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)
1562 if (txq < 1 || txq > dev->num_tx_queues)
1565 if (dev->reg_state == NETREG_REGISTERED) {
1568 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1573 if (txq < dev->real_num_tx_queues)
1574 qdisc_reset_all_tx_gt(dev, txq);
1577 dev->real_num_tx_queues = txq;
1580 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1584 * netif_set_real_num_rx_queues - set actual number of RX queues used
1585 * @dev: Network device
1586 * @rxq: Actual number of RX queues
1588 * This must be called either with the rtnl_lock held or before
1589 * registration of the net device. Returns 0 on success, or a
1590 * negative error code. If called before registration, it always
1593 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1597 if (rxq < 1 || rxq > dev->num_rx_queues)
1600 if (dev->reg_state == NETREG_REGISTERED) {
1603 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1609 dev->real_num_rx_queues = rxq;
1612 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1615 static inline void __netif_reschedule(struct Qdisc *q)
1617 struct softnet_data *sd;
1618 unsigned long flags;
1620 local_irq_save(flags);
1621 sd = &__get_cpu_var(softnet_data);
1622 q->next_sched = NULL;
1623 *sd->output_queue_tailp = q;
1624 sd->output_queue_tailp = &q->next_sched;
1625 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1626 local_irq_restore(flags);
1629 void __netif_schedule(struct Qdisc *q)
1631 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1632 __netif_reschedule(q);
1634 EXPORT_SYMBOL(__netif_schedule);
1636 void dev_kfree_skb_irq(struct sk_buff *skb)
1638 if (atomic_dec_and_test(&skb->users)) {
1639 struct softnet_data *sd;
1640 unsigned long flags;
1642 local_irq_save(flags);
1643 sd = &__get_cpu_var(softnet_data);
1644 skb->next = sd->completion_queue;
1645 sd->completion_queue = skb;
1646 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1647 local_irq_restore(flags);
1650 EXPORT_SYMBOL(dev_kfree_skb_irq);
1652 void dev_kfree_skb_any(struct sk_buff *skb)
1654 if (in_irq() || irqs_disabled())
1655 dev_kfree_skb_irq(skb);
1659 EXPORT_SYMBOL(dev_kfree_skb_any);
1663 * netif_device_detach - mark device as removed
1664 * @dev: network device
1666 * Mark device as removed from system and therefore no longer available.
1668 void netif_device_detach(struct net_device *dev)
1670 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1671 netif_running(dev)) {
1672 netif_tx_stop_all_queues(dev);
1675 EXPORT_SYMBOL(netif_device_detach);
1678 * netif_device_attach - mark device as attached
1679 * @dev: network device
1681 * Mark device as attached from system and restart if needed.
1683 void netif_device_attach(struct net_device *dev)
1685 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1686 netif_running(dev)) {
1687 netif_tx_wake_all_queues(dev);
1688 __netdev_watchdog_up(dev);
1691 EXPORT_SYMBOL(netif_device_attach);
1693 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1695 return ((features & NETIF_F_NO_CSUM) ||
1696 ((features & NETIF_F_V4_CSUM) &&
1697 protocol == htons(ETH_P_IP)) ||
1698 ((features & NETIF_F_V6_CSUM) &&
1699 protocol == htons(ETH_P_IPV6)) ||
1700 ((features & NETIF_F_FCOE_CRC) &&
1701 protocol == htons(ETH_P_FCOE)));
1704 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1706 __be16 protocol = skb->protocol;
1707 int features = dev->features;
1709 if (vlan_tx_tag_present(skb)) {
1710 features &= dev->vlan_features;
1711 } else if (protocol == htons(ETH_P_8021Q)) {
1712 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1713 protocol = veh->h_vlan_encapsulated_proto;
1714 features &= dev->vlan_features;
1717 return can_checksum_protocol(features, protocol);
1721 * skb_dev_set -- assign a new device to a buffer
1722 * @skb: buffer for the new device
1723 * @dev: network device
1725 * If an skb is owned by a device already, we have to reset
1726 * all data private to the namespace a device belongs to
1727 * before assigning it a new device.
1729 #ifdef CONFIG_NET_NS
1730 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1733 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1736 skb_init_secmark(skb);
1740 skb->ipvs_property = 0;
1741 #ifdef CONFIG_NET_SCHED
1747 EXPORT_SYMBOL(skb_set_dev);
1748 #endif /* CONFIG_NET_NS */
1751 * Invalidate hardware checksum when packet is to be mangled, and
1752 * complete checksum manually on outgoing path.
1754 int skb_checksum_help(struct sk_buff *skb)
1757 int ret = 0, offset;
1759 if (skb->ip_summed == CHECKSUM_COMPLETE)
1760 goto out_set_summed;
1762 if (unlikely(skb_shinfo(skb)->gso_size)) {
1763 /* Let GSO fix up the checksum. */
1764 goto out_set_summed;
1767 offset = skb->csum_start - skb_headroom(skb);
1768 BUG_ON(offset >= skb_headlen(skb));
1769 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1771 offset += skb->csum_offset;
1772 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1774 if (skb_cloned(skb) &&
1775 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1776 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1781 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1783 skb->ip_summed = CHECKSUM_NONE;
1787 EXPORT_SYMBOL(skb_checksum_help);
1790 * skb_gso_segment - Perform segmentation on skb.
1791 * @skb: buffer to segment
1792 * @features: features for the output path (see dev->features)
1794 * This function segments the given skb and returns a list of segments.
1796 * It may return NULL if the skb requires no segmentation. This is
1797 * only possible when GSO is used for verifying header integrity.
1799 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1801 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1802 struct packet_type *ptype;
1803 __be16 type = skb->protocol;
1804 int vlan_depth = ETH_HLEN;
1807 while (type == htons(ETH_P_8021Q)) {
1808 struct vlan_hdr *vh;
1810 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1811 return ERR_PTR(-EINVAL);
1813 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1814 type = vh->h_vlan_encapsulated_proto;
1815 vlan_depth += VLAN_HLEN;
1818 skb_reset_mac_header(skb);
1819 skb->mac_len = skb->network_header - skb->mac_header;
1820 __skb_pull(skb, skb->mac_len);
1822 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1823 struct net_device *dev = skb->dev;
1824 struct ethtool_drvinfo info = {};
1826 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1827 dev->ethtool_ops->get_drvinfo(dev, &info);
1829 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1830 info.driver, dev ? dev->features : 0L,
1831 skb->sk ? skb->sk->sk_route_caps : 0L,
1832 skb->len, skb->data_len, skb->ip_summed);
1834 if (skb_header_cloned(skb) &&
1835 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1836 return ERR_PTR(err);
1840 list_for_each_entry_rcu(ptype,
1841 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1842 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1843 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1844 err = ptype->gso_send_check(skb);
1845 segs = ERR_PTR(err);
1846 if (err || skb_gso_ok(skb, features))
1848 __skb_push(skb, (skb->data -
1849 skb_network_header(skb)));
1851 segs = ptype->gso_segment(skb, features);
1857 __skb_push(skb, skb->data - skb_mac_header(skb));
1861 EXPORT_SYMBOL(skb_gso_segment);
1863 /* Take action when hardware reception checksum errors are detected. */
1865 void netdev_rx_csum_fault(struct net_device *dev)
1867 if (net_ratelimit()) {
1868 printk(KERN_ERR "%s: hw csum failure.\n",
1869 dev ? dev->name : "<unknown>");
1873 EXPORT_SYMBOL(netdev_rx_csum_fault);
1876 /* Actually, we should eliminate this check as soon as we know, that:
1877 * 1. IOMMU is present and allows to map all the memory.
1878 * 2. No high memory really exists on this machine.
1881 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1883 #ifdef CONFIG_HIGHMEM
1885 if (!(dev->features & NETIF_F_HIGHDMA)) {
1886 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1887 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1891 if (PCI_DMA_BUS_IS_PHYS) {
1892 struct device *pdev = dev->dev.parent;
1896 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1897 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1898 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1907 void (*destructor)(struct sk_buff *skb);
1910 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1912 static void dev_gso_skb_destructor(struct sk_buff *skb)
1914 struct dev_gso_cb *cb;
1917 struct sk_buff *nskb = skb->next;
1919 skb->next = nskb->next;
1922 } while (skb->next);
1924 cb = DEV_GSO_CB(skb);
1926 cb->destructor(skb);
1930 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1931 * @skb: buffer to segment
1933 * This function segments the given skb and stores the list of segments
1936 static int dev_gso_segment(struct sk_buff *skb)
1938 struct net_device *dev = skb->dev;
1939 struct sk_buff *segs;
1940 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1943 segs = skb_gso_segment(skb, features);
1945 /* Verifying header integrity only. */
1950 return PTR_ERR(segs);
1953 DEV_GSO_CB(skb)->destructor = skb->destructor;
1954 skb->destructor = dev_gso_skb_destructor;
1960 * Try to orphan skb early, right before transmission by the device.
1961 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1962 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1964 static inline void skb_orphan_try(struct sk_buff *skb)
1966 struct sock *sk = skb->sk;
1968 if (sk && !skb_shinfo(skb)->tx_flags) {
1969 /* skb_tx_hash() wont be able to get sk.
1970 * We copy sk_hash into skb->rxhash
1973 skb->rxhash = sk->sk_hash;
1978 int netif_get_vlan_features(struct sk_buff *skb, struct net_device *dev)
1980 __be16 protocol = skb->protocol;
1982 if (protocol == htons(ETH_P_8021Q)) {
1983 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1984 protocol = veh->h_vlan_encapsulated_proto;
1985 } else if (!skb->vlan_tci)
1986 return dev->features;
1988 if (protocol != htons(ETH_P_8021Q))
1989 return dev->features & dev->vlan_features;
1993 EXPORT_SYMBOL(netif_get_vlan_features);
1996 * Returns true if either:
1997 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1998 * 2. skb is fragmented and the device does not support SG, or if
1999 * at least one of fragments is in highmem and device does not
2000 * support DMA from it.
2002 static inline int skb_needs_linearize(struct sk_buff *skb,
2003 struct net_device *dev)
2005 if (skb_is_nonlinear(skb)) {
2006 int features = dev->features;
2008 if (vlan_tx_tag_present(skb))
2009 features &= dev->vlan_features;
2011 return (skb_has_frag_list(skb) &&
2012 !(features & NETIF_F_FRAGLIST)) ||
2013 (skb_shinfo(skb)->nr_frags &&
2014 (!(features & NETIF_F_SG) ||
2015 illegal_highdma(dev, skb)));
2021 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2022 struct netdev_queue *txq)
2024 const struct net_device_ops *ops = dev->netdev_ops;
2025 int rc = NETDEV_TX_OK;
2027 if (likely(!skb->next)) {
2028 if (!list_empty(&ptype_all))
2029 dev_queue_xmit_nit(skb, dev);
2032 * If device doesnt need skb->dst, release it right now while
2033 * its hot in this cpu cache
2035 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2038 skb_orphan_try(skb);
2040 if (vlan_tx_tag_present(skb) &&
2041 !(dev->features & NETIF_F_HW_VLAN_TX)) {
2042 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2049 if (netif_needs_gso(dev, skb)) {
2050 if (unlikely(dev_gso_segment(skb)))
2055 if (skb_needs_linearize(skb, dev) &&
2056 __skb_linearize(skb))
2059 /* If packet is not checksummed and device does not
2060 * support checksumming for this protocol, complete
2061 * checksumming here.
2063 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2064 skb_set_transport_header(skb, skb->csum_start -
2066 if (!dev_can_checksum(dev, skb) &&
2067 skb_checksum_help(skb))
2072 rc = ops->ndo_start_xmit(skb, dev);
2073 trace_net_dev_xmit(skb, rc);
2074 if (rc == NETDEV_TX_OK)
2075 txq_trans_update(txq);
2081 struct sk_buff *nskb = skb->next;
2083 skb->next = nskb->next;
2087 * If device doesnt need nskb->dst, release it right now while
2088 * its hot in this cpu cache
2090 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2093 rc = ops->ndo_start_xmit(nskb, dev);
2094 trace_net_dev_xmit(nskb, rc);
2095 if (unlikely(rc != NETDEV_TX_OK)) {
2096 if (rc & ~NETDEV_TX_MASK)
2097 goto out_kfree_gso_skb;
2098 nskb->next = skb->next;
2102 txq_trans_update(txq);
2103 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2104 return NETDEV_TX_BUSY;
2105 } while (skb->next);
2108 if (likely(skb->next == NULL))
2109 skb->destructor = DEV_GSO_CB(skb)->destructor;
2116 static u32 hashrnd __read_mostly;
2118 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
2122 if (skb_rx_queue_recorded(skb)) {
2123 hash = skb_get_rx_queue(skb);
2124 while (unlikely(hash >= dev->real_num_tx_queues))
2125 hash -= dev->real_num_tx_queues;
2129 if (skb->sk && skb->sk->sk_hash)
2130 hash = skb->sk->sk_hash;
2132 hash = (__force u16) skb->protocol ^ skb->rxhash;
2133 hash = jhash_1word(hash, hashrnd);
2135 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2137 EXPORT_SYMBOL(skb_tx_hash);
2139 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2141 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2142 if (net_ratelimit()) {
2143 pr_warning("%s selects TX queue %d, but "
2144 "real number of TX queues is %d\n",
2145 dev->name, queue_index, dev->real_num_tx_queues);
2152 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2155 struct xps_dev_maps *dev_maps;
2156 struct xps_map *map;
2157 int queue_index = -1;
2160 dev_maps = rcu_dereference(dev->xps_maps);
2162 map = rcu_dereference(
2163 dev_maps->cpu_map[raw_smp_processor_id()]);
2166 queue_index = map->queues[0];
2169 if (skb->sk && skb->sk->sk_hash)
2170 hash = skb->sk->sk_hash;
2172 hash = (__force u16) skb->protocol ^
2174 hash = jhash_1word(hash, hashrnd);
2175 queue_index = map->queues[
2176 ((u64)hash * map->len) >> 32];
2178 if (unlikely(queue_index >= dev->real_num_tx_queues))
2190 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2191 struct sk_buff *skb)
2194 const struct net_device_ops *ops = dev->netdev_ops;
2196 if (dev->real_num_tx_queues == 1)
2198 else if (ops->ndo_select_queue) {
2199 queue_index = ops->ndo_select_queue(dev, skb);
2200 queue_index = dev_cap_txqueue(dev, queue_index);
2202 struct sock *sk = skb->sk;
2203 queue_index = sk_tx_queue_get(sk);
2205 if (queue_index < 0 || skb->ooo_okay ||
2206 queue_index >= dev->real_num_tx_queues) {
2207 int old_index = queue_index;
2209 queue_index = get_xps_queue(dev, skb);
2210 if (queue_index < 0)
2211 queue_index = skb_tx_hash(dev, skb);
2213 if (queue_index != old_index && sk) {
2214 struct dst_entry *dst =
2215 rcu_dereference_check(sk->sk_dst_cache, 1);
2217 if (dst && skb_dst(skb) == dst)
2218 sk_tx_queue_set(sk, queue_index);
2223 skb_set_queue_mapping(skb, queue_index);
2224 return netdev_get_tx_queue(dev, queue_index);
2227 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2228 struct net_device *dev,
2229 struct netdev_queue *txq)
2231 spinlock_t *root_lock = qdisc_lock(q);
2232 bool contended = qdisc_is_running(q);
2236 * Heuristic to force contended enqueues to serialize on a
2237 * separate lock before trying to get qdisc main lock.
2238 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2239 * and dequeue packets faster.
2241 if (unlikely(contended))
2242 spin_lock(&q->busylock);
2244 spin_lock(root_lock);
2245 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2248 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2249 qdisc_run_begin(q)) {
2251 * This is a work-conserving queue; there are no old skbs
2252 * waiting to be sent out; and the qdisc is not running -
2253 * xmit the skb directly.
2255 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2257 __qdisc_update_bstats(q, skb->len);
2258 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2259 if (unlikely(contended)) {
2260 spin_unlock(&q->busylock);
2267 rc = NET_XMIT_SUCCESS;
2270 rc = qdisc_enqueue_root(skb, q);
2271 if (qdisc_run_begin(q)) {
2272 if (unlikely(contended)) {
2273 spin_unlock(&q->busylock);
2279 spin_unlock(root_lock);
2280 if (unlikely(contended))
2281 spin_unlock(&q->busylock);
2285 static DEFINE_PER_CPU(int, xmit_recursion);
2286 #define RECURSION_LIMIT 10
2289 * dev_queue_xmit - transmit a buffer
2290 * @skb: buffer to transmit
2292 * Queue a buffer for transmission to a network device. The caller must
2293 * have set the device and priority and built the buffer before calling
2294 * this function. The function can be called from an interrupt.
2296 * A negative errno code is returned on a failure. A success does not
2297 * guarantee the frame will be transmitted as it may be dropped due
2298 * to congestion or traffic shaping.
2300 * -----------------------------------------------------------------------------------
2301 * I notice this method can also return errors from the queue disciplines,
2302 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2305 * Regardless of the return value, the skb is consumed, so it is currently
2306 * difficult to retry a send to this method. (You can bump the ref count
2307 * before sending to hold a reference for retry if you are careful.)
2309 * When calling this method, interrupts MUST be enabled. This is because
2310 * the BH enable code must have IRQs enabled so that it will not deadlock.
2313 int dev_queue_xmit(struct sk_buff *skb)
2315 struct net_device *dev = skb->dev;
2316 struct netdev_queue *txq;
2320 /* Disable soft irqs for various locks below. Also
2321 * stops preemption for RCU.
2325 txq = dev_pick_tx(dev, skb);
2326 q = rcu_dereference_bh(txq->qdisc);
2328 #ifdef CONFIG_NET_CLS_ACT
2329 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2331 trace_net_dev_queue(skb);
2333 rc = __dev_xmit_skb(skb, q, dev, txq);
2337 /* The device has no queue. Common case for software devices:
2338 loopback, all the sorts of tunnels...
2340 Really, it is unlikely that netif_tx_lock protection is necessary
2341 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2343 However, it is possible, that they rely on protection
2346 Check this and shot the lock. It is not prone from deadlocks.
2347 Either shot noqueue qdisc, it is even simpler 8)
2349 if (dev->flags & IFF_UP) {
2350 int cpu = smp_processor_id(); /* ok because BHs are off */
2352 if (txq->xmit_lock_owner != cpu) {
2354 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2355 goto recursion_alert;
2357 HARD_TX_LOCK(dev, txq, cpu);
2359 if (!netif_tx_queue_stopped(txq)) {
2360 __this_cpu_inc(xmit_recursion);
2361 rc = dev_hard_start_xmit(skb, dev, txq);
2362 __this_cpu_dec(xmit_recursion);
2363 if (dev_xmit_complete(rc)) {
2364 HARD_TX_UNLOCK(dev, txq);
2368 HARD_TX_UNLOCK(dev, txq);
2369 if (net_ratelimit())
2370 printk(KERN_CRIT "Virtual device %s asks to "
2371 "queue packet!\n", dev->name);
2373 /* Recursion is detected! It is possible,
2377 if (net_ratelimit())
2378 printk(KERN_CRIT "Dead loop on virtual device "
2379 "%s, fix it urgently!\n", dev->name);
2384 rcu_read_unlock_bh();
2389 rcu_read_unlock_bh();
2392 EXPORT_SYMBOL(dev_queue_xmit);
2395 /*=======================================================================
2397 =======================================================================*/
2399 int netdev_max_backlog __read_mostly = 1000;
2400 int netdev_tstamp_prequeue __read_mostly = 1;
2401 int netdev_budget __read_mostly = 300;
2402 int weight_p __read_mostly = 64; /* old backlog weight */
2404 /* Called with irq disabled */
2405 static inline void ____napi_schedule(struct softnet_data *sd,
2406 struct napi_struct *napi)
2408 list_add_tail(&napi->poll_list, &sd->poll_list);
2409 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2413 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2414 * and src/dst port numbers. Returns a non-zero hash number on success
2417 __u32 __skb_get_rxhash(struct sk_buff *skb)
2419 int nhoff, hash = 0, poff;
2420 struct ipv6hdr *ip6;
2423 u32 addr1, addr2, ihl;
2429 nhoff = skb_network_offset(skb);
2431 switch (skb->protocol) {
2432 case __constant_htons(ETH_P_IP):
2433 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2436 ip = (struct iphdr *) (skb->data + nhoff);
2437 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2440 ip_proto = ip->protocol;
2441 addr1 = (__force u32) ip->saddr;
2442 addr2 = (__force u32) ip->daddr;
2445 case __constant_htons(ETH_P_IPV6):
2446 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2449 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2450 ip_proto = ip6->nexthdr;
2451 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2452 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2460 poff = proto_ports_offset(ip_proto);
2462 nhoff += ihl * 4 + poff;
2463 if (pskb_may_pull(skb, nhoff + 4)) {
2464 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2465 if (ports.v16[1] < ports.v16[0])
2466 swap(ports.v16[0], ports.v16[1]);
2470 /* get a consistent hash (same value on both flow directions) */
2474 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2481 EXPORT_SYMBOL(__skb_get_rxhash);
2485 /* One global table that all flow-based protocols share. */
2486 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2487 EXPORT_SYMBOL(rps_sock_flow_table);
2490 * get_rps_cpu is called from netif_receive_skb and returns the target
2491 * CPU from the RPS map of the receiving queue for a given skb.
2492 * rcu_read_lock must be held on entry.
2494 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2495 struct rps_dev_flow **rflowp)
2497 struct netdev_rx_queue *rxqueue;
2498 struct rps_map *map;
2499 struct rps_dev_flow_table *flow_table;
2500 struct rps_sock_flow_table *sock_flow_table;
2504 if (skb_rx_queue_recorded(skb)) {
2505 u16 index = skb_get_rx_queue(skb);
2506 if (unlikely(index >= dev->real_num_rx_queues)) {
2507 WARN_ONCE(dev->real_num_rx_queues > 1,
2508 "%s received packet on queue %u, but number "
2509 "of RX queues is %u\n",
2510 dev->name, index, dev->real_num_rx_queues);
2513 rxqueue = dev->_rx + index;
2517 map = rcu_dereference(rxqueue->rps_map);
2519 if (map->len == 1) {
2520 tcpu = map->cpus[0];
2521 if (cpu_online(tcpu))
2525 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2529 skb_reset_network_header(skb);
2530 if (!skb_get_rxhash(skb))
2533 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2534 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2535 if (flow_table && sock_flow_table) {
2537 struct rps_dev_flow *rflow;
2539 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2542 next_cpu = sock_flow_table->ents[skb->rxhash &
2543 sock_flow_table->mask];
2546 * If the desired CPU (where last recvmsg was done) is
2547 * different from current CPU (one in the rx-queue flow
2548 * table entry), switch if one of the following holds:
2549 * - Current CPU is unset (equal to RPS_NO_CPU).
2550 * - Current CPU is offline.
2551 * - The current CPU's queue tail has advanced beyond the
2552 * last packet that was enqueued using this table entry.
2553 * This guarantees that all previous packets for the flow
2554 * have been dequeued, thus preserving in order delivery.
2556 if (unlikely(tcpu != next_cpu) &&
2557 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2558 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2559 rflow->last_qtail)) >= 0)) {
2560 tcpu = rflow->cpu = next_cpu;
2561 if (tcpu != RPS_NO_CPU)
2562 rflow->last_qtail = per_cpu(softnet_data,
2563 tcpu).input_queue_head;
2565 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2573 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2575 if (cpu_online(tcpu)) {
2585 /* Called from hardirq (IPI) context */
2586 static void rps_trigger_softirq(void *data)
2588 struct softnet_data *sd = data;
2590 ____napi_schedule(sd, &sd->backlog);
2594 #endif /* CONFIG_RPS */
2597 * Check if this softnet_data structure is another cpu one
2598 * If yes, queue it to our IPI list and return 1
2601 static int rps_ipi_queued(struct softnet_data *sd)
2604 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2607 sd->rps_ipi_next = mysd->rps_ipi_list;
2608 mysd->rps_ipi_list = sd;
2610 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2613 #endif /* CONFIG_RPS */
2618 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2619 * queue (may be a remote CPU queue).
2621 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2622 unsigned int *qtail)
2624 struct softnet_data *sd;
2625 unsigned long flags;
2627 sd = &per_cpu(softnet_data, cpu);
2629 local_irq_save(flags);
2632 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2633 if (skb_queue_len(&sd->input_pkt_queue)) {
2635 __skb_queue_tail(&sd->input_pkt_queue, skb);
2636 input_queue_tail_incr_save(sd, qtail);
2638 local_irq_restore(flags);
2639 return NET_RX_SUCCESS;
2642 /* Schedule NAPI for backlog device
2643 * We can use non atomic operation since we own the queue lock
2645 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2646 if (!rps_ipi_queued(sd))
2647 ____napi_schedule(sd, &sd->backlog);
2655 local_irq_restore(flags);
2657 atomic_long_inc(&skb->dev->rx_dropped);
2663 * netif_rx - post buffer to the network code
2664 * @skb: buffer to post
2666 * This function receives a packet from a device driver and queues it for
2667 * the upper (protocol) levels to process. It always succeeds. The buffer
2668 * may be dropped during processing for congestion control or by the
2672 * NET_RX_SUCCESS (no congestion)
2673 * NET_RX_DROP (packet was dropped)
2677 int netif_rx(struct sk_buff *skb)
2681 /* if netpoll wants it, pretend we never saw it */
2682 if (netpoll_rx(skb))
2685 if (netdev_tstamp_prequeue)
2686 net_timestamp_check(skb);
2688 trace_netif_rx(skb);
2691 struct rps_dev_flow voidflow, *rflow = &voidflow;
2697 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2699 cpu = smp_processor_id();
2701 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2709 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2715 EXPORT_SYMBOL(netif_rx);
2717 int netif_rx_ni(struct sk_buff *skb)
2722 err = netif_rx(skb);
2723 if (local_softirq_pending())
2729 EXPORT_SYMBOL(netif_rx_ni);
2731 static void net_tx_action(struct softirq_action *h)
2733 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2735 if (sd->completion_queue) {
2736 struct sk_buff *clist;
2738 local_irq_disable();
2739 clist = sd->completion_queue;
2740 sd->completion_queue = NULL;
2744 struct sk_buff *skb = clist;
2745 clist = clist->next;
2747 WARN_ON(atomic_read(&skb->users));
2748 trace_kfree_skb(skb, net_tx_action);
2753 if (sd->output_queue) {
2756 local_irq_disable();
2757 head = sd->output_queue;
2758 sd->output_queue = NULL;
2759 sd->output_queue_tailp = &sd->output_queue;
2763 struct Qdisc *q = head;
2764 spinlock_t *root_lock;
2766 head = head->next_sched;
2768 root_lock = qdisc_lock(q);
2769 if (spin_trylock(root_lock)) {
2770 smp_mb__before_clear_bit();
2771 clear_bit(__QDISC_STATE_SCHED,
2774 spin_unlock(root_lock);
2776 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2778 __netif_reschedule(q);
2780 smp_mb__before_clear_bit();
2781 clear_bit(__QDISC_STATE_SCHED,
2789 static inline int deliver_skb(struct sk_buff *skb,
2790 struct packet_type *pt_prev,
2791 struct net_device *orig_dev)
2793 atomic_inc(&skb->users);
2794 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2797 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2798 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2799 /* This hook is defined here for ATM LANE */
2800 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2801 unsigned char *addr) __read_mostly;
2802 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2805 #ifdef CONFIG_NET_CLS_ACT
2806 /* TODO: Maybe we should just force sch_ingress to be compiled in
2807 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2808 * a compare and 2 stores extra right now if we dont have it on
2809 * but have CONFIG_NET_CLS_ACT
2810 * NOTE: This doesnt stop any functionality; if you dont have
2811 * the ingress scheduler, you just cant add policies on ingress.
2814 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2816 struct net_device *dev = skb->dev;
2817 u32 ttl = G_TC_RTTL(skb->tc_verd);
2818 int result = TC_ACT_OK;
2821 if (unlikely(MAX_RED_LOOP < ttl++)) {
2822 if (net_ratelimit())
2823 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2824 skb->skb_iif, dev->ifindex);
2828 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2829 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2832 if (q != &noop_qdisc) {
2833 spin_lock(qdisc_lock(q));
2834 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2835 result = qdisc_enqueue_root(skb, q);
2836 spin_unlock(qdisc_lock(q));
2842 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2843 struct packet_type **pt_prev,
2844 int *ret, struct net_device *orig_dev)
2846 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
2848 if (!rxq || rxq->qdisc == &noop_qdisc)
2852 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2856 switch (ing_filter(skb, rxq)) {
2870 * netdev_rx_handler_register - register receive handler
2871 * @dev: device to register a handler for
2872 * @rx_handler: receive handler to register
2873 * @rx_handler_data: data pointer that is used by rx handler
2875 * Register a receive hander for a device. This handler will then be
2876 * called from __netif_receive_skb. A negative errno code is returned
2879 * The caller must hold the rtnl_mutex.
2881 int netdev_rx_handler_register(struct net_device *dev,
2882 rx_handler_func_t *rx_handler,
2883 void *rx_handler_data)
2887 if (dev->rx_handler)
2890 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2891 rcu_assign_pointer(dev->rx_handler, rx_handler);
2895 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2898 * netdev_rx_handler_unregister - unregister receive handler
2899 * @dev: device to unregister a handler from
2901 * Unregister a receive hander from a device.
2903 * The caller must hold the rtnl_mutex.
2905 void netdev_rx_handler_unregister(struct net_device *dev)
2909 rcu_assign_pointer(dev->rx_handler, NULL);
2910 rcu_assign_pointer(dev->rx_handler_data, NULL);
2912 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2914 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2915 struct net_device *master)
2917 if (skb->pkt_type == PACKET_HOST) {
2918 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2920 memcpy(dest, master->dev_addr, ETH_ALEN);
2924 /* On bonding slaves other than the currently active slave, suppress
2925 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2926 * ARP on active-backup slaves with arp_validate enabled.
2928 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2930 struct net_device *dev = skb->dev;
2932 if (master->priv_flags & IFF_MASTER_ARPMON)
2933 dev->last_rx = jiffies;
2935 if ((master->priv_flags & IFF_MASTER_ALB) &&
2936 (master->priv_flags & IFF_BRIDGE_PORT)) {
2937 /* Do address unmangle. The local destination address
2938 * will be always the one master has. Provides the right
2939 * functionality in a bridge.
2941 skb_bond_set_mac_by_master(skb, master);
2944 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2945 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2946 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2949 if (master->priv_flags & IFF_MASTER_ALB) {
2950 if (skb->pkt_type != PACKET_BROADCAST &&
2951 skb->pkt_type != PACKET_MULTICAST)
2954 if (master->priv_flags & IFF_MASTER_8023AD &&
2955 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2962 EXPORT_SYMBOL(__skb_bond_should_drop);
2964 static int __netif_receive_skb(struct sk_buff *skb)
2966 struct packet_type *ptype, *pt_prev;
2967 rx_handler_func_t *rx_handler;
2968 struct net_device *orig_dev;
2969 struct net_device *master;
2970 struct net_device *null_or_orig;
2971 struct net_device *orig_or_bond;
2972 int ret = NET_RX_DROP;
2975 if (!netdev_tstamp_prequeue)
2976 net_timestamp_check(skb);
2978 trace_netif_receive_skb(skb);
2980 /* if we've gotten here through NAPI, check netpoll */
2981 if (netpoll_receive_skb(skb))
2985 skb->skb_iif = skb->dev->ifindex;
2988 * bonding note: skbs received on inactive slaves should only
2989 * be delivered to pkt handlers that are exact matches. Also
2990 * the deliver_no_wcard flag will be set. If packet handlers
2991 * are sensitive to duplicate packets these skbs will need to
2992 * be dropped at the handler.
2994 null_or_orig = NULL;
2995 orig_dev = skb->dev;
2996 master = ACCESS_ONCE(orig_dev->master);
2997 if (skb->deliver_no_wcard)
2998 null_or_orig = orig_dev;
3000 if (skb_bond_should_drop(skb, master)) {
3001 skb->deliver_no_wcard = 1;
3002 null_or_orig = orig_dev; /* deliver only exact match */
3007 __this_cpu_inc(softnet_data.processed);
3008 skb_reset_network_header(skb);
3009 skb_reset_transport_header(skb);
3010 skb->mac_len = skb->network_header - skb->mac_header;
3016 #ifdef CONFIG_NET_CLS_ACT
3017 if (skb->tc_verd & TC_NCLS) {
3018 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3023 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3024 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
3025 ptype->dev == orig_dev) {
3027 ret = deliver_skb(skb, pt_prev, orig_dev);
3032 #ifdef CONFIG_NET_CLS_ACT
3033 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3039 /* Handle special case of bridge or macvlan */
3040 rx_handler = rcu_dereference(skb->dev->rx_handler);
3043 ret = deliver_skb(skb, pt_prev, orig_dev);
3046 skb = rx_handler(skb);
3051 if (vlan_tx_tag_present(skb)) {
3053 ret = deliver_skb(skb, pt_prev, orig_dev);
3056 if (vlan_hwaccel_do_receive(&skb)) {
3057 ret = __netif_receive_skb(skb);
3059 } else if (unlikely(!skb))
3064 * Make sure frames received on VLAN interfaces stacked on
3065 * bonding interfaces still make their way to any base bonding
3066 * device that may have registered for a specific ptype. The
3067 * handler may have to adjust skb->dev and orig_dev.
3069 orig_or_bond = orig_dev;
3070 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
3071 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
3072 orig_or_bond = vlan_dev_real_dev(skb->dev);
3075 type = skb->protocol;
3076 list_for_each_entry_rcu(ptype,
3077 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3078 if (ptype->type == type && (ptype->dev == null_or_orig ||
3079 ptype->dev == skb->dev || ptype->dev == orig_dev ||
3080 ptype->dev == orig_or_bond)) {
3082 ret = deliver_skb(skb, pt_prev, orig_dev);
3088 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3090 atomic_long_inc(&skb->dev->rx_dropped);
3092 /* Jamal, now you will not able to escape explaining
3093 * me how you were going to use this. :-)
3104 * netif_receive_skb - process receive buffer from network
3105 * @skb: buffer to process
3107 * netif_receive_skb() is the main receive data processing function.
3108 * It always succeeds. The buffer may be dropped during processing
3109 * for congestion control or by the protocol layers.
3111 * This function may only be called from softirq context and interrupts
3112 * should be enabled.
3114 * Return values (usually ignored):
3115 * NET_RX_SUCCESS: no congestion
3116 * NET_RX_DROP: packet was dropped
3118 int netif_receive_skb(struct sk_buff *skb)
3120 if (netdev_tstamp_prequeue)
3121 net_timestamp_check(skb);
3123 if (skb_defer_rx_timestamp(skb))
3124 return NET_RX_SUCCESS;
3128 struct rps_dev_flow voidflow, *rflow = &voidflow;
3133 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3136 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3140 ret = __netif_receive_skb(skb);
3146 return __netif_receive_skb(skb);
3149 EXPORT_SYMBOL(netif_receive_skb);
3151 /* Network device is going away, flush any packets still pending
3152 * Called with irqs disabled.
3154 static void flush_backlog(void *arg)
3156 struct net_device *dev = arg;
3157 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3158 struct sk_buff *skb, *tmp;
3161 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3162 if (skb->dev == dev) {
3163 __skb_unlink(skb, &sd->input_pkt_queue);
3165 input_queue_head_incr(sd);
3170 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3171 if (skb->dev == dev) {
3172 __skb_unlink(skb, &sd->process_queue);
3174 input_queue_head_incr(sd);
3179 static int napi_gro_complete(struct sk_buff *skb)
3181 struct packet_type *ptype;
3182 __be16 type = skb->protocol;
3183 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3186 if (NAPI_GRO_CB(skb)->count == 1) {
3187 skb_shinfo(skb)->gso_size = 0;
3192 list_for_each_entry_rcu(ptype, head, list) {
3193 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3196 err = ptype->gro_complete(skb);
3202 WARN_ON(&ptype->list == head);
3204 return NET_RX_SUCCESS;
3208 return netif_receive_skb(skb);
3211 inline void napi_gro_flush(struct napi_struct *napi)
3213 struct sk_buff *skb, *next;
3215 for (skb = napi->gro_list; skb; skb = next) {
3218 napi_gro_complete(skb);
3221 napi->gro_count = 0;
3222 napi->gro_list = NULL;
3224 EXPORT_SYMBOL(napi_gro_flush);
3226 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3228 struct sk_buff **pp = NULL;
3229 struct packet_type *ptype;
3230 __be16 type = skb->protocol;
3231 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3234 enum gro_result ret;
3236 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3239 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3243 list_for_each_entry_rcu(ptype, head, list) {
3244 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3247 skb_set_network_header(skb, skb_gro_offset(skb));
3248 mac_len = skb->network_header - skb->mac_header;
3249 skb->mac_len = mac_len;
3250 NAPI_GRO_CB(skb)->same_flow = 0;
3251 NAPI_GRO_CB(skb)->flush = 0;
3252 NAPI_GRO_CB(skb)->free = 0;
3254 pp = ptype->gro_receive(&napi->gro_list, skb);
3259 if (&ptype->list == head)
3262 same_flow = NAPI_GRO_CB(skb)->same_flow;
3263 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3266 struct sk_buff *nskb = *pp;
3270 napi_gro_complete(nskb);
3277 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3281 NAPI_GRO_CB(skb)->count = 1;
3282 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3283 skb->next = napi->gro_list;
3284 napi->gro_list = skb;
3288 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3289 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3291 BUG_ON(skb->end - skb->tail < grow);
3293 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3296 skb->data_len -= grow;
3298 skb_shinfo(skb)->frags[0].page_offset += grow;
3299 skb_shinfo(skb)->frags[0].size -= grow;
3301 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3302 put_page(skb_shinfo(skb)->frags[0].page);
3303 memmove(skb_shinfo(skb)->frags,
3304 skb_shinfo(skb)->frags + 1,
3305 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3316 EXPORT_SYMBOL(dev_gro_receive);
3318 static inline gro_result_t
3319 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3323 for (p = napi->gro_list; p; p = p->next) {
3324 unsigned long diffs;
3326 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3327 diffs |= p->vlan_tci ^ skb->vlan_tci;
3328 diffs |= compare_ether_header(skb_mac_header(p),
3329 skb_gro_mac_header(skb));
3330 NAPI_GRO_CB(p)->same_flow = !diffs;
3331 NAPI_GRO_CB(p)->flush = 0;
3334 return dev_gro_receive(napi, skb);
3337 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3341 if (netif_receive_skb(skb))
3346 case GRO_MERGED_FREE:
3357 EXPORT_SYMBOL(napi_skb_finish);
3359 void skb_gro_reset_offset(struct sk_buff *skb)
3361 NAPI_GRO_CB(skb)->data_offset = 0;
3362 NAPI_GRO_CB(skb)->frag0 = NULL;
3363 NAPI_GRO_CB(skb)->frag0_len = 0;
3365 if (skb->mac_header == skb->tail &&
3366 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3367 NAPI_GRO_CB(skb)->frag0 =
3368 page_address(skb_shinfo(skb)->frags[0].page) +
3369 skb_shinfo(skb)->frags[0].page_offset;
3370 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3373 EXPORT_SYMBOL(skb_gro_reset_offset);
3375 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3377 skb_gro_reset_offset(skb);
3379 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3381 EXPORT_SYMBOL(napi_gro_receive);
3383 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3385 __skb_pull(skb, skb_headlen(skb));
3386 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3392 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3394 struct sk_buff *skb = napi->skb;
3397 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3403 EXPORT_SYMBOL(napi_get_frags);
3405 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3411 skb->protocol = eth_type_trans(skb, skb->dev);
3413 if (ret == GRO_HELD)
3414 skb_gro_pull(skb, -ETH_HLEN);
3415 else if (netif_receive_skb(skb))
3420 case GRO_MERGED_FREE:
3421 napi_reuse_skb(napi, skb);
3430 EXPORT_SYMBOL(napi_frags_finish);
3432 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3434 struct sk_buff *skb = napi->skb;
3441 skb_reset_mac_header(skb);
3442 skb_gro_reset_offset(skb);
3444 off = skb_gro_offset(skb);
3445 hlen = off + sizeof(*eth);
3446 eth = skb_gro_header_fast(skb, off);
3447 if (skb_gro_header_hard(skb, hlen)) {
3448 eth = skb_gro_header_slow(skb, hlen, off);
3449 if (unlikely(!eth)) {
3450 napi_reuse_skb(napi, skb);
3456 skb_gro_pull(skb, sizeof(*eth));
3459 * This works because the only protocols we care about don't require
3460 * special handling. We'll fix it up properly at the end.
3462 skb->protocol = eth->h_proto;
3467 EXPORT_SYMBOL(napi_frags_skb);
3469 gro_result_t napi_gro_frags(struct napi_struct *napi)
3471 struct sk_buff *skb = napi_frags_skb(napi);
3476 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3478 EXPORT_SYMBOL(napi_gro_frags);
3481 * net_rps_action sends any pending IPI's for rps.
3482 * Note: called with local irq disabled, but exits with local irq enabled.
3484 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3487 struct softnet_data *remsd = sd->rps_ipi_list;
3490 sd->rps_ipi_list = NULL;
3494 /* Send pending IPI's to kick RPS processing on remote cpus. */
3496 struct softnet_data *next = remsd->rps_ipi_next;
3498 if (cpu_online(remsd->cpu))
3499 __smp_call_function_single(remsd->cpu,
3508 static int process_backlog(struct napi_struct *napi, int quota)
3511 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3514 /* Check if we have pending ipi, its better to send them now,
3515 * not waiting net_rx_action() end.
3517 if (sd->rps_ipi_list) {
3518 local_irq_disable();
3519 net_rps_action_and_irq_enable(sd);
3522 napi->weight = weight_p;
3523 local_irq_disable();
3524 while (work < quota) {
3525 struct sk_buff *skb;
3528 while ((skb = __skb_dequeue(&sd->process_queue))) {
3530 __netif_receive_skb(skb);
3531 local_irq_disable();
3532 input_queue_head_incr(sd);
3533 if (++work >= quota) {
3540 qlen = skb_queue_len(&sd->input_pkt_queue);
3542 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3543 &sd->process_queue);
3545 if (qlen < quota - work) {
3547 * Inline a custom version of __napi_complete().
3548 * only current cpu owns and manipulates this napi,
3549 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3550 * we can use a plain write instead of clear_bit(),
3551 * and we dont need an smp_mb() memory barrier.
3553 list_del(&napi->poll_list);
3556 quota = work + qlen;
3566 * __napi_schedule - schedule for receive
3567 * @n: entry to schedule
3569 * The entry's receive function will be scheduled to run
3571 void __napi_schedule(struct napi_struct *n)
3573 unsigned long flags;
3575 local_irq_save(flags);
3576 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3577 local_irq_restore(flags);
3579 EXPORT_SYMBOL(__napi_schedule);
3581 void __napi_complete(struct napi_struct *n)
3583 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3584 BUG_ON(n->gro_list);
3586 list_del(&n->poll_list);
3587 smp_mb__before_clear_bit();
3588 clear_bit(NAPI_STATE_SCHED, &n->state);
3590 EXPORT_SYMBOL(__napi_complete);
3592 void napi_complete(struct napi_struct *n)
3594 unsigned long flags;
3597 * don't let napi dequeue from the cpu poll list
3598 * just in case its running on a different cpu
3600 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3604 local_irq_save(flags);
3606 local_irq_restore(flags);
3608 EXPORT_SYMBOL(napi_complete);
3610 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3611 int (*poll)(struct napi_struct *, int), int weight)
3613 INIT_LIST_HEAD(&napi->poll_list);
3614 napi->gro_count = 0;
3615 napi->gro_list = NULL;
3618 napi->weight = weight;
3619 list_add(&napi->dev_list, &dev->napi_list);
3621 #ifdef CONFIG_NETPOLL
3622 spin_lock_init(&napi->poll_lock);
3623 napi->poll_owner = -1;
3625 set_bit(NAPI_STATE_SCHED, &napi->state);
3627 EXPORT_SYMBOL(netif_napi_add);
3629 void netif_napi_del(struct napi_struct *napi)
3631 struct sk_buff *skb, *next;
3633 list_del_init(&napi->dev_list);
3634 napi_free_frags(napi);
3636 for (skb = napi->gro_list; skb; skb = next) {
3642 napi->gro_list = NULL;
3643 napi->gro_count = 0;
3645 EXPORT_SYMBOL(netif_napi_del);
3647 static void net_rx_action(struct softirq_action *h)
3649 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3650 unsigned long time_limit = jiffies + 2;
3651 int budget = netdev_budget;
3654 local_irq_disable();
3656 while (!list_empty(&sd->poll_list)) {
3657 struct napi_struct *n;
3660 /* If softirq window is exhuasted then punt.
3661 * Allow this to run for 2 jiffies since which will allow
3662 * an average latency of 1.5/HZ.
3664 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3669 /* Even though interrupts have been re-enabled, this
3670 * access is safe because interrupts can only add new
3671 * entries to the tail of this list, and only ->poll()
3672 * calls can remove this head entry from the list.
3674 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3676 have = netpoll_poll_lock(n);
3680 /* This NAPI_STATE_SCHED test is for avoiding a race
3681 * with netpoll's poll_napi(). Only the entity which
3682 * obtains the lock and sees NAPI_STATE_SCHED set will
3683 * actually make the ->poll() call. Therefore we avoid
3684 * accidently calling ->poll() when NAPI is not scheduled.
3687 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3688 work = n->poll(n, weight);
3692 WARN_ON_ONCE(work > weight);
3696 local_irq_disable();
3698 /* Drivers must not modify the NAPI state if they
3699 * consume the entire weight. In such cases this code
3700 * still "owns" the NAPI instance and therefore can
3701 * move the instance around on the list at-will.
3703 if (unlikely(work == weight)) {
3704 if (unlikely(napi_disable_pending(n))) {
3707 local_irq_disable();
3709 list_move_tail(&n->poll_list, &sd->poll_list);
3712 netpoll_poll_unlock(have);
3715 net_rps_action_and_irq_enable(sd);
3717 #ifdef CONFIG_NET_DMA
3719 * There may not be any more sk_buffs coming right now, so push
3720 * any pending DMA copies to hardware
3722 dma_issue_pending_all();
3729 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3733 static gifconf_func_t *gifconf_list[NPROTO];
3736 * register_gifconf - register a SIOCGIF handler
3737 * @family: Address family
3738 * @gifconf: Function handler
3740 * Register protocol dependent address dumping routines. The handler
3741 * that is passed must not be freed or reused until it has been replaced
3742 * by another handler.
3744 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3746 if (family >= NPROTO)
3748 gifconf_list[family] = gifconf;
3751 EXPORT_SYMBOL(register_gifconf);
3755 * Map an interface index to its name (SIOCGIFNAME)
3759 * We need this ioctl for efficient implementation of the
3760 * if_indextoname() function required by the IPv6 API. Without
3761 * it, we would have to search all the interfaces to find a
3765 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3767 struct net_device *dev;
3771 * Fetch the caller's info block.
3774 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3778 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3784 strcpy(ifr.ifr_name, dev->name);
3787 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3793 * Perform a SIOCGIFCONF call. This structure will change
3794 * size eventually, and there is nothing I can do about it.
3795 * Thus we will need a 'compatibility mode'.
3798 static int dev_ifconf(struct net *net, char __user *arg)
3801 struct net_device *dev;
3808 * Fetch the caller's info block.
3811 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3818 * Loop over the interfaces, and write an info block for each.
3822 for_each_netdev(net, dev) {
3823 for (i = 0; i < NPROTO; i++) {
3824 if (gifconf_list[i]) {
3827 done = gifconf_list[i](dev, NULL, 0);
3829 done = gifconf_list[i](dev, pos + total,
3839 * All done. Write the updated control block back to the caller.
3841 ifc.ifc_len = total;
3844 * Both BSD and Solaris return 0 here, so we do too.
3846 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3849 #ifdef CONFIG_PROC_FS
3851 * This is invoked by the /proc filesystem handler to display a device
3854 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3857 struct net *net = seq_file_net(seq);
3859 struct net_device *dev;
3863 return SEQ_START_TOKEN;
3866 for_each_netdev_rcu(net, dev)
3873 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3875 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3876 first_net_device(seq_file_net(seq)) :
3877 next_net_device((struct net_device *)v);
3880 return rcu_dereference(dev);
3883 void dev_seq_stop(struct seq_file *seq, void *v)
3889 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3891 struct rtnl_link_stats64 temp;
3892 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3894 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3895 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3896 dev->name, stats->rx_bytes, stats->rx_packets,
3898 stats->rx_dropped + stats->rx_missed_errors,
3899 stats->rx_fifo_errors,
3900 stats->rx_length_errors + stats->rx_over_errors +
3901 stats->rx_crc_errors + stats->rx_frame_errors,
3902 stats->rx_compressed, stats->multicast,
3903 stats->tx_bytes, stats->tx_packets,
3904 stats->tx_errors, stats->tx_dropped,
3905 stats->tx_fifo_errors, stats->collisions,
3906 stats->tx_carrier_errors +
3907 stats->tx_aborted_errors +
3908 stats->tx_window_errors +
3909 stats->tx_heartbeat_errors,
3910 stats->tx_compressed);
3914 * Called from the PROCfs module. This now uses the new arbitrary sized
3915 * /proc/net interface to create /proc/net/dev
3917 static int dev_seq_show(struct seq_file *seq, void *v)
3919 if (v == SEQ_START_TOKEN)
3920 seq_puts(seq, "Inter-| Receive "
3922 " face |bytes packets errs drop fifo frame "
3923 "compressed multicast|bytes packets errs "
3924 "drop fifo colls carrier compressed\n");
3926 dev_seq_printf_stats(seq, v);
3930 static struct softnet_data *softnet_get_online(loff_t *pos)
3932 struct softnet_data *sd = NULL;
3934 while (*pos < nr_cpu_ids)
3935 if (cpu_online(*pos)) {
3936 sd = &per_cpu(softnet_data, *pos);
3943 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3945 return softnet_get_online(pos);
3948 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3951 return softnet_get_online(pos);
3954 static void softnet_seq_stop(struct seq_file *seq, void *v)
3958 static int softnet_seq_show(struct seq_file *seq, void *v)
3960 struct softnet_data *sd = v;
3962 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3963 sd->processed, sd->dropped, sd->time_squeeze, 0,
3964 0, 0, 0, 0, /* was fastroute */
3965 sd->cpu_collision, sd->received_rps);
3969 static const struct seq_operations dev_seq_ops = {
3970 .start = dev_seq_start,
3971 .next = dev_seq_next,
3972 .stop = dev_seq_stop,
3973 .show = dev_seq_show,
3976 static int dev_seq_open(struct inode *inode, struct file *file)
3978 return seq_open_net(inode, file, &dev_seq_ops,
3979 sizeof(struct seq_net_private));
3982 static const struct file_operations dev_seq_fops = {
3983 .owner = THIS_MODULE,
3984 .open = dev_seq_open,
3986 .llseek = seq_lseek,
3987 .release = seq_release_net,
3990 static const struct seq_operations softnet_seq_ops = {
3991 .start = softnet_seq_start,
3992 .next = softnet_seq_next,
3993 .stop = softnet_seq_stop,
3994 .show = softnet_seq_show,
3997 static int softnet_seq_open(struct inode *inode, struct file *file)
3999 return seq_open(file, &softnet_seq_ops);
4002 static const struct file_operations softnet_seq_fops = {
4003 .owner = THIS_MODULE,
4004 .open = softnet_seq_open,
4006 .llseek = seq_lseek,
4007 .release = seq_release,
4010 static void *ptype_get_idx(loff_t pos)
4012 struct packet_type *pt = NULL;
4016 list_for_each_entry_rcu(pt, &ptype_all, list) {
4022 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4023 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4032 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4036 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4039 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4041 struct packet_type *pt;
4042 struct list_head *nxt;
4046 if (v == SEQ_START_TOKEN)
4047 return ptype_get_idx(0);
4050 nxt = pt->list.next;
4051 if (pt->type == htons(ETH_P_ALL)) {
4052 if (nxt != &ptype_all)
4055 nxt = ptype_base[0].next;
4057 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4059 while (nxt == &ptype_base[hash]) {
4060 if (++hash >= PTYPE_HASH_SIZE)
4062 nxt = ptype_base[hash].next;
4065 return list_entry(nxt, struct packet_type, list);
4068 static void ptype_seq_stop(struct seq_file *seq, void *v)
4074 static int ptype_seq_show(struct seq_file *seq, void *v)
4076 struct packet_type *pt = v;
4078 if (v == SEQ_START_TOKEN)
4079 seq_puts(seq, "Type Device Function\n");
4080 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4081 if (pt->type == htons(ETH_P_ALL))
4082 seq_puts(seq, "ALL ");
4084 seq_printf(seq, "%04x", ntohs(pt->type));
4086 seq_printf(seq, " %-8s %pF\n",
4087 pt->dev ? pt->dev->name : "", pt->func);
4093 static const struct seq_operations ptype_seq_ops = {
4094 .start = ptype_seq_start,
4095 .next = ptype_seq_next,
4096 .stop = ptype_seq_stop,
4097 .show = ptype_seq_show,
4100 static int ptype_seq_open(struct inode *inode, struct file *file)
4102 return seq_open_net(inode, file, &ptype_seq_ops,
4103 sizeof(struct seq_net_private));
4106 static const struct file_operations ptype_seq_fops = {
4107 .owner = THIS_MODULE,
4108 .open = ptype_seq_open,
4110 .llseek = seq_lseek,
4111 .release = seq_release_net,
4115 static int __net_init dev_proc_net_init(struct net *net)
4119 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4121 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4123 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4126 if (wext_proc_init(net))
4132 proc_net_remove(net, "ptype");
4134 proc_net_remove(net, "softnet_stat");
4136 proc_net_remove(net, "dev");
4140 static void __net_exit dev_proc_net_exit(struct net *net)
4142 wext_proc_exit(net);
4144 proc_net_remove(net, "ptype");
4145 proc_net_remove(net, "softnet_stat");
4146 proc_net_remove(net, "dev");
4149 static struct pernet_operations __net_initdata dev_proc_ops = {
4150 .init = dev_proc_net_init,
4151 .exit = dev_proc_net_exit,
4154 static int __init dev_proc_init(void)
4156 return register_pernet_subsys(&dev_proc_ops);
4159 #define dev_proc_init() 0
4160 #endif /* CONFIG_PROC_FS */
4164 * netdev_set_master - set up master/slave pair
4165 * @slave: slave device
4166 * @master: new master device
4168 * Changes the master device of the slave. Pass %NULL to break the
4169 * bonding. The caller must hold the RTNL semaphore. On a failure
4170 * a negative errno code is returned. On success the reference counts
4171 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4172 * function returns zero.
4174 int netdev_set_master(struct net_device *slave, struct net_device *master)
4176 struct net_device *old = slave->master;
4186 slave->master = master;
4193 slave->flags |= IFF_SLAVE;
4195 slave->flags &= ~IFF_SLAVE;
4197 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4200 EXPORT_SYMBOL(netdev_set_master);
4202 static void dev_change_rx_flags(struct net_device *dev, int flags)
4204 const struct net_device_ops *ops = dev->netdev_ops;
4206 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4207 ops->ndo_change_rx_flags(dev, flags);
4210 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4212 unsigned short old_flags = dev->flags;
4218 dev->flags |= IFF_PROMISC;
4219 dev->promiscuity += inc;
4220 if (dev->promiscuity == 0) {
4223 * If inc causes overflow, untouch promisc and return error.
4226 dev->flags &= ~IFF_PROMISC;
4228 dev->promiscuity -= inc;
4229 printk(KERN_WARNING "%s: promiscuity touches roof, "
4230 "set promiscuity failed, promiscuity feature "
4231 "of device might be broken.\n", dev->name);
4235 if (dev->flags != old_flags) {
4236 printk(KERN_INFO "device %s %s promiscuous mode\n",
4237 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4239 if (audit_enabled) {
4240 current_uid_gid(&uid, &gid);
4241 audit_log(current->audit_context, GFP_ATOMIC,
4242 AUDIT_ANOM_PROMISCUOUS,
4243 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4244 dev->name, (dev->flags & IFF_PROMISC),
4245 (old_flags & IFF_PROMISC),
4246 audit_get_loginuid(current),
4248 audit_get_sessionid(current));
4251 dev_change_rx_flags(dev, IFF_PROMISC);
4257 * dev_set_promiscuity - update promiscuity count on a device
4261 * Add or remove promiscuity from a device. While the count in the device
4262 * remains above zero the interface remains promiscuous. Once it hits zero
4263 * the device reverts back to normal filtering operation. A negative inc
4264 * value is used to drop promiscuity on the device.
4265 * Return 0 if successful or a negative errno code on error.
4267 int dev_set_promiscuity(struct net_device *dev, int inc)
4269 unsigned short old_flags = dev->flags;
4272 err = __dev_set_promiscuity(dev, inc);
4275 if (dev->flags != old_flags)
4276 dev_set_rx_mode(dev);
4279 EXPORT_SYMBOL(dev_set_promiscuity);
4282 * dev_set_allmulti - update allmulti count on a device
4286 * Add or remove reception of all multicast frames to a device. While the
4287 * count in the device remains above zero the interface remains listening
4288 * to all interfaces. Once it hits zero the device reverts back to normal
4289 * filtering operation. A negative @inc value is used to drop the counter
4290 * when releasing a resource needing all multicasts.
4291 * Return 0 if successful or a negative errno code on error.
4294 int dev_set_allmulti(struct net_device *dev, int inc)
4296 unsigned short old_flags = dev->flags;
4300 dev->flags |= IFF_ALLMULTI;
4301 dev->allmulti += inc;
4302 if (dev->allmulti == 0) {
4305 * If inc causes overflow, untouch allmulti and return error.
4308 dev->flags &= ~IFF_ALLMULTI;
4310 dev->allmulti -= inc;
4311 printk(KERN_WARNING "%s: allmulti touches roof, "
4312 "set allmulti failed, allmulti feature of "
4313 "device might be broken.\n", dev->name);
4317 if (dev->flags ^ old_flags) {
4318 dev_change_rx_flags(dev, IFF_ALLMULTI);
4319 dev_set_rx_mode(dev);
4323 EXPORT_SYMBOL(dev_set_allmulti);
4326 * Upload unicast and multicast address lists to device and
4327 * configure RX filtering. When the device doesn't support unicast
4328 * filtering it is put in promiscuous mode while unicast addresses
4331 void __dev_set_rx_mode(struct net_device *dev)
4333 const struct net_device_ops *ops = dev->netdev_ops;
4335 /* dev_open will call this function so the list will stay sane. */
4336 if (!(dev->flags&IFF_UP))
4339 if (!netif_device_present(dev))
4342 if (ops->ndo_set_rx_mode)
4343 ops->ndo_set_rx_mode(dev);
4345 /* Unicast addresses changes may only happen under the rtnl,
4346 * therefore calling __dev_set_promiscuity here is safe.
4348 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4349 __dev_set_promiscuity(dev, 1);
4350 dev->uc_promisc = 1;
4351 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4352 __dev_set_promiscuity(dev, -1);
4353 dev->uc_promisc = 0;
4356 if (ops->ndo_set_multicast_list)
4357 ops->ndo_set_multicast_list(dev);
4361 void dev_set_rx_mode(struct net_device *dev)
4363 netif_addr_lock_bh(dev);
4364 __dev_set_rx_mode(dev);
4365 netif_addr_unlock_bh(dev);
4369 * dev_get_flags - get flags reported to userspace
4372 * Get the combination of flag bits exported through APIs to userspace.
4374 unsigned dev_get_flags(const struct net_device *dev)
4378 flags = (dev->flags & ~(IFF_PROMISC |
4383 (dev->gflags & (IFF_PROMISC |
4386 if (netif_running(dev)) {
4387 if (netif_oper_up(dev))
4388 flags |= IFF_RUNNING;
4389 if (netif_carrier_ok(dev))
4390 flags |= IFF_LOWER_UP;
4391 if (netif_dormant(dev))
4392 flags |= IFF_DORMANT;
4397 EXPORT_SYMBOL(dev_get_flags);
4399 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4401 int old_flags = dev->flags;
4407 * Set the flags on our device.
4410 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4411 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4413 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4417 * Load in the correct multicast list now the flags have changed.
4420 if ((old_flags ^ flags) & IFF_MULTICAST)
4421 dev_change_rx_flags(dev, IFF_MULTICAST);
4423 dev_set_rx_mode(dev);
4426 * Have we downed the interface. We handle IFF_UP ourselves
4427 * according to user attempts to set it, rather than blindly
4432 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4433 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4436 dev_set_rx_mode(dev);
4439 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4440 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4442 dev->gflags ^= IFF_PROMISC;
4443 dev_set_promiscuity(dev, inc);
4446 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4447 is important. Some (broken) drivers set IFF_PROMISC, when
4448 IFF_ALLMULTI is requested not asking us and not reporting.
4450 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4451 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4453 dev->gflags ^= IFF_ALLMULTI;
4454 dev_set_allmulti(dev, inc);
4460 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4462 unsigned int changes = dev->flags ^ old_flags;
4464 if (changes & IFF_UP) {
4465 if (dev->flags & IFF_UP)
4466 call_netdevice_notifiers(NETDEV_UP, dev);
4468 call_netdevice_notifiers(NETDEV_DOWN, dev);
4471 if (dev->flags & IFF_UP &&
4472 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4473 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4477 * dev_change_flags - change device settings
4479 * @flags: device state flags
4481 * Change settings on device based state flags. The flags are
4482 * in the userspace exported format.
4484 int dev_change_flags(struct net_device *dev, unsigned flags)
4487 int old_flags = dev->flags;
4489 ret = __dev_change_flags(dev, flags);
4493 changes = old_flags ^ dev->flags;
4495 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4497 __dev_notify_flags(dev, old_flags);
4500 EXPORT_SYMBOL(dev_change_flags);
4503 * dev_set_mtu - Change maximum transfer unit
4505 * @new_mtu: new transfer unit
4507 * Change the maximum transfer size of the network device.
4509 int dev_set_mtu(struct net_device *dev, int new_mtu)
4511 const struct net_device_ops *ops = dev->netdev_ops;
4514 if (new_mtu == dev->mtu)
4517 /* MTU must be positive. */
4521 if (!netif_device_present(dev))
4525 if (ops->ndo_change_mtu)
4526 err = ops->ndo_change_mtu(dev, new_mtu);
4530 if (!err && dev->flags & IFF_UP)
4531 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4534 EXPORT_SYMBOL(dev_set_mtu);
4537 * dev_set_mac_address - Change Media Access Control Address
4541 * Change the hardware (MAC) address of the device
4543 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4545 const struct net_device_ops *ops = dev->netdev_ops;
4548 if (!ops->ndo_set_mac_address)
4550 if (sa->sa_family != dev->type)
4552 if (!netif_device_present(dev))
4554 err = ops->ndo_set_mac_address(dev, sa);
4556 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4559 EXPORT_SYMBOL(dev_set_mac_address);
4562 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4564 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4567 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4573 case SIOCGIFFLAGS: /* Get interface flags */
4574 ifr->ifr_flags = (short) dev_get_flags(dev);
4577 case SIOCGIFMETRIC: /* Get the metric on the interface
4578 (currently unused) */
4579 ifr->ifr_metric = 0;
4582 case SIOCGIFMTU: /* Get the MTU of a device */
4583 ifr->ifr_mtu = dev->mtu;
4588 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4590 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4591 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4592 ifr->ifr_hwaddr.sa_family = dev->type;
4600 ifr->ifr_map.mem_start = dev->mem_start;
4601 ifr->ifr_map.mem_end = dev->mem_end;
4602 ifr->ifr_map.base_addr = dev->base_addr;
4603 ifr->ifr_map.irq = dev->irq;
4604 ifr->ifr_map.dma = dev->dma;
4605 ifr->ifr_map.port = dev->if_port;
4609 ifr->ifr_ifindex = dev->ifindex;
4613 ifr->ifr_qlen = dev->tx_queue_len;
4617 /* dev_ioctl() should ensure this case
4629 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4631 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4634 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4635 const struct net_device_ops *ops;
4640 ops = dev->netdev_ops;
4643 case SIOCSIFFLAGS: /* Set interface flags */
4644 return dev_change_flags(dev, ifr->ifr_flags);
4646 case SIOCSIFMETRIC: /* Set the metric on the interface
4647 (currently unused) */
4650 case SIOCSIFMTU: /* Set the MTU of a device */
4651 return dev_set_mtu(dev, ifr->ifr_mtu);
4654 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4656 case SIOCSIFHWBROADCAST:
4657 if (ifr->ifr_hwaddr.sa_family != dev->type)
4659 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4660 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4661 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4665 if (ops->ndo_set_config) {
4666 if (!netif_device_present(dev))
4668 return ops->ndo_set_config(dev, &ifr->ifr_map);
4673 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4674 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4676 if (!netif_device_present(dev))
4678 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4681 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4682 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4684 if (!netif_device_present(dev))
4686 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4689 if (ifr->ifr_qlen < 0)
4691 dev->tx_queue_len = ifr->ifr_qlen;
4695 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4696 return dev_change_name(dev, ifr->ifr_newname);
4699 * Unknown or private ioctl
4702 if ((cmd >= SIOCDEVPRIVATE &&
4703 cmd <= SIOCDEVPRIVATE + 15) ||
4704 cmd == SIOCBONDENSLAVE ||
4705 cmd == SIOCBONDRELEASE ||
4706 cmd == SIOCBONDSETHWADDR ||
4707 cmd == SIOCBONDSLAVEINFOQUERY ||
4708 cmd == SIOCBONDINFOQUERY ||
4709 cmd == SIOCBONDCHANGEACTIVE ||
4710 cmd == SIOCGMIIPHY ||
4711 cmd == SIOCGMIIREG ||
4712 cmd == SIOCSMIIREG ||
4713 cmd == SIOCBRADDIF ||
4714 cmd == SIOCBRDELIF ||
4715 cmd == SIOCSHWTSTAMP ||
4716 cmd == SIOCWANDEV) {
4718 if (ops->ndo_do_ioctl) {
4719 if (netif_device_present(dev))
4720 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4732 * This function handles all "interface"-type I/O control requests. The actual
4733 * 'doing' part of this is dev_ifsioc above.
4737 * dev_ioctl - network device ioctl
4738 * @net: the applicable net namespace
4739 * @cmd: command to issue
4740 * @arg: pointer to a struct ifreq in user space
4742 * Issue ioctl functions to devices. This is normally called by the
4743 * user space syscall interfaces but can sometimes be useful for
4744 * other purposes. The return value is the return from the syscall if
4745 * positive or a negative errno code on error.
4748 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4754 /* One special case: SIOCGIFCONF takes ifconf argument
4755 and requires shared lock, because it sleeps writing
4759 if (cmd == SIOCGIFCONF) {
4761 ret = dev_ifconf(net, (char __user *) arg);
4765 if (cmd == SIOCGIFNAME)
4766 return dev_ifname(net, (struct ifreq __user *)arg);
4768 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4771 ifr.ifr_name[IFNAMSIZ-1] = 0;
4773 colon = strchr(ifr.ifr_name, ':');
4778 * See which interface the caller is talking about.
4783 * These ioctl calls:
4784 * - can be done by all.
4785 * - atomic and do not require locking.
4796 dev_load(net, ifr.ifr_name);
4798 ret = dev_ifsioc_locked(net, &ifr, cmd);
4803 if (copy_to_user(arg, &ifr,
4804 sizeof(struct ifreq)))
4810 dev_load(net, ifr.ifr_name);
4812 ret = dev_ethtool(net, &ifr);
4817 if (copy_to_user(arg, &ifr,
4818 sizeof(struct ifreq)))
4824 * These ioctl calls:
4825 * - require superuser power.
4826 * - require strict serialization.
4832 if (!capable(CAP_NET_ADMIN))
4834 dev_load(net, ifr.ifr_name);
4836 ret = dev_ifsioc(net, &ifr, cmd);
4841 if (copy_to_user(arg, &ifr,
4842 sizeof(struct ifreq)))
4848 * These ioctl calls:
4849 * - require superuser power.
4850 * - require strict serialization.
4851 * - do not return a value
4861 case SIOCSIFHWBROADCAST:
4864 case SIOCBONDENSLAVE:
4865 case SIOCBONDRELEASE:
4866 case SIOCBONDSETHWADDR:
4867 case SIOCBONDCHANGEACTIVE:
4871 if (!capable(CAP_NET_ADMIN))
4874 case SIOCBONDSLAVEINFOQUERY:
4875 case SIOCBONDINFOQUERY:
4876 dev_load(net, ifr.ifr_name);
4878 ret = dev_ifsioc(net, &ifr, cmd);
4883 /* Get the per device memory space. We can add this but
4884 * currently do not support it */
4886 /* Set the per device memory buffer space.
4887 * Not applicable in our case */
4892 * Unknown or private ioctl.
4895 if (cmd == SIOCWANDEV ||
4896 (cmd >= SIOCDEVPRIVATE &&
4897 cmd <= SIOCDEVPRIVATE + 15)) {
4898 dev_load(net, ifr.ifr_name);
4900 ret = dev_ifsioc(net, &ifr, cmd);
4902 if (!ret && copy_to_user(arg, &ifr,
4903 sizeof(struct ifreq)))
4907 /* Take care of Wireless Extensions */
4908 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4909 return wext_handle_ioctl(net, &ifr, cmd, arg);
4916 * dev_new_index - allocate an ifindex
4917 * @net: the applicable net namespace
4919 * Returns a suitable unique value for a new device interface
4920 * number. The caller must hold the rtnl semaphore or the
4921 * dev_base_lock to be sure it remains unique.
4923 static int dev_new_index(struct net *net)
4929 if (!__dev_get_by_index(net, ifindex))
4934 /* Delayed registration/unregisteration */
4935 static LIST_HEAD(net_todo_list);
4937 static void net_set_todo(struct net_device *dev)
4939 list_add_tail(&dev->todo_list, &net_todo_list);
4942 static void rollback_registered_many(struct list_head *head)
4944 struct net_device *dev, *tmp;
4946 BUG_ON(dev_boot_phase);
4949 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4950 /* Some devices call without registering
4951 * for initialization unwind. Remove those
4952 * devices and proceed with the remaining.
4954 if (dev->reg_state == NETREG_UNINITIALIZED) {
4955 pr_debug("unregister_netdevice: device %s/%p never "
4956 "was registered\n", dev->name, dev);
4959 list_del(&dev->unreg_list);
4963 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4965 /* If device is running, close it first. */
4968 /* And unlink it from device chain. */
4969 unlist_netdevice(dev);
4971 dev->reg_state = NETREG_UNREGISTERING;
4976 list_for_each_entry(dev, head, unreg_list) {
4977 /* Shutdown queueing discipline. */
4981 /* Notify protocols, that we are about to destroy
4982 this device. They should clean all the things.
4984 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4986 if (!dev->rtnl_link_ops ||
4987 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4988 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4991 * Flush the unicast and multicast chains
4996 if (dev->netdev_ops->ndo_uninit)
4997 dev->netdev_ops->ndo_uninit(dev);
4999 /* Notifier chain MUST detach us from master device. */
5000 WARN_ON(dev->master);
5002 /* Remove entries from kobject tree */
5003 netdev_unregister_kobject(dev);
5006 /* Process any work delayed until the end of the batch */
5007 dev = list_first_entry(head, struct net_device, unreg_list);
5008 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5012 list_for_each_entry(dev, head, unreg_list)
5016 static void rollback_registered(struct net_device *dev)
5020 list_add(&dev->unreg_list, &single);
5021 rollback_registered_many(&single);
5024 unsigned long netdev_fix_features(unsigned long features, const char *name)
5026 /* Fix illegal SG+CSUM combinations. */
5027 if ((features & NETIF_F_SG) &&
5028 !(features & NETIF_F_ALL_CSUM)) {
5030 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
5031 "checksum feature.\n", name);
5032 features &= ~NETIF_F_SG;
5035 /* TSO requires that SG is present as well. */
5036 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
5038 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
5039 "SG feature.\n", name);
5040 features &= ~NETIF_F_TSO;
5043 if (features & NETIF_F_UFO) {
5044 /* maybe split UFO into V4 and V6? */
5045 if (!((features & NETIF_F_GEN_CSUM) ||
5046 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5047 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5049 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5050 "since no checksum offload features.\n",
5052 features &= ~NETIF_F_UFO;
5055 if (!(features & NETIF_F_SG)) {
5057 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5058 "since no NETIF_F_SG feature.\n", name);
5059 features &= ~NETIF_F_UFO;
5065 EXPORT_SYMBOL(netdev_fix_features);
5068 * netif_stacked_transfer_operstate - transfer operstate
5069 * @rootdev: the root or lower level device to transfer state from
5070 * @dev: the device to transfer operstate to
5072 * Transfer operational state from root to device. This is normally
5073 * called when a stacking relationship exists between the root
5074 * device and the device(a leaf device).
5076 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5077 struct net_device *dev)
5079 if (rootdev->operstate == IF_OPER_DORMANT)
5080 netif_dormant_on(dev);
5082 netif_dormant_off(dev);
5084 if (netif_carrier_ok(rootdev)) {
5085 if (!netif_carrier_ok(dev))
5086 netif_carrier_on(dev);
5088 if (netif_carrier_ok(dev))
5089 netif_carrier_off(dev);
5092 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5095 static int netif_alloc_rx_queues(struct net_device *dev)
5097 unsigned int i, count = dev->num_rx_queues;
5098 struct netdev_rx_queue *rx;
5102 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5104 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5109 for (i = 0; i < count; i++)
5115 static int netif_alloc_netdev_queues(struct net_device *dev)
5117 unsigned int count = dev->num_tx_queues;
5118 struct netdev_queue *tx;
5123 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5125 pr_err("netdev: Unable to allocate %u tx queues.\n",
5131 for (i = 0; i < count; i++) {
5132 netdev_queue_numa_node_write(&tx[i], -1);
5138 static void netdev_init_one_queue(struct net_device *dev,
5139 struct netdev_queue *queue,
5142 /* Initialize queue lock */
5143 spin_lock_init(&queue->_xmit_lock);
5144 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5145 queue->xmit_lock_owner = -1;
5148 static void netdev_init_queues(struct net_device *dev)
5150 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5151 spin_lock_init(&dev->tx_global_lock);
5155 * register_netdevice - register a network device
5156 * @dev: device to register
5158 * Take a completed network device structure and add it to the kernel
5159 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5160 * chain. 0 is returned on success. A negative errno code is returned
5161 * on a failure to set up the device, or if the name is a duplicate.
5163 * Callers must hold the rtnl semaphore. You may want
5164 * register_netdev() instead of this.
5167 * The locking appears insufficient to guarantee two parallel registers
5168 * will not get the same name.
5171 int register_netdevice(struct net_device *dev)
5174 struct net *net = dev_net(dev);
5176 BUG_ON(dev_boot_phase);
5181 /* When net_device's are persistent, this will be fatal. */
5182 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5185 spin_lock_init(&dev->addr_list_lock);
5186 netdev_set_addr_lockdep_class(dev);
5190 netdev_init_queues(dev);
5192 /* Init, if this function is available */
5193 if (dev->netdev_ops->ndo_init) {
5194 ret = dev->netdev_ops->ndo_init(dev);
5202 ret = dev_get_valid_name(dev, dev->name, 0);
5206 dev->ifindex = dev_new_index(net);
5207 if (dev->iflink == -1)
5208 dev->iflink = dev->ifindex;
5210 /* Fix illegal checksum combinations */
5211 if ((dev->features & NETIF_F_HW_CSUM) &&
5212 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5213 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5215 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5218 if ((dev->features & NETIF_F_NO_CSUM) &&
5219 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5220 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5222 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5225 dev->features = netdev_fix_features(dev->features, dev->name);
5227 /* Enable software GSO if SG is supported. */
5228 if (dev->features & NETIF_F_SG)
5229 dev->features |= NETIF_F_GSO;
5231 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5232 * vlan_dev_init() will do the dev->features check, so these features
5233 * are enabled only if supported by underlying device.
5235 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5237 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5238 ret = notifier_to_errno(ret);
5242 ret = netdev_register_kobject(dev);
5245 dev->reg_state = NETREG_REGISTERED;
5248 * Default initial state at registry is that the
5249 * device is present.
5252 set_bit(__LINK_STATE_PRESENT, &dev->state);
5254 dev_init_scheduler(dev);
5256 list_netdevice(dev);
5258 /* Notify protocols, that a new device appeared. */
5259 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5260 ret = notifier_to_errno(ret);
5262 rollback_registered(dev);
5263 dev->reg_state = NETREG_UNREGISTERED;
5266 * Prevent userspace races by waiting until the network
5267 * device is fully setup before sending notifications.
5269 if (!dev->rtnl_link_ops ||
5270 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5271 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5277 if (dev->netdev_ops->ndo_uninit)
5278 dev->netdev_ops->ndo_uninit(dev);
5281 EXPORT_SYMBOL(register_netdevice);
5284 * init_dummy_netdev - init a dummy network device for NAPI
5285 * @dev: device to init
5287 * This takes a network device structure and initialize the minimum
5288 * amount of fields so it can be used to schedule NAPI polls without
5289 * registering a full blown interface. This is to be used by drivers
5290 * that need to tie several hardware interfaces to a single NAPI
5291 * poll scheduler due to HW limitations.
5293 int init_dummy_netdev(struct net_device *dev)
5295 /* Clear everything. Note we don't initialize spinlocks
5296 * are they aren't supposed to be taken by any of the
5297 * NAPI code and this dummy netdev is supposed to be
5298 * only ever used for NAPI polls
5300 memset(dev, 0, sizeof(struct net_device));
5302 /* make sure we BUG if trying to hit standard
5303 * register/unregister code path
5305 dev->reg_state = NETREG_DUMMY;
5307 /* NAPI wants this */
5308 INIT_LIST_HEAD(&dev->napi_list);
5310 /* a dummy interface is started by default */
5311 set_bit(__LINK_STATE_PRESENT, &dev->state);
5312 set_bit(__LINK_STATE_START, &dev->state);
5314 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5315 * because users of this 'device' dont need to change
5321 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5325 * register_netdev - register a network device
5326 * @dev: device to register
5328 * Take a completed network device structure and add it to the kernel
5329 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5330 * chain. 0 is returned on success. A negative errno code is returned
5331 * on a failure to set up the device, or if the name is a duplicate.
5333 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5334 * and expands the device name if you passed a format string to
5337 int register_netdev(struct net_device *dev)
5344 * If the name is a format string the caller wants us to do a
5347 if (strchr(dev->name, '%')) {
5348 err = dev_alloc_name(dev, dev->name);
5353 err = register_netdevice(dev);
5358 EXPORT_SYMBOL(register_netdev);
5360 int netdev_refcnt_read(const struct net_device *dev)
5364 for_each_possible_cpu(i)
5365 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5368 EXPORT_SYMBOL(netdev_refcnt_read);
5371 * netdev_wait_allrefs - wait until all references are gone.
5373 * This is called when unregistering network devices.
5375 * Any protocol or device that holds a reference should register
5376 * for netdevice notification, and cleanup and put back the
5377 * reference if they receive an UNREGISTER event.
5378 * We can get stuck here if buggy protocols don't correctly
5381 static void netdev_wait_allrefs(struct net_device *dev)
5383 unsigned long rebroadcast_time, warning_time;
5386 linkwatch_forget_dev(dev);
5388 rebroadcast_time = warning_time = jiffies;
5389 refcnt = netdev_refcnt_read(dev);
5391 while (refcnt != 0) {
5392 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5395 /* Rebroadcast unregister notification */
5396 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5397 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5398 * should have already handle it the first time */
5400 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5402 /* We must not have linkwatch events
5403 * pending on unregister. If this
5404 * happens, we simply run the queue
5405 * unscheduled, resulting in a noop
5408 linkwatch_run_queue();
5413 rebroadcast_time = jiffies;
5418 refcnt = netdev_refcnt_read(dev);
5420 if (time_after(jiffies, warning_time + 10 * HZ)) {
5421 printk(KERN_EMERG "unregister_netdevice: "
5422 "waiting for %s to become free. Usage "
5425 warning_time = jiffies;
5434 * register_netdevice(x1);
5435 * register_netdevice(x2);
5437 * unregister_netdevice(y1);
5438 * unregister_netdevice(y2);
5444 * We are invoked by rtnl_unlock().
5445 * This allows us to deal with problems:
5446 * 1) We can delete sysfs objects which invoke hotplug
5447 * without deadlocking with linkwatch via keventd.
5448 * 2) Since we run with the RTNL semaphore not held, we can sleep
5449 * safely in order to wait for the netdev refcnt to drop to zero.
5451 * We must not return until all unregister events added during
5452 * the interval the lock was held have been completed.
5454 void netdev_run_todo(void)
5456 struct list_head list;
5458 /* Snapshot list, allow later requests */
5459 list_replace_init(&net_todo_list, &list);
5463 while (!list_empty(&list)) {
5464 struct net_device *dev
5465 = list_first_entry(&list, struct net_device, todo_list);
5466 list_del(&dev->todo_list);
5468 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5469 printk(KERN_ERR "network todo '%s' but state %d\n",
5470 dev->name, dev->reg_state);
5475 dev->reg_state = NETREG_UNREGISTERED;
5477 on_each_cpu(flush_backlog, dev, 1);
5479 netdev_wait_allrefs(dev);
5482 BUG_ON(netdev_refcnt_read(dev));
5483 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5484 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5485 WARN_ON(dev->dn_ptr);
5487 if (dev->destructor)
5488 dev->destructor(dev);
5490 /* Free network device */
5491 kobject_put(&dev->dev.kobj);
5496 * dev_txq_stats_fold - fold tx_queues stats
5497 * @dev: device to get statistics from
5498 * @stats: struct rtnl_link_stats64 to hold results
5500 void dev_txq_stats_fold(const struct net_device *dev,
5501 struct rtnl_link_stats64 *stats)
5503 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5505 struct netdev_queue *txq;
5507 for (i = 0; i < dev->num_tx_queues; i++) {
5508 txq = netdev_get_tx_queue(dev, i);
5509 spin_lock_bh(&txq->_xmit_lock);
5510 tx_bytes += txq->tx_bytes;
5511 tx_packets += txq->tx_packets;
5512 tx_dropped += txq->tx_dropped;
5513 spin_unlock_bh(&txq->_xmit_lock);
5515 if (tx_bytes || tx_packets || tx_dropped) {
5516 stats->tx_bytes = tx_bytes;
5517 stats->tx_packets = tx_packets;
5518 stats->tx_dropped = tx_dropped;
5521 EXPORT_SYMBOL(dev_txq_stats_fold);
5523 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5524 * fields in the same order, with only the type differing.
5526 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5527 const struct net_device_stats *netdev_stats)
5529 #if BITS_PER_LONG == 64
5530 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5531 memcpy(stats64, netdev_stats, sizeof(*stats64));
5533 size_t i, n = sizeof(*stats64) / sizeof(u64);
5534 const unsigned long *src = (const unsigned long *)netdev_stats;
5535 u64 *dst = (u64 *)stats64;
5537 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5538 sizeof(*stats64) / sizeof(u64));
5539 for (i = 0; i < n; i++)
5545 * dev_get_stats - get network device statistics
5546 * @dev: device to get statistics from
5547 * @storage: place to store stats
5549 * Get network statistics from device. Return @storage.
5550 * The device driver may provide its own method by setting
5551 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5552 * otherwise the internal statistics structure is used.
5554 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5555 struct rtnl_link_stats64 *storage)
5557 const struct net_device_ops *ops = dev->netdev_ops;
5559 if (ops->ndo_get_stats64) {
5560 memset(storage, 0, sizeof(*storage));
5561 ops->ndo_get_stats64(dev, storage);
5562 } else if (ops->ndo_get_stats) {
5563 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5565 netdev_stats_to_stats64(storage, &dev->stats);
5566 dev_txq_stats_fold(dev, storage);
5568 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5571 EXPORT_SYMBOL(dev_get_stats);
5573 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5575 struct netdev_queue *queue = dev_ingress_queue(dev);
5577 #ifdef CONFIG_NET_CLS_ACT
5580 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5583 netdev_init_one_queue(dev, queue, NULL);
5584 queue->qdisc = &noop_qdisc;
5585 queue->qdisc_sleeping = &noop_qdisc;
5586 rcu_assign_pointer(dev->ingress_queue, queue);
5592 * alloc_netdev_mq - allocate network device
5593 * @sizeof_priv: size of private data to allocate space for
5594 * @name: device name format string
5595 * @setup: callback to initialize device
5596 * @queue_count: the number of subqueues to allocate
5598 * Allocates a struct net_device with private data area for driver use
5599 * and performs basic initialization. Also allocates subquue structs
5600 * for each queue on the device at the end of the netdevice.
5602 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5603 void (*setup)(struct net_device *), unsigned int queue_count)
5605 struct net_device *dev;
5607 struct net_device *p;
5609 BUG_ON(strlen(name) >= sizeof(dev->name));
5611 if (queue_count < 1) {
5612 pr_err("alloc_netdev: Unable to allocate device "
5613 "with zero queues.\n");
5617 alloc_size = sizeof(struct net_device);
5619 /* ensure 32-byte alignment of private area */
5620 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5621 alloc_size += sizeof_priv;
5623 /* ensure 32-byte alignment of whole construct */
5624 alloc_size += NETDEV_ALIGN - 1;
5626 p = kzalloc(alloc_size, GFP_KERNEL);
5628 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5632 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5633 dev->padded = (char *)dev - (char *)p;
5635 dev->pcpu_refcnt = alloc_percpu(int);
5636 if (!dev->pcpu_refcnt)
5639 if (dev_addr_init(dev))
5645 dev_net_set(dev, &init_net);
5647 dev->num_tx_queues = queue_count;
5648 dev->real_num_tx_queues = queue_count;
5649 if (netif_alloc_netdev_queues(dev))
5653 dev->num_rx_queues = queue_count;
5654 dev->real_num_rx_queues = queue_count;
5655 if (netif_alloc_rx_queues(dev))
5659 dev->gso_max_size = GSO_MAX_SIZE;
5661 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5662 dev->ethtool_ntuple_list.count = 0;
5663 INIT_LIST_HEAD(&dev->napi_list);
5664 INIT_LIST_HEAD(&dev->unreg_list);
5665 INIT_LIST_HEAD(&dev->link_watch_list);
5666 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5668 strcpy(dev->name, name);
5672 free_percpu(dev->pcpu_refcnt);
5682 EXPORT_SYMBOL(alloc_netdev_mq);
5685 * free_netdev - free network device
5688 * This function does the last stage of destroying an allocated device
5689 * interface. The reference to the device object is released.
5690 * If this is the last reference then it will be freed.
5692 void free_netdev(struct net_device *dev)
5694 struct napi_struct *p, *n;
5696 release_net(dev_net(dev));
5703 kfree(rcu_dereference_raw(dev->ingress_queue));
5705 /* Flush device addresses */
5706 dev_addr_flush(dev);
5708 /* Clear ethtool n-tuple list */
5709 ethtool_ntuple_flush(dev);
5711 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5714 free_percpu(dev->pcpu_refcnt);
5715 dev->pcpu_refcnt = NULL;
5717 /* Compatibility with error handling in drivers */
5718 if (dev->reg_state == NETREG_UNINITIALIZED) {
5719 kfree((char *)dev - dev->padded);
5723 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5724 dev->reg_state = NETREG_RELEASED;
5726 /* will free via device release */
5727 put_device(&dev->dev);
5729 EXPORT_SYMBOL(free_netdev);
5732 * synchronize_net - Synchronize with packet receive processing
5734 * Wait for packets currently being received to be done.
5735 * Does not block later packets from starting.
5737 void synchronize_net(void)
5742 EXPORT_SYMBOL(synchronize_net);
5745 * unregister_netdevice_queue - remove device from the kernel
5749 * This function shuts down a device interface and removes it
5750 * from the kernel tables.
5751 * If head not NULL, device is queued to be unregistered later.
5753 * Callers must hold the rtnl semaphore. You may want
5754 * unregister_netdev() instead of this.
5757 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5762 list_move_tail(&dev->unreg_list, head);
5764 rollback_registered(dev);
5765 /* Finish processing unregister after unlock */
5769 EXPORT_SYMBOL(unregister_netdevice_queue);
5772 * unregister_netdevice_many - unregister many devices
5773 * @head: list of devices
5775 void unregister_netdevice_many(struct list_head *head)
5777 struct net_device *dev;
5779 if (!list_empty(head)) {
5780 rollback_registered_many(head);
5781 list_for_each_entry(dev, head, unreg_list)
5785 EXPORT_SYMBOL(unregister_netdevice_many);
5788 * unregister_netdev - remove device from the kernel
5791 * This function shuts down a device interface and removes it
5792 * from the kernel tables.
5794 * This is just a wrapper for unregister_netdevice that takes
5795 * the rtnl semaphore. In general you want to use this and not
5796 * unregister_netdevice.
5798 void unregister_netdev(struct net_device *dev)
5801 unregister_netdevice(dev);
5804 EXPORT_SYMBOL(unregister_netdev);
5807 * dev_change_net_namespace - move device to different nethost namespace
5809 * @net: network namespace
5810 * @pat: If not NULL name pattern to try if the current device name
5811 * is already taken in the destination network namespace.
5813 * This function shuts down a device interface and moves it
5814 * to a new network namespace. On success 0 is returned, on
5815 * a failure a netagive errno code is returned.
5817 * Callers must hold the rtnl semaphore.
5820 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5826 /* Don't allow namespace local devices to be moved. */
5828 if (dev->features & NETIF_F_NETNS_LOCAL)
5831 /* Ensure the device has been registrered */
5833 if (dev->reg_state != NETREG_REGISTERED)
5836 /* Get out if there is nothing todo */
5838 if (net_eq(dev_net(dev), net))
5841 /* Pick the destination device name, and ensure
5842 * we can use it in the destination network namespace.
5845 if (__dev_get_by_name(net, dev->name)) {
5846 /* We get here if we can't use the current device name */
5849 if (dev_get_valid_name(dev, pat, 1))
5854 * And now a mini version of register_netdevice unregister_netdevice.
5857 /* If device is running close it first. */
5860 /* And unlink it from device chain */
5862 unlist_netdevice(dev);
5866 /* Shutdown queueing discipline. */
5869 /* Notify protocols, that we are about to destroy
5870 this device. They should clean all the things.
5872 Note that dev->reg_state stays at NETREG_REGISTERED.
5873 This is wanted because this way 8021q and macvlan know
5874 the device is just moving and can keep their slaves up.
5876 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5877 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5880 * Flush the unicast and multicast chains
5885 /* Actually switch the network namespace */
5886 dev_net_set(dev, net);
5888 /* If there is an ifindex conflict assign a new one */
5889 if (__dev_get_by_index(net, dev->ifindex)) {
5890 int iflink = (dev->iflink == dev->ifindex);
5891 dev->ifindex = dev_new_index(net);
5893 dev->iflink = dev->ifindex;
5896 /* Fixup kobjects */
5897 err = device_rename(&dev->dev, dev->name);
5900 /* Add the device back in the hashes */
5901 list_netdevice(dev);
5903 /* Notify protocols, that a new device appeared. */
5904 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5907 * Prevent userspace races by waiting until the network
5908 * device is fully setup before sending notifications.
5910 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5917 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5919 static int dev_cpu_callback(struct notifier_block *nfb,
5920 unsigned long action,
5923 struct sk_buff **list_skb;
5924 struct sk_buff *skb;
5925 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5926 struct softnet_data *sd, *oldsd;
5928 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5931 local_irq_disable();
5932 cpu = smp_processor_id();
5933 sd = &per_cpu(softnet_data, cpu);
5934 oldsd = &per_cpu(softnet_data, oldcpu);
5936 /* Find end of our completion_queue. */
5937 list_skb = &sd->completion_queue;
5939 list_skb = &(*list_skb)->next;
5940 /* Append completion queue from offline CPU. */
5941 *list_skb = oldsd->completion_queue;
5942 oldsd->completion_queue = NULL;
5944 /* Append output queue from offline CPU. */
5945 if (oldsd->output_queue) {
5946 *sd->output_queue_tailp = oldsd->output_queue;
5947 sd->output_queue_tailp = oldsd->output_queue_tailp;
5948 oldsd->output_queue = NULL;
5949 oldsd->output_queue_tailp = &oldsd->output_queue;
5952 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5955 /* Process offline CPU's input_pkt_queue */
5956 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5958 input_queue_head_incr(oldsd);
5960 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5962 input_queue_head_incr(oldsd);
5970 * netdev_increment_features - increment feature set by one
5971 * @all: current feature set
5972 * @one: new feature set
5973 * @mask: mask feature set
5975 * Computes a new feature set after adding a device with feature set
5976 * @one to the master device with current feature set @all. Will not
5977 * enable anything that is off in @mask. Returns the new feature set.
5979 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5982 /* If device needs checksumming, downgrade to it. */
5983 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5984 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5985 else if (mask & NETIF_F_ALL_CSUM) {
5986 /* If one device supports v4/v6 checksumming, set for all. */
5987 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5988 !(all & NETIF_F_GEN_CSUM)) {
5989 all &= ~NETIF_F_ALL_CSUM;
5990 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5993 /* If one device supports hw checksumming, set for all. */
5994 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5995 all &= ~NETIF_F_ALL_CSUM;
5996 all |= NETIF_F_HW_CSUM;
6000 one |= NETIF_F_ALL_CSUM;
6002 one |= all & NETIF_F_ONE_FOR_ALL;
6003 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
6004 all |= one & mask & NETIF_F_ONE_FOR_ALL;
6008 EXPORT_SYMBOL(netdev_increment_features);
6010 static struct hlist_head *netdev_create_hash(void)
6013 struct hlist_head *hash;
6015 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6017 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6018 INIT_HLIST_HEAD(&hash[i]);
6023 /* Initialize per network namespace state */
6024 static int __net_init netdev_init(struct net *net)
6026 INIT_LIST_HEAD(&net->dev_base_head);
6028 net->dev_name_head = netdev_create_hash();
6029 if (net->dev_name_head == NULL)
6032 net->dev_index_head = netdev_create_hash();
6033 if (net->dev_index_head == NULL)
6039 kfree(net->dev_name_head);
6045 * netdev_drivername - network driver for the device
6046 * @dev: network device
6047 * @buffer: buffer for resulting name
6048 * @len: size of buffer
6050 * Determine network driver for device.
6052 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6054 const struct device_driver *driver;
6055 const struct device *parent;
6057 if (len <= 0 || !buffer)
6061 parent = dev->dev.parent;
6066 driver = parent->driver;
6067 if (driver && driver->name)
6068 strlcpy(buffer, driver->name, len);
6072 static int __netdev_printk(const char *level, const struct net_device *dev,
6073 struct va_format *vaf)
6077 if (dev && dev->dev.parent)
6078 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6079 netdev_name(dev), vaf);
6081 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6083 r = printk("%s(NULL net_device): %pV", level, vaf);
6088 int netdev_printk(const char *level, const struct net_device *dev,
6089 const char *format, ...)
6091 struct va_format vaf;
6095 va_start(args, format);
6100 r = __netdev_printk(level, dev, &vaf);
6105 EXPORT_SYMBOL(netdev_printk);
6107 #define define_netdev_printk_level(func, level) \
6108 int func(const struct net_device *dev, const char *fmt, ...) \
6111 struct va_format vaf; \
6114 va_start(args, fmt); \
6119 r = __netdev_printk(level, dev, &vaf); \
6124 EXPORT_SYMBOL(func);
6126 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6127 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6128 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6129 define_netdev_printk_level(netdev_err, KERN_ERR);
6130 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6131 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6132 define_netdev_printk_level(netdev_info, KERN_INFO);
6134 static void __net_exit netdev_exit(struct net *net)
6136 kfree(net->dev_name_head);
6137 kfree(net->dev_index_head);
6140 static struct pernet_operations __net_initdata netdev_net_ops = {
6141 .init = netdev_init,
6142 .exit = netdev_exit,
6145 static void __net_exit default_device_exit(struct net *net)
6147 struct net_device *dev, *aux;
6149 * Push all migratable network devices back to the
6150 * initial network namespace
6153 for_each_netdev_safe(net, dev, aux) {
6155 char fb_name[IFNAMSIZ];
6157 /* Ignore unmoveable devices (i.e. loopback) */
6158 if (dev->features & NETIF_F_NETNS_LOCAL)
6161 /* Leave virtual devices for the generic cleanup */
6162 if (dev->rtnl_link_ops)
6165 /* Push remaing network devices to init_net */
6166 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6167 err = dev_change_net_namespace(dev, &init_net, fb_name);
6169 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6170 __func__, dev->name, err);
6177 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6179 /* At exit all network devices most be removed from a network
6180 * namespace. Do this in the reverse order of registeration.
6181 * Do this across as many network namespaces as possible to
6182 * improve batching efficiency.
6184 struct net_device *dev;
6186 LIST_HEAD(dev_kill_list);
6189 list_for_each_entry(net, net_list, exit_list) {
6190 for_each_netdev_reverse(net, dev) {
6191 if (dev->rtnl_link_ops)
6192 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6194 unregister_netdevice_queue(dev, &dev_kill_list);
6197 unregister_netdevice_many(&dev_kill_list);
6201 static struct pernet_operations __net_initdata default_device_ops = {
6202 .exit = default_device_exit,
6203 .exit_batch = default_device_exit_batch,
6207 * Initialize the DEV module. At boot time this walks the device list and
6208 * unhooks any devices that fail to initialise (normally hardware not
6209 * present) and leaves us with a valid list of present and active devices.
6214 * This is called single threaded during boot, so no need
6215 * to take the rtnl semaphore.
6217 static int __init net_dev_init(void)
6219 int i, rc = -ENOMEM;
6221 BUG_ON(!dev_boot_phase);
6223 if (dev_proc_init())
6226 if (netdev_kobject_init())
6229 INIT_LIST_HEAD(&ptype_all);
6230 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6231 INIT_LIST_HEAD(&ptype_base[i]);
6233 if (register_pernet_subsys(&netdev_net_ops))
6237 * Initialise the packet receive queues.
6240 for_each_possible_cpu(i) {
6241 struct softnet_data *sd = &per_cpu(softnet_data, i);
6243 memset(sd, 0, sizeof(*sd));
6244 skb_queue_head_init(&sd->input_pkt_queue);
6245 skb_queue_head_init(&sd->process_queue);
6246 sd->completion_queue = NULL;
6247 INIT_LIST_HEAD(&sd->poll_list);
6248 sd->output_queue = NULL;
6249 sd->output_queue_tailp = &sd->output_queue;
6251 sd->csd.func = rps_trigger_softirq;
6257 sd->backlog.poll = process_backlog;
6258 sd->backlog.weight = weight_p;
6259 sd->backlog.gro_list = NULL;
6260 sd->backlog.gro_count = 0;
6265 /* The loopback device is special if any other network devices
6266 * is present in a network namespace the loopback device must
6267 * be present. Since we now dynamically allocate and free the
6268 * loopback device ensure this invariant is maintained by
6269 * keeping the loopback device as the first device on the
6270 * list of network devices. Ensuring the loopback devices
6271 * is the first device that appears and the last network device
6274 if (register_pernet_device(&loopback_net_ops))
6277 if (register_pernet_device(&default_device_ops))
6280 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6281 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6283 hotcpu_notifier(dev_cpu_callback, 0);
6291 subsys_initcall(net_dev_init);
6293 static int __init initialize_hashrnd(void)
6295 get_random_bytes(&hashrnd, sizeof(hashrnd));
6299 late_initcall_sync(initialize_hashrnd);