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;
1799 if (type == htons(ETH_P_8021Q)) {
1800 struct vlan_ethhdr *veh;
1802 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1803 return ERR_PTR(-EINVAL);
1805 veh = (struct vlan_ethhdr *)skb->data;
1806 type = veh->h_vlan_encapsulated_proto;
1809 skb_reset_mac_header(skb);
1810 skb->mac_len = skb->network_header - skb->mac_header;
1811 __skb_pull(skb, skb->mac_len);
1813 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1814 struct net_device *dev = skb->dev;
1815 struct ethtool_drvinfo info = {};
1817 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1818 dev->ethtool_ops->get_drvinfo(dev, &info);
1820 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1821 info.driver, dev ? dev->features : 0L,
1822 skb->sk ? skb->sk->sk_route_caps : 0L,
1823 skb->len, skb->data_len, skb->ip_summed);
1825 if (skb_header_cloned(skb) &&
1826 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1827 return ERR_PTR(err);
1831 list_for_each_entry_rcu(ptype,
1832 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1833 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1834 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1835 err = ptype->gso_send_check(skb);
1836 segs = ERR_PTR(err);
1837 if (err || skb_gso_ok(skb, features))
1839 __skb_push(skb, (skb->data -
1840 skb_network_header(skb)));
1842 segs = ptype->gso_segment(skb, features);
1848 __skb_push(skb, skb->data - skb_mac_header(skb));
1852 EXPORT_SYMBOL(skb_gso_segment);
1854 /* Take action when hardware reception checksum errors are detected. */
1856 void netdev_rx_csum_fault(struct net_device *dev)
1858 if (net_ratelimit()) {
1859 printk(KERN_ERR "%s: hw csum failure.\n",
1860 dev ? dev->name : "<unknown>");
1864 EXPORT_SYMBOL(netdev_rx_csum_fault);
1867 /* Actually, we should eliminate this check as soon as we know, that:
1868 * 1. IOMMU is present and allows to map all the memory.
1869 * 2. No high memory really exists on this machine.
1872 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1874 #ifdef CONFIG_HIGHMEM
1876 if (!(dev->features & NETIF_F_HIGHDMA)) {
1877 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1878 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1882 if (PCI_DMA_BUS_IS_PHYS) {
1883 struct device *pdev = dev->dev.parent;
1887 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1888 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1889 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1898 void (*destructor)(struct sk_buff *skb);
1901 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1903 static void dev_gso_skb_destructor(struct sk_buff *skb)
1905 struct dev_gso_cb *cb;
1908 struct sk_buff *nskb = skb->next;
1910 skb->next = nskb->next;
1913 } while (skb->next);
1915 cb = DEV_GSO_CB(skb);
1917 cb->destructor(skb);
1921 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1922 * @skb: buffer to segment
1924 * This function segments the given skb and stores the list of segments
1927 static int dev_gso_segment(struct sk_buff *skb)
1929 struct net_device *dev = skb->dev;
1930 struct sk_buff *segs;
1931 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1934 segs = skb_gso_segment(skb, features);
1936 /* Verifying header integrity only. */
1941 return PTR_ERR(segs);
1944 DEV_GSO_CB(skb)->destructor = skb->destructor;
1945 skb->destructor = dev_gso_skb_destructor;
1951 * Try to orphan skb early, right before transmission by the device.
1952 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1953 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1955 static inline void skb_orphan_try(struct sk_buff *skb)
1957 struct sock *sk = skb->sk;
1959 if (sk && !skb_shinfo(skb)->tx_flags) {
1960 /* skb_tx_hash() wont be able to get sk.
1961 * We copy sk_hash into skb->rxhash
1964 skb->rxhash = sk->sk_hash;
1970 * Returns true if either:
1971 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1972 * 2. skb is fragmented and the device does not support SG, or if
1973 * at least one of fragments is in highmem and device does not
1974 * support DMA from it.
1976 static inline int skb_needs_linearize(struct sk_buff *skb,
1977 struct net_device *dev)
1979 int features = dev->features;
1981 if (skb->protocol == htons(ETH_P_8021Q) || vlan_tx_tag_present(skb))
1982 features &= dev->vlan_features;
1984 return skb_is_nonlinear(skb) &&
1985 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
1986 (skb_shinfo(skb)->nr_frags && (!(features & NETIF_F_SG) ||
1987 illegal_highdma(dev, skb))));
1990 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1991 struct netdev_queue *txq)
1993 const struct net_device_ops *ops = dev->netdev_ops;
1994 int rc = NETDEV_TX_OK;
1996 if (likely(!skb->next)) {
1997 if (!list_empty(&ptype_all))
1998 dev_queue_xmit_nit(skb, dev);
2001 * If device doesnt need skb->dst, release it right now while
2002 * its hot in this cpu cache
2004 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2007 skb_orphan_try(skb);
2009 if (vlan_tx_tag_present(skb) &&
2010 !(dev->features & NETIF_F_HW_VLAN_TX)) {
2011 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2018 if (netif_needs_gso(dev, skb)) {
2019 if (unlikely(dev_gso_segment(skb)))
2024 if (skb_needs_linearize(skb, dev) &&
2025 __skb_linearize(skb))
2028 /* If packet is not checksummed and device does not
2029 * support checksumming for this protocol, complete
2030 * checksumming here.
2032 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2033 skb_set_transport_header(skb, skb->csum_start -
2035 if (!dev_can_checksum(dev, skb) &&
2036 skb_checksum_help(skb))
2041 rc = ops->ndo_start_xmit(skb, dev);
2042 trace_net_dev_xmit(skb, rc);
2043 if (rc == NETDEV_TX_OK)
2044 txq_trans_update(txq);
2050 struct sk_buff *nskb = skb->next;
2052 skb->next = nskb->next;
2056 * If device doesnt need nskb->dst, release it right now while
2057 * its hot in this cpu cache
2059 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2062 rc = ops->ndo_start_xmit(nskb, dev);
2063 trace_net_dev_xmit(nskb, rc);
2064 if (unlikely(rc != NETDEV_TX_OK)) {
2065 if (rc & ~NETDEV_TX_MASK)
2066 goto out_kfree_gso_skb;
2067 nskb->next = skb->next;
2071 txq_trans_update(txq);
2072 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2073 return NETDEV_TX_BUSY;
2074 } while (skb->next);
2077 if (likely(skb->next == NULL))
2078 skb->destructor = DEV_GSO_CB(skb)->destructor;
2085 static u32 hashrnd __read_mostly;
2087 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
2091 if (skb_rx_queue_recorded(skb)) {
2092 hash = skb_get_rx_queue(skb);
2093 while (unlikely(hash >= dev->real_num_tx_queues))
2094 hash -= dev->real_num_tx_queues;
2098 if (skb->sk && skb->sk->sk_hash)
2099 hash = skb->sk->sk_hash;
2101 hash = (__force u16) skb->protocol ^ skb->rxhash;
2102 hash = jhash_1word(hash, hashrnd);
2104 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2106 EXPORT_SYMBOL(skb_tx_hash);
2108 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2110 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2111 if (net_ratelimit()) {
2112 pr_warning("%s selects TX queue %d, but "
2113 "real number of TX queues is %d\n",
2114 dev->name, queue_index, dev->real_num_tx_queues);
2121 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2122 struct sk_buff *skb)
2125 const struct net_device_ops *ops = dev->netdev_ops;
2127 if (ops->ndo_select_queue) {
2128 queue_index = ops->ndo_select_queue(dev, skb);
2129 queue_index = dev_cap_txqueue(dev, queue_index);
2131 struct sock *sk = skb->sk;
2132 queue_index = sk_tx_queue_get(sk);
2133 if (queue_index < 0 || queue_index >= dev->real_num_tx_queues) {
2136 if (dev->real_num_tx_queues > 1)
2137 queue_index = skb_tx_hash(dev, skb);
2140 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2142 if (dst && skb_dst(skb) == dst)
2143 sk_tx_queue_set(sk, queue_index);
2148 skb_set_queue_mapping(skb, queue_index);
2149 return netdev_get_tx_queue(dev, queue_index);
2152 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2153 struct net_device *dev,
2154 struct netdev_queue *txq)
2156 spinlock_t *root_lock = qdisc_lock(q);
2157 bool contended = qdisc_is_running(q);
2161 * Heuristic to force contended enqueues to serialize on a
2162 * separate lock before trying to get qdisc main lock.
2163 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2164 * and dequeue packets faster.
2166 if (unlikely(contended))
2167 spin_lock(&q->busylock);
2169 spin_lock(root_lock);
2170 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2173 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2174 qdisc_run_begin(q)) {
2176 * This is a work-conserving queue; there are no old skbs
2177 * waiting to be sent out; and the qdisc is not running -
2178 * xmit the skb directly.
2180 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2182 __qdisc_update_bstats(q, skb->len);
2183 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2184 if (unlikely(contended)) {
2185 spin_unlock(&q->busylock);
2192 rc = NET_XMIT_SUCCESS;
2195 rc = qdisc_enqueue_root(skb, q);
2196 if (qdisc_run_begin(q)) {
2197 if (unlikely(contended)) {
2198 spin_unlock(&q->busylock);
2204 spin_unlock(root_lock);
2205 if (unlikely(contended))
2206 spin_unlock(&q->busylock);
2210 static DEFINE_PER_CPU(int, xmit_recursion);
2211 #define RECURSION_LIMIT 10
2214 * dev_queue_xmit - transmit a buffer
2215 * @skb: buffer to transmit
2217 * Queue a buffer for transmission to a network device. The caller must
2218 * have set the device and priority and built the buffer before calling
2219 * this function. The function can be called from an interrupt.
2221 * A negative errno code is returned on a failure. A success does not
2222 * guarantee the frame will be transmitted as it may be dropped due
2223 * to congestion or traffic shaping.
2225 * -----------------------------------------------------------------------------------
2226 * I notice this method can also return errors from the queue disciplines,
2227 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2230 * Regardless of the return value, the skb is consumed, so it is currently
2231 * difficult to retry a send to this method. (You can bump the ref count
2232 * before sending to hold a reference for retry if you are careful.)
2234 * When calling this method, interrupts MUST be enabled. This is because
2235 * the BH enable code must have IRQs enabled so that it will not deadlock.
2238 int dev_queue_xmit(struct sk_buff *skb)
2240 struct net_device *dev = skb->dev;
2241 struct netdev_queue *txq;
2245 /* Disable soft irqs for various locks below. Also
2246 * stops preemption for RCU.
2250 txq = dev_pick_tx(dev, skb);
2251 q = rcu_dereference_bh(txq->qdisc);
2253 #ifdef CONFIG_NET_CLS_ACT
2254 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2256 trace_net_dev_queue(skb);
2258 rc = __dev_xmit_skb(skb, q, dev, txq);
2262 /* The device has no queue. Common case for software devices:
2263 loopback, all the sorts of tunnels...
2265 Really, it is unlikely that netif_tx_lock protection is necessary
2266 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2268 However, it is possible, that they rely on protection
2271 Check this and shot the lock. It is not prone from deadlocks.
2272 Either shot noqueue qdisc, it is even simpler 8)
2274 if (dev->flags & IFF_UP) {
2275 int cpu = smp_processor_id(); /* ok because BHs are off */
2277 if (txq->xmit_lock_owner != cpu) {
2279 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2280 goto recursion_alert;
2282 HARD_TX_LOCK(dev, txq, cpu);
2284 if (!netif_tx_queue_stopped(txq)) {
2285 __this_cpu_inc(xmit_recursion);
2286 rc = dev_hard_start_xmit(skb, dev, txq);
2287 __this_cpu_dec(xmit_recursion);
2288 if (dev_xmit_complete(rc)) {
2289 HARD_TX_UNLOCK(dev, txq);
2293 HARD_TX_UNLOCK(dev, txq);
2294 if (net_ratelimit())
2295 printk(KERN_CRIT "Virtual device %s asks to "
2296 "queue packet!\n", dev->name);
2298 /* Recursion is detected! It is possible,
2302 if (net_ratelimit())
2303 printk(KERN_CRIT "Dead loop on virtual device "
2304 "%s, fix it urgently!\n", dev->name);
2309 rcu_read_unlock_bh();
2314 rcu_read_unlock_bh();
2317 EXPORT_SYMBOL(dev_queue_xmit);
2320 /*=======================================================================
2322 =======================================================================*/
2324 int netdev_max_backlog __read_mostly = 1000;
2325 int netdev_tstamp_prequeue __read_mostly = 1;
2326 int netdev_budget __read_mostly = 300;
2327 int weight_p __read_mostly = 64; /* old backlog weight */
2329 /* Called with irq disabled */
2330 static inline void ____napi_schedule(struct softnet_data *sd,
2331 struct napi_struct *napi)
2333 list_add_tail(&napi->poll_list, &sd->poll_list);
2334 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2338 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2339 * and src/dst port numbers. Returns a non-zero hash number on success
2342 __u32 __skb_get_rxhash(struct sk_buff *skb)
2344 int nhoff, hash = 0, poff;
2345 struct ipv6hdr *ip6;
2348 u32 addr1, addr2, ihl;
2354 nhoff = skb_network_offset(skb);
2356 switch (skb->protocol) {
2357 case __constant_htons(ETH_P_IP):
2358 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2361 ip = (struct iphdr *) (skb->data + nhoff);
2362 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2365 ip_proto = ip->protocol;
2366 addr1 = (__force u32) ip->saddr;
2367 addr2 = (__force u32) ip->daddr;
2370 case __constant_htons(ETH_P_IPV6):
2371 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2374 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2375 ip_proto = ip6->nexthdr;
2376 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2377 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2385 poff = proto_ports_offset(ip_proto);
2387 nhoff += ihl * 4 + poff;
2388 if (pskb_may_pull(skb, nhoff + 4)) {
2389 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2390 if (ports.v16[1] < ports.v16[0])
2391 swap(ports.v16[0], ports.v16[1]);
2395 /* get a consistent hash (same value on both flow directions) */
2399 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2406 EXPORT_SYMBOL(__skb_get_rxhash);
2410 /* One global table that all flow-based protocols share. */
2411 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2412 EXPORT_SYMBOL(rps_sock_flow_table);
2415 * get_rps_cpu is called from netif_receive_skb and returns the target
2416 * CPU from the RPS map of the receiving queue for a given skb.
2417 * rcu_read_lock must be held on entry.
2419 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2420 struct rps_dev_flow **rflowp)
2422 struct netdev_rx_queue *rxqueue;
2423 struct rps_map *map;
2424 struct rps_dev_flow_table *flow_table;
2425 struct rps_sock_flow_table *sock_flow_table;
2429 if (skb_rx_queue_recorded(skb)) {
2430 u16 index = skb_get_rx_queue(skb);
2431 if (unlikely(index >= dev->real_num_rx_queues)) {
2432 WARN_ONCE(dev->real_num_rx_queues > 1,
2433 "%s received packet on queue %u, but number "
2434 "of RX queues is %u\n",
2435 dev->name, index, dev->real_num_rx_queues);
2438 rxqueue = dev->_rx + index;
2442 map = rcu_dereference(rxqueue->rps_map);
2444 if (map->len == 1) {
2445 tcpu = map->cpus[0];
2446 if (cpu_online(tcpu))
2450 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2454 skb_reset_network_header(skb);
2455 if (!skb_get_rxhash(skb))
2458 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2459 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2460 if (flow_table && sock_flow_table) {
2462 struct rps_dev_flow *rflow;
2464 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2467 next_cpu = sock_flow_table->ents[skb->rxhash &
2468 sock_flow_table->mask];
2471 * If the desired CPU (where last recvmsg was done) is
2472 * different from current CPU (one in the rx-queue flow
2473 * table entry), switch if one of the following holds:
2474 * - Current CPU is unset (equal to RPS_NO_CPU).
2475 * - Current CPU is offline.
2476 * - The current CPU's queue tail has advanced beyond the
2477 * last packet that was enqueued using this table entry.
2478 * This guarantees that all previous packets for the flow
2479 * have been dequeued, thus preserving in order delivery.
2481 if (unlikely(tcpu != next_cpu) &&
2482 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2483 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2484 rflow->last_qtail)) >= 0)) {
2485 tcpu = rflow->cpu = next_cpu;
2486 if (tcpu != RPS_NO_CPU)
2487 rflow->last_qtail = per_cpu(softnet_data,
2488 tcpu).input_queue_head;
2490 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2498 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2500 if (cpu_online(tcpu)) {
2510 /* Called from hardirq (IPI) context */
2511 static void rps_trigger_softirq(void *data)
2513 struct softnet_data *sd = data;
2515 ____napi_schedule(sd, &sd->backlog);
2519 #endif /* CONFIG_RPS */
2522 * Check if this softnet_data structure is another cpu one
2523 * If yes, queue it to our IPI list and return 1
2526 static int rps_ipi_queued(struct softnet_data *sd)
2529 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2532 sd->rps_ipi_next = mysd->rps_ipi_list;
2533 mysd->rps_ipi_list = sd;
2535 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2538 #endif /* CONFIG_RPS */
2543 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2544 * queue (may be a remote CPU queue).
2546 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2547 unsigned int *qtail)
2549 struct softnet_data *sd;
2550 unsigned long flags;
2552 sd = &per_cpu(softnet_data, cpu);
2554 local_irq_save(flags);
2557 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2558 if (skb_queue_len(&sd->input_pkt_queue)) {
2560 __skb_queue_tail(&sd->input_pkt_queue, skb);
2561 input_queue_tail_incr_save(sd, qtail);
2563 local_irq_restore(flags);
2564 return NET_RX_SUCCESS;
2567 /* Schedule NAPI for backlog device
2568 * We can use non atomic operation since we own the queue lock
2570 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2571 if (!rps_ipi_queued(sd))
2572 ____napi_schedule(sd, &sd->backlog);
2580 local_irq_restore(flags);
2582 atomic_long_inc(&skb->dev->rx_dropped);
2588 * netif_rx - post buffer to the network code
2589 * @skb: buffer to post
2591 * This function receives a packet from a device driver and queues it for
2592 * the upper (protocol) levels to process. It always succeeds. The buffer
2593 * may be dropped during processing for congestion control or by the
2597 * NET_RX_SUCCESS (no congestion)
2598 * NET_RX_DROP (packet was dropped)
2602 int netif_rx(struct sk_buff *skb)
2606 /* if netpoll wants it, pretend we never saw it */
2607 if (netpoll_rx(skb))
2610 if (netdev_tstamp_prequeue)
2611 net_timestamp_check(skb);
2613 trace_netif_rx(skb);
2616 struct rps_dev_flow voidflow, *rflow = &voidflow;
2622 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2624 cpu = smp_processor_id();
2626 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2634 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2640 EXPORT_SYMBOL(netif_rx);
2642 int netif_rx_ni(struct sk_buff *skb)
2647 err = netif_rx(skb);
2648 if (local_softirq_pending())
2654 EXPORT_SYMBOL(netif_rx_ni);
2656 static void net_tx_action(struct softirq_action *h)
2658 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2660 if (sd->completion_queue) {
2661 struct sk_buff *clist;
2663 local_irq_disable();
2664 clist = sd->completion_queue;
2665 sd->completion_queue = NULL;
2669 struct sk_buff *skb = clist;
2670 clist = clist->next;
2672 WARN_ON(atomic_read(&skb->users));
2673 trace_kfree_skb(skb, net_tx_action);
2678 if (sd->output_queue) {
2681 local_irq_disable();
2682 head = sd->output_queue;
2683 sd->output_queue = NULL;
2684 sd->output_queue_tailp = &sd->output_queue;
2688 struct Qdisc *q = head;
2689 spinlock_t *root_lock;
2691 head = head->next_sched;
2693 root_lock = qdisc_lock(q);
2694 if (spin_trylock(root_lock)) {
2695 smp_mb__before_clear_bit();
2696 clear_bit(__QDISC_STATE_SCHED,
2699 spin_unlock(root_lock);
2701 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2703 __netif_reschedule(q);
2705 smp_mb__before_clear_bit();
2706 clear_bit(__QDISC_STATE_SCHED,
2714 static inline int deliver_skb(struct sk_buff *skb,
2715 struct packet_type *pt_prev,
2716 struct net_device *orig_dev)
2718 atomic_inc(&skb->users);
2719 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2722 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2723 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2724 /* This hook is defined here for ATM LANE */
2725 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2726 unsigned char *addr) __read_mostly;
2727 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2730 #ifdef CONFIG_NET_CLS_ACT
2731 /* TODO: Maybe we should just force sch_ingress to be compiled in
2732 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2733 * a compare and 2 stores extra right now if we dont have it on
2734 * but have CONFIG_NET_CLS_ACT
2735 * NOTE: This doesnt stop any functionality; if you dont have
2736 * the ingress scheduler, you just cant add policies on ingress.
2739 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2741 struct net_device *dev = skb->dev;
2742 u32 ttl = G_TC_RTTL(skb->tc_verd);
2743 int result = TC_ACT_OK;
2746 if (unlikely(MAX_RED_LOOP < ttl++)) {
2747 if (net_ratelimit())
2748 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2749 skb->skb_iif, dev->ifindex);
2753 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2754 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2757 if (q != &noop_qdisc) {
2758 spin_lock(qdisc_lock(q));
2759 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2760 result = qdisc_enqueue_root(skb, q);
2761 spin_unlock(qdisc_lock(q));
2767 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2768 struct packet_type **pt_prev,
2769 int *ret, struct net_device *orig_dev)
2771 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
2773 if (!rxq || rxq->qdisc == &noop_qdisc)
2777 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2781 switch (ing_filter(skb, rxq)) {
2795 * netdev_rx_handler_register - register receive handler
2796 * @dev: device to register a handler for
2797 * @rx_handler: receive handler to register
2798 * @rx_handler_data: data pointer that is used by rx handler
2800 * Register a receive hander for a device. This handler will then be
2801 * called from __netif_receive_skb. A negative errno code is returned
2804 * The caller must hold the rtnl_mutex.
2806 int netdev_rx_handler_register(struct net_device *dev,
2807 rx_handler_func_t *rx_handler,
2808 void *rx_handler_data)
2812 if (dev->rx_handler)
2815 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2816 rcu_assign_pointer(dev->rx_handler, rx_handler);
2820 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2823 * netdev_rx_handler_unregister - unregister receive handler
2824 * @dev: device to unregister a handler from
2826 * Unregister a receive hander from a device.
2828 * The caller must hold the rtnl_mutex.
2830 void netdev_rx_handler_unregister(struct net_device *dev)
2834 rcu_assign_pointer(dev->rx_handler, NULL);
2835 rcu_assign_pointer(dev->rx_handler_data, NULL);
2837 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2839 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2840 struct net_device *master)
2842 if (skb->pkt_type == PACKET_HOST) {
2843 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2845 memcpy(dest, master->dev_addr, ETH_ALEN);
2849 /* On bonding slaves other than the currently active slave, suppress
2850 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2851 * ARP on active-backup slaves with arp_validate enabled.
2853 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2855 struct net_device *dev = skb->dev;
2857 if (master->priv_flags & IFF_MASTER_ARPMON)
2858 dev->last_rx = jiffies;
2860 if ((master->priv_flags & IFF_MASTER_ALB) &&
2861 (master->priv_flags & IFF_BRIDGE_PORT)) {
2862 /* Do address unmangle. The local destination address
2863 * will be always the one master has. Provides the right
2864 * functionality in a bridge.
2866 skb_bond_set_mac_by_master(skb, master);
2869 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2870 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2871 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2874 if (master->priv_flags & IFF_MASTER_ALB) {
2875 if (skb->pkt_type != PACKET_BROADCAST &&
2876 skb->pkt_type != PACKET_MULTICAST)
2879 if (master->priv_flags & IFF_MASTER_8023AD &&
2880 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2887 EXPORT_SYMBOL(__skb_bond_should_drop);
2889 static int __netif_receive_skb(struct sk_buff *skb)
2891 struct packet_type *ptype, *pt_prev;
2892 rx_handler_func_t *rx_handler;
2893 struct net_device *orig_dev;
2894 struct net_device *master;
2895 struct net_device *null_or_orig;
2896 struct net_device *orig_or_bond;
2897 int ret = NET_RX_DROP;
2900 if (!netdev_tstamp_prequeue)
2901 net_timestamp_check(skb);
2903 trace_netif_receive_skb(skb);
2905 /* if we've gotten here through NAPI, check netpoll */
2906 if (netpoll_receive_skb(skb))
2910 skb->skb_iif = skb->dev->ifindex;
2913 * bonding note: skbs received on inactive slaves should only
2914 * be delivered to pkt handlers that are exact matches. Also
2915 * the deliver_no_wcard flag will be set. If packet handlers
2916 * are sensitive to duplicate packets these skbs will need to
2917 * be dropped at the handler.
2919 null_or_orig = NULL;
2920 orig_dev = skb->dev;
2921 master = ACCESS_ONCE(orig_dev->master);
2922 if (skb->deliver_no_wcard)
2923 null_or_orig = orig_dev;
2925 if (skb_bond_should_drop(skb, master)) {
2926 skb->deliver_no_wcard = 1;
2927 null_or_orig = orig_dev; /* deliver only exact match */
2932 __this_cpu_inc(softnet_data.processed);
2933 skb_reset_network_header(skb);
2934 skb_reset_transport_header(skb);
2935 skb->mac_len = skb->network_header - skb->mac_header;
2941 #ifdef CONFIG_NET_CLS_ACT
2942 if (skb->tc_verd & TC_NCLS) {
2943 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2948 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2949 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2950 ptype->dev == orig_dev) {
2952 ret = deliver_skb(skb, pt_prev, orig_dev);
2957 #ifdef CONFIG_NET_CLS_ACT
2958 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2964 /* Handle special case of bridge or macvlan */
2965 rx_handler = rcu_dereference(skb->dev->rx_handler);
2968 ret = deliver_skb(skb, pt_prev, orig_dev);
2971 skb = rx_handler(skb);
2976 if (vlan_tx_tag_present(skb)) {
2978 ret = deliver_skb(skb, pt_prev, orig_dev);
2981 if (vlan_hwaccel_do_receive(&skb)) {
2982 ret = __netif_receive_skb(skb);
2984 } else if (unlikely(!skb))
2989 * Make sure frames received on VLAN interfaces stacked on
2990 * bonding interfaces still make their way to any base bonding
2991 * device that may have registered for a specific ptype. The
2992 * handler may have to adjust skb->dev and orig_dev.
2994 orig_or_bond = orig_dev;
2995 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2996 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2997 orig_or_bond = vlan_dev_real_dev(skb->dev);
3000 type = skb->protocol;
3001 list_for_each_entry_rcu(ptype,
3002 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3003 if (ptype->type == type && (ptype->dev == null_or_orig ||
3004 ptype->dev == skb->dev || ptype->dev == orig_dev ||
3005 ptype->dev == orig_or_bond)) {
3007 ret = deliver_skb(skb, pt_prev, orig_dev);
3013 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3015 atomic_long_inc(&skb->dev->rx_dropped);
3017 /* Jamal, now you will not able to escape explaining
3018 * me how you were going to use this. :-)
3029 * netif_receive_skb - process receive buffer from network
3030 * @skb: buffer to process
3032 * netif_receive_skb() is the main receive data processing function.
3033 * It always succeeds. The buffer may be dropped during processing
3034 * for congestion control or by the protocol layers.
3036 * This function may only be called from softirq context and interrupts
3037 * should be enabled.
3039 * Return values (usually ignored):
3040 * NET_RX_SUCCESS: no congestion
3041 * NET_RX_DROP: packet was dropped
3043 int netif_receive_skb(struct sk_buff *skb)
3045 if (netdev_tstamp_prequeue)
3046 net_timestamp_check(skb);
3048 if (skb_defer_rx_timestamp(skb))
3049 return NET_RX_SUCCESS;
3053 struct rps_dev_flow voidflow, *rflow = &voidflow;
3058 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3061 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3065 ret = __netif_receive_skb(skb);
3071 return __netif_receive_skb(skb);
3074 EXPORT_SYMBOL(netif_receive_skb);
3076 /* Network device is going away, flush any packets still pending
3077 * Called with irqs disabled.
3079 static void flush_backlog(void *arg)
3081 struct net_device *dev = arg;
3082 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3083 struct sk_buff *skb, *tmp;
3086 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3087 if (skb->dev == dev) {
3088 __skb_unlink(skb, &sd->input_pkt_queue);
3090 input_queue_head_incr(sd);
3095 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3096 if (skb->dev == dev) {
3097 __skb_unlink(skb, &sd->process_queue);
3099 input_queue_head_incr(sd);
3104 static int napi_gro_complete(struct sk_buff *skb)
3106 struct packet_type *ptype;
3107 __be16 type = skb->protocol;
3108 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3111 if (NAPI_GRO_CB(skb)->count == 1) {
3112 skb_shinfo(skb)->gso_size = 0;
3117 list_for_each_entry_rcu(ptype, head, list) {
3118 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3121 err = ptype->gro_complete(skb);
3127 WARN_ON(&ptype->list == head);
3129 return NET_RX_SUCCESS;
3133 return netif_receive_skb(skb);
3136 inline void napi_gro_flush(struct napi_struct *napi)
3138 struct sk_buff *skb, *next;
3140 for (skb = napi->gro_list; skb; skb = next) {
3143 napi_gro_complete(skb);
3146 napi->gro_count = 0;
3147 napi->gro_list = NULL;
3149 EXPORT_SYMBOL(napi_gro_flush);
3151 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3153 struct sk_buff **pp = NULL;
3154 struct packet_type *ptype;
3155 __be16 type = skb->protocol;
3156 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3159 enum gro_result ret;
3161 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3164 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3168 list_for_each_entry_rcu(ptype, head, list) {
3169 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3172 skb_set_network_header(skb, skb_gro_offset(skb));
3173 mac_len = skb->network_header - skb->mac_header;
3174 skb->mac_len = mac_len;
3175 NAPI_GRO_CB(skb)->same_flow = 0;
3176 NAPI_GRO_CB(skb)->flush = 0;
3177 NAPI_GRO_CB(skb)->free = 0;
3179 pp = ptype->gro_receive(&napi->gro_list, skb);
3184 if (&ptype->list == head)
3187 same_flow = NAPI_GRO_CB(skb)->same_flow;
3188 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3191 struct sk_buff *nskb = *pp;
3195 napi_gro_complete(nskb);
3202 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3206 NAPI_GRO_CB(skb)->count = 1;
3207 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3208 skb->next = napi->gro_list;
3209 napi->gro_list = skb;
3213 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3214 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3216 BUG_ON(skb->end - skb->tail < grow);
3218 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3221 skb->data_len -= grow;
3223 skb_shinfo(skb)->frags[0].page_offset += grow;
3224 skb_shinfo(skb)->frags[0].size -= grow;
3226 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3227 put_page(skb_shinfo(skb)->frags[0].page);
3228 memmove(skb_shinfo(skb)->frags,
3229 skb_shinfo(skb)->frags + 1,
3230 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3241 EXPORT_SYMBOL(dev_gro_receive);
3243 static inline gro_result_t
3244 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3248 for (p = napi->gro_list; p; p = p->next) {
3249 unsigned long diffs;
3251 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3252 diffs |= p->vlan_tci ^ skb->vlan_tci;
3253 diffs |= compare_ether_header(skb_mac_header(p),
3254 skb_gro_mac_header(skb));
3255 NAPI_GRO_CB(p)->same_flow = !diffs;
3256 NAPI_GRO_CB(p)->flush = 0;
3259 return dev_gro_receive(napi, skb);
3262 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3266 if (netif_receive_skb(skb))
3271 case GRO_MERGED_FREE:
3282 EXPORT_SYMBOL(napi_skb_finish);
3284 void skb_gro_reset_offset(struct sk_buff *skb)
3286 NAPI_GRO_CB(skb)->data_offset = 0;
3287 NAPI_GRO_CB(skb)->frag0 = NULL;
3288 NAPI_GRO_CB(skb)->frag0_len = 0;
3290 if (skb->mac_header == skb->tail &&
3291 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3292 NAPI_GRO_CB(skb)->frag0 =
3293 page_address(skb_shinfo(skb)->frags[0].page) +
3294 skb_shinfo(skb)->frags[0].page_offset;
3295 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3298 EXPORT_SYMBOL(skb_gro_reset_offset);
3300 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3302 skb_gro_reset_offset(skb);
3304 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3306 EXPORT_SYMBOL(napi_gro_receive);
3308 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3310 __skb_pull(skb, skb_headlen(skb));
3311 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3317 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3319 struct sk_buff *skb = napi->skb;
3322 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3328 EXPORT_SYMBOL(napi_get_frags);
3330 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3336 skb->protocol = eth_type_trans(skb, skb->dev);
3338 if (ret == GRO_HELD)
3339 skb_gro_pull(skb, -ETH_HLEN);
3340 else if (netif_receive_skb(skb))
3345 case GRO_MERGED_FREE:
3346 napi_reuse_skb(napi, skb);
3355 EXPORT_SYMBOL(napi_frags_finish);
3357 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3359 struct sk_buff *skb = napi->skb;
3366 skb_reset_mac_header(skb);
3367 skb_gro_reset_offset(skb);
3369 off = skb_gro_offset(skb);
3370 hlen = off + sizeof(*eth);
3371 eth = skb_gro_header_fast(skb, off);
3372 if (skb_gro_header_hard(skb, hlen)) {
3373 eth = skb_gro_header_slow(skb, hlen, off);
3374 if (unlikely(!eth)) {
3375 napi_reuse_skb(napi, skb);
3381 skb_gro_pull(skb, sizeof(*eth));
3384 * This works because the only protocols we care about don't require
3385 * special handling. We'll fix it up properly at the end.
3387 skb->protocol = eth->h_proto;
3392 EXPORT_SYMBOL(napi_frags_skb);
3394 gro_result_t napi_gro_frags(struct napi_struct *napi)
3396 struct sk_buff *skb = napi_frags_skb(napi);
3401 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3403 EXPORT_SYMBOL(napi_gro_frags);
3406 * net_rps_action sends any pending IPI's for rps.
3407 * Note: called with local irq disabled, but exits with local irq enabled.
3409 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3412 struct softnet_data *remsd = sd->rps_ipi_list;
3415 sd->rps_ipi_list = NULL;
3419 /* Send pending IPI's to kick RPS processing on remote cpus. */
3421 struct softnet_data *next = remsd->rps_ipi_next;
3423 if (cpu_online(remsd->cpu))
3424 __smp_call_function_single(remsd->cpu,
3433 static int process_backlog(struct napi_struct *napi, int quota)
3436 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3439 /* Check if we have pending ipi, its better to send them now,
3440 * not waiting net_rx_action() end.
3442 if (sd->rps_ipi_list) {
3443 local_irq_disable();
3444 net_rps_action_and_irq_enable(sd);
3447 napi->weight = weight_p;
3448 local_irq_disable();
3449 while (work < quota) {
3450 struct sk_buff *skb;
3453 while ((skb = __skb_dequeue(&sd->process_queue))) {
3455 __netif_receive_skb(skb);
3456 local_irq_disable();
3457 input_queue_head_incr(sd);
3458 if (++work >= quota) {
3465 qlen = skb_queue_len(&sd->input_pkt_queue);
3467 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3468 &sd->process_queue);
3470 if (qlen < quota - work) {
3472 * Inline a custom version of __napi_complete().
3473 * only current cpu owns and manipulates this napi,
3474 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3475 * we can use a plain write instead of clear_bit(),
3476 * and we dont need an smp_mb() memory barrier.
3478 list_del(&napi->poll_list);
3481 quota = work + qlen;
3491 * __napi_schedule - schedule for receive
3492 * @n: entry to schedule
3494 * The entry's receive function will be scheduled to run
3496 void __napi_schedule(struct napi_struct *n)
3498 unsigned long flags;
3500 local_irq_save(flags);
3501 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3502 local_irq_restore(flags);
3504 EXPORT_SYMBOL(__napi_schedule);
3506 void __napi_complete(struct napi_struct *n)
3508 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3509 BUG_ON(n->gro_list);
3511 list_del(&n->poll_list);
3512 smp_mb__before_clear_bit();
3513 clear_bit(NAPI_STATE_SCHED, &n->state);
3515 EXPORT_SYMBOL(__napi_complete);
3517 void napi_complete(struct napi_struct *n)
3519 unsigned long flags;
3522 * don't let napi dequeue from the cpu poll list
3523 * just in case its running on a different cpu
3525 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3529 local_irq_save(flags);
3531 local_irq_restore(flags);
3533 EXPORT_SYMBOL(napi_complete);
3535 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3536 int (*poll)(struct napi_struct *, int), int weight)
3538 INIT_LIST_HEAD(&napi->poll_list);
3539 napi->gro_count = 0;
3540 napi->gro_list = NULL;
3543 napi->weight = weight;
3544 list_add(&napi->dev_list, &dev->napi_list);
3546 #ifdef CONFIG_NETPOLL
3547 spin_lock_init(&napi->poll_lock);
3548 napi->poll_owner = -1;
3550 set_bit(NAPI_STATE_SCHED, &napi->state);
3552 EXPORT_SYMBOL(netif_napi_add);
3554 void netif_napi_del(struct napi_struct *napi)
3556 struct sk_buff *skb, *next;
3558 list_del_init(&napi->dev_list);
3559 napi_free_frags(napi);
3561 for (skb = napi->gro_list; skb; skb = next) {
3567 napi->gro_list = NULL;
3568 napi->gro_count = 0;
3570 EXPORT_SYMBOL(netif_napi_del);
3572 static void net_rx_action(struct softirq_action *h)
3574 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3575 unsigned long time_limit = jiffies + 2;
3576 int budget = netdev_budget;
3579 local_irq_disable();
3581 while (!list_empty(&sd->poll_list)) {
3582 struct napi_struct *n;
3585 /* If softirq window is exhuasted then punt.
3586 * Allow this to run for 2 jiffies since which will allow
3587 * an average latency of 1.5/HZ.
3589 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3594 /* Even though interrupts have been re-enabled, this
3595 * access is safe because interrupts can only add new
3596 * entries to the tail of this list, and only ->poll()
3597 * calls can remove this head entry from the list.
3599 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3601 have = netpoll_poll_lock(n);
3605 /* This NAPI_STATE_SCHED test is for avoiding a race
3606 * with netpoll's poll_napi(). Only the entity which
3607 * obtains the lock and sees NAPI_STATE_SCHED set will
3608 * actually make the ->poll() call. Therefore we avoid
3609 * accidently calling ->poll() when NAPI is not scheduled.
3612 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3613 work = n->poll(n, weight);
3617 WARN_ON_ONCE(work > weight);
3621 local_irq_disable();
3623 /* Drivers must not modify the NAPI state if they
3624 * consume the entire weight. In such cases this code
3625 * still "owns" the NAPI instance and therefore can
3626 * move the instance around on the list at-will.
3628 if (unlikely(work == weight)) {
3629 if (unlikely(napi_disable_pending(n))) {
3632 local_irq_disable();
3634 list_move_tail(&n->poll_list, &sd->poll_list);
3637 netpoll_poll_unlock(have);
3640 net_rps_action_and_irq_enable(sd);
3642 #ifdef CONFIG_NET_DMA
3644 * There may not be any more sk_buffs coming right now, so push
3645 * any pending DMA copies to hardware
3647 dma_issue_pending_all();
3654 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3658 static gifconf_func_t *gifconf_list[NPROTO];
3661 * register_gifconf - register a SIOCGIF handler
3662 * @family: Address family
3663 * @gifconf: Function handler
3665 * Register protocol dependent address dumping routines. The handler
3666 * that is passed must not be freed or reused until it has been replaced
3667 * by another handler.
3669 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3671 if (family >= NPROTO)
3673 gifconf_list[family] = gifconf;
3676 EXPORT_SYMBOL(register_gifconf);
3680 * Map an interface index to its name (SIOCGIFNAME)
3684 * We need this ioctl for efficient implementation of the
3685 * if_indextoname() function required by the IPv6 API. Without
3686 * it, we would have to search all the interfaces to find a
3690 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3692 struct net_device *dev;
3696 * Fetch the caller's info block.
3699 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3703 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3709 strcpy(ifr.ifr_name, dev->name);
3712 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3718 * Perform a SIOCGIFCONF call. This structure will change
3719 * size eventually, and there is nothing I can do about it.
3720 * Thus we will need a 'compatibility mode'.
3723 static int dev_ifconf(struct net *net, char __user *arg)
3726 struct net_device *dev;
3733 * Fetch the caller's info block.
3736 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3743 * Loop over the interfaces, and write an info block for each.
3747 for_each_netdev(net, dev) {
3748 for (i = 0; i < NPROTO; i++) {
3749 if (gifconf_list[i]) {
3752 done = gifconf_list[i](dev, NULL, 0);
3754 done = gifconf_list[i](dev, pos + total,
3764 * All done. Write the updated control block back to the caller.
3766 ifc.ifc_len = total;
3769 * Both BSD and Solaris return 0 here, so we do too.
3771 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3774 #ifdef CONFIG_PROC_FS
3776 * This is invoked by the /proc filesystem handler to display a device
3779 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3782 struct net *net = seq_file_net(seq);
3784 struct net_device *dev;
3788 return SEQ_START_TOKEN;
3791 for_each_netdev_rcu(net, dev)
3798 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3800 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3801 first_net_device(seq_file_net(seq)) :
3802 next_net_device((struct net_device *)v);
3805 return rcu_dereference(dev);
3808 void dev_seq_stop(struct seq_file *seq, void *v)
3814 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3816 struct rtnl_link_stats64 temp;
3817 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3819 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3820 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3821 dev->name, stats->rx_bytes, stats->rx_packets,
3823 stats->rx_dropped + stats->rx_missed_errors,
3824 stats->rx_fifo_errors,
3825 stats->rx_length_errors + stats->rx_over_errors +
3826 stats->rx_crc_errors + stats->rx_frame_errors,
3827 stats->rx_compressed, stats->multicast,
3828 stats->tx_bytes, stats->tx_packets,
3829 stats->tx_errors, stats->tx_dropped,
3830 stats->tx_fifo_errors, stats->collisions,
3831 stats->tx_carrier_errors +
3832 stats->tx_aborted_errors +
3833 stats->tx_window_errors +
3834 stats->tx_heartbeat_errors,
3835 stats->tx_compressed);
3839 * Called from the PROCfs module. This now uses the new arbitrary sized
3840 * /proc/net interface to create /proc/net/dev
3842 static int dev_seq_show(struct seq_file *seq, void *v)
3844 if (v == SEQ_START_TOKEN)
3845 seq_puts(seq, "Inter-| Receive "
3847 " face |bytes packets errs drop fifo frame "
3848 "compressed multicast|bytes packets errs "
3849 "drop fifo colls carrier compressed\n");
3851 dev_seq_printf_stats(seq, v);
3855 static struct softnet_data *softnet_get_online(loff_t *pos)
3857 struct softnet_data *sd = NULL;
3859 while (*pos < nr_cpu_ids)
3860 if (cpu_online(*pos)) {
3861 sd = &per_cpu(softnet_data, *pos);
3868 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3870 return softnet_get_online(pos);
3873 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3876 return softnet_get_online(pos);
3879 static void softnet_seq_stop(struct seq_file *seq, void *v)
3883 static int softnet_seq_show(struct seq_file *seq, void *v)
3885 struct softnet_data *sd = v;
3887 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3888 sd->processed, sd->dropped, sd->time_squeeze, 0,
3889 0, 0, 0, 0, /* was fastroute */
3890 sd->cpu_collision, sd->received_rps);
3894 static const struct seq_operations dev_seq_ops = {
3895 .start = dev_seq_start,
3896 .next = dev_seq_next,
3897 .stop = dev_seq_stop,
3898 .show = dev_seq_show,
3901 static int dev_seq_open(struct inode *inode, struct file *file)
3903 return seq_open_net(inode, file, &dev_seq_ops,
3904 sizeof(struct seq_net_private));
3907 static const struct file_operations dev_seq_fops = {
3908 .owner = THIS_MODULE,
3909 .open = dev_seq_open,
3911 .llseek = seq_lseek,
3912 .release = seq_release_net,
3915 static const struct seq_operations softnet_seq_ops = {
3916 .start = softnet_seq_start,
3917 .next = softnet_seq_next,
3918 .stop = softnet_seq_stop,
3919 .show = softnet_seq_show,
3922 static int softnet_seq_open(struct inode *inode, struct file *file)
3924 return seq_open(file, &softnet_seq_ops);
3927 static const struct file_operations softnet_seq_fops = {
3928 .owner = THIS_MODULE,
3929 .open = softnet_seq_open,
3931 .llseek = seq_lseek,
3932 .release = seq_release,
3935 static void *ptype_get_idx(loff_t pos)
3937 struct packet_type *pt = NULL;
3941 list_for_each_entry_rcu(pt, &ptype_all, list) {
3947 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3948 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3957 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3961 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3964 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3966 struct packet_type *pt;
3967 struct list_head *nxt;
3971 if (v == SEQ_START_TOKEN)
3972 return ptype_get_idx(0);
3975 nxt = pt->list.next;
3976 if (pt->type == htons(ETH_P_ALL)) {
3977 if (nxt != &ptype_all)
3980 nxt = ptype_base[0].next;
3982 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3984 while (nxt == &ptype_base[hash]) {
3985 if (++hash >= PTYPE_HASH_SIZE)
3987 nxt = ptype_base[hash].next;
3990 return list_entry(nxt, struct packet_type, list);
3993 static void ptype_seq_stop(struct seq_file *seq, void *v)
3999 static int ptype_seq_show(struct seq_file *seq, void *v)
4001 struct packet_type *pt = v;
4003 if (v == SEQ_START_TOKEN)
4004 seq_puts(seq, "Type Device Function\n");
4005 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4006 if (pt->type == htons(ETH_P_ALL))
4007 seq_puts(seq, "ALL ");
4009 seq_printf(seq, "%04x", ntohs(pt->type));
4011 seq_printf(seq, " %-8s %pF\n",
4012 pt->dev ? pt->dev->name : "", pt->func);
4018 static const struct seq_operations ptype_seq_ops = {
4019 .start = ptype_seq_start,
4020 .next = ptype_seq_next,
4021 .stop = ptype_seq_stop,
4022 .show = ptype_seq_show,
4025 static int ptype_seq_open(struct inode *inode, struct file *file)
4027 return seq_open_net(inode, file, &ptype_seq_ops,
4028 sizeof(struct seq_net_private));
4031 static const struct file_operations ptype_seq_fops = {
4032 .owner = THIS_MODULE,
4033 .open = ptype_seq_open,
4035 .llseek = seq_lseek,
4036 .release = seq_release_net,
4040 static int __net_init dev_proc_net_init(struct net *net)
4044 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4046 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4048 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4051 if (wext_proc_init(net))
4057 proc_net_remove(net, "ptype");
4059 proc_net_remove(net, "softnet_stat");
4061 proc_net_remove(net, "dev");
4065 static void __net_exit dev_proc_net_exit(struct net *net)
4067 wext_proc_exit(net);
4069 proc_net_remove(net, "ptype");
4070 proc_net_remove(net, "softnet_stat");
4071 proc_net_remove(net, "dev");
4074 static struct pernet_operations __net_initdata dev_proc_ops = {
4075 .init = dev_proc_net_init,
4076 .exit = dev_proc_net_exit,
4079 static int __init dev_proc_init(void)
4081 return register_pernet_subsys(&dev_proc_ops);
4084 #define dev_proc_init() 0
4085 #endif /* CONFIG_PROC_FS */
4089 * netdev_set_master - set up master/slave pair
4090 * @slave: slave device
4091 * @master: new master device
4093 * Changes the master device of the slave. Pass %NULL to break the
4094 * bonding. The caller must hold the RTNL semaphore. On a failure
4095 * a negative errno code is returned. On success the reference counts
4096 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4097 * function returns zero.
4099 int netdev_set_master(struct net_device *slave, struct net_device *master)
4101 struct net_device *old = slave->master;
4111 slave->master = master;
4118 slave->flags |= IFF_SLAVE;
4120 slave->flags &= ~IFF_SLAVE;
4122 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4125 EXPORT_SYMBOL(netdev_set_master);
4127 static void dev_change_rx_flags(struct net_device *dev, int flags)
4129 const struct net_device_ops *ops = dev->netdev_ops;
4131 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4132 ops->ndo_change_rx_flags(dev, flags);
4135 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4137 unsigned short old_flags = dev->flags;
4143 dev->flags |= IFF_PROMISC;
4144 dev->promiscuity += inc;
4145 if (dev->promiscuity == 0) {
4148 * If inc causes overflow, untouch promisc and return error.
4151 dev->flags &= ~IFF_PROMISC;
4153 dev->promiscuity -= inc;
4154 printk(KERN_WARNING "%s: promiscuity touches roof, "
4155 "set promiscuity failed, promiscuity feature "
4156 "of device might be broken.\n", dev->name);
4160 if (dev->flags != old_flags) {
4161 printk(KERN_INFO "device %s %s promiscuous mode\n",
4162 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4164 if (audit_enabled) {
4165 current_uid_gid(&uid, &gid);
4166 audit_log(current->audit_context, GFP_ATOMIC,
4167 AUDIT_ANOM_PROMISCUOUS,
4168 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4169 dev->name, (dev->flags & IFF_PROMISC),
4170 (old_flags & IFF_PROMISC),
4171 audit_get_loginuid(current),
4173 audit_get_sessionid(current));
4176 dev_change_rx_flags(dev, IFF_PROMISC);
4182 * dev_set_promiscuity - update promiscuity count on a device
4186 * Add or remove promiscuity from a device. While the count in the device
4187 * remains above zero the interface remains promiscuous. Once it hits zero
4188 * the device reverts back to normal filtering operation. A negative inc
4189 * value is used to drop promiscuity on the device.
4190 * Return 0 if successful or a negative errno code on error.
4192 int dev_set_promiscuity(struct net_device *dev, int inc)
4194 unsigned short old_flags = dev->flags;
4197 err = __dev_set_promiscuity(dev, inc);
4200 if (dev->flags != old_flags)
4201 dev_set_rx_mode(dev);
4204 EXPORT_SYMBOL(dev_set_promiscuity);
4207 * dev_set_allmulti - update allmulti count on a device
4211 * Add or remove reception of all multicast frames to a device. While the
4212 * count in the device remains above zero the interface remains listening
4213 * to all interfaces. Once it hits zero the device reverts back to normal
4214 * filtering operation. A negative @inc value is used to drop the counter
4215 * when releasing a resource needing all multicasts.
4216 * Return 0 if successful or a negative errno code on error.
4219 int dev_set_allmulti(struct net_device *dev, int inc)
4221 unsigned short old_flags = dev->flags;
4225 dev->flags |= IFF_ALLMULTI;
4226 dev->allmulti += inc;
4227 if (dev->allmulti == 0) {
4230 * If inc causes overflow, untouch allmulti and return error.
4233 dev->flags &= ~IFF_ALLMULTI;
4235 dev->allmulti -= inc;
4236 printk(KERN_WARNING "%s: allmulti touches roof, "
4237 "set allmulti failed, allmulti feature of "
4238 "device might be broken.\n", dev->name);
4242 if (dev->flags ^ old_flags) {
4243 dev_change_rx_flags(dev, IFF_ALLMULTI);
4244 dev_set_rx_mode(dev);
4248 EXPORT_SYMBOL(dev_set_allmulti);
4251 * Upload unicast and multicast address lists to device and
4252 * configure RX filtering. When the device doesn't support unicast
4253 * filtering it is put in promiscuous mode while unicast addresses
4256 void __dev_set_rx_mode(struct net_device *dev)
4258 const struct net_device_ops *ops = dev->netdev_ops;
4260 /* dev_open will call this function so the list will stay sane. */
4261 if (!(dev->flags&IFF_UP))
4264 if (!netif_device_present(dev))
4267 if (ops->ndo_set_rx_mode)
4268 ops->ndo_set_rx_mode(dev);
4270 /* Unicast addresses changes may only happen under the rtnl,
4271 * therefore calling __dev_set_promiscuity here is safe.
4273 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4274 __dev_set_promiscuity(dev, 1);
4275 dev->uc_promisc = 1;
4276 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4277 __dev_set_promiscuity(dev, -1);
4278 dev->uc_promisc = 0;
4281 if (ops->ndo_set_multicast_list)
4282 ops->ndo_set_multicast_list(dev);
4286 void dev_set_rx_mode(struct net_device *dev)
4288 netif_addr_lock_bh(dev);
4289 __dev_set_rx_mode(dev);
4290 netif_addr_unlock_bh(dev);
4294 * dev_get_flags - get flags reported to userspace
4297 * Get the combination of flag bits exported through APIs to userspace.
4299 unsigned dev_get_flags(const struct net_device *dev)
4303 flags = (dev->flags & ~(IFF_PROMISC |
4308 (dev->gflags & (IFF_PROMISC |
4311 if (netif_running(dev)) {
4312 if (netif_oper_up(dev))
4313 flags |= IFF_RUNNING;
4314 if (netif_carrier_ok(dev))
4315 flags |= IFF_LOWER_UP;
4316 if (netif_dormant(dev))
4317 flags |= IFF_DORMANT;
4322 EXPORT_SYMBOL(dev_get_flags);
4324 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4326 int old_flags = dev->flags;
4332 * Set the flags on our device.
4335 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4336 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4338 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4342 * Load in the correct multicast list now the flags have changed.
4345 if ((old_flags ^ flags) & IFF_MULTICAST)
4346 dev_change_rx_flags(dev, IFF_MULTICAST);
4348 dev_set_rx_mode(dev);
4351 * Have we downed the interface. We handle IFF_UP ourselves
4352 * according to user attempts to set it, rather than blindly
4357 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4358 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4361 dev_set_rx_mode(dev);
4364 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4365 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4367 dev->gflags ^= IFF_PROMISC;
4368 dev_set_promiscuity(dev, inc);
4371 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4372 is important. Some (broken) drivers set IFF_PROMISC, when
4373 IFF_ALLMULTI is requested not asking us and not reporting.
4375 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4376 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4378 dev->gflags ^= IFF_ALLMULTI;
4379 dev_set_allmulti(dev, inc);
4385 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4387 unsigned int changes = dev->flags ^ old_flags;
4389 if (changes & IFF_UP) {
4390 if (dev->flags & IFF_UP)
4391 call_netdevice_notifiers(NETDEV_UP, dev);
4393 call_netdevice_notifiers(NETDEV_DOWN, dev);
4396 if (dev->flags & IFF_UP &&
4397 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4398 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4402 * dev_change_flags - change device settings
4404 * @flags: device state flags
4406 * Change settings on device based state flags. The flags are
4407 * in the userspace exported format.
4409 int dev_change_flags(struct net_device *dev, unsigned flags)
4412 int old_flags = dev->flags;
4414 ret = __dev_change_flags(dev, flags);
4418 changes = old_flags ^ dev->flags;
4420 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4422 __dev_notify_flags(dev, old_flags);
4425 EXPORT_SYMBOL(dev_change_flags);
4428 * dev_set_mtu - Change maximum transfer unit
4430 * @new_mtu: new transfer unit
4432 * Change the maximum transfer size of the network device.
4434 int dev_set_mtu(struct net_device *dev, int new_mtu)
4436 const struct net_device_ops *ops = dev->netdev_ops;
4439 if (new_mtu == dev->mtu)
4442 /* MTU must be positive. */
4446 if (!netif_device_present(dev))
4450 if (ops->ndo_change_mtu)
4451 err = ops->ndo_change_mtu(dev, new_mtu);
4455 if (!err && dev->flags & IFF_UP)
4456 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4459 EXPORT_SYMBOL(dev_set_mtu);
4462 * dev_set_mac_address - Change Media Access Control Address
4466 * Change the hardware (MAC) address of the device
4468 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4470 const struct net_device_ops *ops = dev->netdev_ops;
4473 if (!ops->ndo_set_mac_address)
4475 if (sa->sa_family != dev->type)
4477 if (!netif_device_present(dev))
4479 err = ops->ndo_set_mac_address(dev, sa);
4481 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4484 EXPORT_SYMBOL(dev_set_mac_address);
4487 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4489 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4492 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4498 case SIOCGIFFLAGS: /* Get interface flags */
4499 ifr->ifr_flags = (short) dev_get_flags(dev);
4502 case SIOCGIFMETRIC: /* Get the metric on the interface
4503 (currently unused) */
4504 ifr->ifr_metric = 0;
4507 case SIOCGIFMTU: /* Get the MTU of a device */
4508 ifr->ifr_mtu = dev->mtu;
4513 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4515 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4516 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4517 ifr->ifr_hwaddr.sa_family = dev->type;
4525 ifr->ifr_map.mem_start = dev->mem_start;
4526 ifr->ifr_map.mem_end = dev->mem_end;
4527 ifr->ifr_map.base_addr = dev->base_addr;
4528 ifr->ifr_map.irq = dev->irq;
4529 ifr->ifr_map.dma = dev->dma;
4530 ifr->ifr_map.port = dev->if_port;
4534 ifr->ifr_ifindex = dev->ifindex;
4538 ifr->ifr_qlen = dev->tx_queue_len;
4542 /* dev_ioctl() should ensure this case
4554 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4556 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4559 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4560 const struct net_device_ops *ops;
4565 ops = dev->netdev_ops;
4568 case SIOCSIFFLAGS: /* Set interface flags */
4569 return dev_change_flags(dev, ifr->ifr_flags);
4571 case SIOCSIFMETRIC: /* Set the metric on the interface
4572 (currently unused) */
4575 case SIOCSIFMTU: /* Set the MTU of a device */
4576 return dev_set_mtu(dev, ifr->ifr_mtu);
4579 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4581 case SIOCSIFHWBROADCAST:
4582 if (ifr->ifr_hwaddr.sa_family != dev->type)
4584 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4585 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4586 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4590 if (ops->ndo_set_config) {
4591 if (!netif_device_present(dev))
4593 return ops->ndo_set_config(dev, &ifr->ifr_map);
4598 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4599 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4601 if (!netif_device_present(dev))
4603 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4606 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4607 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4609 if (!netif_device_present(dev))
4611 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4614 if (ifr->ifr_qlen < 0)
4616 dev->tx_queue_len = ifr->ifr_qlen;
4620 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4621 return dev_change_name(dev, ifr->ifr_newname);
4624 * Unknown or private ioctl
4627 if ((cmd >= SIOCDEVPRIVATE &&
4628 cmd <= SIOCDEVPRIVATE + 15) ||
4629 cmd == SIOCBONDENSLAVE ||
4630 cmd == SIOCBONDRELEASE ||
4631 cmd == SIOCBONDSETHWADDR ||
4632 cmd == SIOCBONDSLAVEINFOQUERY ||
4633 cmd == SIOCBONDINFOQUERY ||
4634 cmd == SIOCBONDCHANGEACTIVE ||
4635 cmd == SIOCGMIIPHY ||
4636 cmd == SIOCGMIIREG ||
4637 cmd == SIOCSMIIREG ||
4638 cmd == SIOCBRADDIF ||
4639 cmd == SIOCBRDELIF ||
4640 cmd == SIOCSHWTSTAMP ||
4641 cmd == SIOCWANDEV) {
4643 if (ops->ndo_do_ioctl) {
4644 if (netif_device_present(dev))
4645 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4657 * This function handles all "interface"-type I/O control requests. The actual
4658 * 'doing' part of this is dev_ifsioc above.
4662 * dev_ioctl - network device ioctl
4663 * @net: the applicable net namespace
4664 * @cmd: command to issue
4665 * @arg: pointer to a struct ifreq in user space
4667 * Issue ioctl functions to devices. This is normally called by the
4668 * user space syscall interfaces but can sometimes be useful for
4669 * other purposes. The return value is the return from the syscall if
4670 * positive or a negative errno code on error.
4673 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4679 /* One special case: SIOCGIFCONF takes ifconf argument
4680 and requires shared lock, because it sleeps writing
4684 if (cmd == SIOCGIFCONF) {
4686 ret = dev_ifconf(net, (char __user *) arg);
4690 if (cmd == SIOCGIFNAME)
4691 return dev_ifname(net, (struct ifreq __user *)arg);
4693 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4696 ifr.ifr_name[IFNAMSIZ-1] = 0;
4698 colon = strchr(ifr.ifr_name, ':');
4703 * See which interface the caller is talking about.
4708 * These ioctl calls:
4709 * - can be done by all.
4710 * - atomic and do not require locking.
4721 dev_load(net, ifr.ifr_name);
4723 ret = dev_ifsioc_locked(net, &ifr, cmd);
4728 if (copy_to_user(arg, &ifr,
4729 sizeof(struct ifreq)))
4735 dev_load(net, ifr.ifr_name);
4737 ret = dev_ethtool(net, &ifr);
4742 if (copy_to_user(arg, &ifr,
4743 sizeof(struct ifreq)))
4749 * These ioctl calls:
4750 * - require superuser power.
4751 * - require strict serialization.
4757 if (!capable(CAP_NET_ADMIN))
4759 dev_load(net, ifr.ifr_name);
4761 ret = dev_ifsioc(net, &ifr, cmd);
4766 if (copy_to_user(arg, &ifr,
4767 sizeof(struct ifreq)))
4773 * These ioctl calls:
4774 * - require superuser power.
4775 * - require strict serialization.
4776 * - do not return a value
4786 case SIOCSIFHWBROADCAST:
4789 case SIOCBONDENSLAVE:
4790 case SIOCBONDRELEASE:
4791 case SIOCBONDSETHWADDR:
4792 case SIOCBONDCHANGEACTIVE:
4796 if (!capable(CAP_NET_ADMIN))
4799 case SIOCBONDSLAVEINFOQUERY:
4800 case SIOCBONDINFOQUERY:
4801 dev_load(net, ifr.ifr_name);
4803 ret = dev_ifsioc(net, &ifr, cmd);
4808 /* Get the per device memory space. We can add this but
4809 * currently do not support it */
4811 /* Set the per device memory buffer space.
4812 * Not applicable in our case */
4817 * Unknown or private ioctl.
4820 if (cmd == SIOCWANDEV ||
4821 (cmd >= SIOCDEVPRIVATE &&
4822 cmd <= SIOCDEVPRIVATE + 15)) {
4823 dev_load(net, ifr.ifr_name);
4825 ret = dev_ifsioc(net, &ifr, cmd);
4827 if (!ret && copy_to_user(arg, &ifr,
4828 sizeof(struct ifreq)))
4832 /* Take care of Wireless Extensions */
4833 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4834 return wext_handle_ioctl(net, &ifr, cmd, arg);
4841 * dev_new_index - allocate an ifindex
4842 * @net: the applicable net namespace
4844 * Returns a suitable unique value for a new device interface
4845 * number. The caller must hold the rtnl semaphore or the
4846 * dev_base_lock to be sure it remains unique.
4848 static int dev_new_index(struct net *net)
4854 if (!__dev_get_by_index(net, ifindex))
4859 /* Delayed registration/unregisteration */
4860 static LIST_HEAD(net_todo_list);
4862 static void net_set_todo(struct net_device *dev)
4864 list_add_tail(&dev->todo_list, &net_todo_list);
4867 static void rollback_registered_many(struct list_head *head)
4869 struct net_device *dev, *tmp;
4871 BUG_ON(dev_boot_phase);
4874 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4875 /* Some devices call without registering
4876 * for initialization unwind. Remove those
4877 * devices and proceed with the remaining.
4879 if (dev->reg_state == NETREG_UNINITIALIZED) {
4880 pr_debug("unregister_netdevice: device %s/%p never "
4881 "was registered\n", dev->name, dev);
4884 list_del(&dev->unreg_list);
4888 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4890 /* If device is running, close it first. */
4893 /* And unlink it from device chain. */
4894 unlist_netdevice(dev);
4896 dev->reg_state = NETREG_UNREGISTERING;
4901 list_for_each_entry(dev, head, unreg_list) {
4902 /* Shutdown queueing discipline. */
4906 /* Notify protocols, that we are about to destroy
4907 this device. They should clean all the things.
4909 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4911 if (!dev->rtnl_link_ops ||
4912 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4913 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4916 * Flush the unicast and multicast chains
4921 if (dev->netdev_ops->ndo_uninit)
4922 dev->netdev_ops->ndo_uninit(dev);
4924 /* Notifier chain MUST detach us from master device. */
4925 WARN_ON(dev->master);
4927 /* Remove entries from kobject tree */
4928 netdev_unregister_kobject(dev);
4931 /* Process any work delayed until the end of the batch */
4932 dev = list_first_entry(head, struct net_device, unreg_list);
4933 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4937 list_for_each_entry(dev, head, unreg_list)
4941 static void rollback_registered(struct net_device *dev)
4945 list_add(&dev->unreg_list, &single);
4946 rollback_registered_many(&single);
4949 unsigned long netdev_fix_features(unsigned long features, const char *name)
4951 /* Fix illegal SG+CSUM combinations. */
4952 if ((features & NETIF_F_SG) &&
4953 !(features & NETIF_F_ALL_CSUM)) {
4955 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4956 "checksum feature.\n", name);
4957 features &= ~NETIF_F_SG;
4960 /* TSO requires that SG is present as well. */
4961 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4963 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4964 "SG feature.\n", name);
4965 features &= ~NETIF_F_TSO;
4968 if (features & NETIF_F_UFO) {
4969 if (!(features & NETIF_F_GEN_CSUM)) {
4971 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4972 "since no NETIF_F_HW_CSUM feature.\n",
4974 features &= ~NETIF_F_UFO;
4977 if (!(features & NETIF_F_SG)) {
4979 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4980 "since no NETIF_F_SG feature.\n", name);
4981 features &= ~NETIF_F_UFO;
4987 EXPORT_SYMBOL(netdev_fix_features);
4990 * netif_stacked_transfer_operstate - transfer operstate
4991 * @rootdev: the root or lower level device to transfer state from
4992 * @dev: the device to transfer operstate to
4994 * Transfer operational state from root to device. This is normally
4995 * called when a stacking relationship exists between the root
4996 * device and the device(a leaf device).
4998 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4999 struct net_device *dev)
5001 if (rootdev->operstate == IF_OPER_DORMANT)
5002 netif_dormant_on(dev);
5004 netif_dormant_off(dev);
5006 if (netif_carrier_ok(rootdev)) {
5007 if (!netif_carrier_ok(dev))
5008 netif_carrier_on(dev);
5010 if (netif_carrier_ok(dev))
5011 netif_carrier_off(dev);
5014 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5016 static int netif_alloc_rx_queues(struct net_device *dev)
5019 unsigned int i, count = dev->num_rx_queues;
5020 struct netdev_rx_queue *rx;
5024 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5026 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5032 * Set a pointer to first element in the array which holds the
5035 for (i = 0; i < count; i++)
5041 static int netif_alloc_netdev_queues(struct net_device *dev)
5043 unsigned int count = dev->num_tx_queues;
5044 struct netdev_queue *tx;
5048 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5050 pr_err("netdev: Unable to allocate %u tx queues.\n",
5058 static void netdev_init_one_queue(struct net_device *dev,
5059 struct netdev_queue *queue,
5064 /* Initialize queue lock */
5065 spin_lock_init(&queue->_xmit_lock);
5066 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5067 queue->xmit_lock_owner = -1;
5070 static void netdev_init_queues(struct net_device *dev)
5072 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5073 spin_lock_init(&dev->tx_global_lock);
5077 * register_netdevice - register a network device
5078 * @dev: device to register
5080 * Take a completed network device structure and add it to the kernel
5081 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5082 * chain. 0 is returned on success. A negative errno code is returned
5083 * on a failure to set up the device, or if the name is a duplicate.
5085 * Callers must hold the rtnl semaphore. You may want
5086 * register_netdev() instead of this.
5089 * The locking appears insufficient to guarantee two parallel registers
5090 * will not get the same name.
5093 int register_netdevice(struct net_device *dev)
5096 struct net *net = dev_net(dev);
5098 BUG_ON(dev_boot_phase);
5103 /* When net_device's are persistent, this will be fatal. */
5104 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5107 spin_lock_init(&dev->addr_list_lock);
5108 netdev_set_addr_lockdep_class(dev);
5112 ret = netif_alloc_rx_queues(dev);
5116 ret = netif_alloc_netdev_queues(dev);
5120 netdev_init_queues(dev);
5122 /* Init, if this function is available */
5123 if (dev->netdev_ops->ndo_init) {
5124 ret = dev->netdev_ops->ndo_init(dev);
5132 ret = dev_get_valid_name(dev, dev->name, 0);
5136 dev->ifindex = dev_new_index(net);
5137 if (dev->iflink == -1)
5138 dev->iflink = dev->ifindex;
5140 /* Fix illegal checksum combinations */
5141 if ((dev->features & NETIF_F_HW_CSUM) &&
5142 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5143 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5145 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5148 if ((dev->features & NETIF_F_NO_CSUM) &&
5149 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5150 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5152 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5155 dev->features = netdev_fix_features(dev->features, dev->name);
5157 /* Enable software GSO if SG is supported. */
5158 if (dev->features & NETIF_F_SG)
5159 dev->features |= NETIF_F_GSO;
5161 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5162 * vlan_dev_init() will do the dev->features check, so these features
5163 * are enabled only if supported by underlying device.
5165 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5167 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5168 ret = notifier_to_errno(ret);
5172 ret = netdev_register_kobject(dev);
5175 dev->reg_state = NETREG_REGISTERED;
5178 * Default initial state at registry is that the
5179 * device is present.
5182 set_bit(__LINK_STATE_PRESENT, &dev->state);
5184 dev_init_scheduler(dev);
5186 list_netdevice(dev);
5188 /* Notify protocols, that a new device appeared. */
5189 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5190 ret = notifier_to_errno(ret);
5192 rollback_registered(dev);
5193 dev->reg_state = NETREG_UNREGISTERED;
5196 * Prevent userspace races by waiting until the network
5197 * device is fully setup before sending notifications.
5199 if (!dev->rtnl_link_ops ||
5200 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5201 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5207 if (dev->netdev_ops->ndo_uninit)
5208 dev->netdev_ops->ndo_uninit(dev);
5211 EXPORT_SYMBOL(register_netdevice);
5214 * init_dummy_netdev - init a dummy network device for NAPI
5215 * @dev: device to init
5217 * This takes a network device structure and initialize the minimum
5218 * amount of fields so it can be used to schedule NAPI polls without
5219 * registering a full blown interface. This is to be used by drivers
5220 * that need to tie several hardware interfaces to a single NAPI
5221 * poll scheduler due to HW limitations.
5223 int init_dummy_netdev(struct net_device *dev)
5225 /* Clear everything. Note we don't initialize spinlocks
5226 * are they aren't supposed to be taken by any of the
5227 * NAPI code and this dummy netdev is supposed to be
5228 * only ever used for NAPI polls
5230 memset(dev, 0, sizeof(struct net_device));
5232 /* make sure we BUG if trying to hit standard
5233 * register/unregister code path
5235 dev->reg_state = NETREG_DUMMY;
5237 /* NAPI wants this */
5238 INIT_LIST_HEAD(&dev->napi_list);
5240 /* a dummy interface is started by default */
5241 set_bit(__LINK_STATE_PRESENT, &dev->state);
5242 set_bit(__LINK_STATE_START, &dev->state);
5244 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5245 * because users of this 'device' dont need to change
5251 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5255 * register_netdev - register a network device
5256 * @dev: device to register
5258 * Take a completed network device structure and add it to the kernel
5259 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5260 * chain. 0 is returned on success. A negative errno code is returned
5261 * on a failure to set up the device, or if the name is a duplicate.
5263 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5264 * and expands the device name if you passed a format string to
5267 int register_netdev(struct net_device *dev)
5274 * If the name is a format string the caller wants us to do a
5277 if (strchr(dev->name, '%')) {
5278 err = dev_alloc_name(dev, dev->name);
5283 err = register_netdevice(dev);
5288 EXPORT_SYMBOL(register_netdev);
5290 int netdev_refcnt_read(const struct net_device *dev)
5294 for_each_possible_cpu(i)
5295 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5298 EXPORT_SYMBOL(netdev_refcnt_read);
5301 * netdev_wait_allrefs - wait until all references are gone.
5303 * This is called when unregistering network devices.
5305 * Any protocol or device that holds a reference should register
5306 * for netdevice notification, and cleanup and put back the
5307 * reference if they receive an UNREGISTER event.
5308 * We can get stuck here if buggy protocols don't correctly
5311 static void netdev_wait_allrefs(struct net_device *dev)
5313 unsigned long rebroadcast_time, warning_time;
5316 linkwatch_forget_dev(dev);
5318 rebroadcast_time = warning_time = jiffies;
5319 refcnt = netdev_refcnt_read(dev);
5321 while (refcnt != 0) {
5322 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5325 /* Rebroadcast unregister notification */
5326 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5327 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5328 * should have already handle it the first time */
5330 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5332 /* We must not have linkwatch events
5333 * pending on unregister. If this
5334 * happens, we simply run the queue
5335 * unscheduled, resulting in a noop
5338 linkwatch_run_queue();
5343 rebroadcast_time = jiffies;
5348 refcnt = netdev_refcnt_read(dev);
5350 if (time_after(jiffies, warning_time + 10 * HZ)) {
5351 printk(KERN_EMERG "unregister_netdevice: "
5352 "waiting for %s to become free. Usage "
5355 warning_time = jiffies;
5364 * register_netdevice(x1);
5365 * register_netdevice(x2);
5367 * unregister_netdevice(y1);
5368 * unregister_netdevice(y2);
5374 * We are invoked by rtnl_unlock().
5375 * This allows us to deal with problems:
5376 * 1) We can delete sysfs objects which invoke hotplug
5377 * without deadlocking with linkwatch via keventd.
5378 * 2) Since we run with the RTNL semaphore not held, we can sleep
5379 * safely in order to wait for the netdev refcnt to drop to zero.
5381 * We must not return until all unregister events added during
5382 * the interval the lock was held have been completed.
5384 void netdev_run_todo(void)
5386 struct list_head list;
5388 /* Snapshot list, allow later requests */
5389 list_replace_init(&net_todo_list, &list);
5393 while (!list_empty(&list)) {
5394 struct net_device *dev
5395 = list_first_entry(&list, struct net_device, todo_list);
5396 list_del(&dev->todo_list);
5398 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5399 printk(KERN_ERR "network todo '%s' but state %d\n",
5400 dev->name, dev->reg_state);
5405 dev->reg_state = NETREG_UNREGISTERED;
5407 on_each_cpu(flush_backlog, dev, 1);
5409 netdev_wait_allrefs(dev);
5412 BUG_ON(netdev_refcnt_read(dev));
5413 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5414 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5415 WARN_ON(dev->dn_ptr);
5417 if (dev->destructor)
5418 dev->destructor(dev);
5420 /* Free network device */
5421 kobject_put(&dev->dev.kobj);
5426 * dev_txq_stats_fold - fold tx_queues stats
5427 * @dev: device to get statistics from
5428 * @stats: struct rtnl_link_stats64 to hold results
5430 void dev_txq_stats_fold(const struct net_device *dev,
5431 struct rtnl_link_stats64 *stats)
5433 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5435 struct netdev_queue *txq;
5437 for (i = 0; i < dev->num_tx_queues; i++) {
5438 txq = netdev_get_tx_queue(dev, i);
5439 spin_lock_bh(&txq->_xmit_lock);
5440 tx_bytes += txq->tx_bytes;
5441 tx_packets += txq->tx_packets;
5442 tx_dropped += txq->tx_dropped;
5443 spin_unlock_bh(&txq->_xmit_lock);
5445 if (tx_bytes || tx_packets || tx_dropped) {
5446 stats->tx_bytes = tx_bytes;
5447 stats->tx_packets = tx_packets;
5448 stats->tx_dropped = tx_dropped;
5451 EXPORT_SYMBOL(dev_txq_stats_fold);
5453 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5454 * fields in the same order, with only the type differing.
5456 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5457 const struct net_device_stats *netdev_stats)
5459 #if BITS_PER_LONG == 64
5460 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5461 memcpy(stats64, netdev_stats, sizeof(*stats64));
5463 size_t i, n = sizeof(*stats64) / sizeof(u64);
5464 const unsigned long *src = (const unsigned long *)netdev_stats;
5465 u64 *dst = (u64 *)stats64;
5467 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5468 sizeof(*stats64) / sizeof(u64));
5469 for (i = 0; i < n; i++)
5475 * dev_get_stats - get network device statistics
5476 * @dev: device to get statistics from
5477 * @storage: place to store stats
5479 * Get network statistics from device. Return @storage.
5480 * The device driver may provide its own method by setting
5481 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5482 * otherwise the internal statistics structure is used.
5484 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5485 struct rtnl_link_stats64 *storage)
5487 const struct net_device_ops *ops = dev->netdev_ops;
5489 if (ops->ndo_get_stats64) {
5490 memset(storage, 0, sizeof(*storage));
5491 ops->ndo_get_stats64(dev, storage);
5492 } else if (ops->ndo_get_stats) {
5493 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5495 netdev_stats_to_stats64(storage, &dev->stats);
5496 dev_txq_stats_fold(dev, storage);
5498 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5501 EXPORT_SYMBOL(dev_get_stats);
5503 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5505 struct netdev_queue *queue = dev_ingress_queue(dev);
5507 #ifdef CONFIG_NET_CLS_ACT
5510 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5513 netdev_init_one_queue(dev, queue, NULL);
5514 queue->qdisc = &noop_qdisc;
5515 queue->qdisc_sleeping = &noop_qdisc;
5516 rcu_assign_pointer(dev->ingress_queue, queue);
5522 * alloc_netdev_mq - allocate network device
5523 * @sizeof_priv: size of private data to allocate space for
5524 * @name: device name format string
5525 * @setup: callback to initialize device
5526 * @queue_count: the number of subqueues to allocate
5528 * Allocates a struct net_device with private data area for driver use
5529 * and performs basic initialization. Also allocates subquue structs
5530 * for each queue on the device at the end of the netdevice.
5532 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5533 void (*setup)(struct net_device *), unsigned int queue_count)
5535 struct net_device *dev;
5537 struct net_device *p;
5539 BUG_ON(strlen(name) >= sizeof(dev->name));
5541 if (queue_count < 1) {
5542 pr_err("alloc_netdev: Unable to allocate device "
5543 "with zero queues.\n");
5547 alloc_size = sizeof(struct net_device);
5549 /* ensure 32-byte alignment of private area */
5550 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5551 alloc_size += sizeof_priv;
5553 /* ensure 32-byte alignment of whole construct */
5554 alloc_size += NETDEV_ALIGN - 1;
5556 p = kzalloc(alloc_size, GFP_KERNEL);
5558 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5562 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5563 dev->padded = (char *)dev - (char *)p;
5565 dev->pcpu_refcnt = alloc_percpu(int);
5566 if (!dev->pcpu_refcnt)
5569 if (dev_addr_init(dev))
5575 dev_net_set(dev, &init_net);
5577 dev->num_tx_queues = queue_count;
5578 dev->real_num_tx_queues = queue_count;
5581 dev->num_rx_queues = queue_count;
5582 dev->real_num_rx_queues = queue_count;
5585 dev->gso_max_size = GSO_MAX_SIZE;
5587 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5588 dev->ethtool_ntuple_list.count = 0;
5589 INIT_LIST_HEAD(&dev->napi_list);
5590 INIT_LIST_HEAD(&dev->unreg_list);
5591 INIT_LIST_HEAD(&dev->link_watch_list);
5592 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5594 strcpy(dev->name, name);
5598 free_percpu(dev->pcpu_refcnt);
5603 EXPORT_SYMBOL(alloc_netdev_mq);
5606 * free_netdev - free network device
5609 * This function does the last stage of destroying an allocated device
5610 * interface. The reference to the device object is released.
5611 * If this is the last reference then it will be freed.
5613 void free_netdev(struct net_device *dev)
5615 struct napi_struct *p, *n;
5617 release_net(dev_net(dev));
5621 kfree(rcu_dereference_raw(dev->ingress_queue));
5623 /* Flush device addresses */
5624 dev_addr_flush(dev);
5626 /* Clear ethtool n-tuple list */
5627 ethtool_ntuple_flush(dev);
5629 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5632 free_percpu(dev->pcpu_refcnt);
5633 dev->pcpu_refcnt = NULL;
5635 /* Compatibility with error handling in drivers */
5636 if (dev->reg_state == NETREG_UNINITIALIZED) {
5637 kfree((char *)dev - dev->padded);
5641 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5642 dev->reg_state = NETREG_RELEASED;
5644 /* will free via device release */
5645 put_device(&dev->dev);
5647 EXPORT_SYMBOL(free_netdev);
5650 * synchronize_net - Synchronize with packet receive processing
5652 * Wait for packets currently being received to be done.
5653 * Does not block later packets from starting.
5655 void synchronize_net(void)
5660 EXPORT_SYMBOL(synchronize_net);
5663 * unregister_netdevice_queue - remove device from the kernel
5667 * This function shuts down a device interface and removes it
5668 * from the kernel tables.
5669 * If head not NULL, device is queued to be unregistered later.
5671 * Callers must hold the rtnl semaphore. You may want
5672 * unregister_netdev() instead of this.
5675 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5680 list_move_tail(&dev->unreg_list, head);
5682 rollback_registered(dev);
5683 /* Finish processing unregister after unlock */
5687 EXPORT_SYMBOL(unregister_netdevice_queue);
5690 * unregister_netdevice_many - unregister many devices
5691 * @head: list of devices
5693 void unregister_netdevice_many(struct list_head *head)
5695 struct net_device *dev;
5697 if (!list_empty(head)) {
5698 rollback_registered_many(head);
5699 list_for_each_entry(dev, head, unreg_list)
5703 EXPORT_SYMBOL(unregister_netdevice_many);
5706 * unregister_netdev - remove device from the kernel
5709 * This function shuts down a device interface and removes it
5710 * from the kernel tables.
5712 * This is just a wrapper for unregister_netdevice that takes
5713 * the rtnl semaphore. In general you want to use this and not
5714 * unregister_netdevice.
5716 void unregister_netdev(struct net_device *dev)
5719 unregister_netdevice(dev);
5722 EXPORT_SYMBOL(unregister_netdev);
5725 * dev_change_net_namespace - move device to different nethost namespace
5727 * @net: network namespace
5728 * @pat: If not NULL name pattern to try if the current device name
5729 * is already taken in the destination network namespace.
5731 * This function shuts down a device interface and moves it
5732 * to a new network namespace. On success 0 is returned, on
5733 * a failure a netagive errno code is returned.
5735 * Callers must hold the rtnl semaphore.
5738 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5744 /* Don't allow namespace local devices to be moved. */
5746 if (dev->features & NETIF_F_NETNS_LOCAL)
5749 /* Ensure the device has been registrered */
5751 if (dev->reg_state != NETREG_REGISTERED)
5754 /* Get out if there is nothing todo */
5756 if (net_eq(dev_net(dev), net))
5759 /* Pick the destination device name, and ensure
5760 * we can use it in the destination network namespace.
5763 if (__dev_get_by_name(net, dev->name)) {
5764 /* We get here if we can't use the current device name */
5767 if (dev_get_valid_name(dev, pat, 1))
5772 * And now a mini version of register_netdevice unregister_netdevice.
5775 /* If device is running close it first. */
5778 /* And unlink it from device chain */
5780 unlist_netdevice(dev);
5784 /* Shutdown queueing discipline. */
5787 /* Notify protocols, that we are about to destroy
5788 this device. They should clean all the things.
5790 Note that dev->reg_state stays at NETREG_REGISTERED.
5791 This is wanted because this way 8021q and macvlan know
5792 the device is just moving and can keep their slaves up.
5794 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5795 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5798 * Flush the unicast and multicast chains
5803 /* Actually switch the network namespace */
5804 dev_net_set(dev, net);
5806 /* If there is an ifindex conflict assign a new one */
5807 if (__dev_get_by_index(net, dev->ifindex)) {
5808 int iflink = (dev->iflink == dev->ifindex);
5809 dev->ifindex = dev_new_index(net);
5811 dev->iflink = dev->ifindex;
5814 /* Fixup kobjects */
5815 err = device_rename(&dev->dev, dev->name);
5818 /* Add the device back in the hashes */
5819 list_netdevice(dev);
5821 /* Notify protocols, that a new device appeared. */
5822 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5825 * Prevent userspace races by waiting until the network
5826 * device is fully setup before sending notifications.
5828 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5835 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5837 static int dev_cpu_callback(struct notifier_block *nfb,
5838 unsigned long action,
5841 struct sk_buff **list_skb;
5842 struct sk_buff *skb;
5843 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5844 struct softnet_data *sd, *oldsd;
5846 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5849 local_irq_disable();
5850 cpu = smp_processor_id();
5851 sd = &per_cpu(softnet_data, cpu);
5852 oldsd = &per_cpu(softnet_data, oldcpu);
5854 /* Find end of our completion_queue. */
5855 list_skb = &sd->completion_queue;
5857 list_skb = &(*list_skb)->next;
5858 /* Append completion queue from offline CPU. */
5859 *list_skb = oldsd->completion_queue;
5860 oldsd->completion_queue = NULL;
5862 /* Append output queue from offline CPU. */
5863 if (oldsd->output_queue) {
5864 *sd->output_queue_tailp = oldsd->output_queue;
5865 sd->output_queue_tailp = oldsd->output_queue_tailp;
5866 oldsd->output_queue = NULL;
5867 oldsd->output_queue_tailp = &oldsd->output_queue;
5870 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5873 /* Process offline CPU's input_pkt_queue */
5874 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5876 input_queue_head_incr(oldsd);
5878 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5880 input_queue_head_incr(oldsd);
5888 * netdev_increment_features - increment feature set by one
5889 * @all: current feature set
5890 * @one: new feature set
5891 * @mask: mask feature set
5893 * Computes a new feature set after adding a device with feature set
5894 * @one to the master device with current feature set @all. Will not
5895 * enable anything that is off in @mask. Returns the new feature set.
5897 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5900 /* If device needs checksumming, downgrade to it. */
5901 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5902 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5903 else if (mask & NETIF_F_ALL_CSUM) {
5904 /* If one device supports v4/v6 checksumming, set for all. */
5905 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5906 !(all & NETIF_F_GEN_CSUM)) {
5907 all &= ~NETIF_F_ALL_CSUM;
5908 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5911 /* If one device supports hw checksumming, set for all. */
5912 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5913 all &= ~NETIF_F_ALL_CSUM;
5914 all |= NETIF_F_HW_CSUM;
5918 one |= NETIF_F_ALL_CSUM;
5920 one |= all & NETIF_F_ONE_FOR_ALL;
5921 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5922 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5926 EXPORT_SYMBOL(netdev_increment_features);
5928 static struct hlist_head *netdev_create_hash(void)
5931 struct hlist_head *hash;
5933 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5935 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5936 INIT_HLIST_HEAD(&hash[i]);
5941 /* Initialize per network namespace state */
5942 static int __net_init netdev_init(struct net *net)
5944 INIT_LIST_HEAD(&net->dev_base_head);
5946 net->dev_name_head = netdev_create_hash();
5947 if (net->dev_name_head == NULL)
5950 net->dev_index_head = netdev_create_hash();
5951 if (net->dev_index_head == NULL)
5957 kfree(net->dev_name_head);
5963 * netdev_drivername - network driver for the device
5964 * @dev: network device
5965 * @buffer: buffer for resulting name
5966 * @len: size of buffer
5968 * Determine network driver for device.
5970 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5972 const struct device_driver *driver;
5973 const struct device *parent;
5975 if (len <= 0 || !buffer)
5979 parent = dev->dev.parent;
5984 driver = parent->driver;
5985 if (driver && driver->name)
5986 strlcpy(buffer, driver->name, len);
5990 static int __netdev_printk(const char *level, const struct net_device *dev,
5991 struct va_format *vaf)
5995 if (dev && dev->dev.parent)
5996 r = dev_printk(level, dev->dev.parent, "%s: %pV",
5997 netdev_name(dev), vaf);
5999 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6001 r = printk("%s(NULL net_device): %pV", level, vaf);
6006 int netdev_printk(const char *level, const struct net_device *dev,
6007 const char *format, ...)
6009 struct va_format vaf;
6013 va_start(args, format);
6018 r = __netdev_printk(level, dev, &vaf);
6023 EXPORT_SYMBOL(netdev_printk);
6025 #define define_netdev_printk_level(func, level) \
6026 int func(const struct net_device *dev, const char *fmt, ...) \
6029 struct va_format vaf; \
6032 va_start(args, fmt); \
6037 r = __netdev_printk(level, dev, &vaf); \
6042 EXPORT_SYMBOL(func);
6044 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6045 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6046 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6047 define_netdev_printk_level(netdev_err, KERN_ERR);
6048 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6049 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6050 define_netdev_printk_level(netdev_info, KERN_INFO);
6052 static void __net_exit netdev_exit(struct net *net)
6054 kfree(net->dev_name_head);
6055 kfree(net->dev_index_head);
6058 static struct pernet_operations __net_initdata netdev_net_ops = {
6059 .init = netdev_init,
6060 .exit = netdev_exit,
6063 static void __net_exit default_device_exit(struct net *net)
6065 struct net_device *dev, *aux;
6067 * Push all migratable network devices back to the
6068 * initial network namespace
6071 for_each_netdev_safe(net, dev, aux) {
6073 char fb_name[IFNAMSIZ];
6075 /* Ignore unmoveable devices (i.e. loopback) */
6076 if (dev->features & NETIF_F_NETNS_LOCAL)
6079 /* Leave virtual devices for the generic cleanup */
6080 if (dev->rtnl_link_ops)
6083 /* Push remaing network devices to init_net */
6084 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6085 err = dev_change_net_namespace(dev, &init_net, fb_name);
6087 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6088 __func__, dev->name, err);
6095 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6097 /* At exit all network devices most be removed from a network
6098 * namespace. Do this in the reverse order of registeration.
6099 * Do this across as many network namespaces as possible to
6100 * improve batching efficiency.
6102 struct net_device *dev;
6104 LIST_HEAD(dev_kill_list);
6107 list_for_each_entry(net, net_list, exit_list) {
6108 for_each_netdev_reverse(net, dev) {
6109 if (dev->rtnl_link_ops)
6110 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6112 unregister_netdevice_queue(dev, &dev_kill_list);
6115 unregister_netdevice_many(&dev_kill_list);
6119 static struct pernet_operations __net_initdata default_device_ops = {
6120 .exit = default_device_exit,
6121 .exit_batch = default_device_exit_batch,
6125 * Initialize the DEV module. At boot time this walks the device list and
6126 * unhooks any devices that fail to initialise (normally hardware not
6127 * present) and leaves us with a valid list of present and active devices.
6132 * This is called single threaded during boot, so no need
6133 * to take the rtnl semaphore.
6135 static int __init net_dev_init(void)
6137 int i, rc = -ENOMEM;
6139 BUG_ON(!dev_boot_phase);
6141 if (dev_proc_init())
6144 if (netdev_kobject_init())
6147 INIT_LIST_HEAD(&ptype_all);
6148 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6149 INIT_LIST_HEAD(&ptype_base[i]);
6151 if (register_pernet_subsys(&netdev_net_ops))
6155 * Initialise the packet receive queues.
6158 for_each_possible_cpu(i) {
6159 struct softnet_data *sd = &per_cpu(softnet_data, i);
6161 memset(sd, 0, sizeof(*sd));
6162 skb_queue_head_init(&sd->input_pkt_queue);
6163 skb_queue_head_init(&sd->process_queue);
6164 sd->completion_queue = NULL;
6165 INIT_LIST_HEAD(&sd->poll_list);
6166 sd->output_queue = NULL;
6167 sd->output_queue_tailp = &sd->output_queue;
6169 sd->csd.func = rps_trigger_softirq;
6175 sd->backlog.poll = process_backlog;
6176 sd->backlog.weight = weight_p;
6177 sd->backlog.gro_list = NULL;
6178 sd->backlog.gro_count = 0;
6183 /* The loopback device is special if any other network devices
6184 * is present in a network namespace the loopback device must
6185 * be present. Since we now dynamically allocate and free the
6186 * loopback device ensure this invariant is maintained by
6187 * keeping the loopback device as the first device on the
6188 * list of network devices. Ensuring the loopback devices
6189 * is the first device that appears and the last network device
6192 if (register_pernet_device(&loopback_net_ops))
6195 if (register_pernet_device(&default_device_ops))
6198 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6199 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6201 hotcpu_notifier(dev_cpu_callback, 0);
6209 subsys_initcall(net_dev_init);
6211 static int __init initialize_hashrnd(void)
6213 get_random_bytes(&hashrnd, sizeof(hashrnd));
6217 late_initcall_sync(initialize_hashrnd);