2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
145 * The list of packet types we will receive (as opposed to discard)
146 * and the routines to invoke.
148 * Why 16. Because with 16 the only overlap we get on a hash of the
149 * low nibble of the protocol value is RARP/SNAP/X.25.
151 * NOTE: That is no longer true with the addition of VLAN tags. Not
152 * sure which should go first, but I bet it won't make much
153 * difference if we are running VLANs. The good news is that
154 * this protocol won't be in the list unless compiled in, so
155 * the average user (w/out VLANs) will not be adversely affected.
172 #define PTYPE_HASH_SIZE (16)
173 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
175 static DEFINE_SPINLOCK(ptype_lock);
176 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
177 static struct list_head ptype_all __read_mostly; /* Taps */
180 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
183 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
185 * Writers must hold the rtnl semaphore while they loop through the
186 * dev_base_head list, and hold dev_base_lock for writing when they do the
187 * actual updates. This allows pure readers to access the list even
188 * while a writer is preparing to update it.
190 * To put it another way, dev_base_lock is held for writing only to
191 * protect against pure readers; the rtnl semaphore provides the
192 * protection against other writers.
194 * See, for example usages, register_netdevice() and
195 * unregister_netdevice(), which must be called with the rtnl
198 DEFINE_RWLOCK(dev_base_lock);
199 EXPORT_SYMBOL(dev_base_lock);
201 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
203 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
204 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
207 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
209 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
212 static inline void rps_lock(struct softnet_data *sd)
215 spin_lock(&sd->input_pkt_queue.lock);
219 static inline void rps_unlock(struct softnet_data *sd)
222 spin_unlock(&sd->input_pkt_queue.lock);
226 /* Device list insertion */
227 static int list_netdevice(struct net_device *dev)
229 struct net *net = dev_net(dev);
233 write_lock_bh(&dev_base_lock);
234 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
235 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
236 hlist_add_head_rcu(&dev->index_hlist,
237 dev_index_hash(net, dev->ifindex));
238 write_unlock_bh(&dev_base_lock);
242 /* Device list removal
243 * caller must respect a RCU grace period before freeing/reusing dev
245 static void unlist_netdevice(struct net_device *dev)
249 /* Unlink dev from the device chain */
250 write_lock_bh(&dev_base_lock);
251 list_del_rcu(&dev->dev_list);
252 hlist_del_rcu(&dev->name_hlist);
253 hlist_del_rcu(&dev->index_hlist);
254 write_unlock_bh(&dev_base_lock);
261 static RAW_NOTIFIER_HEAD(netdev_chain);
264 * Device drivers call our routines to queue packets here. We empty the
265 * queue in the local softnet handler.
268 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
269 EXPORT_PER_CPU_SYMBOL(softnet_data);
271 #ifdef CONFIG_LOCKDEP
273 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
274 * according to dev->type
276 static const unsigned short netdev_lock_type[] =
277 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
278 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
279 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
280 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
281 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
282 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
283 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
284 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
285 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
286 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
287 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
288 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
289 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
290 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
291 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
292 ARPHRD_VOID, ARPHRD_NONE};
294 static const char *const netdev_lock_name[] =
295 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
296 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
297 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
298 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
299 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
300 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
301 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
302 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
303 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
304 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
305 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
306 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
307 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
308 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
309 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
310 "_xmit_VOID", "_xmit_NONE"};
312 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
313 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
315 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
319 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
320 if (netdev_lock_type[i] == dev_type)
322 /* the last key is used by default */
323 return ARRAY_SIZE(netdev_lock_type) - 1;
326 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
327 unsigned short dev_type)
331 i = netdev_lock_pos(dev_type);
332 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
333 netdev_lock_name[i]);
336 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
340 i = netdev_lock_pos(dev->type);
341 lockdep_set_class_and_name(&dev->addr_list_lock,
342 &netdev_addr_lock_key[i],
343 netdev_lock_name[i]);
346 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
347 unsigned short dev_type)
350 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
355 /*******************************************************************************
357 Protocol management and registration routines
359 *******************************************************************************/
362 * Add a protocol ID to the list. Now that the input handler is
363 * smarter we can dispense with all the messy stuff that used to be
366 * BEWARE!!! Protocol handlers, mangling input packets,
367 * MUST BE last in hash buckets and checking protocol handlers
368 * MUST start from promiscuous ptype_all chain in net_bh.
369 * It is true now, do not change it.
370 * Explanation follows: if protocol handler, mangling packet, will
371 * be the first on list, it is not able to sense, that packet
372 * is cloned and should be copied-on-write, so that it will
373 * change it and subsequent readers will get broken packet.
377 static inline struct list_head *ptype_head(const struct packet_type *pt)
379 if (pt->type == htons(ETH_P_ALL))
382 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
386 * dev_add_pack - add packet handler
387 * @pt: packet type declaration
389 * Add a protocol handler to the networking stack. The passed &packet_type
390 * is linked into kernel lists and may not be freed until it has been
391 * removed from the kernel lists.
393 * This call does not sleep therefore it can not
394 * guarantee all CPU's that are in middle of receiving packets
395 * will see the new packet type (until the next received packet).
398 void dev_add_pack(struct packet_type *pt)
400 struct list_head *head = ptype_head(pt);
402 spin_lock(&ptype_lock);
403 list_add_rcu(&pt->list, head);
404 spin_unlock(&ptype_lock);
406 EXPORT_SYMBOL(dev_add_pack);
409 * __dev_remove_pack - remove packet handler
410 * @pt: packet type declaration
412 * Remove a protocol handler that was previously added to the kernel
413 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
414 * from the kernel lists and can be freed or reused once this function
417 * The packet type might still be in use by receivers
418 * and must not be freed until after all the CPU's have gone
419 * through a quiescent state.
421 void __dev_remove_pack(struct packet_type *pt)
423 struct list_head *head = ptype_head(pt);
424 struct packet_type *pt1;
426 spin_lock(&ptype_lock);
428 list_for_each_entry(pt1, head, list) {
430 list_del_rcu(&pt->list);
435 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
437 spin_unlock(&ptype_lock);
439 EXPORT_SYMBOL(__dev_remove_pack);
442 * dev_remove_pack - remove packet handler
443 * @pt: packet type declaration
445 * Remove a protocol handler that was previously added to the kernel
446 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
447 * from the kernel lists and can be freed or reused once this function
450 * This call sleeps to guarantee that no CPU is looking at the packet
453 void dev_remove_pack(struct packet_type *pt)
455 __dev_remove_pack(pt);
459 EXPORT_SYMBOL(dev_remove_pack);
461 /******************************************************************************
463 Device Boot-time Settings Routines
465 *******************************************************************************/
467 /* Boot time configuration table */
468 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
471 * netdev_boot_setup_add - add new setup entry
472 * @name: name of the device
473 * @map: configured settings for the device
475 * Adds new setup entry to the dev_boot_setup list. The function
476 * returns 0 on error and 1 on success. This is a generic routine to
479 static int netdev_boot_setup_add(char *name, struct ifmap *map)
481 struct netdev_boot_setup *s;
485 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
486 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
487 memset(s[i].name, 0, sizeof(s[i].name));
488 strlcpy(s[i].name, name, IFNAMSIZ);
489 memcpy(&s[i].map, map, sizeof(s[i].map));
494 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
498 * netdev_boot_setup_check - check boot time settings
499 * @dev: the netdevice
501 * Check boot time settings for the device.
502 * The found settings are set for the device to be used
503 * later in the device probing.
504 * Returns 0 if no settings found, 1 if they are.
506 int netdev_boot_setup_check(struct net_device *dev)
508 struct netdev_boot_setup *s = dev_boot_setup;
511 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
512 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
513 !strcmp(dev->name, s[i].name)) {
514 dev->irq = s[i].map.irq;
515 dev->base_addr = s[i].map.base_addr;
516 dev->mem_start = s[i].map.mem_start;
517 dev->mem_end = s[i].map.mem_end;
523 EXPORT_SYMBOL(netdev_boot_setup_check);
527 * netdev_boot_base - get address from boot time settings
528 * @prefix: prefix for network device
529 * @unit: id for network device
531 * Check boot time settings for the base address of device.
532 * The found settings are set for the device to be used
533 * later in the device probing.
534 * Returns 0 if no settings found.
536 unsigned long netdev_boot_base(const char *prefix, int unit)
538 const struct netdev_boot_setup *s = dev_boot_setup;
542 sprintf(name, "%s%d", prefix, unit);
545 * If device already registered then return base of 1
546 * to indicate not to probe for this interface
548 if (__dev_get_by_name(&init_net, name))
551 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
552 if (!strcmp(name, s[i].name))
553 return s[i].map.base_addr;
558 * Saves at boot time configured settings for any netdevice.
560 int __init netdev_boot_setup(char *str)
565 str = get_options(str, ARRAY_SIZE(ints), ints);
570 memset(&map, 0, sizeof(map));
574 map.base_addr = ints[2];
576 map.mem_start = ints[3];
578 map.mem_end = ints[4];
580 /* Add new entry to the list */
581 return netdev_boot_setup_add(str, &map);
584 __setup("netdev=", netdev_boot_setup);
586 /*******************************************************************************
588 Device Interface Subroutines
590 *******************************************************************************/
593 * __dev_get_by_name - find a device by its name
594 * @net: the applicable net namespace
595 * @name: name to find
597 * Find an interface by name. Must be called under RTNL semaphore
598 * or @dev_base_lock. If the name is found a pointer to the device
599 * is returned. If the name is not found then %NULL is returned. The
600 * reference counters are not incremented so the caller must be
601 * careful with locks.
604 struct net_device *__dev_get_by_name(struct net *net, const char *name)
606 struct hlist_node *p;
607 struct net_device *dev;
608 struct hlist_head *head = dev_name_hash(net, name);
610 hlist_for_each_entry(dev, p, head, name_hlist)
611 if (!strncmp(dev->name, name, IFNAMSIZ))
616 EXPORT_SYMBOL(__dev_get_by_name);
619 * dev_get_by_name_rcu - find a device by its name
620 * @net: the applicable net namespace
621 * @name: name to find
623 * Find an interface by name.
624 * If the name is found a pointer to the device is returned.
625 * If the name is not found then %NULL is returned.
626 * The reference counters are not incremented so the caller must be
627 * careful with locks. The caller must hold RCU lock.
630 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
632 struct hlist_node *p;
633 struct net_device *dev;
634 struct hlist_head *head = dev_name_hash(net, name);
636 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
637 if (!strncmp(dev->name, name, IFNAMSIZ))
642 EXPORT_SYMBOL(dev_get_by_name_rcu);
645 * dev_get_by_name - find a device by its name
646 * @net: the applicable net namespace
647 * @name: name to find
649 * Find an interface by name. This can be called from any
650 * context and does its own locking. The returned handle has
651 * the usage count incremented and the caller must use dev_put() to
652 * release it when it is no longer needed. %NULL is returned if no
653 * matching device is found.
656 struct net_device *dev_get_by_name(struct net *net, const char *name)
658 struct net_device *dev;
661 dev = dev_get_by_name_rcu(net, name);
667 EXPORT_SYMBOL(dev_get_by_name);
670 * __dev_get_by_index - find a device by its ifindex
671 * @net: the applicable net namespace
672 * @ifindex: index of device
674 * Search for an interface by index. Returns %NULL if the device
675 * is not found or a pointer to the device. The device has not
676 * had its reference counter increased so the caller must be careful
677 * about locking. The caller must hold either the RTNL semaphore
681 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
683 struct hlist_node *p;
684 struct net_device *dev;
685 struct hlist_head *head = dev_index_hash(net, ifindex);
687 hlist_for_each_entry(dev, p, head, index_hlist)
688 if (dev->ifindex == ifindex)
693 EXPORT_SYMBOL(__dev_get_by_index);
696 * dev_get_by_index_rcu - find a device by its ifindex
697 * @net: the applicable net namespace
698 * @ifindex: index of device
700 * Search for an interface by index. Returns %NULL if the device
701 * is not found or a pointer to the device. The device has not
702 * had its reference counter increased so the caller must be careful
703 * about locking. The caller must hold RCU lock.
706 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
708 struct hlist_node *p;
709 struct net_device *dev;
710 struct hlist_head *head = dev_index_hash(net, ifindex);
712 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
713 if (dev->ifindex == ifindex)
718 EXPORT_SYMBOL(dev_get_by_index_rcu);
722 * dev_get_by_index - find a device by its ifindex
723 * @net: the applicable net namespace
724 * @ifindex: index of device
726 * Search for an interface by index. Returns NULL if the device
727 * is not found or a pointer to the device. The device returned has
728 * had a reference added and the pointer is safe until the user calls
729 * dev_put to indicate they have finished with it.
732 struct net_device *dev_get_by_index(struct net *net, int ifindex)
734 struct net_device *dev;
737 dev = dev_get_by_index_rcu(net, ifindex);
743 EXPORT_SYMBOL(dev_get_by_index);
746 * dev_getbyhwaddr - find a device by its hardware address
747 * @net: the applicable net namespace
748 * @type: media type of device
749 * @ha: hardware address
751 * Search for an interface by MAC address. Returns NULL if the device
752 * is not found or a pointer to the device. The caller must hold the
753 * rtnl semaphore. The returned device has not had its ref count increased
754 * and the caller must therefore be careful about locking
757 * If the API was consistent this would be __dev_get_by_hwaddr
760 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
762 struct net_device *dev;
766 for_each_netdev(net, dev)
767 if (dev->type == type &&
768 !memcmp(dev->dev_addr, ha, dev->addr_len))
773 EXPORT_SYMBOL(dev_getbyhwaddr);
775 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
777 struct net_device *dev;
780 for_each_netdev(net, dev)
781 if (dev->type == type)
786 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
788 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
790 struct net_device *dev, *ret = NULL;
793 for_each_netdev_rcu(net, dev)
794 if (dev->type == type) {
802 EXPORT_SYMBOL(dev_getfirstbyhwtype);
805 * dev_get_by_flags_rcu - find any device with given flags
806 * @net: the applicable net namespace
807 * @if_flags: IFF_* values
808 * @mask: bitmask of bits in if_flags to check
810 * Search for any interface with the given flags. Returns NULL if a device
811 * is not found or a pointer to the device. Must be called inside
812 * rcu_read_lock(), and result refcount is unchanged.
815 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
818 struct net_device *dev, *ret;
821 for_each_netdev_rcu(net, dev) {
822 if (((dev->flags ^ if_flags) & mask) == 0) {
829 EXPORT_SYMBOL(dev_get_by_flags_rcu);
832 * dev_valid_name - check if name is okay for network device
835 * Network device names need to be valid file names to
836 * to allow sysfs to work. We also disallow any kind of
839 int dev_valid_name(const char *name)
843 if (strlen(name) >= IFNAMSIZ)
845 if (!strcmp(name, ".") || !strcmp(name, ".."))
849 if (*name == '/' || isspace(*name))
855 EXPORT_SYMBOL(dev_valid_name);
858 * __dev_alloc_name - allocate a name for a device
859 * @net: network namespace to allocate the device name in
860 * @name: name format string
861 * @buf: scratch buffer and result name string
863 * Passed a format string - eg "lt%d" it will try and find a suitable
864 * id. It scans list of devices to build up a free map, then chooses
865 * the first empty slot. The caller must hold the dev_base or rtnl lock
866 * while allocating the name and adding the device in order to avoid
868 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
869 * Returns the number of the unit assigned or a negative errno code.
872 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
876 const int max_netdevices = 8*PAGE_SIZE;
877 unsigned long *inuse;
878 struct net_device *d;
880 p = strnchr(name, IFNAMSIZ-1, '%');
883 * Verify the string as this thing may have come from
884 * the user. There must be either one "%d" and no other "%"
887 if (p[1] != 'd' || strchr(p + 2, '%'))
890 /* Use one page as a bit array of possible slots */
891 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
895 for_each_netdev(net, d) {
896 if (!sscanf(d->name, name, &i))
898 if (i < 0 || i >= max_netdevices)
901 /* avoid cases where sscanf is not exact inverse of printf */
902 snprintf(buf, IFNAMSIZ, name, i);
903 if (!strncmp(buf, d->name, IFNAMSIZ))
907 i = find_first_zero_bit(inuse, max_netdevices);
908 free_page((unsigned long) inuse);
912 snprintf(buf, IFNAMSIZ, name, i);
913 if (!__dev_get_by_name(net, buf))
916 /* It is possible to run out of possible slots
917 * when the name is long and there isn't enough space left
918 * for the digits, or if all bits are used.
924 * dev_alloc_name - allocate a name for a device
926 * @name: name format string
928 * Passed a format string - eg "lt%d" it will try and find a suitable
929 * id. It scans list of devices to build up a free map, then chooses
930 * the first empty slot. The caller must hold the dev_base or rtnl lock
931 * while allocating the name and adding the device in order to avoid
933 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
934 * Returns the number of the unit assigned or a negative errno code.
937 int dev_alloc_name(struct net_device *dev, const char *name)
943 BUG_ON(!dev_net(dev));
945 ret = __dev_alloc_name(net, name, buf);
947 strlcpy(dev->name, buf, IFNAMSIZ);
950 EXPORT_SYMBOL(dev_alloc_name);
952 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
956 BUG_ON(!dev_net(dev));
959 if (!dev_valid_name(name))
962 if (fmt && strchr(name, '%'))
963 return dev_alloc_name(dev, name);
964 else if (__dev_get_by_name(net, name))
966 else if (dev->name != name)
967 strlcpy(dev->name, name, IFNAMSIZ);
973 * dev_change_name - change name of a device
975 * @newname: name (or format string) must be at least IFNAMSIZ
977 * Change name of a device, can pass format strings "eth%d".
980 int dev_change_name(struct net_device *dev, const char *newname)
982 char oldname[IFNAMSIZ];
988 BUG_ON(!dev_net(dev));
991 if (dev->flags & IFF_UP)
994 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
997 memcpy(oldname, dev->name, IFNAMSIZ);
999 err = dev_get_valid_name(dev, newname, 1);
1004 ret = device_rename(&dev->dev, dev->name);
1006 memcpy(dev->name, oldname, IFNAMSIZ);
1010 write_lock_bh(&dev_base_lock);
1011 hlist_del(&dev->name_hlist);
1012 write_unlock_bh(&dev_base_lock);
1016 write_lock_bh(&dev_base_lock);
1017 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1018 write_unlock_bh(&dev_base_lock);
1020 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1021 ret = notifier_to_errno(ret);
1024 /* err >= 0 after dev_alloc_name() or stores the first errno */
1027 memcpy(dev->name, oldname, IFNAMSIZ);
1031 "%s: name change rollback failed: %d.\n",
1040 * dev_set_alias - change ifalias of a device
1042 * @alias: name up to IFALIASZ
1043 * @len: limit of bytes to copy from info
1045 * Set ifalias for a device,
1047 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1051 if (len >= IFALIASZ)
1056 kfree(dev->ifalias);
1057 dev->ifalias = NULL;
1062 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1066 strlcpy(dev->ifalias, alias, len+1);
1072 * netdev_features_change - device changes features
1073 * @dev: device to cause notification
1075 * Called to indicate a device has changed features.
1077 void netdev_features_change(struct net_device *dev)
1079 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1081 EXPORT_SYMBOL(netdev_features_change);
1084 * netdev_state_change - device changes state
1085 * @dev: device to cause notification
1087 * Called to indicate a device has changed state. This function calls
1088 * the notifier chains for netdev_chain and sends a NEWLINK message
1089 * to the routing socket.
1091 void netdev_state_change(struct net_device *dev)
1093 if (dev->flags & IFF_UP) {
1094 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1095 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1098 EXPORT_SYMBOL(netdev_state_change);
1100 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1102 return call_netdevice_notifiers(event, dev);
1104 EXPORT_SYMBOL(netdev_bonding_change);
1107 * dev_load - load a network module
1108 * @net: the applicable net namespace
1109 * @name: name of interface
1111 * If a network interface is not present and the process has suitable
1112 * privileges this function loads the module. If module loading is not
1113 * available in this kernel then it becomes a nop.
1116 void dev_load(struct net *net, const char *name)
1118 struct net_device *dev;
1121 dev = dev_get_by_name_rcu(net, name);
1124 if (!dev && capable(CAP_NET_ADMIN))
1125 request_module("%s", name);
1127 EXPORT_SYMBOL(dev_load);
1129 static int __dev_open(struct net_device *dev)
1131 const struct net_device_ops *ops = dev->netdev_ops;
1137 * Is it even present?
1139 if (!netif_device_present(dev))
1142 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1143 ret = notifier_to_errno(ret);
1148 * Call device private open method
1150 set_bit(__LINK_STATE_START, &dev->state);
1152 if (ops->ndo_validate_addr)
1153 ret = ops->ndo_validate_addr(dev);
1155 if (!ret && ops->ndo_open)
1156 ret = ops->ndo_open(dev);
1159 * If it went open OK then:
1163 clear_bit(__LINK_STATE_START, &dev->state);
1168 dev->flags |= IFF_UP;
1173 net_dmaengine_get();
1176 * Initialize multicasting status
1178 dev_set_rx_mode(dev);
1181 * Wakeup transmit queue engine
1190 * dev_open - prepare an interface for use.
1191 * @dev: device to open
1193 * Takes a device from down to up state. The device's private open
1194 * function is invoked and then the multicast lists are loaded. Finally
1195 * the device is moved into the up state and a %NETDEV_UP message is
1196 * sent to the netdev notifier chain.
1198 * Calling this function on an active interface is a nop. On a failure
1199 * a negative errno code is returned.
1201 int dev_open(struct net_device *dev)
1208 if (dev->flags & IFF_UP)
1214 ret = __dev_open(dev);
1219 * ... and announce new interface.
1221 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1222 call_netdevice_notifiers(NETDEV_UP, dev);
1226 EXPORT_SYMBOL(dev_open);
1228 static int __dev_close(struct net_device *dev)
1230 const struct net_device_ops *ops = dev->netdev_ops;
1236 * Tell people we are going down, so that they can
1237 * prepare to death, when device is still operating.
1239 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1241 clear_bit(__LINK_STATE_START, &dev->state);
1243 /* Synchronize to scheduled poll. We cannot touch poll list,
1244 * it can be even on different cpu. So just clear netif_running().
1246 * dev->stop() will invoke napi_disable() on all of it's
1247 * napi_struct instances on this device.
1249 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1251 dev_deactivate(dev);
1254 * Call the device specific close. This cannot fail.
1255 * Only if device is UP
1257 * We allow it to be called even after a DETACH hot-plug
1264 * Device is now down.
1267 dev->flags &= ~IFF_UP;
1272 net_dmaengine_put();
1278 * dev_close - shutdown an interface.
1279 * @dev: device to shutdown
1281 * This function moves an active device into down state. A
1282 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1283 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1286 int dev_close(struct net_device *dev)
1288 if (!(dev->flags & IFF_UP))
1294 * Tell people we are down
1296 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1297 call_netdevice_notifiers(NETDEV_DOWN, dev);
1301 EXPORT_SYMBOL(dev_close);
1305 * dev_disable_lro - disable Large Receive Offload on a device
1308 * Disable Large Receive Offload (LRO) on a net device. Must be
1309 * called under RTNL. This is needed if received packets may be
1310 * forwarded to another interface.
1312 void dev_disable_lro(struct net_device *dev)
1314 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1315 dev->ethtool_ops->set_flags) {
1316 u32 flags = dev->ethtool_ops->get_flags(dev);
1317 if (flags & ETH_FLAG_LRO) {
1318 flags &= ~ETH_FLAG_LRO;
1319 dev->ethtool_ops->set_flags(dev, flags);
1322 WARN_ON(dev->features & NETIF_F_LRO);
1324 EXPORT_SYMBOL(dev_disable_lro);
1327 static int dev_boot_phase = 1;
1330 * Device change register/unregister. These are not inline or static
1331 * as we export them to the world.
1335 * register_netdevice_notifier - register a network notifier block
1338 * Register a notifier to be called when network device events occur.
1339 * The notifier passed is linked into the kernel structures and must
1340 * not be reused until it has been unregistered. A negative errno code
1341 * is returned on a failure.
1343 * When registered all registration and up events are replayed
1344 * to the new notifier to allow device to have a race free
1345 * view of the network device list.
1348 int register_netdevice_notifier(struct notifier_block *nb)
1350 struct net_device *dev;
1351 struct net_device *last;
1356 err = raw_notifier_chain_register(&netdev_chain, nb);
1362 for_each_netdev(net, dev) {
1363 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1364 err = notifier_to_errno(err);
1368 if (!(dev->flags & IFF_UP))
1371 nb->notifier_call(nb, NETDEV_UP, dev);
1382 for_each_netdev(net, dev) {
1386 if (dev->flags & IFF_UP) {
1387 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1388 nb->notifier_call(nb, NETDEV_DOWN, dev);
1390 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1391 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1395 raw_notifier_chain_unregister(&netdev_chain, nb);
1398 EXPORT_SYMBOL(register_netdevice_notifier);
1401 * unregister_netdevice_notifier - unregister a network notifier block
1404 * Unregister a notifier previously registered by
1405 * register_netdevice_notifier(). The notifier is unlinked into the
1406 * kernel structures and may then be reused. A negative errno code
1407 * is returned on a failure.
1410 int unregister_netdevice_notifier(struct notifier_block *nb)
1415 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1419 EXPORT_SYMBOL(unregister_netdevice_notifier);
1422 * call_netdevice_notifiers - call all network notifier blocks
1423 * @val: value passed unmodified to notifier function
1424 * @dev: net_device pointer passed unmodified to notifier function
1426 * Call all network notifier blocks. Parameters and return value
1427 * are as for raw_notifier_call_chain().
1430 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1433 return raw_notifier_call_chain(&netdev_chain, val, dev);
1436 /* When > 0 there are consumers of rx skb time stamps */
1437 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1439 void net_enable_timestamp(void)
1441 atomic_inc(&netstamp_needed);
1443 EXPORT_SYMBOL(net_enable_timestamp);
1445 void net_disable_timestamp(void)
1447 atomic_dec(&netstamp_needed);
1449 EXPORT_SYMBOL(net_disable_timestamp);
1451 static inline void net_timestamp_set(struct sk_buff *skb)
1453 if (atomic_read(&netstamp_needed))
1454 __net_timestamp(skb);
1456 skb->tstamp.tv64 = 0;
1459 static inline void net_timestamp_check(struct sk_buff *skb)
1461 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1462 __net_timestamp(skb);
1466 * dev_forward_skb - loopback an skb to another netif
1468 * @dev: destination network device
1469 * @skb: buffer to forward
1472 * NET_RX_SUCCESS (no congestion)
1473 * NET_RX_DROP (packet was dropped, but freed)
1475 * dev_forward_skb can be used for injecting an skb from the
1476 * start_xmit function of one device into the receive queue
1477 * of another device.
1479 * The receiving device may be in another namespace, so
1480 * we have to clear all information in the skb that could
1481 * impact namespace isolation.
1483 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1488 if (unlikely(!(dev->flags & IFF_UP) ||
1489 (skb->len > (dev->mtu + dev->hard_header_len + VLAN_HLEN)))) {
1490 atomic_long_inc(&dev->rx_dropped);
1494 skb_set_dev(skb, dev);
1495 skb->tstamp.tv64 = 0;
1496 skb->pkt_type = PACKET_HOST;
1497 skb->protocol = eth_type_trans(skb, dev);
1498 return netif_rx(skb);
1500 EXPORT_SYMBOL_GPL(dev_forward_skb);
1503 * Support routine. Sends outgoing frames to any network
1504 * taps currently in use.
1507 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1509 struct packet_type *ptype;
1511 #ifdef CONFIG_NET_CLS_ACT
1512 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1513 net_timestamp_set(skb);
1515 net_timestamp_set(skb);
1519 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1520 /* Never send packets back to the socket
1521 * they originated from - MvS (miquels@drinkel.ow.org)
1523 if ((ptype->dev == dev || !ptype->dev) &&
1524 (ptype->af_packet_priv == NULL ||
1525 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1526 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1530 /* skb->nh should be correctly
1531 set by sender, so that the second statement is
1532 just protection against buggy protocols.
1534 skb_reset_mac_header(skb2);
1536 if (skb_network_header(skb2) < skb2->data ||
1537 skb2->network_header > skb2->tail) {
1538 if (net_ratelimit())
1539 printk(KERN_CRIT "protocol %04x is "
1541 ntohs(skb2->protocol),
1543 skb_reset_network_header(skb2);
1546 skb2->transport_header = skb2->network_header;
1547 skb2->pkt_type = PACKET_OUTGOING;
1548 ptype->func(skb2, skb->dev, ptype, skb->dev);
1555 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1556 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1558 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1560 if (txq < 1 || txq > dev->num_tx_queues)
1563 if (dev->reg_state == NETREG_REGISTERED) {
1566 if (txq < dev->real_num_tx_queues)
1567 qdisc_reset_all_tx_gt(dev, txq);
1570 dev->real_num_tx_queues = txq;
1573 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1577 * netif_set_real_num_rx_queues - set actual number of RX queues used
1578 * @dev: Network device
1579 * @rxq: Actual number of RX queues
1581 * This must be called either with the rtnl_lock held or before
1582 * registration of the net device. Returns 0 on success, or a
1583 * negative error code. If called before registration, it always
1586 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1590 if (rxq < 1 || rxq > dev->num_rx_queues)
1593 if (dev->reg_state == NETREG_REGISTERED) {
1596 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1602 dev->real_num_rx_queues = rxq;
1605 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1608 static inline void __netif_reschedule(struct Qdisc *q)
1610 struct softnet_data *sd;
1611 unsigned long flags;
1613 local_irq_save(flags);
1614 sd = &__get_cpu_var(softnet_data);
1615 q->next_sched = NULL;
1616 *sd->output_queue_tailp = q;
1617 sd->output_queue_tailp = &q->next_sched;
1618 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1619 local_irq_restore(flags);
1622 void __netif_schedule(struct Qdisc *q)
1624 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1625 __netif_reschedule(q);
1627 EXPORT_SYMBOL(__netif_schedule);
1629 void dev_kfree_skb_irq(struct sk_buff *skb)
1631 if (atomic_dec_and_test(&skb->users)) {
1632 struct softnet_data *sd;
1633 unsigned long flags;
1635 local_irq_save(flags);
1636 sd = &__get_cpu_var(softnet_data);
1637 skb->next = sd->completion_queue;
1638 sd->completion_queue = skb;
1639 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1640 local_irq_restore(flags);
1643 EXPORT_SYMBOL(dev_kfree_skb_irq);
1645 void dev_kfree_skb_any(struct sk_buff *skb)
1647 if (in_irq() || irqs_disabled())
1648 dev_kfree_skb_irq(skb);
1652 EXPORT_SYMBOL(dev_kfree_skb_any);
1656 * netif_device_detach - mark device as removed
1657 * @dev: network device
1659 * Mark device as removed from system and therefore no longer available.
1661 void netif_device_detach(struct net_device *dev)
1663 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1664 netif_running(dev)) {
1665 netif_tx_stop_all_queues(dev);
1668 EXPORT_SYMBOL(netif_device_detach);
1671 * netif_device_attach - mark device as attached
1672 * @dev: network device
1674 * Mark device as attached from system and restart if needed.
1676 void netif_device_attach(struct net_device *dev)
1678 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1679 netif_running(dev)) {
1680 netif_tx_wake_all_queues(dev);
1681 __netdev_watchdog_up(dev);
1684 EXPORT_SYMBOL(netif_device_attach);
1686 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1688 return ((features & NETIF_F_NO_CSUM) ||
1689 ((features & NETIF_F_V4_CSUM) &&
1690 protocol == htons(ETH_P_IP)) ||
1691 ((features & NETIF_F_V6_CSUM) &&
1692 protocol == htons(ETH_P_IPV6)) ||
1693 ((features & NETIF_F_FCOE_CRC) &&
1694 protocol == htons(ETH_P_FCOE)));
1697 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1699 __be16 protocol = skb->protocol;
1700 int features = dev->features;
1702 if (vlan_tx_tag_present(skb)) {
1703 features &= dev->vlan_features;
1704 } else if (protocol == htons(ETH_P_8021Q)) {
1705 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1706 protocol = veh->h_vlan_encapsulated_proto;
1707 features &= dev->vlan_features;
1710 return can_checksum_protocol(features, protocol);
1714 * skb_dev_set -- assign a new device to a buffer
1715 * @skb: buffer for the new device
1716 * @dev: network device
1718 * If an skb is owned by a device already, we have to reset
1719 * all data private to the namespace a device belongs to
1720 * before assigning it a new device.
1722 #ifdef CONFIG_NET_NS
1723 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1726 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1729 skb_init_secmark(skb);
1733 skb->ipvs_property = 0;
1734 #ifdef CONFIG_NET_SCHED
1740 EXPORT_SYMBOL(skb_set_dev);
1741 #endif /* CONFIG_NET_NS */
1744 * Invalidate hardware checksum when packet is to be mangled, and
1745 * complete checksum manually on outgoing path.
1747 int skb_checksum_help(struct sk_buff *skb)
1750 int ret = 0, offset;
1752 if (skb->ip_summed == CHECKSUM_COMPLETE)
1753 goto out_set_summed;
1755 if (unlikely(skb_shinfo(skb)->gso_size)) {
1756 /* Let GSO fix up the checksum. */
1757 goto out_set_summed;
1760 offset = skb->csum_start - skb_headroom(skb);
1761 BUG_ON(offset >= skb_headlen(skb));
1762 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1764 offset += skb->csum_offset;
1765 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1767 if (skb_cloned(skb) &&
1768 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1769 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1774 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1776 skb->ip_summed = CHECKSUM_NONE;
1780 EXPORT_SYMBOL(skb_checksum_help);
1783 * skb_gso_segment - Perform segmentation on skb.
1784 * @skb: buffer to segment
1785 * @features: features for the output path (see dev->features)
1787 * This function segments the given skb and returns a list of segments.
1789 * It may return NULL if the skb requires no segmentation. This is
1790 * only possible when GSO is used for verifying header integrity.
1792 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1794 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1795 struct packet_type *ptype;
1796 __be16 type = skb->protocol;
1797 int vlan_depth = ETH_HLEN;
1800 while (type == htons(ETH_P_8021Q)) {
1801 struct vlan_hdr *vh;
1803 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1804 return ERR_PTR(-EINVAL);
1806 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1807 type = vh->h_vlan_encapsulated_proto;
1808 vlan_depth += VLAN_HLEN;
1811 skb_reset_mac_header(skb);
1812 skb->mac_len = skb->network_header - skb->mac_header;
1813 __skb_pull(skb, skb->mac_len);
1815 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1816 struct net_device *dev = skb->dev;
1817 struct ethtool_drvinfo info = {};
1819 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1820 dev->ethtool_ops->get_drvinfo(dev, &info);
1822 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1823 info.driver, dev ? dev->features : 0L,
1824 skb->sk ? skb->sk->sk_route_caps : 0L,
1825 skb->len, skb->data_len, skb->ip_summed);
1827 if (skb_header_cloned(skb) &&
1828 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1829 return ERR_PTR(err);
1833 list_for_each_entry_rcu(ptype,
1834 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1835 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1836 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1837 err = ptype->gso_send_check(skb);
1838 segs = ERR_PTR(err);
1839 if (err || skb_gso_ok(skb, features))
1841 __skb_push(skb, (skb->data -
1842 skb_network_header(skb)));
1844 segs = ptype->gso_segment(skb, features);
1850 __skb_push(skb, skb->data - skb_mac_header(skb));
1854 EXPORT_SYMBOL(skb_gso_segment);
1856 /* Take action when hardware reception checksum errors are detected. */
1858 void netdev_rx_csum_fault(struct net_device *dev)
1860 if (net_ratelimit()) {
1861 printk(KERN_ERR "%s: hw csum failure.\n",
1862 dev ? dev->name : "<unknown>");
1866 EXPORT_SYMBOL(netdev_rx_csum_fault);
1869 /* Actually, we should eliminate this check as soon as we know, that:
1870 * 1. IOMMU is present and allows to map all the memory.
1871 * 2. No high memory really exists on this machine.
1874 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1876 #ifdef CONFIG_HIGHMEM
1878 if (!(dev->features & NETIF_F_HIGHDMA)) {
1879 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1880 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1884 if (PCI_DMA_BUS_IS_PHYS) {
1885 struct device *pdev = dev->dev.parent;
1889 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1890 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1891 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1900 void (*destructor)(struct sk_buff *skb);
1903 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1905 static void dev_gso_skb_destructor(struct sk_buff *skb)
1907 struct dev_gso_cb *cb;
1910 struct sk_buff *nskb = skb->next;
1912 skb->next = nskb->next;
1915 } while (skb->next);
1917 cb = DEV_GSO_CB(skb);
1919 cb->destructor(skb);
1923 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1924 * @skb: buffer to segment
1926 * This function segments the given skb and stores the list of segments
1929 static int dev_gso_segment(struct sk_buff *skb)
1931 struct net_device *dev = skb->dev;
1932 struct sk_buff *segs;
1933 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1936 segs = skb_gso_segment(skb, features);
1938 /* Verifying header integrity only. */
1943 return PTR_ERR(segs);
1946 DEV_GSO_CB(skb)->destructor = skb->destructor;
1947 skb->destructor = dev_gso_skb_destructor;
1953 * Try to orphan skb early, right before transmission by the device.
1954 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1955 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1957 static inline void skb_orphan_try(struct sk_buff *skb)
1959 struct sock *sk = skb->sk;
1961 if (sk && !skb_shinfo(skb)->tx_flags) {
1962 /* skb_tx_hash() wont be able to get sk.
1963 * We copy sk_hash into skb->rxhash
1966 skb->rxhash = sk->sk_hash;
1971 int netif_get_vlan_features(struct sk_buff *skb, struct net_device *dev)
1973 __be16 protocol = skb->protocol;
1975 if (protocol == htons(ETH_P_8021Q)) {
1976 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1977 protocol = veh->h_vlan_encapsulated_proto;
1978 } else if (!skb->vlan_tci)
1979 return dev->features;
1981 if (protocol != htons(ETH_P_8021Q))
1982 return dev->features & dev->vlan_features;
1986 EXPORT_SYMBOL(netif_get_vlan_features);
1989 * Returns true if either:
1990 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1991 * 2. skb is fragmented and the device does not support SG, or if
1992 * at least one of fragments is in highmem and device does not
1993 * support DMA from it.
1995 static inline int skb_needs_linearize(struct sk_buff *skb,
1996 struct net_device *dev)
1998 if (skb_is_nonlinear(skb)) {
1999 int features = dev->features;
2001 if (vlan_tx_tag_present(skb))
2002 features &= dev->vlan_features;
2004 return (skb_has_frag_list(skb) &&
2005 !(features & NETIF_F_FRAGLIST)) ||
2006 (skb_shinfo(skb)->nr_frags &&
2007 (!(features & NETIF_F_SG) ||
2008 illegal_highdma(dev, skb)));
2014 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2015 struct netdev_queue *txq)
2017 const struct net_device_ops *ops = dev->netdev_ops;
2018 int rc = NETDEV_TX_OK;
2020 if (likely(!skb->next)) {
2021 if (!list_empty(&ptype_all))
2022 dev_queue_xmit_nit(skb, dev);
2025 * If device doesnt need skb->dst, release it right now while
2026 * its hot in this cpu cache
2028 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2031 skb_orphan_try(skb);
2033 if (vlan_tx_tag_present(skb) &&
2034 !(dev->features & NETIF_F_HW_VLAN_TX)) {
2035 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2042 if (netif_needs_gso(dev, skb)) {
2043 if (unlikely(dev_gso_segment(skb)))
2048 if (skb_needs_linearize(skb, dev) &&
2049 __skb_linearize(skb))
2052 /* If packet is not checksummed and device does not
2053 * support checksumming for this protocol, complete
2054 * checksumming here.
2056 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2057 skb_set_transport_header(skb, skb->csum_start -
2059 if (!dev_can_checksum(dev, skb) &&
2060 skb_checksum_help(skb))
2065 rc = ops->ndo_start_xmit(skb, dev);
2066 trace_net_dev_xmit(skb, rc);
2067 if (rc == NETDEV_TX_OK)
2068 txq_trans_update(txq);
2074 struct sk_buff *nskb = skb->next;
2076 skb->next = nskb->next;
2080 * If device doesnt need nskb->dst, release it right now while
2081 * its hot in this cpu cache
2083 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2086 rc = ops->ndo_start_xmit(nskb, dev);
2087 trace_net_dev_xmit(nskb, rc);
2088 if (unlikely(rc != NETDEV_TX_OK)) {
2089 if (rc & ~NETDEV_TX_MASK)
2090 goto out_kfree_gso_skb;
2091 nskb->next = skb->next;
2095 txq_trans_update(txq);
2096 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2097 return NETDEV_TX_BUSY;
2098 } while (skb->next);
2101 if (likely(skb->next == NULL))
2102 skb->destructor = DEV_GSO_CB(skb)->destructor;
2109 static u32 hashrnd __read_mostly;
2111 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
2115 if (skb_rx_queue_recorded(skb)) {
2116 hash = skb_get_rx_queue(skb);
2117 while (unlikely(hash >= dev->real_num_tx_queues))
2118 hash -= dev->real_num_tx_queues;
2122 if (skb->sk && skb->sk->sk_hash)
2123 hash = skb->sk->sk_hash;
2125 hash = (__force u16) skb->protocol ^ skb->rxhash;
2126 hash = jhash_1word(hash, hashrnd);
2128 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2130 EXPORT_SYMBOL(skb_tx_hash);
2132 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2134 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2135 if (net_ratelimit()) {
2136 pr_warning("%s selects TX queue %d, but "
2137 "real number of TX queues is %d\n",
2138 dev->name, queue_index, dev->real_num_tx_queues);
2145 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2146 struct sk_buff *skb)
2149 const struct net_device_ops *ops = dev->netdev_ops;
2151 if (ops->ndo_select_queue) {
2152 queue_index = ops->ndo_select_queue(dev, skb);
2153 queue_index = dev_cap_txqueue(dev, queue_index);
2155 struct sock *sk = skb->sk;
2156 queue_index = sk_tx_queue_get(sk);
2157 if (queue_index < 0 || queue_index >= dev->real_num_tx_queues) {
2160 if (dev->real_num_tx_queues > 1)
2161 queue_index = skb_tx_hash(dev, skb);
2164 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2166 if (dst && skb_dst(skb) == dst)
2167 sk_tx_queue_set(sk, queue_index);
2172 skb_set_queue_mapping(skb, queue_index);
2173 return netdev_get_tx_queue(dev, queue_index);
2176 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2177 struct net_device *dev,
2178 struct netdev_queue *txq)
2180 spinlock_t *root_lock = qdisc_lock(q);
2181 bool contended = qdisc_is_running(q);
2185 * Heuristic to force contended enqueues to serialize on a
2186 * separate lock before trying to get qdisc main lock.
2187 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2188 * and dequeue packets faster.
2190 if (unlikely(contended))
2191 spin_lock(&q->busylock);
2193 spin_lock(root_lock);
2194 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2197 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2198 qdisc_run_begin(q)) {
2200 * This is a work-conserving queue; there are no old skbs
2201 * waiting to be sent out; and the qdisc is not running -
2202 * xmit the skb directly.
2204 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2206 __qdisc_update_bstats(q, skb->len);
2207 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2208 if (unlikely(contended)) {
2209 spin_unlock(&q->busylock);
2216 rc = NET_XMIT_SUCCESS;
2219 rc = qdisc_enqueue_root(skb, q);
2220 if (qdisc_run_begin(q)) {
2221 if (unlikely(contended)) {
2222 spin_unlock(&q->busylock);
2228 spin_unlock(root_lock);
2229 if (unlikely(contended))
2230 spin_unlock(&q->busylock);
2234 static DEFINE_PER_CPU(int, xmit_recursion);
2235 #define RECURSION_LIMIT 10
2238 * dev_queue_xmit - transmit a buffer
2239 * @skb: buffer to transmit
2241 * Queue a buffer for transmission to a network device. The caller must
2242 * have set the device and priority and built the buffer before calling
2243 * this function. The function can be called from an interrupt.
2245 * A negative errno code is returned on a failure. A success does not
2246 * guarantee the frame will be transmitted as it may be dropped due
2247 * to congestion or traffic shaping.
2249 * -----------------------------------------------------------------------------------
2250 * I notice this method can also return errors from the queue disciplines,
2251 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2254 * Regardless of the return value, the skb is consumed, so it is currently
2255 * difficult to retry a send to this method. (You can bump the ref count
2256 * before sending to hold a reference for retry if you are careful.)
2258 * When calling this method, interrupts MUST be enabled. This is because
2259 * the BH enable code must have IRQs enabled so that it will not deadlock.
2262 int dev_queue_xmit(struct sk_buff *skb)
2264 struct net_device *dev = skb->dev;
2265 struct netdev_queue *txq;
2269 /* Disable soft irqs for various locks below. Also
2270 * stops preemption for RCU.
2274 txq = dev_pick_tx(dev, skb);
2275 q = rcu_dereference_bh(txq->qdisc);
2277 #ifdef CONFIG_NET_CLS_ACT
2278 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2280 trace_net_dev_queue(skb);
2282 rc = __dev_xmit_skb(skb, q, dev, txq);
2286 /* The device has no queue. Common case for software devices:
2287 loopback, all the sorts of tunnels...
2289 Really, it is unlikely that netif_tx_lock protection is necessary
2290 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2292 However, it is possible, that they rely on protection
2295 Check this and shot the lock. It is not prone from deadlocks.
2296 Either shot noqueue qdisc, it is even simpler 8)
2298 if (dev->flags & IFF_UP) {
2299 int cpu = smp_processor_id(); /* ok because BHs are off */
2301 if (txq->xmit_lock_owner != cpu) {
2303 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2304 goto recursion_alert;
2306 HARD_TX_LOCK(dev, txq, cpu);
2308 if (!netif_tx_queue_stopped(txq)) {
2309 __this_cpu_inc(xmit_recursion);
2310 rc = dev_hard_start_xmit(skb, dev, txq);
2311 __this_cpu_dec(xmit_recursion);
2312 if (dev_xmit_complete(rc)) {
2313 HARD_TX_UNLOCK(dev, txq);
2317 HARD_TX_UNLOCK(dev, txq);
2318 if (net_ratelimit())
2319 printk(KERN_CRIT "Virtual device %s asks to "
2320 "queue packet!\n", dev->name);
2322 /* Recursion is detected! It is possible,
2326 if (net_ratelimit())
2327 printk(KERN_CRIT "Dead loop on virtual device "
2328 "%s, fix it urgently!\n", dev->name);
2333 rcu_read_unlock_bh();
2338 rcu_read_unlock_bh();
2341 EXPORT_SYMBOL(dev_queue_xmit);
2344 /*=======================================================================
2346 =======================================================================*/
2348 int netdev_max_backlog __read_mostly = 1000;
2349 int netdev_tstamp_prequeue __read_mostly = 1;
2350 int netdev_budget __read_mostly = 300;
2351 int weight_p __read_mostly = 64; /* old backlog weight */
2353 /* Called with irq disabled */
2354 static inline void ____napi_schedule(struct softnet_data *sd,
2355 struct napi_struct *napi)
2357 list_add_tail(&napi->poll_list, &sd->poll_list);
2358 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2362 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2363 * and src/dst port numbers. Returns a non-zero hash number on success
2366 __u32 __skb_get_rxhash(struct sk_buff *skb)
2368 int nhoff, hash = 0, poff;
2369 struct ipv6hdr *ip6;
2372 u32 addr1, addr2, ihl;
2378 nhoff = skb_network_offset(skb);
2380 switch (skb->protocol) {
2381 case __constant_htons(ETH_P_IP):
2382 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2385 ip = (struct iphdr *) (skb->data + nhoff);
2386 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2389 ip_proto = ip->protocol;
2390 addr1 = (__force u32) ip->saddr;
2391 addr2 = (__force u32) ip->daddr;
2394 case __constant_htons(ETH_P_IPV6):
2395 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2398 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2399 ip_proto = ip6->nexthdr;
2400 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2401 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2409 poff = proto_ports_offset(ip_proto);
2411 nhoff += ihl * 4 + poff;
2412 if (pskb_may_pull(skb, nhoff + 4)) {
2413 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2414 if (ports.v16[1] < ports.v16[0])
2415 swap(ports.v16[0], ports.v16[1]);
2419 /* get a consistent hash (same value on both flow directions) */
2423 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2430 EXPORT_SYMBOL(__skb_get_rxhash);
2434 /* One global table that all flow-based protocols share. */
2435 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2436 EXPORT_SYMBOL(rps_sock_flow_table);
2439 * get_rps_cpu is called from netif_receive_skb and returns the target
2440 * CPU from the RPS map of the receiving queue for a given skb.
2441 * rcu_read_lock must be held on entry.
2443 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2444 struct rps_dev_flow **rflowp)
2446 struct netdev_rx_queue *rxqueue;
2447 struct rps_map *map;
2448 struct rps_dev_flow_table *flow_table;
2449 struct rps_sock_flow_table *sock_flow_table;
2453 if (skb_rx_queue_recorded(skb)) {
2454 u16 index = skb_get_rx_queue(skb);
2455 if (unlikely(index >= dev->real_num_rx_queues)) {
2456 WARN_ONCE(dev->real_num_rx_queues > 1,
2457 "%s received packet on queue %u, but number "
2458 "of RX queues is %u\n",
2459 dev->name, index, dev->real_num_rx_queues);
2462 rxqueue = dev->_rx + index;
2466 map = rcu_dereference(rxqueue->rps_map);
2468 if (map->len == 1) {
2469 tcpu = map->cpus[0];
2470 if (cpu_online(tcpu))
2474 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2478 skb_reset_network_header(skb);
2479 if (!skb_get_rxhash(skb))
2482 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2483 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2484 if (flow_table && sock_flow_table) {
2486 struct rps_dev_flow *rflow;
2488 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2491 next_cpu = sock_flow_table->ents[skb->rxhash &
2492 sock_flow_table->mask];
2495 * If the desired CPU (where last recvmsg was done) is
2496 * different from current CPU (one in the rx-queue flow
2497 * table entry), switch if one of the following holds:
2498 * - Current CPU is unset (equal to RPS_NO_CPU).
2499 * - Current CPU is offline.
2500 * - The current CPU's queue tail has advanced beyond the
2501 * last packet that was enqueued using this table entry.
2502 * This guarantees that all previous packets for the flow
2503 * have been dequeued, thus preserving in order delivery.
2505 if (unlikely(tcpu != next_cpu) &&
2506 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2507 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2508 rflow->last_qtail)) >= 0)) {
2509 tcpu = rflow->cpu = next_cpu;
2510 if (tcpu != RPS_NO_CPU)
2511 rflow->last_qtail = per_cpu(softnet_data,
2512 tcpu).input_queue_head;
2514 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2522 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2524 if (cpu_online(tcpu)) {
2534 /* Called from hardirq (IPI) context */
2535 static void rps_trigger_softirq(void *data)
2537 struct softnet_data *sd = data;
2539 ____napi_schedule(sd, &sd->backlog);
2543 #endif /* CONFIG_RPS */
2546 * Check if this softnet_data structure is another cpu one
2547 * If yes, queue it to our IPI list and return 1
2550 static int rps_ipi_queued(struct softnet_data *sd)
2553 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2556 sd->rps_ipi_next = mysd->rps_ipi_list;
2557 mysd->rps_ipi_list = sd;
2559 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2562 #endif /* CONFIG_RPS */
2567 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2568 * queue (may be a remote CPU queue).
2570 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2571 unsigned int *qtail)
2573 struct softnet_data *sd;
2574 unsigned long flags;
2576 sd = &per_cpu(softnet_data, cpu);
2578 local_irq_save(flags);
2581 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2582 if (skb_queue_len(&sd->input_pkt_queue)) {
2584 __skb_queue_tail(&sd->input_pkt_queue, skb);
2585 input_queue_tail_incr_save(sd, qtail);
2587 local_irq_restore(flags);
2588 return NET_RX_SUCCESS;
2591 /* Schedule NAPI for backlog device
2592 * We can use non atomic operation since we own the queue lock
2594 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2595 if (!rps_ipi_queued(sd))
2596 ____napi_schedule(sd, &sd->backlog);
2604 local_irq_restore(flags);
2606 atomic_long_inc(&skb->dev->rx_dropped);
2612 * netif_rx - post buffer to the network code
2613 * @skb: buffer to post
2615 * This function receives a packet from a device driver and queues it for
2616 * the upper (protocol) levels to process. It always succeeds. The buffer
2617 * may be dropped during processing for congestion control or by the
2621 * NET_RX_SUCCESS (no congestion)
2622 * NET_RX_DROP (packet was dropped)
2626 int netif_rx(struct sk_buff *skb)
2630 /* if netpoll wants it, pretend we never saw it */
2631 if (netpoll_rx(skb))
2634 if (netdev_tstamp_prequeue)
2635 net_timestamp_check(skb);
2637 trace_netif_rx(skb);
2640 struct rps_dev_flow voidflow, *rflow = &voidflow;
2646 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2648 cpu = smp_processor_id();
2650 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2658 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2664 EXPORT_SYMBOL(netif_rx);
2666 int netif_rx_ni(struct sk_buff *skb)
2671 err = netif_rx(skb);
2672 if (local_softirq_pending())
2678 EXPORT_SYMBOL(netif_rx_ni);
2680 static void net_tx_action(struct softirq_action *h)
2682 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2684 if (sd->completion_queue) {
2685 struct sk_buff *clist;
2687 local_irq_disable();
2688 clist = sd->completion_queue;
2689 sd->completion_queue = NULL;
2693 struct sk_buff *skb = clist;
2694 clist = clist->next;
2696 WARN_ON(atomic_read(&skb->users));
2697 trace_kfree_skb(skb, net_tx_action);
2702 if (sd->output_queue) {
2705 local_irq_disable();
2706 head = sd->output_queue;
2707 sd->output_queue = NULL;
2708 sd->output_queue_tailp = &sd->output_queue;
2712 struct Qdisc *q = head;
2713 spinlock_t *root_lock;
2715 head = head->next_sched;
2717 root_lock = qdisc_lock(q);
2718 if (spin_trylock(root_lock)) {
2719 smp_mb__before_clear_bit();
2720 clear_bit(__QDISC_STATE_SCHED,
2723 spin_unlock(root_lock);
2725 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2727 __netif_reschedule(q);
2729 smp_mb__before_clear_bit();
2730 clear_bit(__QDISC_STATE_SCHED,
2738 static inline int deliver_skb(struct sk_buff *skb,
2739 struct packet_type *pt_prev,
2740 struct net_device *orig_dev)
2742 atomic_inc(&skb->users);
2743 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2746 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2747 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2748 /* This hook is defined here for ATM LANE */
2749 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2750 unsigned char *addr) __read_mostly;
2751 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2754 #ifdef CONFIG_NET_CLS_ACT
2755 /* TODO: Maybe we should just force sch_ingress to be compiled in
2756 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2757 * a compare and 2 stores extra right now if we dont have it on
2758 * but have CONFIG_NET_CLS_ACT
2759 * NOTE: This doesnt stop any functionality; if you dont have
2760 * the ingress scheduler, you just cant add policies on ingress.
2763 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2765 struct net_device *dev = skb->dev;
2766 u32 ttl = G_TC_RTTL(skb->tc_verd);
2767 int result = TC_ACT_OK;
2770 if (unlikely(MAX_RED_LOOP < ttl++)) {
2771 if (net_ratelimit())
2772 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2773 skb->skb_iif, dev->ifindex);
2777 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2778 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2781 if (q != &noop_qdisc) {
2782 spin_lock(qdisc_lock(q));
2783 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2784 result = qdisc_enqueue_root(skb, q);
2785 spin_unlock(qdisc_lock(q));
2791 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2792 struct packet_type **pt_prev,
2793 int *ret, struct net_device *orig_dev)
2795 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
2797 if (!rxq || rxq->qdisc == &noop_qdisc)
2801 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2805 switch (ing_filter(skb, rxq)) {
2819 * netdev_rx_handler_register - register receive handler
2820 * @dev: device to register a handler for
2821 * @rx_handler: receive handler to register
2822 * @rx_handler_data: data pointer that is used by rx handler
2824 * Register a receive hander for a device. This handler will then be
2825 * called from __netif_receive_skb. A negative errno code is returned
2828 * The caller must hold the rtnl_mutex.
2830 int netdev_rx_handler_register(struct net_device *dev,
2831 rx_handler_func_t *rx_handler,
2832 void *rx_handler_data)
2836 if (dev->rx_handler)
2839 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2840 rcu_assign_pointer(dev->rx_handler, rx_handler);
2844 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2847 * netdev_rx_handler_unregister - unregister receive handler
2848 * @dev: device to unregister a handler from
2850 * Unregister a receive hander from a device.
2852 * The caller must hold the rtnl_mutex.
2854 void netdev_rx_handler_unregister(struct net_device *dev)
2858 rcu_assign_pointer(dev->rx_handler, NULL);
2859 rcu_assign_pointer(dev->rx_handler_data, NULL);
2861 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2863 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2864 struct net_device *master)
2866 if (skb->pkt_type == PACKET_HOST) {
2867 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2869 memcpy(dest, master->dev_addr, ETH_ALEN);
2873 /* On bonding slaves other than the currently active slave, suppress
2874 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2875 * ARP on active-backup slaves with arp_validate enabled.
2877 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2879 struct net_device *dev = skb->dev;
2881 if (master->priv_flags & IFF_MASTER_ARPMON)
2882 dev->last_rx = jiffies;
2884 if ((master->priv_flags & IFF_MASTER_ALB) &&
2885 (master->priv_flags & IFF_BRIDGE_PORT)) {
2886 /* Do address unmangle. The local destination address
2887 * will be always the one master has. Provides the right
2888 * functionality in a bridge.
2890 skb_bond_set_mac_by_master(skb, master);
2893 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2894 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2895 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2898 if (master->priv_flags & IFF_MASTER_ALB) {
2899 if (skb->pkt_type != PACKET_BROADCAST &&
2900 skb->pkt_type != PACKET_MULTICAST)
2903 if (master->priv_flags & IFF_MASTER_8023AD &&
2904 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2911 EXPORT_SYMBOL(__skb_bond_should_drop);
2913 static int __netif_receive_skb(struct sk_buff *skb)
2915 struct packet_type *ptype, *pt_prev;
2916 rx_handler_func_t *rx_handler;
2917 struct net_device *orig_dev;
2918 struct net_device *master;
2919 struct net_device *null_or_orig;
2920 struct net_device *orig_or_bond;
2921 int ret = NET_RX_DROP;
2924 if (!netdev_tstamp_prequeue)
2925 net_timestamp_check(skb);
2927 trace_netif_receive_skb(skb);
2929 /* if we've gotten here through NAPI, check netpoll */
2930 if (netpoll_receive_skb(skb))
2934 skb->skb_iif = skb->dev->ifindex;
2937 * bonding note: skbs received on inactive slaves should only
2938 * be delivered to pkt handlers that are exact matches. Also
2939 * the deliver_no_wcard flag will be set. If packet handlers
2940 * are sensitive to duplicate packets these skbs will need to
2941 * be dropped at the handler.
2943 null_or_orig = NULL;
2944 orig_dev = skb->dev;
2945 master = ACCESS_ONCE(orig_dev->master);
2946 if (skb->deliver_no_wcard)
2947 null_or_orig = orig_dev;
2949 if (skb_bond_should_drop(skb, master)) {
2950 skb->deliver_no_wcard = 1;
2951 null_or_orig = orig_dev; /* deliver only exact match */
2956 __this_cpu_inc(softnet_data.processed);
2957 skb_reset_network_header(skb);
2958 skb_reset_transport_header(skb);
2959 skb->mac_len = skb->network_header - skb->mac_header;
2965 #ifdef CONFIG_NET_CLS_ACT
2966 if (skb->tc_verd & TC_NCLS) {
2967 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2972 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2973 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2974 ptype->dev == orig_dev) {
2976 ret = deliver_skb(skb, pt_prev, orig_dev);
2981 #ifdef CONFIG_NET_CLS_ACT
2982 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2988 /* Handle special case of bridge or macvlan */
2989 rx_handler = rcu_dereference(skb->dev->rx_handler);
2992 ret = deliver_skb(skb, pt_prev, orig_dev);
2995 skb = rx_handler(skb);
3000 if (vlan_tx_tag_present(skb)) {
3002 ret = deliver_skb(skb, pt_prev, orig_dev);
3005 if (vlan_hwaccel_do_receive(&skb)) {
3006 ret = __netif_receive_skb(skb);
3008 } else if (unlikely(!skb))
3013 * Make sure frames received on VLAN interfaces stacked on
3014 * bonding interfaces still make their way to any base bonding
3015 * device that may have registered for a specific ptype. The
3016 * handler may have to adjust skb->dev and orig_dev.
3018 orig_or_bond = orig_dev;
3019 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
3020 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
3021 orig_or_bond = vlan_dev_real_dev(skb->dev);
3024 type = skb->protocol;
3025 list_for_each_entry_rcu(ptype,
3026 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3027 if (ptype->type == type && (ptype->dev == null_or_orig ||
3028 ptype->dev == skb->dev || ptype->dev == orig_dev ||
3029 ptype->dev == orig_or_bond)) {
3031 ret = deliver_skb(skb, pt_prev, orig_dev);
3037 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3039 atomic_long_inc(&skb->dev->rx_dropped);
3041 /* Jamal, now you will not able to escape explaining
3042 * me how you were going to use this. :-)
3053 * netif_receive_skb - process receive buffer from network
3054 * @skb: buffer to process
3056 * netif_receive_skb() is the main receive data processing function.
3057 * It always succeeds. The buffer may be dropped during processing
3058 * for congestion control or by the protocol layers.
3060 * This function may only be called from softirq context and interrupts
3061 * should be enabled.
3063 * Return values (usually ignored):
3064 * NET_RX_SUCCESS: no congestion
3065 * NET_RX_DROP: packet was dropped
3067 int netif_receive_skb(struct sk_buff *skb)
3069 if (netdev_tstamp_prequeue)
3070 net_timestamp_check(skb);
3072 if (skb_defer_rx_timestamp(skb))
3073 return NET_RX_SUCCESS;
3077 struct rps_dev_flow voidflow, *rflow = &voidflow;
3082 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3085 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3089 ret = __netif_receive_skb(skb);
3095 return __netif_receive_skb(skb);
3098 EXPORT_SYMBOL(netif_receive_skb);
3100 /* Network device is going away, flush any packets still pending
3101 * Called with irqs disabled.
3103 static void flush_backlog(void *arg)
3105 struct net_device *dev = arg;
3106 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3107 struct sk_buff *skb, *tmp;
3110 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3111 if (skb->dev == dev) {
3112 __skb_unlink(skb, &sd->input_pkt_queue);
3114 input_queue_head_incr(sd);
3119 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3120 if (skb->dev == dev) {
3121 __skb_unlink(skb, &sd->process_queue);
3123 input_queue_head_incr(sd);
3128 static int napi_gro_complete(struct sk_buff *skb)
3130 struct packet_type *ptype;
3131 __be16 type = skb->protocol;
3132 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3135 if (NAPI_GRO_CB(skb)->count == 1) {
3136 skb_shinfo(skb)->gso_size = 0;
3141 list_for_each_entry_rcu(ptype, head, list) {
3142 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3145 err = ptype->gro_complete(skb);
3151 WARN_ON(&ptype->list == head);
3153 return NET_RX_SUCCESS;
3157 return netif_receive_skb(skb);
3160 inline void napi_gro_flush(struct napi_struct *napi)
3162 struct sk_buff *skb, *next;
3164 for (skb = napi->gro_list; skb; skb = next) {
3167 napi_gro_complete(skb);
3170 napi->gro_count = 0;
3171 napi->gro_list = NULL;
3173 EXPORT_SYMBOL(napi_gro_flush);
3175 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3177 struct sk_buff **pp = NULL;
3178 struct packet_type *ptype;
3179 __be16 type = skb->protocol;
3180 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3183 enum gro_result ret;
3185 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3188 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3192 list_for_each_entry_rcu(ptype, head, list) {
3193 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3196 skb_set_network_header(skb, skb_gro_offset(skb));
3197 mac_len = skb->network_header - skb->mac_header;
3198 skb->mac_len = mac_len;
3199 NAPI_GRO_CB(skb)->same_flow = 0;
3200 NAPI_GRO_CB(skb)->flush = 0;
3201 NAPI_GRO_CB(skb)->free = 0;
3203 pp = ptype->gro_receive(&napi->gro_list, skb);
3208 if (&ptype->list == head)
3211 same_flow = NAPI_GRO_CB(skb)->same_flow;
3212 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3215 struct sk_buff *nskb = *pp;
3219 napi_gro_complete(nskb);
3226 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3230 NAPI_GRO_CB(skb)->count = 1;
3231 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3232 skb->next = napi->gro_list;
3233 napi->gro_list = skb;
3237 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3238 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3240 BUG_ON(skb->end - skb->tail < grow);
3242 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3245 skb->data_len -= grow;
3247 skb_shinfo(skb)->frags[0].page_offset += grow;
3248 skb_shinfo(skb)->frags[0].size -= grow;
3250 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3251 put_page(skb_shinfo(skb)->frags[0].page);
3252 memmove(skb_shinfo(skb)->frags,
3253 skb_shinfo(skb)->frags + 1,
3254 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3265 EXPORT_SYMBOL(dev_gro_receive);
3267 static inline gro_result_t
3268 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3272 for (p = napi->gro_list; p; p = p->next) {
3273 unsigned long diffs;
3275 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3276 diffs |= p->vlan_tci ^ skb->vlan_tci;
3277 diffs |= compare_ether_header(skb_mac_header(p),
3278 skb_gro_mac_header(skb));
3279 NAPI_GRO_CB(p)->same_flow = !diffs;
3280 NAPI_GRO_CB(p)->flush = 0;
3283 return dev_gro_receive(napi, skb);
3286 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3290 if (netif_receive_skb(skb))
3295 case GRO_MERGED_FREE:
3306 EXPORT_SYMBOL(napi_skb_finish);
3308 void skb_gro_reset_offset(struct sk_buff *skb)
3310 NAPI_GRO_CB(skb)->data_offset = 0;
3311 NAPI_GRO_CB(skb)->frag0 = NULL;
3312 NAPI_GRO_CB(skb)->frag0_len = 0;
3314 if (skb->mac_header == skb->tail &&
3315 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3316 NAPI_GRO_CB(skb)->frag0 =
3317 page_address(skb_shinfo(skb)->frags[0].page) +
3318 skb_shinfo(skb)->frags[0].page_offset;
3319 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3322 EXPORT_SYMBOL(skb_gro_reset_offset);
3324 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3326 skb_gro_reset_offset(skb);
3328 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3330 EXPORT_SYMBOL(napi_gro_receive);
3332 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3334 __skb_pull(skb, skb_headlen(skb));
3335 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3341 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3343 struct sk_buff *skb = napi->skb;
3346 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3352 EXPORT_SYMBOL(napi_get_frags);
3354 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3360 skb->protocol = eth_type_trans(skb, skb->dev);
3362 if (ret == GRO_HELD)
3363 skb_gro_pull(skb, -ETH_HLEN);
3364 else if (netif_receive_skb(skb))
3369 case GRO_MERGED_FREE:
3370 napi_reuse_skb(napi, skb);
3379 EXPORT_SYMBOL(napi_frags_finish);
3381 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3383 struct sk_buff *skb = napi->skb;
3390 skb_reset_mac_header(skb);
3391 skb_gro_reset_offset(skb);
3393 off = skb_gro_offset(skb);
3394 hlen = off + sizeof(*eth);
3395 eth = skb_gro_header_fast(skb, off);
3396 if (skb_gro_header_hard(skb, hlen)) {
3397 eth = skb_gro_header_slow(skb, hlen, off);
3398 if (unlikely(!eth)) {
3399 napi_reuse_skb(napi, skb);
3405 skb_gro_pull(skb, sizeof(*eth));
3408 * This works because the only protocols we care about don't require
3409 * special handling. We'll fix it up properly at the end.
3411 skb->protocol = eth->h_proto;
3416 EXPORT_SYMBOL(napi_frags_skb);
3418 gro_result_t napi_gro_frags(struct napi_struct *napi)
3420 struct sk_buff *skb = napi_frags_skb(napi);
3425 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3427 EXPORT_SYMBOL(napi_gro_frags);
3430 * net_rps_action sends any pending IPI's for rps.
3431 * Note: called with local irq disabled, but exits with local irq enabled.
3433 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3436 struct softnet_data *remsd = sd->rps_ipi_list;
3439 sd->rps_ipi_list = NULL;
3443 /* Send pending IPI's to kick RPS processing on remote cpus. */
3445 struct softnet_data *next = remsd->rps_ipi_next;
3447 if (cpu_online(remsd->cpu))
3448 __smp_call_function_single(remsd->cpu,
3457 static int process_backlog(struct napi_struct *napi, int quota)
3460 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3463 /* Check if we have pending ipi, its better to send them now,
3464 * not waiting net_rx_action() end.
3466 if (sd->rps_ipi_list) {
3467 local_irq_disable();
3468 net_rps_action_and_irq_enable(sd);
3471 napi->weight = weight_p;
3472 local_irq_disable();
3473 while (work < quota) {
3474 struct sk_buff *skb;
3477 while ((skb = __skb_dequeue(&sd->process_queue))) {
3479 __netif_receive_skb(skb);
3480 local_irq_disable();
3481 input_queue_head_incr(sd);
3482 if (++work >= quota) {
3489 qlen = skb_queue_len(&sd->input_pkt_queue);
3491 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3492 &sd->process_queue);
3494 if (qlen < quota - work) {
3496 * Inline a custom version of __napi_complete().
3497 * only current cpu owns and manipulates this napi,
3498 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3499 * we can use a plain write instead of clear_bit(),
3500 * and we dont need an smp_mb() memory barrier.
3502 list_del(&napi->poll_list);
3505 quota = work + qlen;
3515 * __napi_schedule - schedule for receive
3516 * @n: entry to schedule
3518 * The entry's receive function will be scheduled to run
3520 void __napi_schedule(struct napi_struct *n)
3522 unsigned long flags;
3524 local_irq_save(flags);
3525 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3526 local_irq_restore(flags);
3528 EXPORT_SYMBOL(__napi_schedule);
3530 void __napi_complete(struct napi_struct *n)
3532 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3533 BUG_ON(n->gro_list);
3535 list_del(&n->poll_list);
3536 smp_mb__before_clear_bit();
3537 clear_bit(NAPI_STATE_SCHED, &n->state);
3539 EXPORT_SYMBOL(__napi_complete);
3541 void napi_complete(struct napi_struct *n)
3543 unsigned long flags;
3546 * don't let napi dequeue from the cpu poll list
3547 * just in case its running on a different cpu
3549 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3553 local_irq_save(flags);
3555 local_irq_restore(flags);
3557 EXPORT_SYMBOL(napi_complete);
3559 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3560 int (*poll)(struct napi_struct *, int), int weight)
3562 INIT_LIST_HEAD(&napi->poll_list);
3563 napi->gro_count = 0;
3564 napi->gro_list = NULL;
3567 napi->weight = weight;
3568 list_add(&napi->dev_list, &dev->napi_list);
3570 #ifdef CONFIG_NETPOLL
3571 spin_lock_init(&napi->poll_lock);
3572 napi->poll_owner = -1;
3574 set_bit(NAPI_STATE_SCHED, &napi->state);
3576 EXPORT_SYMBOL(netif_napi_add);
3578 void netif_napi_del(struct napi_struct *napi)
3580 struct sk_buff *skb, *next;
3582 list_del_init(&napi->dev_list);
3583 napi_free_frags(napi);
3585 for (skb = napi->gro_list; skb; skb = next) {
3591 napi->gro_list = NULL;
3592 napi->gro_count = 0;
3594 EXPORT_SYMBOL(netif_napi_del);
3596 static void net_rx_action(struct softirq_action *h)
3598 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3599 unsigned long time_limit = jiffies + 2;
3600 int budget = netdev_budget;
3603 local_irq_disable();
3605 while (!list_empty(&sd->poll_list)) {
3606 struct napi_struct *n;
3609 /* If softirq window is exhuasted then punt.
3610 * Allow this to run for 2 jiffies since which will allow
3611 * an average latency of 1.5/HZ.
3613 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3618 /* Even though interrupts have been re-enabled, this
3619 * access is safe because interrupts can only add new
3620 * entries to the tail of this list, and only ->poll()
3621 * calls can remove this head entry from the list.
3623 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3625 have = netpoll_poll_lock(n);
3629 /* This NAPI_STATE_SCHED test is for avoiding a race
3630 * with netpoll's poll_napi(). Only the entity which
3631 * obtains the lock and sees NAPI_STATE_SCHED set will
3632 * actually make the ->poll() call. Therefore we avoid
3633 * accidently calling ->poll() when NAPI is not scheduled.
3636 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3637 work = n->poll(n, weight);
3641 WARN_ON_ONCE(work > weight);
3645 local_irq_disable();
3647 /* Drivers must not modify the NAPI state if they
3648 * consume the entire weight. In such cases this code
3649 * still "owns" the NAPI instance and therefore can
3650 * move the instance around on the list at-will.
3652 if (unlikely(work == weight)) {
3653 if (unlikely(napi_disable_pending(n))) {
3656 local_irq_disable();
3658 list_move_tail(&n->poll_list, &sd->poll_list);
3661 netpoll_poll_unlock(have);
3664 net_rps_action_and_irq_enable(sd);
3666 #ifdef CONFIG_NET_DMA
3668 * There may not be any more sk_buffs coming right now, so push
3669 * any pending DMA copies to hardware
3671 dma_issue_pending_all();
3678 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3682 static gifconf_func_t *gifconf_list[NPROTO];
3685 * register_gifconf - register a SIOCGIF handler
3686 * @family: Address family
3687 * @gifconf: Function handler
3689 * Register protocol dependent address dumping routines. The handler
3690 * that is passed must not be freed or reused until it has been replaced
3691 * by another handler.
3693 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3695 if (family >= NPROTO)
3697 gifconf_list[family] = gifconf;
3700 EXPORT_SYMBOL(register_gifconf);
3704 * Map an interface index to its name (SIOCGIFNAME)
3708 * We need this ioctl for efficient implementation of the
3709 * if_indextoname() function required by the IPv6 API. Without
3710 * it, we would have to search all the interfaces to find a
3714 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3716 struct net_device *dev;
3720 * Fetch the caller's info block.
3723 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3727 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3733 strcpy(ifr.ifr_name, dev->name);
3736 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3742 * Perform a SIOCGIFCONF call. This structure will change
3743 * size eventually, and there is nothing I can do about it.
3744 * Thus we will need a 'compatibility mode'.
3747 static int dev_ifconf(struct net *net, char __user *arg)
3750 struct net_device *dev;
3757 * Fetch the caller's info block.
3760 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3767 * Loop over the interfaces, and write an info block for each.
3771 for_each_netdev(net, dev) {
3772 for (i = 0; i < NPROTO; i++) {
3773 if (gifconf_list[i]) {
3776 done = gifconf_list[i](dev, NULL, 0);
3778 done = gifconf_list[i](dev, pos + total,
3788 * All done. Write the updated control block back to the caller.
3790 ifc.ifc_len = total;
3793 * Both BSD and Solaris return 0 here, so we do too.
3795 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3798 #ifdef CONFIG_PROC_FS
3800 * This is invoked by the /proc filesystem handler to display a device
3803 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3806 struct net *net = seq_file_net(seq);
3808 struct net_device *dev;
3812 return SEQ_START_TOKEN;
3815 for_each_netdev_rcu(net, dev)
3822 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3824 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3825 first_net_device(seq_file_net(seq)) :
3826 next_net_device((struct net_device *)v);
3829 return rcu_dereference(dev);
3832 void dev_seq_stop(struct seq_file *seq, void *v)
3838 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3840 struct rtnl_link_stats64 temp;
3841 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3843 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3844 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3845 dev->name, stats->rx_bytes, stats->rx_packets,
3847 stats->rx_dropped + stats->rx_missed_errors,
3848 stats->rx_fifo_errors,
3849 stats->rx_length_errors + stats->rx_over_errors +
3850 stats->rx_crc_errors + stats->rx_frame_errors,
3851 stats->rx_compressed, stats->multicast,
3852 stats->tx_bytes, stats->tx_packets,
3853 stats->tx_errors, stats->tx_dropped,
3854 stats->tx_fifo_errors, stats->collisions,
3855 stats->tx_carrier_errors +
3856 stats->tx_aborted_errors +
3857 stats->tx_window_errors +
3858 stats->tx_heartbeat_errors,
3859 stats->tx_compressed);
3863 * Called from the PROCfs module. This now uses the new arbitrary sized
3864 * /proc/net interface to create /proc/net/dev
3866 static int dev_seq_show(struct seq_file *seq, void *v)
3868 if (v == SEQ_START_TOKEN)
3869 seq_puts(seq, "Inter-| Receive "
3871 " face |bytes packets errs drop fifo frame "
3872 "compressed multicast|bytes packets errs "
3873 "drop fifo colls carrier compressed\n");
3875 dev_seq_printf_stats(seq, v);
3879 static struct softnet_data *softnet_get_online(loff_t *pos)
3881 struct softnet_data *sd = NULL;
3883 while (*pos < nr_cpu_ids)
3884 if (cpu_online(*pos)) {
3885 sd = &per_cpu(softnet_data, *pos);
3892 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3894 return softnet_get_online(pos);
3897 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3900 return softnet_get_online(pos);
3903 static void softnet_seq_stop(struct seq_file *seq, void *v)
3907 static int softnet_seq_show(struct seq_file *seq, void *v)
3909 struct softnet_data *sd = v;
3911 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3912 sd->processed, sd->dropped, sd->time_squeeze, 0,
3913 0, 0, 0, 0, /* was fastroute */
3914 sd->cpu_collision, sd->received_rps);
3918 static const struct seq_operations dev_seq_ops = {
3919 .start = dev_seq_start,
3920 .next = dev_seq_next,
3921 .stop = dev_seq_stop,
3922 .show = dev_seq_show,
3925 static int dev_seq_open(struct inode *inode, struct file *file)
3927 return seq_open_net(inode, file, &dev_seq_ops,
3928 sizeof(struct seq_net_private));
3931 static const struct file_operations dev_seq_fops = {
3932 .owner = THIS_MODULE,
3933 .open = dev_seq_open,
3935 .llseek = seq_lseek,
3936 .release = seq_release_net,
3939 static const struct seq_operations softnet_seq_ops = {
3940 .start = softnet_seq_start,
3941 .next = softnet_seq_next,
3942 .stop = softnet_seq_stop,
3943 .show = softnet_seq_show,
3946 static int softnet_seq_open(struct inode *inode, struct file *file)
3948 return seq_open(file, &softnet_seq_ops);
3951 static const struct file_operations softnet_seq_fops = {
3952 .owner = THIS_MODULE,
3953 .open = softnet_seq_open,
3955 .llseek = seq_lseek,
3956 .release = seq_release,
3959 static void *ptype_get_idx(loff_t pos)
3961 struct packet_type *pt = NULL;
3965 list_for_each_entry_rcu(pt, &ptype_all, list) {
3971 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3972 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3981 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3985 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3988 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3990 struct packet_type *pt;
3991 struct list_head *nxt;
3995 if (v == SEQ_START_TOKEN)
3996 return ptype_get_idx(0);
3999 nxt = pt->list.next;
4000 if (pt->type == htons(ETH_P_ALL)) {
4001 if (nxt != &ptype_all)
4004 nxt = ptype_base[0].next;
4006 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4008 while (nxt == &ptype_base[hash]) {
4009 if (++hash >= PTYPE_HASH_SIZE)
4011 nxt = ptype_base[hash].next;
4014 return list_entry(nxt, struct packet_type, list);
4017 static void ptype_seq_stop(struct seq_file *seq, void *v)
4023 static int ptype_seq_show(struct seq_file *seq, void *v)
4025 struct packet_type *pt = v;
4027 if (v == SEQ_START_TOKEN)
4028 seq_puts(seq, "Type Device Function\n");
4029 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4030 if (pt->type == htons(ETH_P_ALL))
4031 seq_puts(seq, "ALL ");
4033 seq_printf(seq, "%04x", ntohs(pt->type));
4035 seq_printf(seq, " %-8s %pF\n",
4036 pt->dev ? pt->dev->name : "", pt->func);
4042 static const struct seq_operations ptype_seq_ops = {
4043 .start = ptype_seq_start,
4044 .next = ptype_seq_next,
4045 .stop = ptype_seq_stop,
4046 .show = ptype_seq_show,
4049 static int ptype_seq_open(struct inode *inode, struct file *file)
4051 return seq_open_net(inode, file, &ptype_seq_ops,
4052 sizeof(struct seq_net_private));
4055 static const struct file_operations ptype_seq_fops = {
4056 .owner = THIS_MODULE,
4057 .open = ptype_seq_open,
4059 .llseek = seq_lseek,
4060 .release = seq_release_net,
4064 static int __net_init dev_proc_net_init(struct net *net)
4068 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4070 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4072 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4075 if (wext_proc_init(net))
4081 proc_net_remove(net, "ptype");
4083 proc_net_remove(net, "softnet_stat");
4085 proc_net_remove(net, "dev");
4089 static void __net_exit dev_proc_net_exit(struct net *net)
4091 wext_proc_exit(net);
4093 proc_net_remove(net, "ptype");
4094 proc_net_remove(net, "softnet_stat");
4095 proc_net_remove(net, "dev");
4098 static struct pernet_operations __net_initdata dev_proc_ops = {
4099 .init = dev_proc_net_init,
4100 .exit = dev_proc_net_exit,
4103 static int __init dev_proc_init(void)
4105 return register_pernet_subsys(&dev_proc_ops);
4108 #define dev_proc_init() 0
4109 #endif /* CONFIG_PROC_FS */
4113 * netdev_set_master - set up master/slave pair
4114 * @slave: slave device
4115 * @master: new master device
4117 * Changes the master device of the slave. Pass %NULL to break the
4118 * bonding. The caller must hold the RTNL semaphore. On a failure
4119 * a negative errno code is returned. On success the reference counts
4120 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4121 * function returns zero.
4123 int netdev_set_master(struct net_device *slave, struct net_device *master)
4125 struct net_device *old = slave->master;
4135 slave->master = master;
4142 slave->flags |= IFF_SLAVE;
4144 slave->flags &= ~IFF_SLAVE;
4146 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4149 EXPORT_SYMBOL(netdev_set_master);
4151 static void dev_change_rx_flags(struct net_device *dev, int flags)
4153 const struct net_device_ops *ops = dev->netdev_ops;
4155 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4156 ops->ndo_change_rx_flags(dev, flags);
4159 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4161 unsigned short old_flags = dev->flags;
4167 dev->flags |= IFF_PROMISC;
4168 dev->promiscuity += inc;
4169 if (dev->promiscuity == 0) {
4172 * If inc causes overflow, untouch promisc and return error.
4175 dev->flags &= ~IFF_PROMISC;
4177 dev->promiscuity -= inc;
4178 printk(KERN_WARNING "%s: promiscuity touches roof, "
4179 "set promiscuity failed, promiscuity feature "
4180 "of device might be broken.\n", dev->name);
4184 if (dev->flags != old_flags) {
4185 printk(KERN_INFO "device %s %s promiscuous mode\n",
4186 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4188 if (audit_enabled) {
4189 current_uid_gid(&uid, &gid);
4190 audit_log(current->audit_context, GFP_ATOMIC,
4191 AUDIT_ANOM_PROMISCUOUS,
4192 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4193 dev->name, (dev->flags & IFF_PROMISC),
4194 (old_flags & IFF_PROMISC),
4195 audit_get_loginuid(current),
4197 audit_get_sessionid(current));
4200 dev_change_rx_flags(dev, IFF_PROMISC);
4206 * dev_set_promiscuity - update promiscuity count on a device
4210 * Add or remove promiscuity from a device. While the count in the device
4211 * remains above zero the interface remains promiscuous. Once it hits zero
4212 * the device reverts back to normal filtering operation. A negative inc
4213 * value is used to drop promiscuity on the device.
4214 * Return 0 if successful or a negative errno code on error.
4216 int dev_set_promiscuity(struct net_device *dev, int inc)
4218 unsigned short old_flags = dev->flags;
4221 err = __dev_set_promiscuity(dev, inc);
4224 if (dev->flags != old_flags)
4225 dev_set_rx_mode(dev);
4228 EXPORT_SYMBOL(dev_set_promiscuity);
4231 * dev_set_allmulti - update allmulti count on a device
4235 * Add or remove reception of all multicast frames to a device. While the
4236 * count in the device remains above zero the interface remains listening
4237 * to all interfaces. Once it hits zero the device reverts back to normal
4238 * filtering operation. A negative @inc value is used to drop the counter
4239 * when releasing a resource needing all multicasts.
4240 * Return 0 if successful or a negative errno code on error.
4243 int dev_set_allmulti(struct net_device *dev, int inc)
4245 unsigned short old_flags = dev->flags;
4249 dev->flags |= IFF_ALLMULTI;
4250 dev->allmulti += inc;
4251 if (dev->allmulti == 0) {
4254 * If inc causes overflow, untouch allmulti and return error.
4257 dev->flags &= ~IFF_ALLMULTI;
4259 dev->allmulti -= inc;
4260 printk(KERN_WARNING "%s: allmulti touches roof, "
4261 "set allmulti failed, allmulti feature of "
4262 "device might be broken.\n", dev->name);
4266 if (dev->flags ^ old_flags) {
4267 dev_change_rx_flags(dev, IFF_ALLMULTI);
4268 dev_set_rx_mode(dev);
4272 EXPORT_SYMBOL(dev_set_allmulti);
4275 * Upload unicast and multicast address lists to device and
4276 * configure RX filtering. When the device doesn't support unicast
4277 * filtering it is put in promiscuous mode while unicast addresses
4280 void __dev_set_rx_mode(struct net_device *dev)
4282 const struct net_device_ops *ops = dev->netdev_ops;
4284 /* dev_open will call this function so the list will stay sane. */
4285 if (!(dev->flags&IFF_UP))
4288 if (!netif_device_present(dev))
4291 if (ops->ndo_set_rx_mode)
4292 ops->ndo_set_rx_mode(dev);
4294 /* Unicast addresses changes may only happen under the rtnl,
4295 * therefore calling __dev_set_promiscuity here is safe.
4297 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4298 __dev_set_promiscuity(dev, 1);
4299 dev->uc_promisc = 1;
4300 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4301 __dev_set_promiscuity(dev, -1);
4302 dev->uc_promisc = 0;
4305 if (ops->ndo_set_multicast_list)
4306 ops->ndo_set_multicast_list(dev);
4310 void dev_set_rx_mode(struct net_device *dev)
4312 netif_addr_lock_bh(dev);
4313 __dev_set_rx_mode(dev);
4314 netif_addr_unlock_bh(dev);
4318 * dev_get_flags - get flags reported to userspace
4321 * Get the combination of flag bits exported through APIs to userspace.
4323 unsigned dev_get_flags(const struct net_device *dev)
4327 flags = (dev->flags & ~(IFF_PROMISC |
4332 (dev->gflags & (IFF_PROMISC |
4335 if (netif_running(dev)) {
4336 if (netif_oper_up(dev))
4337 flags |= IFF_RUNNING;
4338 if (netif_carrier_ok(dev))
4339 flags |= IFF_LOWER_UP;
4340 if (netif_dormant(dev))
4341 flags |= IFF_DORMANT;
4346 EXPORT_SYMBOL(dev_get_flags);
4348 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4350 int old_flags = dev->flags;
4356 * Set the flags on our device.
4359 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4360 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4362 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4366 * Load in the correct multicast list now the flags have changed.
4369 if ((old_flags ^ flags) & IFF_MULTICAST)
4370 dev_change_rx_flags(dev, IFF_MULTICAST);
4372 dev_set_rx_mode(dev);
4375 * Have we downed the interface. We handle IFF_UP ourselves
4376 * according to user attempts to set it, rather than blindly
4381 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4382 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4385 dev_set_rx_mode(dev);
4388 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4389 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4391 dev->gflags ^= IFF_PROMISC;
4392 dev_set_promiscuity(dev, inc);
4395 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4396 is important. Some (broken) drivers set IFF_PROMISC, when
4397 IFF_ALLMULTI is requested not asking us and not reporting.
4399 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4400 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4402 dev->gflags ^= IFF_ALLMULTI;
4403 dev_set_allmulti(dev, inc);
4409 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4411 unsigned int changes = dev->flags ^ old_flags;
4413 if (changes & IFF_UP) {
4414 if (dev->flags & IFF_UP)
4415 call_netdevice_notifiers(NETDEV_UP, dev);
4417 call_netdevice_notifiers(NETDEV_DOWN, dev);
4420 if (dev->flags & IFF_UP &&
4421 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4422 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4426 * dev_change_flags - change device settings
4428 * @flags: device state flags
4430 * Change settings on device based state flags. The flags are
4431 * in the userspace exported format.
4433 int dev_change_flags(struct net_device *dev, unsigned flags)
4436 int old_flags = dev->flags;
4438 ret = __dev_change_flags(dev, flags);
4442 changes = old_flags ^ dev->flags;
4444 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4446 __dev_notify_flags(dev, old_flags);
4449 EXPORT_SYMBOL(dev_change_flags);
4452 * dev_set_mtu - Change maximum transfer unit
4454 * @new_mtu: new transfer unit
4456 * Change the maximum transfer size of the network device.
4458 int dev_set_mtu(struct net_device *dev, int new_mtu)
4460 const struct net_device_ops *ops = dev->netdev_ops;
4463 if (new_mtu == dev->mtu)
4466 /* MTU must be positive. */
4470 if (!netif_device_present(dev))
4474 if (ops->ndo_change_mtu)
4475 err = ops->ndo_change_mtu(dev, new_mtu);
4479 if (!err && dev->flags & IFF_UP)
4480 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4483 EXPORT_SYMBOL(dev_set_mtu);
4486 * dev_set_mac_address - Change Media Access Control Address
4490 * Change the hardware (MAC) address of the device
4492 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4494 const struct net_device_ops *ops = dev->netdev_ops;
4497 if (!ops->ndo_set_mac_address)
4499 if (sa->sa_family != dev->type)
4501 if (!netif_device_present(dev))
4503 err = ops->ndo_set_mac_address(dev, sa);
4505 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4508 EXPORT_SYMBOL(dev_set_mac_address);
4511 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4513 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4516 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4522 case SIOCGIFFLAGS: /* Get interface flags */
4523 ifr->ifr_flags = (short) dev_get_flags(dev);
4526 case SIOCGIFMETRIC: /* Get the metric on the interface
4527 (currently unused) */
4528 ifr->ifr_metric = 0;
4531 case SIOCGIFMTU: /* Get the MTU of a device */
4532 ifr->ifr_mtu = dev->mtu;
4537 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4539 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4540 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4541 ifr->ifr_hwaddr.sa_family = dev->type;
4549 ifr->ifr_map.mem_start = dev->mem_start;
4550 ifr->ifr_map.mem_end = dev->mem_end;
4551 ifr->ifr_map.base_addr = dev->base_addr;
4552 ifr->ifr_map.irq = dev->irq;
4553 ifr->ifr_map.dma = dev->dma;
4554 ifr->ifr_map.port = dev->if_port;
4558 ifr->ifr_ifindex = dev->ifindex;
4562 ifr->ifr_qlen = dev->tx_queue_len;
4566 /* dev_ioctl() should ensure this case
4578 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4580 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4583 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4584 const struct net_device_ops *ops;
4589 ops = dev->netdev_ops;
4592 case SIOCSIFFLAGS: /* Set interface flags */
4593 return dev_change_flags(dev, ifr->ifr_flags);
4595 case SIOCSIFMETRIC: /* Set the metric on the interface
4596 (currently unused) */
4599 case SIOCSIFMTU: /* Set the MTU of a device */
4600 return dev_set_mtu(dev, ifr->ifr_mtu);
4603 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4605 case SIOCSIFHWBROADCAST:
4606 if (ifr->ifr_hwaddr.sa_family != dev->type)
4608 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4609 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4610 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4614 if (ops->ndo_set_config) {
4615 if (!netif_device_present(dev))
4617 return ops->ndo_set_config(dev, &ifr->ifr_map);
4622 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4623 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4625 if (!netif_device_present(dev))
4627 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4630 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4631 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4633 if (!netif_device_present(dev))
4635 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4638 if (ifr->ifr_qlen < 0)
4640 dev->tx_queue_len = ifr->ifr_qlen;
4644 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4645 return dev_change_name(dev, ifr->ifr_newname);
4648 * Unknown or private ioctl
4651 if ((cmd >= SIOCDEVPRIVATE &&
4652 cmd <= SIOCDEVPRIVATE + 15) ||
4653 cmd == SIOCBONDENSLAVE ||
4654 cmd == SIOCBONDRELEASE ||
4655 cmd == SIOCBONDSETHWADDR ||
4656 cmd == SIOCBONDSLAVEINFOQUERY ||
4657 cmd == SIOCBONDINFOQUERY ||
4658 cmd == SIOCBONDCHANGEACTIVE ||
4659 cmd == SIOCGMIIPHY ||
4660 cmd == SIOCGMIIREG ||
4661 cmd == SIOCSMIIREG ||
4662 cmd == SIOCBRADDIF ||
4663 cmd == SIOCBRDELIF ||
4664 cmd == SIOCSHWTSTAMP ||
4665 cmd == SIOCWANDEV) {
4667 if (ops->ndo_do_ioctl) {
4668 if (netif_device_present(dev))
4669 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4681 * This function handles all "interface"-type I/O control requests. The actual
4682 * 'doing' part of this is dev_ifsioc above.
4686 * dev_ioctl - network device ioctl
4687 * @net: the applicable net namespace
4688 * @cmd: command to issue
4689 * @arg: pointer to a struct ifreq in user space
4691 * Issue ioctl functions to devices. This is normally called by the
4692 * user space syscall interfaces but can sometimes be useful for
4693 * other purposes. The return value is the return from the syscall if
4694 * positive or a negative errno code on error.
4697 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4703 /* One special case: SIOCGIFCONF takes ifconf argument
4704 and requires shared lock, because it sleeps writing
4708 if (cmd == SIOCGIFCONF) {
4710 ret = dev_ifconf(net, (char __user *) arg);
4714 if (cmd == SIOCGIFNAME)
4715 return dev_ifname(net, (struct ifreq __user *)arg);
4717 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4720 ifr.ifr_name[IFNAMSIZ-1] = 0;
4722 colon = strchr(ifr.ifr_name, ':');
4727 * See which interface the caller is talking about.
4732 * These ioctl calls:
4733 * - can be done by all.
4734 * - atomic and do not require locking.
4745 dev_load(net, ifr.ifr_name);
4747 ret = dev_ifsioc_locked(net, &ifr, cmd);
4752 if (copy_to_user(arg, &ifr,
4753 sizeof(struct ifreq)))
4759 dev_load(net, ifr.ifr_name);
4761 ret = dev_ethtool(net, &ifr);
4766 if (copy_to_user(arg, &ifr,
4767 sizeof(struct ifreq)))
4773 * These ioctl calls:
4774 * - require superuser power.
4775 * - require strict serialization.
4781 if (!capable(CAP_NET_ADMIN))
4783 dev_load(net, ifr.ifr_name);
4785 ret = dev_ifsioc(net, &ifr, cmd);
4790 if (copy_to_user(arg, &ifr,
4791 sizeof(struct ifreq)))
4797 * These ioctl calls:
4798 * - require superuser power.
4799 * - require strict serialization.
4800 * - do not return a value
4810 case SIOCSIFHWBROADCAST:
4813 case SIOCBONDENSLAVE:
4814 case SIOCBONDRELEASE:
4815 case SIOCBONDSETHWADDR:
4816 case SIOCBONDCHANGEACTIVE:
4820 if (!capable(CAP_NET_ADMIN))
4823 case SIOCBONDSLAVEINFOQUERY:
4824 case SIOCBONDINFOQUERY:
4825 dev_load(net, ifr.ifr_name);
4827 ret = dev_ifsioc(net, &ifr, cmd);
4832 /* Get the per device memory space. We can add this but
4833 * currently do not support it */
4835 /* Set the per device memory buffer space.
4836 * Not applicable in our case */
4841 * Unknown or private ioctl.
4844 if (cmd == SIOCWANDEV ||
4845 (cmd >= SIOCDEVPRIVATE &&
4846 cmd <= SIOCDEVPRIVATE + 15)) {
4847 dev_load(net, ifr.ifr_name);
4849 ret = dev_ifsioc(net, &ifr, cmd);
4851 if (!ret && copy_to_user(arg, &ifr,
4852 sizeof(struct ifreq)))
4856 /* Take care of Wireless Extensions */
4857 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4858 return wext_handle_ioctl(net, &ifr, cmd, arg);
4865 * dev_new_index - allocate an ifindex
4866 * @net: the applicable net namespace
4868 * Returns a suitable unique value for a new device interface
4869 * number. The caller must hold the rtnl semaphore or the
4870 * dev_base_lock to be sure it remains unique.
4872 static int dev_new_index(struct net *net)
4878 if (!__dev_get_by_index(net, ifindex))
4883 /* Delayed registration/unregisteration */
4884 static LIST_HEAD(net_todo_list);
4886 static void net_set_todo(struct net_device *dev)
4888 list_add_tail(&dev->todo_list, &net_todo_list);
4891 static void rollback_registered_many(struct list_head *head)
4893 struct net_device *dev, *tmp;
4895 BUG_ON(dev_boot_phase);
4898 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4899 /* Some devices call without registering
4900 * for initialization unwind. Remove those
4901 * devices and proceed with the remaining.
4903 if (dev->reg_state == NETREG_UNINITIALIZED) {
4904 pr_debug("unregister_netdevice: device %s/%p never "
4905 "was registered\n", dev->name, dev);
4908 list_del(&dev->unreg_list);
4912 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4914 /* If device is running, close it first. */
4917 /* And unlink it from device chain. */
4918 unlist_netdevice(dev);
4920 dev->reg_state = NETREG_UNREGISTERING;
4925 list_for_each_entry(dev, head, unreg_list) {
4926 /* Shutdown queueing discipline. */
4930 /* Notify protocols, that we are about to destroy
4931 this device. They should clean all the things.
4933 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4935 if (!dev->rtnl_link_ops ||
4936 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4937 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4940 * Flush the unicast and multicast chains
4945 if (dev->netdev_ops->ndo_uninit)
4946 dev->netdev_ops->ndo_uninit(dev);
4948 /* Notifier chain MUST detach us from master device. */
4949 WARN_ON(dev->master);
4951 /* Remove entries from kobject tree */
4952 netdev_unregister_kobject(dev);
4955 /* Process any work delayed until the end of the batch */
4956 dev = list_first_entry(head, struct net_device, unreg_list);
4957 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4961 list_for_each_entry(dev, head, unreg_list)
4965 static void rollback_registered(struct net_device *dev)
4969 list_add(&dev->unreg_list, &single);
4970 rollback_registered_many(&single);
4973 unsigned long netdev_fix_features(unsigned long features, const char *name)
4975 /* Fix illegal SG+CSUM combinations. */
4976 if ((features & NETIF_F_SG) &&
4977 !(features & NETIF_F_ALL_CSUM)) {
4979 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4980 "checksum feature.\n", name);
4981 features &= ~NETIF_F_SG;
4984 /* TSO requires that SG is present as well. */
4985 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4987 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4988 "SG feature.\n", name);
4989 features &= ~NETIF_F_TSO;
4992 if (features & NETIF_F_UFO) {
4993 if (!(features & NETIF_F_GEN_CSUM)) {
4995 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4996 "since no NETIF_F_HW_CSUM feature.\n",
4998 features &= ~NETIF_F_UFO;
5001 if (!(features & NETIF_F_SG)) {
5003 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5004 "since no NETIF_F_SG feature.\n", name);
5005 features &= ~NETIF_F_UFO;
5011 EXPORT_SYMBOL(netdev_fix_features);
5014 * netif_stacked_transfer_operstate - transfer operstate
5015 * @rootdev: the root or lower level device to transfer state from
5016 * @dev: the device to transfer operstate to
5018 * Transfer operational state from root to device. This is normally
5019 * called when a stacking relationship exists between the root
5020 * device and the device(a leaf device).
5022 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5023 struct net_device *dev)
5025 if (rootdev->operstate == IF_OPER_DORMANT)
5026 netif_dormant_on(dev);
5028 netif_dormant_off(dev);
5030 if (netif_carrier_ok(rootdev)) {
5031 if (!netif_carrier_ok(dev))
5032 netif_carrier_on(dev);
5034 if (netif_carrier_ok(dev))
5035 netif_carrier_off(dev);
5038 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5040 static int netif_alloc_rx_queues(struct net_device *dev)
5043 unsigned int i, count = dev->num_rx_queues;
5044 struct netdev_rx_queue *rx;
5048 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5050 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5055 for (i = 0; i < count; i++)
5061 static int netif_alloc_netdev_queues(struct net_device *dev)
5063 unsigned int count = dev->num_tx_queues;
5064 struct netdev_queue *tx;
5068 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5070 pr_err("netdev: Unable to allocate %u tx queues.\n",
5078 static void netdev_init_one_queue(struct net_device *dev,
5079 struct netdev_queue *queue,
5084 /* Initialize queue lock */
5085 spin_lock_init(&queue->_xmit_lock);
5086 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5087 queue->xmit_lock_owner = -1;
5090 static void netdev_init_queues(struct net_device *dev)
5092 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5093 spin_lock_init(&dev->tx_global_lock);
5097 * register_netdevice - register a network device
5098 * @dev: device to register
5100 * Take a completed network device structure and add it to the kernel
5101 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5102 * chain. 0 is returned on success. A negative errno code is returned
5103 * on a failure to set up the device, or if the name is a duplicate.
5105 * Callers must hold the rtnl semaphore. You may want
5106 * register_netdev() instead of this.
5109 * The locking appears insufficient to guarantee two parallel registers
5110 * will not get the same name.
5113 int register_netdevice(struct net_device *dev)
5116 struct net *net = dev_net(dev);
5118 BUG_ON(dev_boot_phase);
5123 /* When net_device's are persistent, this will be fatal. */
5124 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5127 spin_lock_init(&dev->addr_list_lock);
5128 netdev_set_addr_lockdep_class(dev);
5132 netdev_init_queues(dev);
5134 /* Init, if this function is available */
5135 if (dev->netdev_ops->ndo_init) {
5136 ret = dev->netdev_ops->ndo_init(dev);
5144 ret = dev_get_valid_name(dev, dev->name, 0);
5148 dev->ifindex = dev_new_index(net);
5149 if (dev->iflink == -1)
5150 dev->iflink = dev->ifindex;
5152 /* Fix illegal checksum combinations */
5153 if ((dev->features & NETIF_F_HW_CSUM) &&
5154 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5155 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5157 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5160 if ((dev->features & NETIF_F_NO_CSUM) &&
5161 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5162 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5164 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5167 dev->features = netdev_fix_features(dev->features, dev->name);
5169 /* Enable software GSO if SG is supported. */
5170 if (dev->features & NETIF_F_SG)
5171 dev->features |= NETIF_F_GSO;
5173 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5174 * vlan_dev_init() will do the dev->features check, so these features
5175 * are enabled only if supported by underlying device.
5177 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5179 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5180 ret = notifier_to_errno(ret);
5184 ret = netdev_register_kobject(dev);
5187 dev->reg_state = NETREG_REGISTERED;
5190 * Default initial state at registry is that the
5191 * device is present.
5194 set_bit(__LINK_STATE_PRESENT, &dev->state);
5196 dev_init_scheduler(dev);
5198 list_netdevice(dev);
5200 /* Notify protocols, that a new device appeared. */
5201 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5202 ret = notifier_to_errno(ret);
5204 rollback_registered(dev);
5205 dev->reg_state = NETREG_UNREGISTERED;
5208 * Prevent userspace races by waiting until the network
5209 * device is fully setup before sending notifications.
5211 if (!dev->rtnl_link_ops ||
5212 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5213 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5219 if (dev->netdev_ops->ndo_uninit)
5220 dev->netdev_ops->ndo_uninit(dev);
5223 EXPORT_SYMBOL(register_netdevice);
5226 * init_dummy_netdev - init a dummy network device for NAPI
5227 * @dev: device to init
5229 * This takes a network device structure and initialize the minimum
5230 * amount of fields so it can be used to schedule NAPI polls without
5231 * registering a full blown interface. This is to be used by drivers
5232 * that need to tie several hardware interfaces to a single NAPI
5233 * poll scheduler due to HW limitations.
5235 int init_dummy_netdev(struct net_device *dev)
5237 /* Clear everything. Note we don't initialize spinlocks
5238 * are they aren't supposed to be taken by any of the
5239 * NAPI code and this dummy netdev is supposed to be
5240 * only ever used for NAPI polls
5242 memset(dev, 0, sizeof(struct net_device));
5244 /* make sure we BUG if trying to hit standard
5245 * register/unregister code path
5247 dev->reg_state = NETREG_DUMMY;
5249 /* NAPI wants this */
5250 INIT_LIST_HEAD(&dev->napi_list);
5252 /* a dummy interface is started by default */
5253 set_bit(__LINK_STATE_PRESENT, &dev->state);
5254 set_bit(__LINK_STATE_START, &dev->state);
5256 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5257 * because users of this 'device' dont need to change
5263 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5267 * register_netdev - register a network device
5268 * @dev: device to register
5270 * Take a completed network device structure and add it to the kernel
5271 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5272 * chain. 0 is returned on success. A negative errno code is returned
5273 * on a failure to set up the device, or if the name is a duplicate.
5275 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5276 * and expands the device name if you passed a format string to
5279 int register_netdev(struct net_device *dev)
5286 * If the name is a format string the caller wants us to do a
5289 if (strchr(dev->name, '%')) {
5290 err = dev_alloc_name(dev, dev->name);
5295 err = register_netdevice(dev);
5300 EXPORT_SYMBOL(register_netdev);
5302 int netdev_refcnt_read(const struct net_device *dev)
5306 for_each_possible_cpu(i)
5307 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5310 EXPORT_SYMBOL(netdev_refcnt_read);
5313 * netdev_wait_allrefs - wait until all references are gone.
5315 * This is called when unregistering network devices.
5317 * Any protocol or device that holds a reference should register
5318 * for netdevice notification, and cleanup and put back the
5319 * reference if they receive an UNREGISTER event.
5320 * We can get stuck here if buggy protocols don't correctly
5323 static void netdev_wait_allrefs(struct net_device *dev)
5325 unsigned long rebroadcast_time, warning_time;
5328 linkwatch_forget_dev(dev);
5330 rebroadcast_time = warning_time = jiffies;
5331 refcnt = netdev_refcnt_read(dev);
5333 while (refcnt != 0) {
5334 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5337 /* Rebroadcast unregister notification */
5338 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5339 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5340 * should have already handle it the first time */
5342 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5344 /* We must not have linkwatch events
5345 * pending on unregister. If this
5346 * happens, we simply run the queue
5347 * unscheduled, resulting in a noop
5350 linkwatch_run_queue();
5355 rebroadcast_time = jiffies;
5360 refcnt = netdev_refcnt_read(dev);
5362 if (time_after(jiffies, warning_time + 10 * HZ)) {
5363 printk(KERN_EMERG "unregister_netdevice: "
5364 "waiting for %s to become free. Usage "
5367 warning_time = jiffies;
5376 * register_netdevice(x1);
5377 * register_netdevice(x2);
5379 * unregister_netdevice(y1);
5380 * unregister_netdevice(y2);
5386 * We are invoked by rtnl_unlock().
5387 * This allows us to deal with problems:
5388 * 1) We can delete sysfs objects which invoke hotplug
5389 * without deadlocking with linkwatch via keventd.
5390 * 2) Since we run with the RTNL semaphore not held, we can sleep
5391 * safely in order to wait for the netdev refcnt to drop to zero.
5393 * We must not return until all unregister events added during
5394 * the interval the lock was held have been completed.
5396 void netdev_run_todo(void)
5398 struct list_head list;
5400 /* Snapshot list, allow later requests */
5401 list_replace_init(&net_todo_list, &list);
5405 while (!list_empty(&list)) {
5406 struct net_device *dev
5407 = list_first_entry(&list, struct net_device, todo_list);
5408 list_del(&dev->todo_list);
5410 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5411 printk(KERN_ERR "network todo '%s' but state %d\n",
5412 dev->name, dev->reg_state);
5417 dev->reg_state = NETREG_UNREGISTERED;
5419 on_each_cpu(flush_backlog, dev, 1);
5421 netdev_wait_allrefs(dev);
5424 BUG_ON(netdev_refcnt_read(dev));
5425 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5426 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5427 WARN_ON(dev->dn_ptr);
5429 if (dev->destructor)
5430 dev->destructor(dev);
5432 /* Free network device */
5433 kobject_put(&dev->dev.kobj);
5438 * dev_txq_stats_fold - fold tx_queues stats
5439 * @dev: device to get statistics from
5440 * @stats: struct rtnl_link_stats64 to hold results
5442 void dev_txq_stats_fold(const struct net_device *dev,
5443 struct rtnl_link_stats64 *stats)
5445 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5447 struct netdev_queue *txq;
5449 for (i = 0; i < dev->num_tx_queues; i++) {
5450 txq = netdev_get_tx_queue(dev, i);
5451 spin_lock_bh(&txq->_xmit_lock);
5452 tx_bytes += txq->tx_bytes;
5453 tx_packets += txq->tx_packets;
5454 tx_dropped += txq->tx_dropped;
5455 spin_unlock_bh(&txq->_xmit_lock);
5457 if (tx_bytes || tx_packets || tx_dropped) {
5458 stats->tx_bytes = tx_bytes;
5459 stats->tx_packets = tx_packets;
5460 stats->tx_dropped = tx_dropped;
5463 EXPORT_SYMBOL(dev_txq_stats_fold);
5465 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5466 * fields in the same order, with only the type differing.
5468 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5469 const struct net_device_stats *netdev_stats)
5471 #if BITS_PER_LONG == 64
5472 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5473 memcpy(stats64, netdev_stats, sizeof(*stats64));
5475 size_t i, n = sizeof(*stats64) / sizeof(u64);
5476 const unsigned long *src = (const unsigned long *)netdev_stats;
5477 u64 *dst = (u64 *)stats64;
5479 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5480 sizeof(*stats64) / sizeof(u64));
5481 for (i = 0; i < n; i++)
5487 * dev_get_stats - get network device statistics
5488 * @dev: device to get statistics from
5489 * @storage: place to store stats
5491 * Get network statistics from device. Return @storage.
5492 * The device driver may provide its own method by setting
5493 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5494 * otherwise the internal statistics structure is used.
5496 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5497 struct rtnl_link_stats64 *storage)
5499 const struct net_device_ops *ops = dev->netdev_ops;
5501 if (ops->ndo_get_stats64) {
5502 memset(storage, 0, sizeof(*storage));
5503 ops->ndo_get_stats64(dev, storage);
5504 } else if (ops->ndo_get_stats) {
5505 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5507 netdev_stats_to_stats64(storage, &dev->stats);
5508 dev_txq_stats_fold(dev, storage);
5510 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5513 EXPORT_SYMBOL(dev_get_stats);
5515 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5517 struct netdev_queue *queue = dev_ingress_queue(dev);
5519 #ifdef CONFIG_NET_CLS_ACT
5522 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5525 netdev_init_one_queue(dev, queue, NULL);
5526 queue->qdisc = &noop_qdisc;
5527 queue->qdisc_sleeping = &noop_qdisc;
5528 rcu_assign_pointer(dev->ingress_queue, queue);
5534 * alloc_netdev_mq - allocate network device
5535 * @sizeof_priv: size of private data to allocate space for
5536 * @name: device name format string
5537 * @setup: callback to initialize device
5538 * @queue_count: the number of subqueues to allocate
5540 * Allocates a struct net_device with private data area for driver use
5541 * and performs basic initialization. Also allocates subquue structs
5542 * for each queue on the device at the end of the netdevice.
5544 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5545 void (*setup)(struct net_device *), unsigned int queue_count)
5547 struct net_device *dev;
5549 struct net_device *p;
5551 BUG_ON(strlen(name) >= sizeof(dev->name));
5553 if (queue_count < 1) {
5554 pr_err("alloc_netdev: Unable to allocate device "
5555 "with zero queues.\n");
5559 alloc_size = sizeof(struct net_device);
5561 /* ensure 32-byte alignment of private area */
5562 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5563 alloc_size += sizeof_priv;
5565 /* ensure 32-byte alignment of whole construct */
5566 alloc_size += NETDEV_ALIGN - 1;
5568 p = kzalloc(alloc_size, GFP_KERNEL);
5570 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5574 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5575 dev->padded = (char *)dev - (char *)p;
5577 dev->pcpu_refcnt = alloc_percpu(int);
5578 if (!dev->pcpu_refcnt)
5581 if (dev_addr_init(dev))
5587 dev_net_set(dev, &init_net);
5589 dev->num_tx_queues = queue_count;
5590 dev->real_num_tx_queues = queue_count;
5591 if (netif_alloc_netdev_queues(dev))
5595 dev->num_rx_queues = queue_count;
5596 dev->real_num_rx_queues = queue_count;
5597 if (netif_alloc_rx_queues(dev))
5601 dev->gso_max_size = GSO_MAX_SIZE;
5603 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5604 dev->ethtool_ntuple_list.count = 0;
5605 INIT_LIST_HEAD(&dev->napi_list);
5606 INIT_LIST_HEAD(&dev->unreg_list);
5607 INIT_LIST_HEAD(&dev->link_watch_list);
5608 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5610 strcpy(dev->name, name);
5614 free_percpu(dev->pcpu_refcnt);
5624 EXPORT_SYMBOL(alloc_netdev_mq);
5627 * free_netdev - free network device
5630 * This function does the last stage of destroying an allocated device
5631 * interface. The reference to the device object is released.
5632 * If this is the last reference then it will be freed.
5634 void free_netdev(struct net_device *dev)
5636 struct napi_struct *p, *n;
5638 release_net(dev_net(dev));
5645 kfree(rcu_dereference_raw(dev->ingress_queue));
5647 /* Flush device addresses */
5648 dev_addr_flush(dev);
5650 /* Clear ethtool n-tuple list */
5651 ethtool_ntuple_flush(dev);
5653 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5656 free_percpu(dev->pcpu_refcnt);
5657 dev->pcpu_refcnt = NULL;
5659 /* Compatibility with error handling in drivers */
5660 if (dev->reg_state == NETREG_UNINITIALIZED) {
5661 kfree((char *)dev - dev->padded);
5665 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5666 dev->reg_state = NETREG_RELEASED;
5668 /* will free via device release */
5669 put_device(&dev->dev);
5671 EXPORT_SYMBOL(free_netdev);
5674 * synchronize_net - Synchronize with packet receive processing
5676 * Wait for packets currently being received to be done.
5677 * Does not block later packets from starting.
5679 void synchronize_net(void)
5684 EXPORT_SYMBOL(synchronize_net);
5687 * unregister_netdevice_queue - remove device from the kernel
5691 * This function shuts down a device interface and removes it
5692 * from the kernel tables.
5693 * If head not NULL, device is queued to be unregistered later.
5695 * Callers must hold the rtnl semaphore. You may want
5696 * unregister_netdev() instead of this.
5699 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5704 list_move_tail(&dev->unreg_list, head);
5706 rollback_registered(dev);
5707 /* Finish processing unregister after unlock */
5711 EXPORT_SYMBOL(unregister_netdevice_queue);
5714 * unregister_netdevice_many - unregister many devices
5715 * @head: list of devices
5717 void unregister_netdevice_many(struct list_head *head)
5719 struct net_device *dev;
5721 if (!list_empty(head)) {
5722 rollback_registered_many(head);
5723 list_for_each_entry(dev, head, unreg_list)
5727 EXPORT_SYMBOL(unregister_netdevice_many);
5730 * unregister_netdev - remove device from the kernel
5733 * This function shuts down a device interface and removes it
5734 * from the kernel tables.
5736 * This is just a wrapper for unregister_netdevice that takes
5737 * the rtnl semaphore. In general you want to use this and not
5738 * unregister_netdevice.
5740 void unregister_netdev(struct net_device *dev)
5743 unregister_netdevice(dev);
5746 EXPORT_SYMBOL(unregister_netdev);
5749 * dev_change_net_namespace - move device to different nethost namespace
5751 * @net: network namespace
5752 * @pat: If not NULL name pattern to try if the current device name
5753 * is already taken in the destination network namespace.
5755 * This function shuts down a device interface and moves it
5756 * to a new network namespace. On success 0 is returned, on
5757 * a failure a netagive errno code is returned.
5759 * Callers must hold the rtnl semaphore.
5762 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5768 /* Don't allow namespace local devices to be moved. */
5770 if (dev->features & NETIF_F_NETNS_LOCAL)
5773 /* Ensure the device has been registrered */
5775 if (dev->reg_state != NETREG_REGISTERED)
5778 /* Get out if there is nothing todo */
5780 if (net_eq(dev_net(dev), net))
5783 /* Pick the destination device name, and ensure
5784 * we can use it in the destination network namespace.
5787 if (__dev_get_by_name(net, dev->name)) {
5788 /* We get here if we can't use the current device name */
5791 if (dev_get_valid_name(dev, pat, 1))
5796 * And now a mini version of register_netdevice unregister_netdevice.
5799 /* If device is running close it first. */
5802 /* And unlink it from device chain */
5804 unlist_netdevice(dev);
5808 /* Shutdown queueing discipline. */
5811 /* Notify protocols, that we are about to destroy
5812 this device. They should clean all the things.
5814 Note that dev->reg_state stays at NETREG_REGISTERED.
5815 This is wanted because this way 8021q and macvlan know
5816 the device is just moving and can keep their slaves up.
5818 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5819 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5822 * Flush the unicast and multicast chains
5827 /* Actually switch the network namespace */
5828 dev_net_set(dev, net);
5830 /* If there is an ifindex conflict assign a new one */
5831 if (__dev_get_by_index(net, dev->ifindex)) {
5832 int iflink = (dev->iflink == dev->ifindex);
5833 dev->ifindex = dev_new_index(net);
5835 dev->iflink = dev->ifindex;
5838 /* Fixup kobjects */
5839 err = device_rename(&dev->dev, dev->name);
5842 /* Add the device back in the hashes */
5843 list_netdevice(dev);
5845 /* Notify protocols, that a new device appeared. */
5846 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5849 * Prevent userspace races by waiting until the network
5850 * device is fully setup before sending notifications.
5852 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5859 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5861 static int dev_cpu_callback(struct notifier_block *nfb,
5862 unsigned long action,
5865 struct sk_buff **list_skb;
5866 struct sk_buff *skb;
5867 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5868 struct softnet_data *sd, *oldsd;
5870 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5873 local_irq_disable();
5874 cpu = smp_processor_id();
5875 sd = &per_cpu(softnet_data, cpu);
5876 oldsd = &per_cpu(softnet_data, oldcpu);
5878 /* Find end of our completion_queue. */
5879 list_skb = &sd->completion_queue;
5881 list_skb = &(*list_skb)->next;
5882 /* Append completion queue from offline CPU. */
5883 *list_skb = oldsd->completion_queue;
5884 oldsd->completion_queue = NULL;
5886 /* Append output queue from offline CPU. */
5887 if (oldsd->output_queue) {
5888 *sd->output_queue_tailp = oldsd->output_queue;
5889 sd->output_queue_tailp = oldsd->output_queue_tailp;
5890 oldsd->output_queue = NULL;
5891 oldsd->output_queue_tailp = &oldsd->output_queue;
5894 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5897 /* Process offline CPU's input_pkt_queue */
5898 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5900 input_queue_head_incr(oldsd);
5902 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5904 input_queue_head_incr(oldsd);
5912 * netdev_increment_features - increment feature set by one
5913 * @all: current feature set
5914 * @one: new feature set
5915 * @mask: mask feature set
5917 * Computes a new feature set after adding a device with feature set
5918 * @one to the master device with current feature set @all. Will not
5919 * enable anything that is off in @mask. Returns the new feature set.
5921 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5924 /* If device needs checksumming, downgrade to it. */
5925 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5926 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5927 else if (mask & NETIF_F_ALL_CSUM) {
5928 /* If one device supports v4/v6 checksumming, set for all. */
5929 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5930 !(all & NETIF_F_GEN_CSUM)) {
5931 all &= ~NETIF_F_ALL_CSUM;
5932 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5935 /* If one device supports hw checksumming, set for all. */
5936 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5937 all &= ~NETIF_F_ALL_CSUM;
5938 all |= NETIF_F_HW_CSUM;
5942 one |= NETIF_F_ALL_CSUM;
5944 one |= all & NETIF_F_ONE_FOR_ALL;
5945 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5946 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5950 EXPORT_SYMBOL(netdev_increment_features);
5952 static struct hlist_head *netdev_create_hash(void)
5955 struct hlist_head *hash;
5957 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5959 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5960 INIT_HLIST_HEAD(&hash[i]);
5965 /* Initialize per network namespace state */
5966 static int __net_init netdev_init(struct net *net)
5968 INIT_LIST_HEAD(&net->dev_base_head);
5970 net->dev_name_head = netdev_create_hash();
5971 if (net->dev_name_head == NULL)
5974 net->dev_index_head = netdev_create_hash();
5975 if (net->dev_index_head == NULL)
5981 kfree(net->dev_name_head);
5987 * netdev_drivername - network driver for the device
5988 * @dev: network device
5989 * @buffer: buffer for resulting name
5990 * @len: size of buffer
5992 * Determine network driver for device.
5994 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5996 const struct device_driver *driver;
5997 const struct device *parent;
5999 if (len <= 0 || !buffer)
6003 parent = dev->dev.parent;
6008 driver = parent->driver;
6009 if (driver && driver->name)
6010 strlcpy(buffer, driver->name, len);
6014 static int __netdev_printk(const char *level, const struct net_device *dev,
6015 struct va_format *vaf)
6019 if (dev && dev->dev.parent)
6020 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6021 netdev_name(dev), vaf);
6023 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6025 r = printk("%s(NULL net_device): %pV", level, vaf);
6030 int netdev_printk(const char *level, const struct net_device *dev,
6031 const char *format, ...)
6033 struct va_format vaf;
6037 va_start(args, format);
6042 r = __netdev_printk(level, dev, &vaf);
6047 EXPORT_SYMBOL(netdev_printk);
6049 #define define_netdev_printk_level(func, level) \
6050 int func(const struct net_device *dev, const char *fmt, ...) \
6053 struct va_format vaf; \
6056 va_start(args, fmt); \
6061 r = __netdev_printk(level, dev, &vaf); \
6066 EXPORT_SYMBOL(func);
6068 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6069 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6070 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6071 define_netdev_printk_level(netdev_err, KERN_ERR);
6072 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6073 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6074 define_netdev_printk_level(netdev_info, KERN_INFO);
6076 static void __net_exit netdev_exit(struct net *net)
6078 kfree(net->dev_name_head);
6079 kfree(net->dev_index_head);
6082 static struct pernet_operations __net_initdata netdev_net_ops = {
6083 .init = netdev_init,
6084 .exit = netdev_exit,
6087 static void __net_exit default_device_exit(struct net *net)
6089 struct net_device *dev, *aux;
6091 * Push all migratable network devices back to the
6092 * initial network namespace
6095 for_each_netdev_safe(net, dev, aux) {
6097 char fb_name[IFNAMSIZ];
6099 /* Ignore unmoveable devices (i.e. loopback) */
6100 if (dev->features & NETIF_F_NETNS_LOCAL)
6103 /* Leave virtual devices for the generic cleanup */
6104 if (dev->rtnl_link_ops)
6107 /* Push remaing network devices to init_net */
6108 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6109 err = dev_change_net_namespace(dev, &init_net, fb_name);
6111 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6112 __func__, dev->name, err);
6119 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6121 /* At exit all network devices most be removed from a network
6122 * namespace. Do this in the reverse order of registeration.
6123 * Do this across as many network namespaces as possible to
6124 * improve batching efficiency.
6126 struct net_device *dev;
6128 LIST_HEAD(dev_kill_list);
6131 list_for_each_entry(net, net_list, exit_list) {
6132 for_each_netdev_reverse(net, dev) {
6133 if (dev->rtnl_link_ops)
6134 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6136 unregister_netdevice_queue(dev, &dev_kill_list);
6139 unregister_netdevice_many(&dev_kill_list);
6143 static struct pernet_operations __net_initdata default_device_ops = {
6144 .exit = default_device_exit,
6145 .exit_batch = default_device_exit_batch,
6149 * Initialize the DEV module. At boot time this walks the device list and
6150 * unhooks any devices that fail to initialise (normally hardware not
6151 * present) and leaves us with a valid list of present and active devices.
6156 * This is called single threaded during boot, so no need
6157 * to take the rtnl semaphore.
6159 static int __init net_dev_init(void)
6161 int i, rc = -ENOMEM;
6163 BUG_ON(!dev_boot_phase);
6165 if (dev_proc_init())
6168 if (netdev_kobject_init())
6171 INIT_LIST_HEAD(&ptype_all);
6172 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6173 INIT_LIST_HEAD(&ptype_base[i]);
6175 if (register_pernet_subsys(&netdev_net_ops))
6179 * Initialise the packet receive queues.
6182 for_each_possible_cpu(i) {
6183 struct softnet_data *sd = &per_cpu(softnet_data, i);
6185 memset(sd, 0, sizeof(*sd));
6186 skb_queue_head_init(&sd->input_pkt_queue);
6187 skb_queue_head_init(&sd->process_queue);
6188 sd->completion_queue = NULL;
6189 INIT_LIST_HEAD(&sd->poll_list);
6190 sd->output_queue = NULL;
6191 sd->output_queue_tailp = &sd->output_queue;
6193 sd->csd.func = rps_trigger_softirq;
6199 sd->backlog.poll = process_backlog;
6200 sd->backlog.weight = weight_p;
6201 sd->backlog.gro_list = NULL;
6202 sd->backlog.gro_count = 0;
6207 /* The loopback device is special if any other network devices
6208 * is present in a network namespace the loopback device must
6209 * be present. Since we now dynamically allocate and free the
6210 * loopback device ensure this invariant is maintained by
6211 * keeping the loopback device as the first device on the
6212 * list of network devices. Ensuring the loopback devices
6213 * is the first device that appears and the last network device
6216 if (register_pernet_device(&loopback_net_ops))
6219 if (register_pernet_device(&default_device_ops))
6222 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6223 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6225 hotcpu_notifier(dev_cpu_callback, 0);
6233 subsys_initcall(net_dev_init);
6235 static int __init initialize_hashrnd(void)
6237 get_random_bytes(&hashrnd, sizeof(hashrnd));
6241 late_initcall_sync(initialize_hashrnd);