_receive_skb. A negative errno code is returned
2827 * The caller must hold the rtnl_mutex.
2829 int netdev_rx_handler_register(struct net_device *dev,
2830 rx_handler_func_t *rx_handler,
2831 void *rx_handler_data)
2835 if (dev->rx_handler)
2838 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2839 rcu_assign_pointer(dev->rx_handler, rx_handler);
2843 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2846 * netdev_rx_handler_unregister - unregister receive handler
2847 * @dev: device to unregister a handler from
2849 * Unregister a receive hander from a device.
2851 * The caller must hold the rtnl_mutex.
2853 void netdev_rx_handler_unregister(struct net_device *dev)
2857 rcu_assign_pointer(dev->rx_handler, NULL);
2858 rcu_assign_pointer(dev->rx_handler_data, NULL);
2860 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2862 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2863 struct net_device *master)
2865 if (skb->pkt_type == PACKET_HOST) {
2866 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2868 memcpy(dest, master->dev_addr, ETH_ALEN);
2872 /* On bonding slaves other than the currently active slave, suppress
2873 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2874 * ARP on active-backup slaves with arp_validate enabled.
2876 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2878 struct net_device *dev = skb->dev;
2880 if (master->priv_flags & IFF_MASTER_ARPMON)
2881 dev->last_rx = jiffies;
2883 if ((master->priv_flags & IFF_MASTER_ALB) &&
2884 (master->priv_flags & IFF_BRIDGE_PORT)) {
2885 /* Do address unmangle. The local destination address
2886 * will be always the one master has. Provides the right
2887 * functionality in a bridge.
2889 skb_bond_set_mac_by_master(skb, master);
2892 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2893 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2894 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2897 if (master->priv_flags & IFF_MASTER_ALB) {
2898 if (skb->pkt_type != PACKET_BROADCAST &&
2899 skb->pkt_type != PACKET_MULTICAST)
2902 if (master->priv_flags & IFF_MASTER_8023AD &&
2903 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2910 EXPORT_SYMBOL(__skb_bond_should_drop);
2912 static int __netif_receive_skb(struct sk_buff *skb)
2914 struct packet_type *ptype, *pt_prev;
2915 rx_handler_func_t *rx_handler;
2916 struct net_device *orig_dev;
2917 struct net_device *master;
2918 struct net_device *null_or_orig;
2919 struct net_device *orig_or_bond;
2920 int ret = NET_RX_DROP;
2923 if (!netdev_tstamp_prequeue)
2924 net_timestamp_check(skb);
2926 trace_netif_receive_skb(skb);
2928 /* if we've gotten here through NAPI, check netpoll */
2929 if (netpoll_receive_skb(skb))
2933 skb->skb_iif = skb->dev->ifindex;
2936 * bonding note: skbs received on inactive slaves should only
2937 * be delivered to pkt handlers that are exact matches. Also
2938 * the deliver_no_wcard flag will be set. If packet handlers
2939 * are sensitive to duplicate packets these skbs will need to
2940 * be dropped at the handler.
2942 null_or_orig = NULL;
2943 orig_dev = skb->dev;
2944 master = ACCESS_ONCE(orig_dev->master);
2945 if (skb->deliver_no_wcard)
2946 null_or_orig = orig_dev;
2948 if (skb_bond_should_drop(skb, master)) {
2949 skb->deliver_no_wcard = 1;
2950 null_or_orig = orig_dev; /* deliver only exact match */
2955 __this_cpu_inc(softnet_data.processed);
2956 skb_reset_network_header(skb);
2957 skb_reset_transport_header(skb);
2958 skb->mac_len = skb->network_header - skb->mac_header;
2964 #ifdef CONFIG_NET_CLS_ACT
2965 if (skb->tc_verd & TC_NCLS) {
2966 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2971 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2972 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2973 ptype->dev == orig_dev) {
2975 ret = deliver_skb(skb, pt_prev, orig_dev);
2980 #ifdef CONFIG_NET_CLS_ACT
2981 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2987 /* Handle special case of bridge or macvlan */
2988 rx_handler = rcu_dereference(skb->dev->rx_handler);
2991 ret = deliver_skb(skb, pt_prev, orig_dev);
2994 skb = rx_handler(skb);
2999 if (vlan_tx_tag_present(skb)) {
3001 ret = deliver_skb(skb, pt_prev, orig_dev);
3004 if (vlan_hwaccel_do_receive(&skb)) {
3005 ret = __netif_receive_skb(skb);
3007 } else if (unlikely(!skb))
3012 * Make sure frames received on VLAN interfaces stacked on
3013 * bonding interfaces still make their way to any base bonding
3014 * device that may have registered for a specific ptype. The
3015 * handler may have to adjust skb->dev and orig_dev.
3017 orig_or_bond = orig_dev;
3018 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
3019 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
3020 orig_or_bond = vlan_dev_real_dev(skb->dev);
3023 type = skb->protocol;
3024 list_for_each_entry_rcu(ptype,
3025 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3026 if (ptype->type == type && (ptype->dev == null_or_orig ||
3027 ptype->dev == skb->dev || ptype->dev == orig_dev ||
3028 ptype->dev == orig_or_bond)) {
3030 ret = deliver_skb(skb, pt_prev, orig_dev);
3036 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3038 atomic_long_inc(&skb->dev->rx_dropped);
3040 /* Jamal, now you will not able to escape explaining
3041 * me how you were going to use this. :-)
3052 * netif_receive_skb - process receive buffer from network
3053 * @skb: buffer to process
3055 * netif_receive_skb() is the main receive data processing function.
3056 * It always succeeds. The buffer may be dropped during processing
3057 * for congestion control or by the protocol layers.
3059 * This function may only be called from softirq context and interrupts
3060 * should be enabled.
3062 * Return values (usually ignored):
3063 * NET_RX_SUCCESS: no congestion
3064 * NET_RX_DROP: packet was dropped
3066 int netif_receive_skb(struct sk_buff *skb)
3068 if (netdev_tstamp_prequeue)
3069 net_timestamp_check(skb);
3071 if (skb_defer_rx_timestamp(skb))
3072 return NET_RX_SUCCESS;
3076 struct rps_dev_flow voidflow, *rflow = &voidflow;
3081 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3084 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3088 ret = __netif_receive_skb(skb);
3094 return __netif_receive_skb(skb);
3097 EXPORT_SYMBOL(netif_receive_skb);
3099 /* Network device is going away, flush any packets still pending
3100 * Called with irqs disabled.
3102 static void flush_backlog(void *arg)
3104 struct net_device *dev = arg;
3105 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3106 struct sk_buff *skb, *tmp;
3109 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3110 if (skb->dev == dev) {
3111 __skb_unlink(skb, &sd->input_pkt_queue);
3113 input_queue_head_incr(sd);
3118 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3119 if (skb->dev == dev) {
3120 __skb_unlink(skb, &sd->process_queue);
3122 input_queue_head_incr(sd);
3127 static int napi_gro_complete(struct sk_buff *skb)
3129 struct packet_type *ptype;
3130 __be16 type = skb->protocol;
3131 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3134 if (NAPI_GRO_CB(skb)->count == 1) {
3135 skb_shinfo(skb)->gso_size = 0;
3140 list_for_each_entry_rcu(ptype, head, list) {
3141 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3144 err = ptype->gro_complete(skb);
3150 WARN_ON(&ptype->list == head);
3152 return NET_RX_SUCCESS;
3156 return netif_receive_skb(skb);
3159 inline void napi_gro_flush(struct napi_struct *napi)
3161 struct sk_buff *skb, *next;
3163 for (skb = napi->gro_list; skb; skb = next) {
3166 napi_gro_complete(skb);
3169 napi->gro_count = 0;
3170 napi->gro_list = NULL;
3172 EXPORT_SYMBOL(napi_gro_flush);
3174 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3176 struct sk_buff **pp = NULL;
3177 struct packet_type *ptype;
3178 __be16 type = skb->protocol;
3179 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3182 enum gro_result ret;
3184 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3187 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3191 list_for_each_entry_rcu(ptype, head, list) {
3192 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3195 skb_set_network_header(skb, skb_gro_offset(skb));
3196 mac_len = skb->network_header - skb->mac_header;
3197 skb->mac_len = mac_len;
3198 NAPI_GRO_CB(skb)->same_flow = 0;
3199 NAPI_GRO_CB(skb)->flush = 0;
3200 NAPI_GRO_CB(skb)->free = 0;
3202 pp = ptype->gro_receive(&napi->gro_list, skb);
3207 if (&ptype->list == head)
3210 same_flow = NAPI_GRO_CB(skb)->same_flow;
3211 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3214 struct sk_buff *nskb = *pp;
3218 napi_gro_complete(nskb);
3225 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3229 NAPI_GRO_CB(skb)->count = 1;
3230 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3231 skb->next = napi->gro_list;
3232 napi->gro_list = skb;
3236 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3237 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3239 BUG_ON(skb->end - skb->tail < grow);
3241 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3244 skb->data_len -= grow;
3246 skb_shinfo(skb)->frags[0].page_offset += grow;
3247 skb_shinfo(skb)->frags[0].size -= grow;
3249 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3250 put_page(skb_shinfo(skb)->frags[0].page);
3251 memmove(skb_shinfo(skb)->frags,
3252 skb_shinfo(skb)->frags + 1,
3253 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3264 EXPORT_SYMBOL(dev_gro_receive);
3266 static inline gro_result_t
3267 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3271 for (p = napi->gro_list; p; p = p->next) {
3272 unsigned long diffs;
3274 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3275 diffs |= p->vlan_tci ^ skb->vlan_tci;
3276 diffs |= compare_ether_header(skb_mac_header(p),
3277 skb_gro_mac_header(skb));
3278 NAPI_GRO_CB(p)->same_flow = !diffs;
3279 NAPI_GRO_CB(p)->flush = 0;
3282 return dev_gro_receive(napi, skb);
3285 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3289 if (netif_receive_skb(skb))
3294 case GRO_MERGED_FREE:
3305 EXPORT_SYMBOL(napi_skb_finish);
3307 void skb_gro_reset_offset(struct sk_buff *skb)
3309 NAPI_GRO_CB(skb)->data_offset = 0;
3310 NAPI_GRO_CB(skb)->frag0 = NULL;
3311 NAPI_GRO_CB(skb)->frag0_len = 0;
3313 if (skb->mac_header == skb->tail &&
3314 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3315 NAPI_GRO_CB(skb)->frag0 =
3316 page_address(skb_shinfo(skb)->frags[0].page) +
3317 skb_shinfo(skb)->frags[0].page_offset;
3318 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3321 EXPORT_SYMBOL(skb_gro_reset_offset);
3323 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3325 skb_gro_reset_offset(skb);
3327 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3329 EXPORT_SYMBOL(napi_gro_receive);
3331 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3333 __skb_pull(skb, skb_headlen(skb));
3334 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3340 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3342 struct sk_buff *skb = napi->skb;
3345 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3351 EXPORT_SYMBOL(napi_get_frags);
3353 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3359 skb->protocol = eth_type_trans(skb, skb->dev);
3361 if (ret == GRO_HELD)
3362 skb_gro_pull(skb, -ETH_HLEN);
3363 else if (netif_receive_skb(skb))
3368 case GRO_MERGED_FREE:
3369 napi_reuse_skb(napi, skb);
3378 EXPORT_SYMBOL(napi_frags_finish);
3380 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3382 struct sk_buff *skb = napi->skb;
3389 skb_reset_mac_header(skb);
3390 skb_gro_reset_offset(skb);
3392 off = skb_gro_offset(skb);
3393 hlen = off + sizeof(*eth);
3394 eth = skb_gro_header_fast(skb, off);
3395 if (skb_gro_header_hard(skb, hlen)) {
3396 eth = skb_gro_header_slow(skb, hlen, off);
3397 if (unlikely(!eth)) {
3398 napi_reuse_skb(napi, skb);
3404 skb_gro_pull(skb, sizeof(*eth));
3407 * This works because the only protocols we care about don't require
3408 * special handling. We'll fix it up properly at the end.
3410 skb->protocol = eth->h_proto;
3415 EXPORT_SYMBOL(napi_frags_skb);
3417 gro_result_t napi_gro_frags(struct napi_struct *napi)
3419 struct sk_buff *skb = napi_frags_skb(napi);
3424 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3426 EXPORT_SYMBOL(napi_gro_frags);
3429 * net_rps_action sends any pending IPI's for rps.
3430 * Note: called with local irq disabled, but exits with local irq enabled.
3432 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3435 struct softnet_data *remsd = sd->rps_ipi_list;
3438 sd->rps_ipi_list = NULL;
3442 /* Send pending IPI's to kick RPS processing on remote cpus. */
3444 struct softnet_data *next = remsd->rps_ipi_next;
3446 if (cpu_online(remsd->cpu))
3447 __smp_call_function_single(remsd->cpu,
3456 static int process_backlog(struct napi_struct *napi, int quota)
3459 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3462 /* Check if we have pending ipi, its better to send them now,
3463 * not waiting net_rx_action() end.
3465 if (sd->rps_ipi_list) {
3466 local_irq_disable();
3467 net_rps_action_and_irq_enable(sd);
3470 napi->weight = weight_p;
3471 local_irq_disable();
3472 while (work < quota) {
3473 struct sk_buff *skb;
3476 while ((skb = __skb_dequeue(&sd->process_queue))) {
3478 __netif_receive_skb(skb);
3479 local_irq_disable();
3480 input_queue_head_incr(sd);
3481 if (++work >= quota) {
3488 qlen = skb_queue_len(&sd->input_pkt_queue);
3490 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3491 &sd->process_queue);
3493 if (qlen < quota - work) {
3495 * Inline a custom version of __napi_complete().
3496 * only current cpu owns and manipulates this napi,
3497 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3498 * we can use a plain write instead of clear_bit(),
3499 * and we dont need an smp_mb() memory barrier.
3501 list_del(&napi->poll_list);
3504 quota = work + qlen;
3514 * __napi_schedule - schedule for receive
3515 * @n: entry to schedule
3517 * The entry's receive function will be scheduled to run
3519 void __napi_schedule(struct napi_struct *n)
3521 unsigned long flags;
3523 local_irq_save(flags);
3524 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3525 local_irq_restore(flags);
3527 EXPORT_SYMBOL(__napi_schedule);
3529 void __napi_complete(struct napi_struct *n)
3531 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3532 BUG_ON(n->gro_list);
3534 list_del(&n->poll_list);
3535 smp_mb__before_clear_bit();
3536 clear_bit(NAPI_STATE_SCHED, &n->state);
3538 EXPORT_SYMBOL(__napi_complete);
3540 void napi_complete(struct napi_struct *n)
3542 unsigned long flags;
3545 * don't let napi dequeue from the cpu poll list
3546 * just in case its running on a different cpu
3548 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3552 local_irq_save(flags);
3554 local_irq_restore(flags);
3556 EXPORT_SYMBOL(napi_complete);
3558 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3559 int (*poll)(struct napi_struct *, int), int weight)
3561 INIT_LIST_HEAD(&napi->poll_list);
3562 napi->gro_count = 0;
3563 napi->gro_list = NULL;
3566 napi->weight = weight;
3567 list_add(&napi->dev_list, &dev->napi_list);
3569 #ifdef CONFIG_NETPOLL
3570 spin_lock_init(&napi->poll_lock);
3571 napi->poll_owner = -1;
3573 set_bit(NAPI_STATE_SCHED, &napi->state);
3575 EXPORT_SYMBOL(netif_napi_add);
3577 void netif_napi_del(struct napi_struct *napi)
3579 struct sk_buff *skb, *next;
3581 list_del_init(&napi->dev_list);
3582 napi_free_frags(napi);
3584 for (skb = napi->gro_list; skb; skb = next) {
3590 napi->gro_list = NULL;
3591 napi->gro_count = 0;
3593 EXPORT_SYMBOL(netif_napi_del);
3595 static void net_rx_action(struct softirq_action *h)
3597 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3598 unsigned long time_limit = jiffies + 2;
3599 int budget = netdev_budget;
3602 local_irq_disable();
3604 while (!list_empty(&sd->poll_list)) {
3605 struct napi_struct *n;
3608 /* If softirq window is exhuasted then punt.
3609 * Allow this to run for 2 jiffies since which will allow
3610 * an average latency of 1.5/HZ.
3612 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3617 /* Even though interrupts have been re-enabled, this
3618 * access is safe because interrupts can only add new
3619 * entries to the tail of this list, and only ->poll()
3620 * calls can remove this head entry from the list.
3622 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3624 have = netpoll_poll_lock(n);
3628 /* This NAPI_STATE_SCHED test is for avoiding a race
3629 * with netpoll's poll_napi(). Only the entity which
3630 * obtains the lock and sees NAPI_STATE_SCHED set will
3631 * actually make the ->poll() call. Therefore we avoid
3632 * accidently calling ->poll() when NAPI is not scheduled.
3635 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3636 work = n->poll(n, weight);
3640 WARN_ON_ONCE(work > weight);
3644 local_irq_disable();
3646 /* Drivers must not modify the NAPI state if they
3647 * consume the entire weight. In such cases this code
3648 * still "owns" the NAPI instance and therefore can
3649 * move the instance around on the list at-will.
3651 if (unlikely(work == weight)) {
3652 if (unlikely(napi_disable_pending(n))) {
3655 local_irq_disable();
3657 list_move_tail(&n->poll_list, &sd->poll_list);
3660 netpoll_poll_unlock(have);
3663 net_rps_action_and_irq_enable(sd);
3665 #ifdef CONFIG_NET_DMA
3667 * There may not be any more sk_buffs coming right now, so push
3668 * any pending DMA copies to hardware
3670 dma_issue_pending_all();
3677 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3681 static gifconf_func_t *gifconf_list[NPROTO];
3684 * register_gifconf - register a SIOCGIF handler
3685 * @family: Address family
3686 * @gifconf: Function handler
3688 * Register protocol dependent address dumping routines. The handler
3689 * that is passed must not be freed or reused until it has been replaced
3690 * by another handler.
3692 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3694 if (family >= NPROTO)
3696 gifconf_list[family] = gifconf;
3699 EXPORT_SYMBOL(register_gifconf);
3703 * Map an interface index to its name (SIOCGIFNAME)
3707 * We need this ioctl for efficient implementation of the
3708 * if_indextoname() function required by the IPv6 API. Without
3709 * it, we would have to search all the interfaces to find a
3713 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3715 struct net_device *dev;
3719 * Fetch the caller's info block.
3722 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3726 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3732 strcpy(ifr.ifr_name, dev->name);
3735 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3741 * Perform a SIOCGIFCONF call. This structure will change
3742 * size eventually, and there is nothing I can do about it.
3743 * Thus we will need a 'compatibility mode'.
3746 static int dev_ifconf(struct net *net, char __user *arg)
3749 struct net_device *dev;
3756 * Fetch the caller's info block.
3759 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3766 * Loop over the interfaces, and write an info block for each.
3770 for_each_netdev(net, dev) {
3771 for (i = 0; i < NPROTO; i++) {
3772 if (gifconf_list[i]) {
3775 done = gifconf_list[i](dev, NULL, 0);
3777 done = gifconf_list[i](dev, pos + total,
3787 * All done. Write the updated control block back to the caller.
3789 ifc.ifc_len = total;
3792 * Both BSD and Solaris return 0 here, so we do too.
3794 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3797 #ifdef CONFIG_PROC_FS
3799 * This is invoked by the /proc filesystem handler to display a device
3802 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3805 struct net *net = seq_file_net(seq);
3807 struct net_device *dev;
3811 return SEQ_START_TOKEN;
3814 for_each_netdev_rcu(net, dev)
3821 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3823 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3824 first_net_device(seq_file_net(seq)) :
3825 next_net_device((struct net_device *)v);
3828 return rcu_dereference(dev);
3831 void dev_seq_stop(struct seq_file *seq, void *v)
3837 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3839 struct rtnl_link_stats64 temp;
3840 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3842 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3843 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3844 dev->name, stats->rx_bytes, stats->rx_packets,
3846 stats->rx_dropped + stats->rx_missed_errors,
3847 stats->rx_fifo_errors,
3848 stats->rx_length_errors + stats->rx_over_errors +
3849 stats->rx_crc_errors + stats->rx_frame_errors,
3850 stats->rx_compressed, stats->multicast,
3851 stats->tx_bytes, stats->tx_packets,
3852 stats->tx_errors, stats->tx_dropped,
3853 stats->tx_fifo_errors, stats->collisions,
3854 stats->tx_carrier_errors +
3855 stats->tx_aborted_errors +
3856 stats->tx_window_errors +
3857 stats->tx_heartbeat_errors,
3858 stats->tx_compressed);
3862 * Called from the PROCfs module. This now uses the new arbitrary sized
3863 * /proc/net interface to create /proc/net/dev
3865 static int dev_seq_show(struct seq_file *seq, void *v)
3867 if (v == SEQ_START_TOKEN)
3868 seq_puts(seq, "Inter-| Receive "
3870 " face |bytes packets errs drop fifo frame "
3871 "compressed multicast|bytes packets errs "
3872 "drop fifo colls carrier compressed\n");
3874 dev_seq_printf_stats(seq, v);
3878 static struct softnet_data *softnet_get_online(loff_t *pos)
3880 struct softnet_data *sd = NULL;
3882 while (*pos < nr_cpu_ids)
3883 if (cpu_online(*pos)) {
3884 sd = &per_cpu(softnet_data, *pos);
3891 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3893 return softnet_get_online(pos);
3896 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3899 return softnet_get_online(pos);
3902 static void softnet_seq_stop(struct seq_file *seq, void *v)
3906 static int softnet_seq_show(struct seq_file *seq, void *v)
3908 struct softnet_data *sd = v;
3910 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3911 sd->processed, sd->dropped, sd->time_squeeze, 0,
3912 0, 0, 0, 0, /* was fastroute */
3913 sd->cpu_collision, sd->received_rps);
3917 static const struct seq_operations dev_seq_ops = {
3918 .start = dev_seq_start,
3919 .next = dev_seq_next,
3920 .stop = dev_seq_stop,
3921 .show = dev_seq_show,
3924 static int dev_seq_open(struct inode *inode, struct file *file)
3926 return seq_open_net(inode, file, &dev_seq_ops,
3927 sizeof(struct seq_net_private));
3930 static const struct file_operations dev_seq_fops = {
3931 .owner = THIS_MODULE,
3932 .open = dev_seq_open,
3934 .llseek = seq_lseek,
3935 .release = seq_release_net,
3938 static const struct seq_operations softnet_seq_ops = {
3939 .start = softnet_seq_start,
3940 .next = softnet_seq_next,
3941 .stop = softnet_seq_stop,
3942 .show = softnet_seq_show,
3945 static int softnet_seq_open(struct inode *inode, struct file *file)
3947 return seq_open(file, &softnet_seq_ops);
3950 static const struct file_operations softnet_seq_fops = {
3951 .owner = THIS_MODULE,
3952 .open = softnet_seq_open,
3954 .llseek = seq_lseek,
3955 .release = seq_release,
3958 static void *ptype_get_idx(loff_t pos)
3960 struct packet_type *pt = NULL;
3964 list_for_each_entry_rcu(pt, &ptype_all, list) {
3970 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3971 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3980 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3984 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3987 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3989 struct packet_type *pt;
3990 struct list_head *nxt;
3994 if (v == SEQ_START_TOKEN)
3995 return ptype_get_idx(0);
3998 nxt = pt->list.next;
3999 if (pt->type == htons(ETH_P_ALL)) {
4000 if (nxt != &ptype_all)
4003 nxt = ptype_base[0].next;
4005 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4007 while (nxt == &ptype_base[hash]) {
4008 if (++hash >= PTYPE_HASH_SIZE)
4010 nxt = ptype_base[hash].next;
4013 return list_entry(nxt, struct packet_type, list);
4016 static void ptype_seq_stop(struct seq_file *seq, void *v)
4022 static int ptype_seq_show(struct seq_file *seq, void *v)
4024 struct packet_type *pt = v;
4026 if (v == SEQ_START_TOKEN)
4027 seq_puts(seq, "Type Device Function\n");
4028 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4029 if (pt->type == htons(ETH_P_ALL))
4030 seq_puts(seq, "ALL ");
4032 seq_printf(seq, "%04x", ntohs(pt->type));
4034 seq_printf(seq, " %-8s %pF\n",
4035 pt->dev ? pt->dev->name : "", pt->func);
4041 static const struct seq_operations ptype_seq_ops = {
4042 .start = ptype_seq_start,
4043 .next = ptype_seq_next,
4044 .stop = ptype_seq_stop,
4045 .show = ptype_seq_show,
4048 static int ptype_seq_open(struct inode *inode, struct file *file)
4050 return seq_open_net(inode, file, &ptype_seq_ops,
4051 sizeof(struct seq_net_private));
4054 static const struct file_operations ptype_seq_fops = {
4055 .owner = THIS_MODULE,
4056 .open = ptype_seq_open,
4058 .llseek = seq_lseek,
4059 .release = seq_release_net,
4063 static int __net_init dev_proc_net_init(struct net *net)
4067 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4069 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4071 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4074 if (wext_proc_init(net))
4080 proc_net_remove(net, "ptype");
4082 proc_net_remove(net, "softnet_stat");
4084 proc_net_remove(net, "dev");
4088 static void __net_exit dev_proc_net_exit(struct net *net)
4090 wext_proc_exit(net);
4092 proc_net_remove(net, "ptype");
4093 proc_net_remove(net, "softnet_stat");
4094 proc_net_remove(net, "dev");
4097 static struct pernet_operations __net_initdata dev_proc_ops = {
4098 .init = dev_proc_net_init,
4099 .exit = dev_proc_net_exit,
4102 static int __init dev_proc_init(void)
4104 return register_pernet_subsys(&dev_proc_ops);
4107 #define dev_proc_init() 0
4108 #endif /* CONFIG_PROC_FS */
4112 * netdev_set_master - set up master/slave pair
4113 * @slave: slave device
4114 * @master: new master device
4116 * Changes the master device of the slave. Pass %NULL to break the
4117 * bonding. The caller must hold the RTNL semaphore. On a failure
4118 * a negative errno code is returned. On success the reference counts
4119 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4120 * function returns zero.
4122 int netdev_set_master(struct net_device *slave, struct net_device *master)
4124 struct net_device *old = slave->master;
4134 slave->master = master;
4141 slave->flags |= IFF_SLAVE;
4143 slave->flags &= ~IFF_SLAVE;
4145 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4148 EXPORT_SYMBOL(netdev_set_master);
4150 static void dev_change_rx_flags(struct net_device *dev, int flags)
4152 const struct net_device_ops *ops = dev->netdev_ops;
4154 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4155 ops->ndo_change_rx_flags(dev, flags);
4158 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4160 unsigned short old_flags = dev->flags;
4166 dev->flags |= IFF_PROMISC;
4167 dev->promiscuity += inc;
4168 if (dev->promiscuity == 0) {
4171 * If inc causes overflow, untouch promisc and return error.
4174 dev->flags &= ~IFF_PROMISC;
4176 dev->promiscuity -= inc;
4177 printk(KERN_WARNING "%s: promiscuity touches roof, "
4178 "set promiscuity failed, promiscuity feature "
4179 "of device might be broken.\n", dev->name);
4183 if (dev->flags != old_flags) {
4184 printk(KERN_INFO "device %s %s promiscuous mode\n",
4185 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4187 if (audit_enabled) {
4188 current_uid_gid(&uid, &gid);
4189 audit_log(current->audit_context, GFP_ATOMIC,
4190 AUDIT_ANOM_PROMISCUOUS,
4191 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4192 dev->name, (dev->flags & IFF_PROMISC),
4193 (old_flags & IFF_PROMISC),
4194 audit_get_loginuid(current),
4196 audit_get_sessionid(current));
4199 dev_change_rx_flags(dev, IFF_PROMISC);
4205 * dev_set_promiscuity - update promiscuity count on a device
4209 * Add or remove promiscuity from a device. While the count in the device
4210 * remains above zero the interface remains promiscuous. Once it hits zero
4211 * the device reverts back to normal filtering operation. A negative inc
4212 * value is used to drop promiscuity on the device.
4213 * Return 0 if successful or a negative errno code on error.
4215 int dev_set_promiscuity(struct net_device *dev, int inc)
4217 unsigned short old_flags = dev->flags;
4220 err = __dev_set_promiscuity(dev, inc);
4223 if (dev->flags != old_flags)
4224 dev_set_rx_mode(dev);
4227 EXPORT_SYMBOL(dev_set_promiscuity);
4230 * dev_set_allmulti - update allmulti count on a device
4234 * Add or remove reception of all multicast frames to a device. While the
4235 * count in the device remains above zero the interface remains listening
4236 * to all interfaces. Once it hits zero the device reverts back to normal
4237 * filtering operation. A negative @inc value is used to drop the counter
4238 * when releasing a resource needing all multicasts.
4239 * Return 0 if successful or a negative errno code on error.
4242 int dev_set_allmulti(struct net_device *dev, int inc)
4244 unsigned short old_flags = dev->flags;
4248 dev->flags |= IFF_ALLMULTI;
4249 dev->allmulti += inc;
4250 if (dev->allmulti == 0) {
4253 * If inc causes overflow, untouch allmulti and return error.
4256 dev->flags &= ~IFF_ALLMULTI;
4258 dev->allmulti -= inc;
4259 printk(KERN_WARNING "%s: allmulti touches roof, "
4260 "set allmulti failed, allmulti feature of "
4261 "device might be broken.\n", dev->name);
4265 if (dev->flags ^ old_flags) {
4266 dev_change_rx_flags(dev, IFF_ALLMULTI);
4267 dev_set_rx_mode(dev);
4271 EXPORT_SYMBOL(dev_set_allmulti);
4274 * Upload unicast and multicast address lists to device and
4275 * configure RX filtering. When the device doesn't support unicast
4276 * filtering it is put in promiscuous mode while unicast addresses
4279 void __dev_set_rx_mode(struct net_device *dev)
4281 const struct net_device_ops *ops = dev->netdev_ops;
4283 /* dev_open will call this function so the list will stay sane. */
4284 if (!(dev->flags&IFF_UP))
4287 if (!netif_device_present(dev))
4290 if (ops->ndo_set_rx_mode)
4291 ops->ndo_set_rx_mode(dev);
4293 /* Unicast addresses changes may only happen under the rtnl,
4294 * therefore calling __dev_set_promiscuity here is safe.
4296 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4297 __dev_set_promiscuity(dev, 1);
4298 dev->uc_promisc = 1;
4299 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4300 __dev_set_promiscuity(dev, -1);
4301 dev->uc_promisc = 0;
4304 if (ops->ndo_set_multicast_list)
4305 ops->ndo_set_multicast_list(dev);
4309 void dev_set_rx_mode(struct net_device *dev)
4311 netif_addr_lock_bh(dev);
4312 __dev_set_rx_mode(dev);
4313 netif_addr_unlock_bh(dev);
4317 * dev_get_flags - get flags reported to userspace
4320 * Get the combination of flag bits exported through APIs to userspace.
4322 unsigned dev_get_flags(const struct net_device *dev)
4326 flags = (dev->flags & ~(IFF_PROMISC |
4331 (dev->gflags & (IFF_PROMISC |
4334 if (netif_running(dev)) {
4335 if (netif_oper_up(dev))
4336 flags |= IFF_RUNNING;
4337 if (netif_carrier_ok(dev))
4338 flags |= IFF_LOWER_UP;
4339 if (netif_dormant(dev))
4340 flags |= IFF_DORMANT;
4345 EXPORT_SYMBOL(dev_get_flags);
4347 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4349 int old_flags = dev->flags;
4355 * Set the flags on our device.
4358 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4359 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4361 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4365 * Load in the correct multicast list now the flags have changed.
4368 if ((old_flags ^ flags) & IFF_MULTICAST)
4369 dev_change_rx_flags(dev, IFF_MULTICAST);
4371 dev_set_rx_mode(dev);
4374 * Have we downed the interface. We handle IFF_UP ourselves
4375 * according to user attempts to set it, rather than blindly
4380 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4381 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4384 dev_set_rx_mode(dev);
4387 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4388 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4390 dev->gflags ^= IFF_PROMISC;
4391 dev_set_promiscuity(dev, inc);
4394 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4395 is important. Some (broken) drivers set IFF_PROMISC, when
4396 IFF_ALLMULTI is requested not asking us and not reporting.
4398 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4399 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4401 dev->gflags ^= IFF_ALLMULTI;
4402 dev_set_allmulti(dev, inc);
4408 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4410 unsigned int changes = dev->flags ^ old_flags;
4412 if (changes & IFF_UP) {
4413 if (dev->flags & IFF_UP)
4414 call_netdevice_notifiers(NETDEV_UP, dev);
4416 call_netdevice_notifiers(NETDEV_DOWN, dev);
4419 if (dev->flags & IFF_UP &&
4420 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4421 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4425 * dev_change_flags - change device settings
4427 * @flags: device state flags
4429 * Change settings on device based state flags. The flags are
4430 * in the userspace exported format.
4432 int dev_change_flags(struct net_device *dev, unsigned flags)
4435 int old_flags = dev->flags;
4437 ret = __dev_change_flags(dev, flags);
4441 changes = old_flags ^ dev->flags;
4443 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4445 __dev_notify_flags(dev, old_flags);
4448 EXPORT_SYMBOL(dev_change_flags);
4451 * dev_set_mtu - Change maximum transfer unit
4453 * @new_mtu: new transfer unit
4455 * Change the maximum transfer size of the network device.
4457 int dev_set_mtu(struct net_device *dev, int new_mtu)
4459 const struct net_device_ops *ops = dev->netdev_ops;
4462 if (new_mtu == dev->mtu)
4465 /* MTU must be positive. */
4469 if (!netif_device_present(dev))
4473 if (ops->ndo_change_mtu)
4474 err = ops->ndo_change_mtu(dev, new_mtu);
4478 if (!err && dev->flags & IFF_UP)
4479 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4482 EXPORT_SYMBOL(dev_set_mtu);
4485 * dev_set_mac_address - Change Media Access Control Address
4489 * Change the hardware (MAC) address of the device
4491 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4493 const struct net_device_ops *ops = dev->netdev_ops;
4496 if (!ops->ndo_set_mac_address)
4498 if (sa->sa_family != dev->type)
4500 if (!netif_device_present(dev))
4502 err = ops->ndo_set_mac_address(dev, sa);
4504 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4507 EXPORT_SYMBOL(dev_set_mac_address);
4510 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4512 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4515 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4521 case SIOCGIFFLAGS: /* Get interface flags */
4522 ifr->ifr_flags = (short) dev_get_flags(dev);
4525 case SIOCGIFMETRIC: /* Get the metric on the interface
4526 (currently unused) */
4527 ifr->ifr_metric = 0;
4530 case SIOCGIFMTU: /* Get the MTU of a device */
4531 ifr->ifr_mtu = dev->mtu;
4536 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4538 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4539 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4540 ifr->ifr_hwaddr.sa_family = dev->type;
4548 ifr->ifr_map.mem_start = dev->mem_start;
4549 ifr->ifr_map.mem_end = dev->mem_end;
4550 ifr->ifr_map.base_addr = dev->base_addr;
4551 ifr->ifr_map.irq = dev->irq;
4552 ifr->ifr_map.dma = dev->dma;
4553 ifr->ifr_map.port = dev->if_port;
4557 ifr->ifr_ifindex = dev->ifindex;
4561 ifr->ifr_qlen = dev->tx_queue_len;
4565 /* dev_ioctl() should ensure this case
4577 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4579 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4582 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4583 const struct net_device_ops *ops;
4588 ops = dev->netdev_ops;
4591 case SIOCSIFFLAGS: /* Set interface flags */
4592 return dev_change_flags(dev, ifr->ifr_flags);
4594 case SIOCSIFMETRIC: /* Set the metric on the interface
4595 (currently unused) */
4598 case SIOCSIFMTU: /* Set the MTU of a device */
4599 return dev_set_mtu(dev, ifr->ifr_mtu);
4602 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4604 case SIOCSIFHWBROADCAST:
4605 if (ifr->ifr_hwaddr.sa_family != dev->type)
4607 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4608 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4609 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4613 if (ops->ndo_set_config) {
4614 if (!netif_device_present(dev))
4616 return ops->ndo_set_config(dev, &ifr->ifr_map);
4621 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4622 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4624 if (!netif_device_present(dev))
4626 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4629 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4630 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4632 if (!netif_device_present(dev))
4634 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4637 if (ifr->ifr_qlen < 0)
4639 dev->tx_queue_len = ifr->ifr_qlen;
4643 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4644 return dev_change_name(dev, ifr->ifr_newname);
4647 * Unknown or private ioctl
4650 if ((cmd >= SIOCDEVPRIVATE &&
4651 cmd <= SIOCDEVPRIVATE + 15) ||
4652 cmd == SIOCBONDENSLAVE ||
4653 cmd == SIOCBONDRELEASE ||
4654 cmd == SIOCBONDSETHWADDR ||
4655 cmd == SIOCBONDSLAVEINFOQUERY ||
4656 cmd == SIOCBONDINFOQUERY ||
4657 cmd == SIOCBONDCHANGEACTIVE ||
4658 cmd == SIOCGMIIPHY ||
4659 cmd == SIOCGMIIREG ||
4660 cmd == SIOCSMIIREG ||
4661 cmd == SIOCBRADDIF ||
4662 cmd == SIOCBRDELIF ||
4663 cmd == SIOCSHWTSTAMP ||
4664 cmd == SIOCWANDEV) {
4666 if (ops->ndo_do_ioctl) {
4667 if (netif_device_present(dev))
4668 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4680 * This function handles all "interface"-type I/O control requests. The actual
4681 * 'doing' part of this is dev_ifsioc above.
4685 * dev_ioctl - network device ioctl
4686 * @net: the applicable net namespace
4687 * @cmd: command to issue
4688 * @arg: pointer to a struct ifreq in user space
4690 * Issue ioctl functions to devices. This is normally called by the
4691 * user space syscall interfaces but can sometimes be useful for
4692 * other purposes. The return value is the return from the syscall if
4693 * positive or a negative errno code on error.
4696 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4702 /* One special case: SIOCGIFCONF takes ifconf argument
4703 and requires shared lock, because it sleeps writing
4707 if (cmd == SIOCGIFCONF) {
4709 ret = dev_ifconf(net, (char __user *) arg);
4713 if (cmd == SIOCGIFNAME)
4714 return dev_ifname(net, (struct ifreq __user *)arg);
4716 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4719 ifr.ifr_name[IFNAMSIZ-1] = 0;
4721 colon = strchr(ifr.ifr_name, ':');
4726 * See which interface the caller is talking about.
4731 * These ioctl calls:
4732 * - can be done by all.
4733 * - atomic and do not require locking.
4744 dev_load(net, ifr.ifr_name);
4746 ret = dev_ifsioc_locked(net, &ifr, cmd);
4751 if (copy_to_user(arg, &ifr,
4752 sizeof(struct ifreq)))
4758 dev_load(net, ifr.ifr_name);
4760 ret = dev_ethtool(net, &ifr);
4765 if (copy_to_user(arg, &ifr,
4766 sizeof(struct ifreq)))
4772 * These ioctl calls:
4773 * - require superuser power.
4774 * - require strict serialization.
4780 if (!capable(CAP_NET_ADMIN))
4782 dev_load(net, ifr.ifr_name);
4784 ret = dev_ifsioc(net, &ifr, cmd);
4789 if (copy_to_user(arg, &ifr,
4790 sizeof(struct ifreq)))
4796 * These ioctl calls:
4797 * - require superuser power.
4798 * - require strict serialization.
4799 * - do not return a value
4809 case SIOCSIFHWBROADCAST:
4812 case SIOCBONDENSLAVE:
4813 case SIOCBONDRELEASE:
4814 case SIOCBONDSETHWADDR:
4815 case SIOCBONDCHANGEACTIVE:
4819 if (!capable(CAP_NET_ADMIN))
4822 case SIOCBONDSLAVEINFOQUERY:
4823 case SIOCBONDINFOQUERY:
4824 dev_load(net, ifr.ifr_name);
4826 ret = dev_ifsioc(net, &ifr, cmd);
4831 /* Get the per device memory space. We can add this but
4832 * currently do not support it */
4834 /* Set the per device memory buffer space.
4835 * Not applicable in our case */
4840 * Unknown or private ioctl.
4843 if (cmd == SIOCWANDEV ||
4844 (cmd >= SIOCDEVPRIVATE &&
4845 cmd <= SIOCDEVPRIVATE + 15)) {
4846 dev_load(net, ifr.ifr_name);
4848 ret = dev_ifsioc(net, &ifr, cmd);
4850 if (!ret && copy_to_user(arg, &ifr,
4851 sizeof(struct ifreq)))
4855 /* Take care of Wireless Extensions */
4856 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4857 return wext_handle_ioctl(net, &ifr, cmd, arg);
4864 * dev_new_index - allocate an ifindex
4865 * @net: the applicable net namespace
4867 * Returns a suitable unique value for a new device interface
4868 * number. The caller must hold the rtnl semaphore or the
4869 * dev_base_lock to be sure it remains unique.
4871 static int dev_new_index(struct net *net)
4877 if (!__dev_get_by_index(net, ifindex))
4882 /* Delayed registration/unregisteration */
4883 static LIST_HEAD(net_todo_list);
4885 static void net_set_todo(struct net_device *dev)
4887 list_add_tail(&dev->todo_list, &net_todo_list);
4890 static void rollback_registered_many(struct list_head *head)
4892 struct net_device *dev, *tmp;
4894 BUG_ON(dev_boot_phase);
4897 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4898 /* Some devices call without registering
4899 * for initialization unwind. Remove those
4900 * devices and proceed with the remaining.
4902 if (dev->reg_state == NETREG_UNINITIALIZED) {
4903 pr_debug("unregister_netdevice: device %s/%p never "
4904 "was registered\n", dev->name, dev);
4907 list_del(&dev->unreg_list);
4911 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4913 /* If device is running, close it first. */
4916 /* And unlink it from device chain. */
4917 unlist_netdevice(dev);
4919 dev->reg_state = NETREG_UNREGISTERING;
4924 list_for_each_entry(dev, head, unreg_list) {
4925 /* Shutdown queueing discipline. */
4929 /* Notify protocols, that we are about to destroy
4930 this device. They should clean all the things.
4932 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4934 if (!dev->rtnl_link_ops ||
4935 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4936 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4939 * Flush the unicast and multicast chains
4944 if (dev->netdev_ops->ndo_uninit)
4945 dev->netdev_ops->ndo_uninit(dev);
4947 /* Notifier chain MUST detach us from master device. */
4948 WARN_ON(dev->master);
4950 /* Remove entries from kobject tree */
4951 netdev_unregister_kobject(dev);
4954 /* Process any work delayed until the end of the batch */
4955 dev = list_first_entry(head, struct net_device, unreg_list);
4956 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4960 list_for_each_entry(dev, head, unreg_list)
4964 static void rollback_registered(struct net_device *dev)
4968 list_add(&dev->unreg_list, &single);
4969 rollback_registered_many(&single);
4972 unsigned long netdev_fix_features(unsigned long features, const char *name)
4974 /* Fix illegal SG+CSUM combinations. */
4975 if ((features & NETIF_F_SG) &&
4976 !(features & NETIF_F_ALL_CSUM)) {
4978 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4979 "checksum feature.\n", name);
4980 features &= ~NETIF_F_SG;
4983 /* TSO requires that SG is present as well. */
4984 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4986 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4987 "SG feature.\n", name);
4988 features &= ~NETIF_F_TSO;
4991 if (features & NETIF_F_UFO) {
4992 if (!(features & NETIF_F_GEN_CSUM)) {
4994 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4995 "since no NETIF_F_HW_CSUM feature.\n",
4997 features &= ~NETIF_F_UFO;
5000 if (!(features & NETIF_F_SG)) {
5002 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5003 "since no NETIF_F_SG feature.\n", name);
5004 features &= ~NETIF_F_UFO;
5010 EXPORT_SYMBOL(netdev_fix_features);
5013 * netif_stacked_transfer_operstate - transfer operstate
5014 * @rootdev: the root or lower level device to transfer state from
5015 * @dev: the device to transfer operstate to
5017 * Transfer operational state from root to device. This is normally
5018 * called when a stacking relationship exists between the root
5019 * device and the device(a leaf device).
5021 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5022 struct net_device *dev)
5024 if (rootdev->operstate == IF_OPER_DORMANT)
5025 netif_dormant_on(dev);
5027 netif_dormant_off(dev);
5029 if (netif_carrier_ok(rootdev)) {
5030 if (!netif_carrier_ok(dev))
5031 netif_carrier_on(dev);
5033 if (netif_carrier_ok(dev))
5034 netif_carrier_off(dev);
5037 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5039 static int netif_alloc_rx_queues(struct net_device *dev)
5042 unsigned int i, count = dev->num_rx_queues;
5043 struct netdev_rx_queue *rx;
5047 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5049 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5055 * Set a pointer to first element in the array which holds the
5058 for (i = 0; i < count; i++)
5064 static int netif_alloc_netdev_queues(struct net_device *dev)
5066 unsigned int count = dev->num_tx_queues;
5067 struct netdev_queue *tx;
5071 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5073 pr_err("netdev: Unable to allocate %u tx queues.\n",
5081 static void netdev_init_one_queue(struct net_device *dev,
5082 struct netdev_queue *queue,
5087 /* Initialize queue lock */
5088 spin_lock_init(&queue->_xmit_lock);
5089 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5090 queue->xmit_lock_owner = -1;
5093 static void netdev_init_queues(struct net_device *dev)
5095 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5096 spin_lock_init(&dev->tx_global_lock);
5100 * register_netdevice - register a network device
5101 * @dev: device to register
5103 * Take a completed network device structure and add it to the kernel
5104 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5105 * chain. 0 is returned on success. A negative errno code is returned
5106 * on a failure to set up the device, or if the name is a duplicate.
5108 * Callers must hold the rtnl semaphore. You may want
5109 * register_netdev() instead of this.
5112 * The locking appears insufficient to guarantee two parallel registers
5113 * will not get the same name.
5116 int register_netdevice(struct net_device *dev)
5119 struct net *net = dev_net(dev);
5121 BUG_ON(dev_boot_phase);
5126 /* When net_device's are persistent, this will be fatal. */
5127 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5130 spin_lock_init(&dev->addr_list_lock);
5131 netdev_set_addr_lockdep_class(dev);
5135 ret = netif_alloc_rx_queues(dev);
5139 netdev_init_queues(dev);
5141 /* Init, if this function is available */
5142 if (dev->netdev_ops->ndo_init) {
5143 ret = dev->netdev_ops->ndo_init(dev);
5151 ret = dev_get_valid_name(dev, dev->name, 0);
5155 dev->ifindex = dev_new_index(net);
5156 if (dev->iflink == -1)
5157 dev->iflink = dev->ifindex;
5159 /* Fix illegal checksum combinations */
5160 if ((dev->features & NETIF_F_HW_CSUM) &&
5161 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5162 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5164 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5167 if ((dev->features & NETIF_F_NO_CSUM) &&
5168 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5169 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5171 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5174 dev->features = netdev_fix_features(dev->features, dev->name);
5176 /* Enable software GSO if SG is supported. */
5177 if (dev->features & NETIF_F_SG)
5178 dev->features |= NETIF_F_GSO;
5180 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5181 * vlan_dev_init() will do the dev->features check, so these features
5182 * are enabled only if supported by underlying device.
5184 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5186 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5187 ret = notifier_to_errno(ret);
5191 ret = netdev_register_kobject(dev);
5194 dev->reg_state = NETREG_REGISTERED;
5197 * Default initial state at registry is that the
5198 * device is present.
5201 set_bit(__LINK_STATE_PRESENT, &dev->state);
5203 dev_init_scheduler(dev);
5205 list_netdevice(dev);
5207 /* Notify protocols, that a new device appeared. */
5208 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5209 ret = notifier_to_errno(ret);
5211 rollback_registered(dev);
5212 dev->reg_state = NETREG_UNREGISTERED;
5215 * Prevent userspace races by waiting until the network
5216 * device is fully setup before sending notifications.
5218 if (!dev->rtnl_link_ops ||
5219 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5220 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5226 if (dev->netdev_ops->ndo_uninit)
5227 dev->netdev_ops->ndo_uninit(dev);
5230 EXPORT_SYMBOL(register_netdevice);
5233 * init_dummy_netdev - init a dummy network device for NAPI
5234 * @dev: device to init
5236 * This takes a network device structure and initialize the minimum
5237 * amount of fields so it can be used to schedule NAPI polls without
5238 * registering a full blown interface. This is to be used by drivers
5239 * that need to tie several hardware interfaces to a single NAPI
5240 * poll scheduler due to HW limitations.
5242 int init_dummy_netdev(struct net_device *dev)
5244 /* Clear everything. Note we don't initialize spinlocks
5245 * are they aren't supposed to be taken by any of the
5246 * NAPI code and this dummy netdev is supposed to be
5247 * only ever used for NAPI polls
5249 memset(dev, 0, sizeof(struct net_device));
5251 /* make sure we BUG if trying to hit standard
5252 * register/unregister code path
5254 dev->reg_state = NETREG_DUMMY;
5256 /* NAPI wants this */
5257 INIT_LIST_HEAD(&dev->napi_list);
5259 /* a dummy interface is started by default */
5260 set_bit(__LINK_STATE_PRESENT, &dev->state);
5261 set_bit(__LINK_STATE_START, &dev->state);
5263 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5264 * because users of this 'device' dont need to change
5270 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5274 * register_netdev - register a network device
5275 * @dev: device to register
5277 * Take a completed network device structure and add it to the kernel
5278 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5279 * chain. 0 is returned on success. A negative errno code is returned
5280 * on a failure to set up the device, or if the name is a duplicate.
5282 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5283 * and expands the device name if you passed a format string to
5286 int register_netdev(struct net_device *dev)
5293 * If the name is a format string the caller wants us to do a
5296 if (strchr(dev->name, '%')) {
5297 err = dev_alloc_name(dev, dev->name);
5302 err = register_netdevice(dev);
5307 EXPORT_SYMBOL(register_netdev);
5309 int netdev_refcnt_read(const struct net_device *dev)
5313 for_each_possible_cpu(i)
5314 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5317 EXPORT_SYMBOL(netdev_refcnt_read);
5320 * netdev_wait_allrefs - wait until all references are gone.
5322 * This is called when unregistering network devices.
5324 * Any protocol or device that holds a reference should register
5325 * for netdevice notification, and cleanup and put back the
5326 * reference if they receive an UNREGISTER event.
5327 * We can get stuck here if buggy protocols don't correctly
5330 static void netdev_wait_allrefs(struct net_device *dev)
5332 unsigned long rebroadcast_time, warning_time;
5335 linkwatch_forget_dev(dev);
5337 rebroadcast_time = warning_time = jiffies;
5338 refcnt = netdev_refcnt_read(dev);
5340 while (refcnt != 0) {
5341 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5344 /* Rebroadcast unregister notification */
5345 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5346 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5347 * should have already handle it the first time */
5349 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5351 /* We must not have linkwatch events
5352 * pending on unregister. If this
5353 * happens, we simply run the queue
5354 * unscheduled, resulting in a noop
5357 linkwatch_run_queue();
5362 rebroadcast_time = jiffies;
5367 refcnt = netdev_refcnt_read(dev);
5369 if (time_after(jiffies, warning_time + 10 * HZ)) {
5370 printk(KERN_EMERG "unregister_netdevice: "
5371 "waiting for %s to become free. Usage "
5374 warning_time = jiffies;
5383 * register_netdevice(x1);
5384 * register_netdevice(x2);
5386 * unregister_netdevice(y1);
5387 * unregister_netdevice(y2);
5393 * We are invoked by rtnl_unlock().
5394 * This allows us to deal with problems:
5395 * 1) We can delete sysfs objects which invoke hotplug
5396 * without deadlocking with linkwatch via keventd.
5397 * 2) Since we run with the RTNL semaphore not held, we can sleep
5398 * safely in order to wait for the netdev refcnt to drop to zero.
5400 * We must not return until all unregister events added during
5401 * the interval the lock was held have been completed.
5403 void netdev_run_todo(void)
5405 struct list_head list;
5407 /* Snapshot list, allow later requests */
5408 list_replace_init(&net_todo_list, &list);
5412 while (!list_empty(&list)) {
5413 struct net_device *dev
5414 = list_first_entry(&list, struct net_device, todo_list);
5415 list_del(&dev->todo_list);
5417 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5418 printk(KERN_ERR "network todo '%s' but state %d\n",
5419 dev->name, dev->reg_state);
5424 dev->reg_state = NETREG_UNREGISTERED;
5426 on_each_cpu(flush_backlog, dev, 1);
5428 netdev_wait_allrefs(dev);
5431 BUG_ON(netdev_refcnt_read(dev));
5432 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5433 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5434 WARN_ON(dev->dn_ptr);
5436 if (dev->destructor)
5437 dev->destructor(dev);
5439 /* Free network device */
5440 kobject_put(&dev->dev.kobj);
5445 * dev_txq_stats_fold - fold tx_queues stats
5446 * @dev: device to get statistics from
5447 * @stats: struct rtnl_link_stats64 to hold results
5449 void dev_txq_stats_fold(const struct net_device *dev,
5450 struct rtnl_link_stats64 *stats)
5452 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5454 struct netdev_queue *txq;
5456 for (i = 0; i < dev->num_tx_queues; i++) {
5457 txq = netdev_get_tx_queue(dev, i);
5458 spin_lock_bh(&txq->_xmit_lock);
5459 tx_bytes += txq->tx_bytes;
5460 tx_packets += txq->tx_packets;
5461 tx_dropped += txq->tx_dropped;
5462 spin_unlock_bh(&txq->_xmit_lock);
5464 if (tx_bytes || tx_packets || tx_dropped) {
5465 stats->tx_bytes = tx_bytes;
5466 stats->tx_packets = tx_packets;
5467 stats->tx_dropped = tx_dropped;
5470 EXPORT_SYMBOL(dev_txq_stats_fold);
5472 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5473 * fields in the same order, with only the type differing.
5475 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5476 const struct net_device_stats *netdev_stats)
5478 #if BITS_PER_LONG == 64
5479 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5480 memcpy(stats64, netdev_stats, sizeof(*stats64));
5482 size_t i, n = sizeof(*stats64) / sizeof(u64);
5483 const unsigned long *src = (const unsigned long *)netdev_stats;
5484 u64 *dst = (u64 *)stats64;
5486 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5487 sizeof(*stats64) / sizeof(u64));
5488 for (i = 0; i < n; i++)
5494 * dev_get_stats - get network device statistics
5495 * @dev: device to get statistics from
5496 * @storage: place to store stats
5498 * Get network statistics from device. Return @storage.
5499 * The device driver may provide its own method by setting
5500 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5501 * otherwise the internal statistics structure is used.
5503 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5504 struct rtnl_link_stats64 *storage)
5506 const struct net_device_ops *ops = dev->netdev_ops;
5508 if (ops->ndo_get_stats64) {
5509 memset(storage, 0, sizeof(*storage));
5510 ops->ndo_get_stats64(dev, storage);
5511 } else if (ops->ndo_get_stats) {
5512 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5514 netdev_stats_to_stats64(storage, &dev->stats);
5515 dev_txq_stats_fold(dev, storage);
5517 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5520 EXPORT_SYMBOL(dev_get_stats);
5522 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5524 struct netdev_queue *queue = dev_ingress_queue(dev);
5526 #ifdef CONFIG_NET_CLS_ACT
5529 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5532 netdev_init_one_queue(dev, queue, NULL);
5533 queue->qdisc = &noop_qdisc;
5534 queue->qdisc_sleeping = &noop_qdisc;
5535 rcu_assign_pointer(dev->ingress_queue, queue);
5541 * alloc_netdev_mq - allocate network device
5542 * @sizeof_priv: size of private data to allocate space for
5543 * @name: device name format string
5544 * @setup: callback to initialize device
5545 * @queue_count: the number of subqueues to allocate
5547 * Allocates a struct net_device with private data area for driver use
5548 * and performs basic initialization. Also allocates subquue structs
5549 * for each queue on the device at the end of the netdevice.
5551 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5552 void (*setup)(struct net_device *), unsigned int queue_count)
5554 struct net_device *dev;
5556 struct net_device *p;
5558 BUG_ON(strlen(name) >= sizeof(dev->name));
5560 if (queue_count < 1) {
5561 pr_err("alloc_netdev: Unable to allocate device "
5562 "with zero queues.\n");
5566 alloc_size = sizeof(struct net_device);
5568 /* ensure 32-byte alignment of private area */
5569 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5570 alloc_size += sizeof_priv;
5572 /* ensure 32-byte alignment of whole construct */
5573 alloc_size += NETDEV_ALIGN - 1;
5575 p = kzalloc(alloc_size, GFP_KERNEL);
5577 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5581 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5582 dev->padded = (char *)dev - (char *)p;
5584 dev->pcpu_refcnt = alloc_percpu(int);
5585 if (!dev->pcpu_refcnt)
5588 if (dev_addr_init(dev))
5594 dev_net_set(dev, &init_net);
5596 dev->num_tx_queues = queue_count;
5597 dev->real_num_tx_queues = queue_count;
5598 if (netif_alloc_netdev_queues(dev))
5602 dev->num_rx_queues = queue_count;
5603 dev->real_num_rx_queues = queue_count;
5606 dev->gso_max_size = GSO_MAX_SIZE;
5608 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5609 dev->ethtool_ntuple_list.count = 0;
5610 INIT_LIST_HEAD(&dev->napi_list);
5611 INIT_LIST_HEAD(&dev->unreg_list);
5612 INIT_LIST_HEAD(&dev->link_watch_list);
5613 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5615 strcpy(dev->name, name);
5619 free_percpu(dev->pcpu_refcnt);
5625 EXPORT_SYMBOL(alloc_netdev_mq);
5628 * free_netdev - free network device
5631 * This function does the last stage of destroying an allocated device
5632 * interface. The reference to the device object is released.
5633 * If this is the last reference then it will be freed.
5635 void free_netdev(struct net_device *dev)
5637 struct napi_struct *p, *n;
5639 release_net(dev_net(dev));
5643 kfree(rcu_dereference_raw(dev->ingress_queue));
5645 /* Flush device addresses */
5646 dev_addr_flush(dev);
5648 /* Clear ethtool n-tuple list */
5649 ethtool_ntuple_flush(dev);
5651 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5654 free_percpu(dev->pcpu_refcnt);
5655 dev->pcpu_refcnt = NULL;
5657 /* Compatibility with error handling in drivers */
5658 if (dev->reg_state == NETREG_UNINITIALIZED) {
5659 kfree((char *)dev - dev->padded);
5663 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5664 dev->reg_state = NETREG_RELEASED;
5666 /* will free via device release */
5667 put_device(&dev->dev);
5669 EXPORT_SYMBOL(free_netdev);
5672 * synchronize_net - Synchronize with packet receive processing
5674 * Wait for packets currently being received to be done.
5675 * Does not block later packets from starting.
5677 void synchronize_net(void)
5682 EXPORT_SYMBOL(synchronize_net);
5685 * unregister_netdevice_queue - remove device from the kernel
5689 * This function shuts down a device interface and removes it
5690 * from the kernel tables.
5691 * If head not NULL, device is queued to be unregistered later.
5693 * Callers must hold the rtnl semaphore. You may want
5694 * unregister_netdev() instead of this.
5697 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5702 list_move_tail(&dev->unreg_list, head);
5704 rollback_registered(dev);
5705 /* Finish processing unregister after unlock */
5709 EXPORT_SYMBOL(unregister_netdevice_queue);
5712 * unregister_netdevice_many - unregister many devices
5713 * @head: list of devices
5715 void unregister_netdevice_many(struct list_head *head)
5717 struct net_device *dev;
5719 if (!list_empty(head)) {
5720 rollback_registered_many(head);
5721 list_for_each_entry(dev, head, unreg_list)
5725 EXPORT_SYMBOL(unregister_netdevice_many);
5728 * unregister_netdev - remove device from the kernel
5731 * This function shuts down a device interface and removes it
5732 * from the kernel tables.
5734 * This is just a wrapper for unregister_netdevice that takes
5735 * the rtnl semaphore. In general you want to use this and not
5736 * unregister_netdevice.
5738 void unregister_netdev(struct net_device *dev)
5741 unregister_netdevice(dev);
5744 EXPORT_SYMBOL(unregister_netdev);
5747 * dev_change_net_namespace - move device to different nethost namespace
5749 * @net: network namespace
5750 * @pat: If not NULL name pattern to try if the current device name
5751 * is already taken in the destination network namespace.
5753 * This function shuts down a device interface and moves it
5754 * to a new network namespace. On success 0 is returned, on
5755 * a failure a netagive errno code is returned.
5757 * Callers must hold the rtnl semaphore.
5760 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5766 /* Don't allow namespace local devices to be moved. */
5768 if (dev->features & NETIF_F_NETNS_LOCAL)
5771 /* Ensure the device has been registrered */
5773 if (dev->reg_state != NETREG_REGISTERED)
5776 /* Get out if there is nothing todo */
5778 if (net_eq(dev_net(dev), net))
5781 /* Pick the destination device name, and ensure
5782 * we can use it in the destination network namespace.
5785 if (__dev_get_by_name(net, dev->name)) {
5786 /* We get here if we can't use the current device name */
5789 if (dev_get_valid_name(dev, pat, 1))
5794 * And now a mini version of register_netdevice unregister_netdevice.
5797 /* If device is running close it first. */
5800 /* And unlink it from device chain */
5802 unlist_netdevice(dev);
5806 /* Shutdown queueing discipline. */
5809 /* Notify protocols, that we are about to destroy
5810 this device. They should clean all the things.
5812 Note that dev->reg_state stays at NETREG_REGISTERED.
5813 This is wanted because this way 8021q and macvlan know
5814 the device is just moving and can keep their slaves up.
5816 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5817 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5820 * Flush the unicast and multicast chains
5825 /* Actually switch the network namespace */
5826 dev_net_set(dev, net);
5828 /* If there is an ifindex conflict assign a new one */
5829 if (__dev_get_by_index(net, dev->ifindex)) {
5830 int iflink = (dev->iflink == dev->ifindex);
5831 dev->ifindex = dev_new_index(net);
5833 dev->iflink = dev->ifindex;
5836 /* Fixup kobjects */
5837 err = device_rename(&dev->dev, dev->name);
5840 /* Add the device back in the hashes */
5841 list_netdevice(dev);
5843 /* Notify protocols, that a new device appeared. */
5844 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5847 * Prevent userspace races by waiting until the network
5848 * device is fully setup before sending notifications.
5850 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5857 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5859 static int dev_cpu_callback(struct notifier_block *nfb,
5860 unsigned long action,
5863 struct sk_buff **list_skb;
5864 struct sk_buff *skb;
5865 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5866 struct softnet_data *sd, *oldsd;
5868 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5871 local_irq_disable();
5872 cpu = smp_processor_id();
5873 sd = &per_cpu(softnet_data, cpu);
5874 oldsd = &per_cpu(softnet_data, oldcpu);
5876 /* Find end of our completion_queue. */
5877 list_skb = &sd->completion_queue;
5879 list_skb = &(*list_skb)->next;
5880 /* Append completion queue from offline CPU. */
5881 *list_skb = oldsd->completion_queue;
5882 oldsd->completion_queue = NULL;
5884 /* Append output queue from offline CPU. */
5885 if (oldsd->output_queue) {
5886 *sd->output_queue_tailp = oldsd->output_queue;
5887 sd->output_queue_tailp = oldsd->output_queue_tailp;
5888 oldsd->output_queue = NULL;
5889 oldsd->output_queue_tailp = &oldsd->output_queue;
5892 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5895 /* Process offline CPU's input_pkt_queue */
5896 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5898 input_queue_head_incr(oldsd);
5900 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5902 input_queue_head_incr(oldsd);
5910 * netdev_increment_features - increment feature set by one
5911 * @all: current feature set
5912 * @one: new feature set
5913 * @mask: mask feature set
5915 * Computes a new feature set after adding a device with feature set
5916 * @one to the master device with current feature set @all. Will not
5917 * enable anything that is off in @mask. Returns the new feature set.
5919 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5922 /* If device needs checksumming, downgrade to it. */
5923 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5924 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5925 else if (mask & NETIF_F_ALL_CSUM) {
5926 /* If one device supports v4/v6 checksumming, set for all. */
5927 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5928 !(all & NETIF_F_GEN_CSUM)) {
5929 all &= ~NETIF_F_ALL_CSUM;
5930 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5933 /* If one device supports hw checksumming, set for all. */
5934 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5935 all &= ~NETIF_F_ALL_CSUM;
5936 all |= NETIF_F_HW_CSUM;
5940 one |= NETIF_F_ALL_CSUM;
5942 one |= all & NETIF_F_ONE_FOR_ALL;
5943 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5944 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5948 EXPORT_SYMBOL(netdev_increment_features);
5950 static struct hlist_head *netdev_create_hash(void)
5953 struct hlist_head *hash;
5955 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5957 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5958 INIT_HLIST_HEAD(&hash[i]);
5963 /* Initialize per network namespace state */
5964 static int __net_init netdev_init(struct net *net)
5966 INIT_LIST_HEAD(&net->dev_base_head);
5968 net->dev_name_head = netdev_create_hash();
5969 if (net->dev_name_head == NULL)
5972 net->dev_index_head = netdev_create_hash();
5973 if (net->dev_index_head == NULL)
5979 kfree(net->dev_name_head);
5985 * netdev_drivername - network driver for the device
5986 * @dev: network device
5987 * @buffer: buffer for resulting name
5988 * @len: size of buffer
5990 * Determine network driver for device.
5992 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5994 const struct device_driver *driver;
5995 const struct device *parent;
5997 if (len <= 0 || !buffer)
6001 parent = dev->dev.parent;
6006 driver = parent->driver;
6007 if (driver && driver->name)
6008 strlcpy(buffer, driver->name, len);
6012 static int __netdev_printk(const char *level, const struct net_device *dev,
6013 struct va_format *vaf)
6017 if (dev && dev->dev.parent)
6018 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6019 netdev_name(dev), vaf);
6021 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6023 r = printk("%s(NULL net_device): %pV", level, vaf);
6028 int netdev_printk(const char *level, const struct net_device *dev,
6029 const char *format, ...)
6031 struct va_format vaf;
6035 va_start(args, format);
6040 r = __netdev_printk(level, dev, &vaf);
6045 EXPORT_SYMBOL(netdev_printk);
6047 #define define_netdev_printk_level(func, level) \
6048 int func(const struct net_device *dev, const char *fmt, ...) \
6051 struct va_format vaf; \
6054 va_start(args, fmt); \
6059 r = __netdev_printk(level, dev, &vaf); \
6064 EXPORT_SYMBOL(func);
6066 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6067 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6068 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6069 define_netdev_printk_level(netdev_err, KERN_ERR);
6070 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6071 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6072 define_netdev_printk_level(netdev_info, KERN_INFO);
6074 static void __net_exit netdev_exit(struct net *net)
6076 kfree(net->dev_name_head);
6077 kfree(net->dev_index_head);
6080 static struct pernet_operations __net_initdata netdev_net_ops = {
6081 .init = netdev_init,
6082 .exit = netdev_exit,
6085 static void __net_exit default_device_exit(struct net *net)
6087 struct net_device *dev, *aux;
6089 * Push all migratable network devices back to the
6090 * initial network namespace
6093 for_each_netdev_safe(net, dev, aux) {
6095 char fb_name[IFNAMSIZ];
6097 /* Ignore unmoveable devices (i.e. loopback) */
6098 if (dev->features & NETIF_F_NETNS_LOCAL)
6101 /* Leave virtual devices for the generic cleanup */
6102 if (dev->rtnl_link_ops)
6105 /* Push remaing network devices to init_net */
6106 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6107 err = dev_change_net_namespace(dev, &init_net, fb_name);
6109 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6110 __func__, dev->name, err);
6117 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6119 /* At exit all network devices most be removed from a network
6120 * namespace. Do this in the reverse order of registeration.
6121 * Do this across as many network namespaces as possible to
6122 * improve batching efficiency.
6124 struct net_device *dev;
6126 LIST_HEAD(dev_kill_list);
6129 list_for_each_entry(net, net_list, exit_list) {
6130 for_each_netdev_reverse(net, dev) {
6131 if (dev->rtnl_link_ops)
6132 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6134 unregister_netdevice_queue(dev, &dev_kill_list);
6137 unregister_netdevice_many(&dev_kill_list);
6141 static struct pernet_operations __net_initdata default_device_ops = {
6142 .exit = default_device_exit,
6143 .exit_batch = default_device_exit_batch,
6147 * Initialize the DEV module. At boot time this walks the device list and
6148 * unhooks any devices that fail to initialise (normally hardware not
6149 * present) and leaves us with a valid list of present and active devices.
6154 * This is called single threaded during boot, so no need
6155 * to take the rtnl semaphore.
6157 static int __init net_dev_init(void)
6159 int i, rc = -ENOMEM;
6161 BUG_ON(!dev_boot_phase);
6163 if (dev_proc_init())
6166 if (netdev_kobject_init())
6169 INIT_LIST_HEAD(&ptype_all);
6170 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6171 INIT_LIST_HEAD(&ptype_base[i]);
6173 if (register_pernet_subsys(&netdev_net_ops))
6177 * Initialise the packet receive queues.
6180 for_each_possible_cpu(i) {
6181 struct softnet_data *sd = &per_cpu(softnet_data, i);
6183 memset(sd, 0, sizeof(*sd));
6184 skb_queue_head_init(&sd->input_pkt_queue);
6185 skb_queue_head_init(&sd->process_queue);
6186 sd->completion_queue = NULL;
6187 INIT_LIST_HEAD(&sd->poll_list);
6188 sd->output_queue = NULL;
6189 sd->output_queue_tailp = &sd->output_queue;
6191 sd->csd.func = rps_trigger_softirq;
6197 sd->backlog.poll = process_backlog;
6198 sd->backlog.weight = weight_p;
6199 sd->backlog.gro_list = NULL;
6200 sd->backlog.gro_count = 0;
6205 /* The loopback device is special if any other network devices
6206 * is present in a network namespace the loopback device must
6207 * be present. Since we now dynamically allocate and free the
6208 * loopback device ensure this invariant is maintained by
6209 * keeping the loopback device as the first device on the
6210 * list of network devices. Ensuring the loopback devices
6211 * is the first device that appears and the last network device
6214 if (register_pernet_device(&loopback_net_ops))
6217 if (register_pernet_device(&default_device_ops))
6220 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6221 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6223 hotcpu_notifier(dev_cpu_callback, 0);
6231 subsys_initcall(net_dev_init);
6233 static int __init initialize_hashrnd(void)
6235 get_random_bytes(&hashrnd, sizeof(hashrnd));
6239 late_initcall_sync(initialize_hashrnd);