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>
135 #include <linux/cpu_rmap.h>
137 #include "net-sysfs.h"
139 /* Instead of increasing this, you should create a hash table. */
140 #define MAX_GRO_SKBS 8
142 /* This should be increased if a protocol with a bigger head is added. */
143 #define GRO_MAX_HEAD (MAX_HEADER + 128)
146 * The list of packet types we will receive (as opposed to discard)
147 * and the routines to invoke.
149 * Why 16. Because with 16 the only overlap we get on a hash of the
150 * low nibble of the protocol value is RARP/SNAP/X.25.
152 * NOTE: That is no longer true with the addition of VLAN tags. Not
153 * sure which should go first, but I bet it won't make much
154 * difference if we are running VLANs. The good news is that
155 * this protocol won't be in the list unless compiled in, so
156 * the average user (w/out VLANs) will not be adversely affected.
173 #define PTYPE_HASH_SIZE (16)
174 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
176 static DEFINE_SPINLOCK(ptype_lock);
177 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
178 static struct list_head ptype_all __read_mostly; /* Taps */
181 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
184 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
186 * Writers must hold the rtnl semaphore while they loop through the
187 * dev_base_head list, and hold dev_base_lock for writing when they do the
188 * actual updates. This allows pure readers to access the list even
189 * while a writer is preparing to update it.
191 * To put it another way, dev_base_lock is held for writing only to
192 * protect against pure readers; the rtnl semaphore provides the
193 * protection against other writers.
195 * See, for example usages, register_netdevice() and
196 * unregister_netdevice(), which must be called with the rtnl
199 DEFINE_RWLOCK(dev_base_lock);
200 EXPORT_SYMBOL(dev_base_lock);
202 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
204 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
205 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
208 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
210 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
213 static inline void rps_lock(struct softnet_data *sd)
216 spin_lock(&sd->input_pkt_queue.lock);
220 static inline void rps_unlock(struct softnet_data *sd)
223 spin_unlock(&sd->input_pkt_queue.lock);
227 /* Device list insertion */
228 static int list_netdevice(struct net_device *dev)
230 struct net *net = dev_net(dev);
234 write_lock_bh(&dev_base_lock);
235 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
236 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
237 hlist_add_head_rcu(&dev->index_hlist,
238 dev_index_hash(net, dev->ifindex));
239 write_unlock_bh(&dev_base_lock);
243 /* Device list removal
244 * caller must respect a RCU grace period before freeing/reusing dev
246 static void unlist_netdevice(struct net_device *dev)
250 /* Unlink dev from the device chain */
251 write_lock_bh(&dev_base_lock);
252 list_del_rcu(&dev->dev_list);
253 hlist_del_rcu(&dev->name_hlist);
254 hlist_del_rcu(&dev->index_hlist);
255 write_unlock_bh(&dev_base_lock);
262 static RAW_NOTIFIER_HEAD(netdev_chain);
265 * Device drivers call our routines to queue packets here. We empty the
266 * queue in the local softnet handler.
269 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
270 EXPORT_PER_CPU_SYMBOL(softnet_data);
272 #ifdef CONFIG_LOCKDEP
274 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
275 * according to dev->type
277 static const unsigned short netdev_lock_type[] =
278 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
279 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
280 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
281 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
282 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
283 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
284 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
285 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
286 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
287 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
288 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
289 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
290 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
291 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
292 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
293 ARPHRD_VOID, ARPHRD_NONE};
295 static const char *const netdev_lock_name[] =
296 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
297 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
298 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
299 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
300 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
301 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
302 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
303 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
304 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
305 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
306 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
307 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
308 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
309 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
310 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
311 "_xmit_VOID", "_xmit_NONE"};
313 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
316 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
320 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
321 if (netdev_lock_type[i] == dev_type)
323 /* the last key is used by default */
324 return ARRAY_SIZE(netdev_lock_type) - 1;
327 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
328 unsigned short dev_type)
332 i = netdev_lock_pos(dev_type);
333 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
334 netdev_lock_name[i]);
337 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
341 i = netdev_lock_pos(dev->type);
342 lockdep_set_class_and_name(&dev->addr_list_lock,
343 &netdev_addr_lock_key[i],
344 netdev_lock_name[i]);
347 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
348 unsigned short dev_type)
351 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
356 /*******************************************************************************
358 Protocol management and registration routines
360 *******************************************************************************/
363 * Add a protocol ID to the list. Now that the input handler is
364 * smarter we can dispense with all the messy stuff that used to be
367 * BEWARE!!! Protocol handlers, mangling input packets,
368 * MUST BE last in hash buckets and checking protocol handlers
369 * MUST start from promiscuous ptype_all chain in net_bh.
370 * It is true now, do not change it.
371 * Explanation follows: if protocol handler, mangling packet, will
372 * be the first on list, it is not able to sense, that packet
373 * is cloned and should be copied-on-write, so that it will
374 * change it and subsequent readers will get broken packet.
378 static inline struct list_head *ptype_head(const struct packet_type *pt)
380 if (pt->type == htons(ETH_P_ALL))
383 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
387 * dev_add_pack - add packet handler
388 * @pt: packet type declaration
390 * Add a protocol handler to the networking stack. The passed &packet_type
391 * is linked into kernel lists and may not be freed until it has been
392 * removed from the kernel lists.
394 * This call does not sleep therefore it can not
395 * guarantee all CPU's that are in middle of receiving packets
396 * will see the new packet type (until the next received packet).
399 void dev_add_pack(struct packet_type *pt)
401 struct list_head *head = ptype_head(pt);
403 spin_lock(&ptype_lock);
404 list_add_rcu(&pt->list, head);
405 spin_unlock(&ptype_lock);
407 EXPORT_SYMBOL(dev_add_pack);
410 * __dev_remove_pack - remove packet handler
411 * @pt: packet type declaration
413 * Remove a protocol handler that was previously added to the kernel
414 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
415 * from the kernel lists and can be freed or reused once this function
418 * The packet type might still be in use by receivers
419 * and must not be freed until after all the CPU's have gone
420 * through a quiescent state.
422 void __dev_remove_pack(struct packet_type *pt)
424 struct list_head *head = ptype_head(pt);
425 struct packet_type *pt1;
427 spin_lock(&ptype_lock);
429 list_for_each_entry(pt1, head, list) {
431 list_del_rcu(&pt->list);
436 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
438 spin_unlock(&ptype_lock);
440 EXPORT_SYMBOL(__dev_remove_pack);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type *pt)
456 __dev_remove_pack(pt);
460 EXPORT_SYMBOL(dev_remove_pack);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name, struct ifmap *map)
482 struct netdev_boot_setup *s;
486 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
487 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
488 memset(s[i].name, 0, sizeof(s[i].name));
489 strlcpy(s[i].name, name, IFNAMSIZ);
490 memcpy(&s[i].map, map, sizeof(s[i].map));
495 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device *dev)
509 struct netdev_boot_setup *s = dev_boot_setup;
512 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
513 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
514 !strcmp(dev->name, s[i].name)) {
515 dev->irq = s[i].map.irq;
516 dev->base_addr = s[i].map.base_addr;
517 dev->mem_start = s[i].map.mem_start;
518 dev->mem_end = s[i].map.mem_end;
524 EXPORT_SYMBOL(netdev_boot_setup_check);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix, int unit)
539 const struct netdev_boot_setup *s = dev_boot_setup;
543 sprintf(name, "%s%d", prefix, unit);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net, name))
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
553 if (!strcmp(name, s[i].name))
554 return s[i].map.base_addr;
559 * Saves at boot time configured settings for any netdevice.
561 int __init netdev_boot_setup(char *str)
566 str = get_options(str, ARRAY_SIZE(ints), ints);
571 memset(&map, 0, sizeof(map));
575 map.base_addr = ints[2];
577 map.mem_start = ints[3];
579 map.mem_end = ints[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str, &map);
585 __setup("netdev=", netdev_boot_setup);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device *__dev_get_by_name(struct net *net, const char *name)
607 struct hlist_node *p;
608 struct net_device *dev;
609 struct hlist_head *head = dev_name_hash(net, name);
611 hlist_for_each_entry(dev, p, head, name_hlist)
612 if (!strncmp(dev->name, name, IFNAMSIZ))
617 EXPORT_SYMBOL(__dev_get_by_name);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
633 struct hlist_node *p;
634 struct net_device *dev;
635 struct hlist_head *head = dev_name_hash(net, name);
637 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
638 if (!strncmp(dev->name, name, IFNAMSIZ))
643 EXPORT_SYMBOL(dev_get_by_name_rcu);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device *dev_get_by_name(struct net *net, const char *name)
659 struct net_device *dev;
662 dev = dev_get_by_name_rcu(net, name);
668 EXPORT_SYMBOL(dev_get_by_name);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
684 struct hlist_node *p;
685 struct net_device *dev;
686 struct hlist_head *head = dev_index_hash(net, ifindex);
688 hlist_for_each_entry(dev, p, head, index_hlist)
689 if (dev->ifindex == ifindex)
694 EXPORT_SYMBOL(__dev_get_by_index);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
709 struct hlist_node *p;
710 struct net_device *dev;
711 struct hlist_head *head = dev_index_hash(net, ifindex);
713 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
714 if (dev->ifindex == ifindex)
719 EXPORT_SYMBOL(dev_get_by_index_rcu);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device *dev_get_by_index(struct net *net, int ifindex)
735 struct net_device *dev;
738 dev = dev_get_by_index_rcu(net, ifindex);
744 EXPORT_SYMBOL(dev_get_by_index);
747 * dev_getbyhwaddr_rcu - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device.
754 * The caller must hold RCU or RTNL.
755 * The returned device has not had its ref count increased
756 * and the caller must therefore be careful about locking
760 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
763 struct net_device *dev;
765 for_each_netdev_rcu(net, dev)
766 if (dev->type == type &&
767 !memcmp(dev->dev_addr, ha, dev->addr_len))
772 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
774 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
776 struct net_device *dev;
779 for_each_netdev(net, dev)
780 if (dev->type == type)
785 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
787 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
789 struct net_device *dev, *ret = NULL;
792 for_each_netdev_rcu(net, dev)
793 if (dev->type == type) {
801 EXPORT_SYMBOL(dev_getfirstbyhwtype);
804 * dev_get_by_flags_rcu - find any device with given flags
805 * @net: the applicable net namespace
806 * @if_flags: IFF_* values
807 * @mask: bitmask of bits in if_flags to check
809 * Search for any interface with the given flags. Returns NULL if a device
810 * is not found or a pointer to the device. Must be called inside
811 * rcu_read_lock(), and result refcount is unchanged.
814 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
817 struct net_device *dev, *ret;
820 for_each_netdev_rcu(net, dev) {
821 if (((dev->flags ^ if_flags) & mask) == 0) {
828 EXPORT_SYMBOL(dev_get_by_flags_rcu);
831 * dev_valid_name - check if name is okay for network device
834 * Network device names need to be valid file names to
835 * to allow sysfs to work. We also disallow any kind of
838 int dev_valid_name(const char *name)
842 if (strlen(name) >= IFNAMSIZ)
844 if (!strcmp(name, ".") || !strcmp(name, ".."))
848 if (*name == '/' || isspace(*name))
854 EXPORT_SYMBOL(dev_valid_name);
857 * __dev_alloc_name - allocate a name for a device
858 * @net: network namespace to allocate the device name in
859 * @name: name format string
860 * @buf: scratch buffer and result name string
862 * Passed a format string - eg "lt%d" it will try and find a suitable
863 * id. It scans list of devices to build up a free map, then chooses
864 * the first empty slot. The caller must hold the dev_base or rtnl lock
865 * while allocating the name and adding the device in order to avoid
867 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
868 * Returns the number of the unit assigned or a negative errno code.
871 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
875 const int max_netdevices = 8*PAGE_SIZE;
876 unsigned long *inuse;
877 struct net_device *d;
879 p = strnchr(name, IFNAMSIZ-1, '%');
882 * Verify the string as this thing may have come from
883 * the user. There must be either one "%d" and no other "%"
886 if (p[1] != 'd' || strchr(p + 2, '%'))
889 /* Use one page as a bit array of possible slots */
890 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
894 for_each_netdev(net, d) {
895 if (!sscanf(d->name, name, &i))
897 if (i < 0 || i >= max_netdevices)
900 /* avoid cases where sscanf is not exact inverse of printf */
901 snprintf(buf, IFNAMSIZ, name, i);
902 if (!strncmp(buf, d->name, IFNAMSIZ))
906 i = find_first_zero_bit(inuse, max_netdevices);
907 free_page((unsigned long) inuse);
911 snprintf(buf, IFNAMSIZ, name, i);
912 if (!__dev_get_by_name(net, buf))
915 /* It is possible to run out of possible slots
916 * when the name is long and there isn't enough space left
917 * for the digits, or if all bits are used.
923 * dev_alloc_name - allocate a name for a device
925 * @name: name format string
927 * Passed a format string - eg "lt%d" it will try and find a suitable
928 * id. It scans list of devices to build up a free map, then chooses
929 * the first empty slot. The caller must hold the dev_base or rtnl lock
930 * while allocating the name and adding the device in order to avoid
932 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
933 * Returns the number of the unit assigned or a negative errno code.
936 int dev_alloc_name(struct net_device *dev, const char *name)
942 BUG_ON(!dev_net(dev));
944 ret = __dev_alloc_name(net, name, buf);
946 strlcpy(dev->name, buf, IFNAMSIZ);
949 EXPORT_SYMBOL(dev_alloc_name);
951 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
955 BUG_ON(!dev_net(dev));
958 if (!dev_valid_name(name))
961 if (fmt && strchr(name, '%'))
962 return dev_alloc_name(dev, name);
963 else if (__dev_get_by_name(net, name))
965 else if (dev->name != name)
966 strlcpy(dev->name, name, IFNAMSIZ);
972 * dev_change_name - change name of a device
974 * @newname: name (or format string) must be at least IFNAMSIZ
976 * Change name of a device, can pass format strings "eth%d".
979 int dev_change_name(struct net_device *dev, const char *newname)
981 char oldname[IFNAMSIZ];
987 BUG_ON(!dev_net(dev));
990 if (dev->flags & IFF_UP)
993 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
996 memcpy(oldname, dev->name, IFNAMSIZ);
998 err = dev_get_valid_name(dev, newname, 1);
1003 ret = device_rename(&dev->dev, dev->name);
1005 memcpy(dev->name, oldname, IFNAMSIZ);
1009 write_lock_bh(&dev_base_lock);
1010 hlist_del(&dev->name_hlist);
1011 write_unlock_bh(&dev_base_lock);
1015 write_lock_bh(&dev_base_lock);
1016 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1017 write_unlock_bh(&dev_base_lock);
1019 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1020 ret = notifier_to_errno(ret);
1023 /* err >= 0 after dev_alloc_name() or stores the first errno */
1026 memcpy(dev->name, oldname, IFNAMSIZ);
1030 "%s: name change rollback failed: %d.\n",
1039 * dev_set_alias - change ifalias of a device
1041 * @alias: name up to IFALIASZ
1042 * @len: limit of bytes to copy from info
1044 * Set ifalias for a device,
1046 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1050 if (len >= IFALIASZ)
1055 kfree(dev->ifalias);
1056 dev->ifalias = NULL;
1061 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1065 strlcpy(dev->ifalias, alias, len+1);
1071 * netdev_features_change - device changes features
1072 * @dev: device to cause notification
1074 * Called to indicate a device has changed features.
1076 void netdev_features_change(struct net_device *dev)
1078 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1080 EXPORT_SYMBOL(netdev_features_change);
1083 * netdev_state_change - device changes state
1084 * @dev: device to cause notification
1086 * Called to indicate a device has changed state. This function calls
1087 * the notifier chains for netdev_chain and sends a NEWLINK message
1088 * to the routing socket.
1090 void netdev_state_change(struct net_device *dev)
1092 if (dev->flags & IFF_UP) {
1093 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1094 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1097 EXPORT_SYMBOL(netdev_state_change);
1099 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1101 return call_netdevice_notifiers(event, dev);
1103 EXPORT_SYMBOL(netdev_bonding_change);
1106 * dev_load - load a network module
1107 * @net: the applicable net namespace
1108 * @name: name of interface
1110 * If a network interface is not present and the process has suitable
1111 * privileges this function loads the module. If module loading is not
1112 * available in this kernel then it becomes a nop.
1115 void dev_load(struct net *net, const char *name)
1117 struct net_device *dev;
1120 dev = dev_get_by_name_rcu(net, name);
1123 if (!dev && capable(CAP_NET_ADMIN))
1124 request_module("%s", name);
1126 EXPORT_SYMBOL(dev_load);
1128 static int __dev_open(struct net_device *dev)
1130 const struct net_device_ops *ops = dev->netdev_ops;
1136 * Is it even present?
1138 if (!netif_device_present(dev))
1141 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1142 ret = notifier_to_errno(ret);
1147 * Call device private open method
1149 set_bit(__LINK_STATE_START, &dev->state);
1151 if (ops->ndo_validate_addr)
1152 ret = ops->ndo_validate_addr(dev);
1154 if (!ret && ops->ndo_open)
1155 ret = ops->ndo_open(dev);
1158 * If it went open OK then:
1162 clear_bit(__LINK_STATE_START, &dev->state);
1167 dev->flags |= IFF_UP;
1172 net_dmaengine_get();
1175 * Initialize multicasting status
1177 dev_set_rx_mode(dev);
1180 * Wakeup transmit queue engine
1189 * dev_open - prepare an interface for use.
1190 * @dev: device to open
1192 * Takes a device from down to up state. The device's private open
1193 * function is invoked and then the multicast lists are loaded. Finally
1194 * the device is moved into the up state and a %NETDEV_UP message is
1195 * sent to the netdev notifier chain.
1197 * Calling this function on an active interface is a nop. On a failure
1198 * a negative errno code is returned.
1200 int dev_open(struct net_device *dev)
1207 if (dev->flags & IFF_UP)
1213 ret = __dev_open(dev);
1218 * ... and announce new interface.
1220 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1221 call_netdevice_notifiers(NETDEV_UP, dev);
1225 EXPORT_SYMBOL(dev_open);
1227 static int __dev_close_many(struct list_head *head)
1229 struct net_device *dev;
1234 list_for_each_entry(dev, head, unreg_list) {
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, it
1244 * 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(). */
1252 dev_deactivate_many(head);
1254 list_for_each_entry(dev, head, unreg_list) {
1255 const struct net_device_ops *ops = dev->netdev_ops;
1258 * Call the device specific close. This cannot fail.
1259 * Only if device is UP
1261 * We allow it to be called even after a DETACH hot-plug
1268 * Device is now down.
1271 dev->flags &= ~IFF_UP;
1276 net_dmaengine_put();
1282 static int __dev_close(struct net_device *dev)
1286 list_add(&dev->unreg_list, &single);
1287 return __dev_close_many(&single);
1290 static int dev_close_many(struct list_head *head)
1292 struct net_device *dev, *tmp;
1293 LIST_HEAD(tmp_list);
1295 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1296 if (!(dev->flags & IFF_UP))
1297 list_move(&dev->unreg_list, &tmp_list);
1299 __dev_close_many(head);
1302 * Tell people we are down
1304 list_for_each_entry(dev, head, unreg_list) {
1305 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1306 call_netdevice_notifiers(NETDEV_DOWN, dev);
1309 /* rollback_registered_many needs the complete original list */
1310 list_splice(&tmp_list, head);
1315 * dev_close - shutdown an interface.
1316 * @dev: device to shutdown
1318 * This function moves an active device into down state. A
1319 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1320 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1323 int dev_close(struct net_device *dev)
1327 list_add(&dev->unreg_list, &single);
1328 dev_close_many(&single);
1332 EXPORT_SYMBOL(dev_close);
1336 * dev_disable_lro - disable Large Receive Offload on a device
1339 * Disable Large Receive Offload (LRO) on a net device. Must be
1340 * called under RTNL. This is needed if received packets may be
1341 * forwarded to another interface.
1343 void dev_disable_lro(struct net_device *dev)
1345 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1346 dev->ethtool_ops->set_flags) {
1347 u32 flags = dev->ethtool_ops->get_flags(dev);
1348 if (flags & ETH_FLAG_LRO) {
1349 flags &= ~ETH_FLAG_LRO;
1350 dev->ethtool_ops->set_flags(dev, flags);
1353 WARN_ON(dev->features & NETIF_F_LRO);
1355 EXPORT_SYMBOL(dev_disable_lro);
1358 static int dev_boot_phase = 1;
1361 * Device change register/unregister. These are not inline or static
1362 * as we export them to the world.
1366 * register_netdevice_notifier - register a network notifier block
1369 * Register a notifier to be called when network device events occur.
1370 * The notifier passed is linked into the kernel structures and must
1371 * not be reused until it has been unregistered. A negative errno code
1372 * is returned on a failure.
1374 * When registered all registration and up events are replayed
1375 * to the new notifier to allow device to have a race free
1376 * view of the network device list.
1379 int register_netdevice_notifier(struct notifier_block *nb)
1381 struct net_device *dev;
1382 struct net_device *last;
1387 err = raw_notifier_chain_register(&netdev_chain, nb);
1393 for_each_netdev(net, dev) {
1394 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1395 err = notifier_to_errno(err);
1399 if (!(dev->flags & IFF_UP))
1402 nb->notifier_call(nb, NETDEV_UP, dev);
1413 for_each_netdev(net, dev) {
1417 if (dev->flags & IFF_UP) {
1418 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1419 nb->notifier_call(nb, NETDEV_DOWN, dev);
1421 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1422 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1426 raw_notifier_chain_unregister(&netdev_chain, nb);
1429 EXPORT_SYMBOL(register_netdevice_notifier);
1432 * unregister_netdevice_notifier - unregister a network notifier block
1435 * Unregister a notifier previously registered by
1436 * register_netdevice_notifier(). The notifier is unlinked into the
1437 * kernel structures and may then be reused. A negative errno code
1438 * is returned on a failure.
1441 int unregister_netdevice_notifier(struct notifier_block *nb)
1446 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1450 EXPORT_SYMBOL(unregister_netdevice_notifier);
1453 * call_netdevice_notifiers - call all network notifier blocks
1454 * @val: value passed unmodified to notifier function
1455 * @dev: net_device pointer passed unmodified to notifier function
1457 * Call all network notifier blocks. Parameters and return value
1458 * are as for raw_notifier_call_chain().
1461 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1464 return raw_notifier_call_chain(&netdev_chain, val, dev);
1467 /* When > 0 there are consumers of rx skb time stamps */
1468 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1470 void net_enable_timestamp(void)
1472 atomic_inc(&netstamp_needed);
1474 EXPORT_SYMBOL(net_enable_timestamp);
1476 void net_disable_timestamp(void)
1478 atomic_dec(&netstamp_needed);
1480 EXPORT_SYMBOL(net_disable_timestamp);
1482 static inline void net_timestamp_set(struct sk_buff *skb)
1484 if (atomic_read(&netstamp_needed))
1485 __net_timestamp(skb);
1487 skb->tstamp.tv64 = 0;
1490 static inline void net_timestamp_check(struct sk_buff *skb)
1492 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1493 __net_timestamp(skb);
1497 * dev_forward_skb - loopback an skb to another netif
1499 * @dev: destination network device
1500 * @skb: buffer to forward
1503 * NET_RX_SUCCESS (no congestion)
1504 * NET_RX_DROP (packet was dropped, but freed)
1506 * dev_forward_skb can be used for injecting an skb from the
1507 * start_xmit function of one device into the receive queue
1508 * of another device.
1510 * The receiving device may be in another namespace, so
1511 * we have to clear all information in the skb that could
1512 * impact namespace isolation.
1514 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1519 if (unlikely(!(dev->flags & IFF_UP) ||
1520 (skb->len > (dev->mtu + dev->hard_header_len + VLAN_HLEN)))) {
1521 atomic_long_inc(&dev->rx_dropped);
1525 skb_set_dev(skb, dev);
1526 skb->tstamp.tv64 = 0;
1527 skb->pkt_type = PACKET_HOST;
1528 skb->protocol = eth_type_trans(skb, dev);
1529 return netif_rx(skb);
1531 EXPORT_SYMBOL_GPL(dev_forward_skb);
1533 static inline int deliver_skb(struct sk_buff *skb,
1534 struct packet_type *pt_prev,
1535 struct net_device *orig_dev)
1537 atomic_inc(&skb->users);
1538 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1542 * Support routine. Sends outgoing frames to any network
1543 * taps currently in use.
1546 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1548 struct packet_type *ptype;
1549 struct sk_buff *skb2 = NULL;
1550 struct packet_type *pt_prev = NULL;
1553 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1554 /* Never send packets back to the socket
1555 * they originated from - MvS (miquels@drinkel.ow.org)
1557 if ((ptype->dev == dev || !ptype->dev) &&
1558 (ptype->af_packet_priv == NULL ||
1559 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1561 deliver_skb(skb2, pt_prev, skb->dev);
1566 skb2 = skb_clone(skb, GFP_ATOMIC);
1570 net_timestamp_set(skb2);
1572 /* skb->nh should be correctly
1573 set by sender, so that the second statement is
1574 just protection against buggy protocols.
1576 skb_reset_mac_header(skb2);
1578 if (skb_network_header(skb2) < skb2->data ||
1579 skb2->network_header > skb2->tail) {
1580 if (net_ratelimit())
1581 printk(KERN_CRIT "protocol %04x is "
1583 ntohs(skb2->protocol),
1585 skb_reset_network_header(skb2);
1588 skb2->transport_header = skb2->network_header;
1589 skb2->pkt_type = PACKET_OUTGOING;
1594 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1598 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1599 * @dev: Network device
1600 * @txq: number of queues available
1602 * If real_num_tx_queues is changed the tc mappings may no longer be
1603 * valid. To resolve this verify the tc mapping remains valid and if
1604 * not NULL the mapping. With no priorities mapping to this
1605 * offset/count pair it will no longer be used. In the worst case TC0
1606 * is invalid nothing can be done so disable priority mappings. If is
1607 * expected that drivers will fix this mapping if they can before
1608 * calling netif_set_real_num_tx_queues.
1610 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1613 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1615 /* If TC0 is invalidated disable TC mapping */
1616 if (tc->offset + tc->count > txq) {
1617 pr_warning("Number of in use tx queues changed "
1618 "invalidating tc mappings. Priority "
1619 "traffic classification disabled!\n");
1624 /* Invalidated prio to tc mappings set to TC0 */
1625 for (i = 1; i < TC_BITMASK + 1; i++) {
1626 int q = netdev_get_prio_tc_map(dev, i);
1628 tc = &dev->tc_to_txq[q];
1629 if (tc->offset + tc->count > txq) {
1630 pr_warning("Number of in use tx queues "
1631 "changed. Priority %i to tc "
1632 "mapping %i is no longer valid "
1633 "setting map to 0\n",
1635 netdev_set_prio_tc_map(dev, i, 0);
1641 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1642 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1644 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1648 if (txq < 1 || txq > dev->num_tx_queues)
1651 if (dev->reg_state == NETREG_REGISTERED) {
1654 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1660 netif_setup_tc(dev, txq);
1662 if (txq < dev->real_num_tx_queues)
1663 qdisc_reset_all_tx_gt(dev, txq);
1666 dev->real_num_tx_queues = txq;
1669 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1673 * netif_set_real_num_rx_queues - set actual number of RX queues used
1674 * @dev: Network device
1675 * @rxq: Actual number of RX queues
1677 * This must be called either with the rtnl_lock held or before
1678 * registration of the net device. Returns 0 on success, or a
1679 * negative error code. If called before registration, it always
1682 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1686 if (rxq < 1 || rxq > dev->num_rx_queues)
1689 if (dev->reg_state == NETREG_REGISTERED) {
1692 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1698 dev->real_num_rx_queues = rxq;
1701 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1704 static inline void __netif_reschedule(struct Qdisc *q)
1706 struct softnet_data *sd;
1707 unsigned long flags;
1709 local_irq_save(flags);
1710 sd = &__get_cpu_var(softnet_data);
1711 q->next_sched = NULL;
1712 *sd->output_queue_tailp = q;
1713 sd->output_queue_tailp = &q->next_sched;
1714 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1715 local_irq_restore(flags);
1718 void __netif_schedule(struct Qdisc *q)
1720 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1721 __netif_reschedule(q);
1723 EXPORT_SYMBOL(__netif_schedule);
1725 void dev_kfree_skb_irq(struct sk_buff *skb)
1727 if (atomic_dec_and_test(&skb->users)) {
1728 struct softnet_data *sd;
1729 unsigned long flags;
1731 local_irq_save(flags);
1732 sd = &__get_cpu_var(softnet_data);
1733 skb->next = sd->completion_queue;
1734 sd->completion_queue = skb;
1735 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1736 local_irq_restore(flags);
1739 EXPORT_SYMBOL(dev_kfree_skb_irq);
1741 void dev_kfree_skb_any(struct sk_buff *skb)
1743 if (in_irq() || irqs_disabled())
1744 dev_kfree_skb_irq(skb);
1748 EXPORT_SYMBOL(dev_kfree_skb_any);
1752 * netif_device_detach - mark device as removed
1753 * @dev: network device
1755 * Mark device as removed from system and therefore no longer available.
1757 void netif_device_detach(struct net_device *dev)
1759 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1760 netif_running(dev)) {
1761 netif_tx_stop_all_queues(dev);
1764 EXPORT_SYMBOL(netif_device_detach);
1767 * netif_device_attach - mark device as attached
1768 * @dev: network device
1770 * Mark device as attached from system and restart if needed.
1772 void netif_device_attach(struct net_device *dev)
1774 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1775 netif_running(dev)) {
1776 netif_tx_wake_all_queues(dev);
1777 __netdev_watchdog_up(dev);
1780 EXPORT_SYMBOL(netif_device_attach);
1783 * skb_dev_set -- assign a new device to a buffer
1784 * @skb: buffer for the new device
1785 * @dev: network device
1787 * If an skb is owned by a device already, we have to reset
1788 * all data private to the namespace a device belongs to
1789 * before assigning it a new device.
1791 #ifdef CONFIG_NET_NS
1792 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1795 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1798 skb_init_secmark(skb);
1802 skb->ipvs_property = 0;
1803 #ifdef CONFIG_NET_SCHED
1809 EXPORT_SYMBOL(skb_set_dev);
1810 #endif /* CONFIG_NET_NS */
1813 * Invalidate hardware checksum when packet is to be mangled, and
1814 * complete checksum manually on outgoing path.
1816 int skb_checksum_help(struct sk_buff *skb)
1819 int ret = 0, offset;
1821 if (skb->ip_summed == CHECKSUM_COMPLETE)
1822 goto out_set_summed;
1824 if (unlikely(skb_shinfo(skb)->gso_size)) {
1825 /* Let GSO fix up the checksum. */
1826 goto out_set_summed;
1829 offset = skb_checksum_start_offset(skb);
1830 BUG_ON(offset >= skb_headlen(skb));
1831 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1833 offset += skb->csum_offset;
1834 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1836 if (skb_cloned(skb) &&
1837 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1838 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1843 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1845 skb->ip_summed = CHECKSUM_NONE;
1849 EXPORT_SYMBOL(skb_checksum_help);
1852 * skb_gso_segment - Perform segmentation on skb.
1853 * @skb: buffer to segment
1854 * @features: features for the output path (see dev->features)
1856 * This function segments the given skb and returns a list of segments.
1858 * It may return NULL if the skb requires no segmentation. This is
1859 * only possible when GSO is used for verifying header integrity.
1861 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1863 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1864 struct packet_type *ptype;
1865 __be16 type = skb->protocol;
1866 int vlan_depth = ETH_HLEN;
1869 while (type == htons(ETH_P_8021Q)) {
1870 struct vlan_hdr *vh;
1872 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1873 return ERR_PTR(-EINVAL);
1875 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1876 type = vh->h_vlan_encapsulated_proto;
1877 vlan_depth += VLAN_HLEN;
1880 skb_reset_mac_header(skb);
1881 skb->mac_len = skb->network_header - skb->mac_header;
1882 __skb_pull(skb, skb->mac_len);
1884 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1885 struct net_device *dev = skb->dev;
1886 struct ethtool_drvinfo info = {};
1888 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1889 dev->ethtool_ops->get_drvinfo(dev, &info);
1891 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1892 info.driver, dev ? dev->features : 0L,
1893 skb->sk ? skb->sk->sk_route_caps : 0L,
1894 skb->len, skb->data_len, skb->ip_summed);
1896 if (skb_header_cloned(skb) &&
1897 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1898 return ERR_PTR(err);
1902 list_for_each_entry_rcu(ptype,
1903 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1904 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1905 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1906 err = ptype->gso_send_check(skb);
1907 segs = ERR_PTR(err);
1908 if (err || skb_gso_ok(skb, features))
1910 __skb_push(skb, (skb->data -
1911 skb_network_header(skb)));
1913 segs = ptype->gso_segment(skb, features);
1919 __skb_push(skb, skb->data - skb_mac_header(skb));
1923 EXPORT_SYMBOL(skb_gso_segment);
1925 /* Take action when hardware reception checksum errors are detected. */
1927 void netdev_rx_csum_fault(struct net_device *dev)
1929 if (net_ratelimit()) {
1930 printk(KERN_ERR "%s: hw csum failure.\n",
1931 dev ? dev->name : "<unknown>");
1935 EXPORT_SYMBOL(netdev_rx_csum_fault);
1938 /* Actually, we should eliminate this check as soon as we know, that:
1939 * 1. IOMMU is present and allows to map all the memory.
1940 * 2. No high memory really exists on this machine.
1943 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1945 #ifdef CONFIG_HIGHMEM
1947 if (!(dev->features & NETIF_F_HIGHDMA)) {
1948 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1949 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1953 if (PCI_DMA_BUS_IS_PHYS) {
1954 struct device *pdev = dev->dev.parent;
1958 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1959 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1960 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1969 void (*destructor)(struct sk_buff *skb);
1972 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1974 static void dev_gso_skb_destructor(struct sk_buff *skb)
1976 struct dev_gso_cb *cb;
1979 struct sk_buff *nskb = skb->next;
1981 skb->next = nskb->next;
1984 } while (skb->next);
1986 cb = DEV_GSO_CB(skb);
1988 cb->destructor(skb);
1992 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1993 * @skb: buffer to segment
1994 * @features: device features as applicable to this skb
1996 * This function segments the given skb and stores the list of segments
1999 static int dev_gso_segment(struct sk_buff *skb, int features)
2001 struct sk_buff *segs;
2003 segs = skb_gso_segment(skb, features);
2005 /* Verifying header integrity only. */
2010 return PTR_ERR(segs);
2013 DEV_GSO_CB(skb)->destructor = skb->destructor;
2014 skb->destructor = dev_gso_skb_destructor;
2020 * Try to orphan skb early, right before transmission by the device.
2021 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2022 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2024 static inline void skb_orphan_try(struct sk_buff *skb)
2026 struct sock *sk = skb->sk;
2028 if (sk && !skb_shinfo(skb)->tx_flags) {
2029 /* skb_tx_hash() wont be able to get sk.
2030 * We copy sk_hash into skb->rxhash
2033 skb->rxhash = sk->sk_hash;
2038 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2040 return ((features & NETIF_F_GEN_CSUM) ||
2041 ((features & NETIF_F_V4_CSUM) &&
2042 protocol == htons(ETH_P_IP)) ||
2043 ((features & NETIF_F_V6_CSUM) &&
2044 protocol == htons(ETH_P_IPV6)) ||
2045 ((features & NETIF_F_FCOE_CRC) &&
2046 protocol == htons(ETH_P_FCOE)));
2049 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2051 if (!can_checksum_protocol(features, protocol)) {
2052 features &= ~NETIF_F_ALL_CSUM;
2053 features &= ~NETIF_F_SG;
2054 } else if (illegal_highdma(skb->dev, skb)) {
2055 features &= ~NETIF_F_SG;
2061 u32 netif_skb_features(struct sk_buff *skb)
2063 __be16 protocol = skb->protocol;
2064 u32 features = skb->dev->features;
2066 if (protocol == htons(ETH_P_8021Q)) {
2067 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2068 protocol = veh->h_vlan_encapsulated_proto;
2069 } else if (!vlan_tx_tag_present(skb)) {
2070 return harmonize_features(skb, protocol, features);
2073 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2075 if (protocol != htons(ETH_P_8021Q)) {
2076 return harmonize_features(skb, protocol, features);
2078 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2079 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2080 return harmonize_features(skb, protocol, features);
2083 EXPORT_SYMBOL(netif_skb_features);
2086 * Returns true if either:
2087 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2088 * 2. skb is fragmented and the device does not support SG, or if
2089 * at least one of fragments is in highmem and device does not
2090 * support DMA from it.
2092 static inline int skb_needs_linearize(struct sk_buff *skb,
2095 return skb_is_nonlinear(skb) &&
2096 ((skb_has_frag_list(skb) &&
2097 !(features & NETIF_F_FRAGLIST)) ||
2098 (skb_shinfo(skb)->nr_frags &&
2099 !(features & NETIF_F_SG)));
2102 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2103 struct netdev_queue *txq)
2105 const struct net_device_ops *ops = dev->netdev_ops;
2106 int rc = NETDEV_TX_OK;
2108 if (likely(!skb->next)) {
2112 * If device doesnt need skb->dst, release it right now while
2113 * its hot in this cpu cache
2115 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2118 if (!list_empty(&ptype_all))
2119 dev_queue_xmit_nit(skb, dev);
2121 skb_orphan_try(skb);
2123 features = netif_skb_features(skb);
2125 if (vlan_tx_tag_present(skb) &&
2126 !(features & NETIF_F_HW_VLAN_TX)) {
2127 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2134 if (netif_needs_gso(skb, features)) {
2135 if (unlikely(dev_gso_segment(skb, features)))
2140 if (skb_needs_linearize(skb, features) &&
2141 __skb_linearize(skb))
2144 /* If packet is not checksummed and device does not
2145 * support checksumming for this protocol, complete
2146 * checksumming here.
2148 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2149 skb_set_transport_header(skb,
2150 skb_checksum_start_offset(skb));
2151 if (!(features & NETIF_F_ALL_CSUM) &&
2152 skb_checksum_help(skb))
2157 rc = ops->ndo_start_xmit(skb, dev);
2158 trace_net_dev_xmit(skb, rc);
2159 if (rc == NETDEV_TX_OK)
2160 txq_trans_update(txq);
2166 struct sk_buff *nskb = skb->next;
2168 skb->next = nskb->next;
2172 * If device doesnt need nskb->dst, release it right now while
2173 * its hot in this cpu cache
2175 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2178 rc = ops->ndo_start_xmit(nskb, dev);
2179 trace_net_dev_xmit(nskb, rc);
2180 if (unlikely(rc != NETDEV_TX_OK)) {
2181 if (rc & ~NETDEV_TX_MASK)
2182 goto out_kfree_gso_skb;
2183 nskb->next = skb->next;
2187 txq_trans_update(txq);
2188 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2189 return NETDEV_TX_BUSY;
2190 } while (skb->next);
2193 if (likely(skb->next == NULL))
2194 skb->destructor = DEV_GSO_CB(skb)->destructor;
2201 static u32 hashrnd __read_mostly;
2204 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2205 * to be used as a distribution range.
2207 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2208 unsigned int num_tx_queues)
2212 u16 qcount = num_tx_queues;
2214 if (skb_rx_queue_recorded(skb)) {
2215 hash = skb_get_rx_queue(skb);
2216 while (unlikely(hash >= num_tx_queues))
2217 hash -= num_tx_queues;
2222 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2223 qoffset = dev->tc_to_txq[tc].offset;
2224 qcount = dev->tc_to_txq[tc].count;
2227 if (skb->sk && skb->sk->sk_hash)
2228 hash = skb->sk->sk_hash;
2230 hash = (__force u16) skb->protocol ^ skb->rxhash;
2231 hash = jhash_1word(hash, hashrnd);
2233 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2235 EXPORT_SYMBOL(__skb_tx_hash);
2237 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2239 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2240 if (net_ratelimit()) {
2241 pr_warning("%s selects TX queue %d, but "
2242 "real number of TX queues is %d\n",
2243 dev->name, queue_index, dev->real_num_tx_queues);
2250 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2253 struct xps_dev_maps *dev_maps;
2254 struct xps_map *map;
2255 int queue_index = -1;
2258 dev_maps = rcu_dereference(dev->xps_maps);
2260 map = rcu_dereference(
2261 dev_maps->cpu_map[raw_smp_processor_id()]);
2264 queue_index = map->queues[0];
2267 if (skb->sk && skb->sk->sk_hash)
2268 hash = skb->sk->sk_hash;
2270 hash = (__force u16) skb->protocol ^
2272 hash = jhash_1word(hash, hashrnd);
2273 queue_index = map->queues[
2274 ((u64)hash * map->len) >> 32];
2276 if (unlikely(queue_index >= dev->real_num_tx_queues))
2288 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2289 struct sk_buff *skb)
2292 const struct net_device_ops *ops = dev->netdev_ops;
2294 if (dev->real_num_tx_queues == 1)
2296 else if (ops->ndo_select_queue) {
2297 queue_index = ops->ndo_select_queue(dev, skb);
2298 queue_index = dev_cap_txqueue(dev, queue_index);
2300 struct sock *sk = skb->sk;
2301 queue_index = sk_tx_queue_get(sk);
2303 if (queue_index < 0 || skb->ooo_okay ||
2304 queue_index >= dev->real_num_tx_queues) {
2305 int old_index = queue_index;
2307 queue_index = get_xps_queue(dev, skb);
2308 if (queue_index < 0)
2309 queue_index = skb_tx_hash(dev, skb);
2311 if (queue_index != old_index && sk) {
2312 struct dst_entry *dst =
2313 rcu_dereference_check(sk->sk_dst_cache, 1);
2315 if (dst && skb_dst(skb) == dst)
2316 sk_tx_queue_set(sk, queue_index);
2321 skb_set_queue_mapping(skb, queue_index);
2322 return netdev_get_tx_queue(dev, queue_index);
2325 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2326 struct net_device *dev,
2327 struct netdev_queue *txq)
2329 spinlock_t *root_lock = qdisc_lock(q);
2333 qdisc_skb_cb(skb)->pkt_len = skb->len;
2334 qdisc_calculate_pkt_len(skb, q);
2336 * Heuristic to force contended enqueues to serialize on a
2337 * separate lock before trying to get qdisc main lock.
2338 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2339 * and dequeue packets faster.
2341 contended = qdisc_is_running(q);
2342 if (unlikely(contended))
2343 spin_lock(&q->busylock);
2345 spin_lock(root_lock);
2346 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2349 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2350 qdisc_run_begin(q)) {
2352 * This is a work-conserving queue; there are no old skbs
2353 * waiting to be sent out; and the qdisc is not running -
2354 * xmit the skb directly.
2356 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2359 qdisc_bstats_update(q, skb);
2361 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2362 if (unlikely(contended)) {
2363 spin_unlock(&q->busylock);
2370 rc = NET_XMIT_SUCCESS;
2373 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2374 if (qdisc_run_begin(q)) {
2375 if (unlikely(contended)) {
2376 spin_unlock(&q->busylock);
2382 spin_unlock(root_lock);
2383 if (unlikely(contended))
2384 spin_unlock(&q->busylock);
2388 static DEFINE_PER_CPU(int, xmit_recursion);
2389 #define RECURSION_LIMIT 10
2392 * dev_queue_xmit - transmit a buffer
2393 * @skb: buffer to transmit
2395 * Queue a buffer for transmission to a network device. The caller must
2396 * have set the device and priority and built the buffer before calling
2397 * this function. The function can be called from an interrupt.
2399 * A negative errno code is returned on a failure. A success does not
2400 * guarantee the frame will be transmitted as it may be dropped due
2401 * to congestion or traffic shaping.
2403 * -----------------------------------------------------------------------------------
2404 * I notice this method can also return errors from the queue disciplines,
2405 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2408 * Regardless of the return value, the skb is consumed, so it is currently
2409 * difficult to retry a send to this method. (You can bump the ref count
2410 * before sending to hold a reference for retry if you are careful.)
2412 * When calling this method, interrupts MUST be enabled. This is because
2413 * the BH enable code must have IRQs enabled so that it will not deadlock.
2416 int dev_queue_xmit(struct sk_buff *skb)
2418 struct net_device *dev = skb->dev;
2419 struct netdev_queue *txq;
2423 /* Disable soft irqs for various locks below. Also
2424 * stops preemption for RCU.
2428 txq = dev_pick_tx(dev, skb);
2429 q = rcu_dereference_bh(txq->qdisc);
2431 #ifdef CONFIG_NET_CLS_ACT
2432 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2434 trace_net_dev_queue(skb);
2436 rc = __dev_xmit_skb(skb, q, dev, txq);
2440 /* The device has no queue. Common case for software devices:
2441 loopback, all the sorts of tunnels...
2443 Really, it is unlikely that netif_tx_lock protection is necessary
2444 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2446 However, it is possible, that they rely on protection
2449 Check this and shot the lock. It is not prone from deadlocks.
2450 Either shot noqueue qdisc, it is even simpler 8)
2452 if (dev->flags & IFF_UP) {
2453 int cpu = smp_processor_id(); /* ok because BHs are off */
2455 if (txq->xmit_lock_owner != cpu) {
2457 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2458 goto recursion_alert;
2460 HARD_TX_LOCK(dev, txq, cpu);
2462 if (!netif_tx_queue_stopped(txq)) {
2463 __this_cpu_inc(xmit_recursion);
2464 rc = dev_hard_start_xmit(skb, dev, txq);
2465 __this_cpu_dec(xmit_recursion);
2466 if (dev_xmit_complete(rc)) {
2467 HARD_TX_UNLOCK(dev, txq);
2471 HARD_TX_UNLOCK(dev, txq);
2472 if (net_ratelimit())
2473 printk(KERN_CRIT "Virtual device %s asks to "
2474 "queue packet!\n", dev->name);
2476 /* Recursion is detected! It is possible,
2480 if (net_ratelimit())
2481 printk(KERN_CRIT "Dead loop on virtual device "
2482 "%s, fix it urgently!\n", dev->name);
2487 rcu_read_unlock_bh();
2492 rcu_read_unlock_bh();
2495 EXPORT_SYMBOL(dev_queue_xmit);
2498 /*=======================================================================
2500 =======================================================================*/
2502 int netdev_max_backlog __read_mostly = 1000;
2503 int netdev_tstamp_prequeue __read_mostly = 1;
2504 int netdev_budget __read_mostly = 300;
2505 int weight_p __read_mostly = 64; /* old backlog weight */
2507 /* Called with irq disabled */
2508 static inline void ____napi_schedule(struct softnet_data *sd,
2509 struct napi_struct *napi)
2511 list_add_tail(&napi->poll_list, &sd->poll_list);
2512 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2516 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2517 * and src/dst port numbers. Returns a non-zero hash number on success
2520 __u32 __skb_get_rxhash(struct sk_buff *skb)
2522 int nhoff, hash = 0, poff;
2523 struct ipv6hdr *ip6;
2526 u32 addr1, addr2, ihl;
2532 nhoff = skb_network_offset(skb);
2534 switch (skb->protocol) {
2535 case __constant_htons(ETH_P_IP):
2536 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2539 ip = (struct iphdr *) (skb->data + nhoff);
2540 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2543 ip_proto = ip->protocol;
2544 addr1 = (__force u32) ip->saddr;
2545 addr2 = (__force u32) ip->daddr;
2548 case __constant_htons(ETH_P_IPV6):
2549 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2552 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2553 ip_proto = ip6->nexthdr;
2554 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2555 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2563 poff = proto_ports_offset(ip_proto);
2565 nhoff += ihl * 4 + poff;
2566 if (pskb_may_pull(skb, nhoff + 4)) {
2567 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2568 if (ports.v16[1] < ports.v16[0])
2569 swap(ports.v16[0], ports.v16[1]);
2573 /* get a consistent hash (same value on both flow directions) */
2577 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2584 EXPORT_SYMBOL(__skb_get_rxhash);
2588 /* One global table that all flow-based protocols share. */
2589 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2590 EXPORT_SYMBOL(rps_sock_flow_table);
2592 static struct rps_dev_flow *
2593 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2594 struct rps_dev_flow *rflow, u16 next_cpu)
2598 tcpu = rflow->cpu = next_cpu;
2599 if (tcpu != RPS_NO_CPU) {
2600 #ifdef CONFIG_RFS_ACCEL
2601 struct netdev_rx_queue *rxqueue;
2602 struct rps_dev_flow_table *flow_table;
2603 struct rps_dev_flow *old_rflow;
2608 /* Should we steer this flow to a different hardware queue? */
2609 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap)
2611 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2612 if (rxq_index == skb_get_rx_queue(skb))
2615 rxqueue = dev->_rx + rxq_index;
2616 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2619 flow_id = skb->rxhash & flow_table->mask;
2620 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2621 rxq_index, flow_id);
2625 rflow = &flow_table->flows[flow_id];
2626 rflow->cpu = next_cpu;
2628 if (old_rflow->filter == rflow->filter)
2629 old_rflow->filter = RPS_NO_FILTER;
2633 per_cpu(softnet_data, tcpu).input_queue_head;
2640 * get_rps_cpu is called from netif_receive_skb and returns the target
2641 * CPU from the RPS map of the receiving queue for a given skb.
2642 * rcu_read_lock must be held on entry.
2644 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2645 struct rps_dev_flow **rflowp)
2647 struct netdev_rx_queue *rxqueue;
2648 struct rps_map *map;
2649 struct rps_dev_flow_table *flow_table;
2650 struct rps_sock_flow_table *sock_flow_table;
2654 if (skb_rx_queue_recorded(skb)) {
2655 u16 index = skb_get_rx_queue(skb);
2656 if (unlikely(index >= dev->real_num_rx_queues)) {
2657 WARN_ONCE(dev->real_num_rx_queues > 1,
2658 "%s received packet on queue %u, but number "
2659 "of RX queues is %u\n",
2660 dev->name, index, dev->real_num_rx_queues);
2663 rxqueue = dev->_rx + index;
2667 map = rcu_dereference(rxqueue->rps_map);
2669 if (map->len == 1 &&
2670 !rcu_dereference_raw(rxqueue->rps_flow_table)) {
2671 tcpu = map->cpus[0];
2672 if (cpu_online(tcpu))
2676 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2680 skb_reset_network_header(skb);
2681 if (!skb_get_rxhash(skb))
2684 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2685 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2686 if (flow_table && sock_flow_table) {
2688 struct rps_dev_flow *rflow;
2690 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2693 next_cpu = sock_flow_table->ents[skb->rxhash &
2694 sock_flow_table->mask];
2697 * If the desired CPU (where last recvmsg was done) is
2698 * different from current CPU (one in the rx-queue flow
2699 * table entry), switch if one of the following holds:
2700 * - Current CPU is unset (equal to RPS_NO_CPU).
2701 * - Current CPU is offline.
2702 * - The current CPU's queue tail has advanced beyond the
2703 * last packet that was enqueued using this table entry.
2704 * This guarantees that all previous packets for the flow
2705 * have been dequeued, thus preserving in order delivery.
2707 if (unlikely(tcpu != next_cpu) &&
2708 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2709 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2710 rflow->last_qtail)) >= 0))
2711 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2713 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2721 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2723 if (cpu_online(tcpu)) {
2733 #ifdef CONFIG_RFS_ACCEL
2736 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2737 * @dev: Device on which the filter was set
2738 * @rxq_index: RX queue index
2739 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2740 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2742 * Drivers that implement ndo_rx_flow_steer() should periodically call
2743 * this function for each installed filter and remove the filters for
2744 * which it returns %true.
2746 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2747 u32 flow_id, u16 filter_id)
2749 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2750 struct rps_dev_flow_table *flow_table;
2751 struct rps_dev_flow *rflow;
2756 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2757 if (flow_table && flow_id <= flow_table->mask) {
2758 rflow = &flow_table->flows[flow_id];
2759 cpu = ACCESS_ONCE(rflow->cpu);
2760 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2761 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2762 rflow->last_qtail) <
2763 (int)(10 * flow_table->mask)))
2769 EXPORT_SYMBOL(rps_may_expire_flow);
2771 #endif /* CONFIG_RFS_ACCEL */
2773 /* Called from hardirq (IPI) context */
2774 static void rps_trigger_softirq(void *data)
2776 struct softnet_data *sd = data;
2778 ____napi_schedule(sd, &sd->backlog);
2782 #endif /* CONFIG_RPS */
2785 * Check if this softnet_data structure is another cpu one
2786 * If yes, queue it to our IPI list and return 1
2789 static int rps_ipi_queued(struct softnet_data *sd)
2792 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2795 sd->rps_ipi_next = mysd->rps_ipi_list;
2796 mysd->rps_ipi_list = sd;
2798 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2801 #endif /* CONFIG_RPS */
2806 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2807 * queue (may be a remote CPU queue).
2809 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2810 unsigned int *qtail)
2812 struct softnet_data *sd;
2813 unsigned long flags;
2815 sd = &per_cpu(softnet_data, cpu);
2817 local_irq_save(flags);
2820 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2821 if (skb_queue_len(&sd->input_pkt_queue)) {
2823 __skb_queue_tail(&sd->input_pkt_queue, skb);
2824 input_queue_tail_incr_save(sd, qtail);
2826 local_irq_restore(flags);
2827 return NET_RX_SUCCESS;
2830 /* Schedule NAPI for backlog device
2831 * We can use non atomic operation since we own the queue lock
2833 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2834 if (!rps_ipi_queued(sd))
2835 ____napi_schedule(sd, &sd->backlog);
2843 local_irq_restore(flags);
2845 atomic_long_inc(&skb->dev->rx_dropped);
2851 * netif_rx - post buffer to the network code
2852 * @skb: buffer to post
2854 * This function receives a packet from a device driver and queues it for
2855 * the upper (protocol) levels to process. It always succeeds. The buffer
2856 * may be dropped during processing for congestion control or by the
2860 * NET_RX_SUCCESS (no congestion)
2861 * NET_RX_DROP (packet was dropped)
2865 int netif_rx(struct sk_buff *skb)
2869 /* if netpoll wants it, pretend we never saw it */
2870 if (netpoll_rx(skb))
2873 if (netdev_tstamp_prequeue)
2874 net_timestamp_check(skb);
2876 trace_netif_rx(skb);
2879 struct rps_dev_flow voidflow, *rflow = &voidflow;
2885 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2887 cpu = smp_processor_id();
2889 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2897 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2903 EXPORT_SYMBOL(netif_rx);
2905 int netif_rx_ni(struct sk_buff *skb)
2910 err = netif_rx(skb);
2911 if (local_softirq_pending())
2917 EXPORT_SYMBOL(netif_rx_ni);
2919 static void net_tx_action(struct softirq_action *h)
2921 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2923 if (sd->completion_queue) {
2924 struct sk_buff *clist;
2926 local_irq_disable();
2927 clist = sd->completion_queue;
2928 sd->completion_queue = NULL;
2932 struct sk_buff *skb = clist;
2933 clist = clist->next;
2935 WARN_ON(atomic_read(&skb->users));
2936 trace_kfree_skb(skb, net_tx_action);
2941 if (sd->output_queue) {
2944 local_irq_disable();
2945 head = sd->output_queue;
2946 sd->output_queue = NULL;
2947 sd->output_queue_tailp = &sd->output_queue;
2951 struct Qdisc *q = head;
2952 spinlock_t *root_lock;
2954 head = head->next_sched;
2956 root_lock = qdisc_lock(q);
2957 if (spin_trylock(root_lock)) {
2958 smp_mb__before_clear_bit();
2959 clear_bit(__QDISC_STATE_SCHED,
2962 spin_unlock(root_lock);
2964 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2966 __netif_reschedule(q);
2968 smp_mb__before_clear_bit();
2969 clear_bit(__QDISC_STATE_SCHED,
2977 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2978 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2979 /* This hook is defined here for ATM LANE */
2980 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2981 unsigned char *addr) __read_mostly;
2982 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2985 #ifdef CONFIG_NET_CLS_ACT
2986 /* TODO: Maybe we should just force sch_ingress to be compiled in
2987 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2988 * a compare and 2 stores extra right now if we dont have it on
2989 * but have CONFIG_NET_CLS_ACT
2990 * NOTE: This doesnt stop any functionality; if you dont have
2991 * the ingress scheduler, you just cant add policies on ingress.
2994 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2996 struct net_device *dev = skb->dev;
2997 u32 ttl = G_TC_RTTL(skb->tc_verd);
2998 int result = TC_ACT_OK;
3001 if (unlikely(MAX_RED_LOOP < ttl++)) {
3002 if (net_ratelimit())
3003 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3004 skb->skb_iif, dev->ifindex);
3008 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3009 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3012 if (q != &noop_qdisc) {
3013 spin_lock(qdisc_lock(q));
3014 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3015 result = qdisc_enqueue_root(skb, q);
3016 spin_unlock(qdisc_lock(q));
3022 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3023 struct packet_type **pt_prev,
3024 int *ret, struct net_device *orig_dev)
3026 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3028 if (!rxq || rxq->qdisc == &noop_qdisc)
3032 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3036 switch (ing_filter(skb, rxq)) {
3050 * netdev_rx_handler_register - register receive handler
3051 * @dev: device to register a handler for
3052 * @rx_handler: receive handler to register
3053 * @rx_handler_data: data pointer that is used by rx handler
3055 * Register a receive hander for a device. This handler will then be
3056 * called from __netif_receive_skb. A negative errno code is returned
3059 * The caller must hold the rtnl_mutex.
3061 int netdev_rx_handler_register(struct net_device *dev,
3062 rx_handler_func_t *rx_handler,
3063 void *rx_handler_data)
3067 if (dev->rx_handler)
3070 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3071 rcu_assign_pointer(dev->rx_handler, rx_handler);
3075 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3078 * netdev_rx_handler_unregister - unregister receive handler
3079 * @dev: device to unregister a handler from
3081 * Unregister a receive hander from a device.
3083 * The caller must hold the rtnl_mutex.
3085 void netdev_rx_handler_unregister(struct net_device *dev)
3089 rcu_assign_pointer(dev->rx_handler, NULL);
3090 rcu_assign_pointer(dev->rx_handler_data, NULL);
3092 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3094 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
3095 struct net_device *master)
3097 if (skb->pkt_type == PACKET_HOST) {
3098 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
3100 memcpy(dest, master->dev_addr, ETH_ALEN);
3104 /* On bonding slaves other than the currently active slave, suppress
3105 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
3106 * ARP on active-backup slaves with arp_validate enabled.
3108 static int __skb_bond_should_drop(struct sk_buff *skb,
3109 struct net_device *master)
3111 struct net_device *dev = skb->dev;
3113 if (master->priv_flags & IFF_MASTER_ARPMON)
3114 dev->last_rx = jiffies;
3116 if ((master->priv_flags & IFF_MASTER_ALB) &&
3117 (master->priv_flags & IFF_BRIDGE_PORT)) {
3118 /* Do address unmangle. The local destination address
3119 * will be always the one master has. Provides the right
3120 * functionality in a bridge.
3122 skb_bond_set_mac_by_master(skb, master);
3125 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
3126 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
3127 skb->protocol == __cpu_to_be16(ETH_P_ARP))
3130 if (master->priv_flags & IFF_MASTER_ALB) {
3131 if (skb->pkt_type != PACKET_BROADCAST &&
3132 skb->pkt_type != PACKET_MULTICAST)
3135 if (master->priv_flags & IFF_MASTER_8023AD &&
3136 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
3144 static int __netif_receive_skb(struct sk_buff *skb)
3146 struct packet_type *ptype, *pt_prev;
3147 rx_handler_func_t *rx_handler;
3148 struct net_device *orig_dev;
3149 struct net_device *null_or_orig;
3150 struct net_device *orig_or_bond;
3151 int ret = NET_RX_DROP;
3154 if (!netdev_tstamp_prequeue)
3155 net_timestamp_check(skb);
3157 trace_netif_receive_skb(skb);
3159 /* if we've gotten here through NAPI, check netpoll */
3160 if (netpoll_receive_skb(skb))
3164 skb->skb_iif = skb->dev->ifindex;
3167 * bonding note: skbs received on inactive slaves should only
3168 * be delivered to pkt handlers that are exact matches. Also
3169 * the deliver_no_wcard flag will be set. If packet handlers
3170 * are sensitive to duplicate packets these skbs will need to
3171 * be dropped at the handler.
3173 null_or_orig = NULL;
3174 orig_dev = skb->dev;
3175 if (skb->deliver_no_wcard)
3176 null_or_orig = orig_dev;
3177 else if (netif_is_bond_slave(orig_dev)) {
3178 struct net_device *bond_master = ACCESS_ONCE(orig_dev->master);
3180 if (likely(bond_master)) {
3181 if (__skb_bond_should_drop(skb, bond_master)) {
3182 skb->deliver_no_wcard = 1;
3183 /* deliver only exact match */
3184 null_or_orig = orig_dev;
3186 skb->dev = bond_master;
3190 __this_cpu_inc(softnet_data.processed);
3191 skb_reset_network_header(skb);
3192 skb_reset_transport_header(skb);
3193 skb->mac_len = skb->network_header - skb->mac_header;
3199 #ifdef CONFIG_NET_CLS_ACT
3200 if (skb->tc_verd & TC_NCLS) {
3201 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3206 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3207 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
3208 ptype->dev == orig_dev) {
3210 ret = deliver_skb(skb, pt_prev, orig_dev);
3215 #ifdef CONFIG_NET_CLS_ACT
3216 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3222 /* Handle special case of bridge or macvlan */
3223 rx_handler = rcu_dereference(skb->dev->rx_handler);
3226 ret = deliver_skb(skb, pt_prev, orig_dev);
3229 skb = rx_handler(skb);
3234 if (vlan_tx_tag_present(skb)) {
3236 ret = deliver_skb(skb, pt_prev, orig_dev);
3239 if (vlan_hwaccel_do_receive(&skb)) {
3240 ret = __netif_receive_skb(skb);
3242 } else if (unlikely(!skb))
3247 * Make sure frames received on VLAN interfaces stacked on
3248 * bonding interfaces still make their way to any base bonding
3249 * device that may have registered for a specific ptype. The
3250 * handler may have to adjust skb->dev and orig_dev.
3252 orig_or_bond = orig_dev;
3253 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
3254 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
3255 orig_or_bond = vlan_dev_real_dev(skb->dev);
3258 type = skb->protocol;
3259 list_for_each_entry_rcu(ptype,
3260 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3261 if (ptype->type == type && (ptype->dev == null_or_orig ||
3262 ptype->dev == skb->dev || ptype->dev == orig_dev ||
3263 ptype->dev == orig_or_bond)) {
3265 ret = deliver_skb(skb, pt_prev, orig_dev);
3271 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3273 atomic_long_inc(&skb->dev->rx_dropped);
3275 /* Jamal, now you will not able to escape explaining
3276 * me how you were going to use this. :-)
3287 * netif_receive_skb - process receive buffer from network
3288 * @skb: buffer to process
3290 * netif_receive_skb() is the main receive data processing function.
3291 * It always succeeds. The buffer may be dropped during processing
3292 * for congestion control or by the protocol layers.
3294 * This function may only be called from softirq context and interrupts
3295 * should be enabled.
3297 * Return values (usually ignored):
3298 * NET_RX_SUCCESS: no congestion
3299 * NET_RX_DROP: packet was dropped
3301 int netif_receive_skb(struct sk_buff *skb)
3303 if (netdev_tstamp_prequeue)
3304 net_timestamp_check(skb);
3306 if (skb_defer_rx_timestamp(skb))
3307 return NET_RX_SUCCESS;
3311 struct rps_dev_flow voidflow, *rflow = &voidflow;
3316 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3319 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3323 ret = __netif_receive_skb(skb);
3329 return __netif_receive_skb(skb);
3332 EXPORT_SYMBOL(netif_receive_skb);
3334 /* Network device is going away, flush any packets still pending
3335 * Called with irqs disabled.
3337 static void flush_backlog(void *arg)
3339 struct net_device *dev = arg;
3340 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3341 struct sk_buff *skb, *tmp;
3344 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3345 if (skb->dev == dev) {
3346 __skb_unlink(skb, &sd->input_pkt_queue);
3348 input_queue_head_incr(sd);
3353 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3354 if (skb->dev == dev) {
3355 __skb_unlink(skb, &sd->process_queue);
3357 input_queue_head_incr(sd);
3362 static int napi_gro_complete(struct sk_buff *skb)
3364 struct packet_type *ptype;
3365 __be16 type = skb->protocol;
3366 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3369 if (NAPI_GRO_CB(skb)->count == 1) {
3370 skb_shinfo(skb)->gso_size = 0;
3375 list_for_each_entry_rcu(ptype, head, list) {
3376 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3379 err = ptype->gro_complete(skb);
3385 WARN_ON(&ptype->list == head);
3387 return NET_RX_SUCCESS;
3391 return netif_receive_skb(skb);
3394 inline void napi_gro_flush(struct napi_struct *napi)
3396 struct sk_buff *skb, *next;
3398 for (skb = napi->gro_list; skb; skb = next) {
3401 napi_gro_complete(skb);
3404 napi->gro_count = 0;
3405 napi->gro_list = NULL;
3407 EXPORT_SYMBOL(napi_gro_flush);
3409 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3411 struct sk_buff **pp = NULL;
3412 struct packet_type *ptype;
3413 __be16 type = skb->protocol;
3414 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3417 enum gro_result ret;
3419 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3422 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3426 list_for_each_entry_rcu(ptype, head, list) {
3427 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3430 skb_set_network_header(skb, skb_gro_offset(skb));
3431 mac_len = skb->network_header - skb->mac_header;
3432 skb->mac_len = mac_len;
3433 NAPI_GRO_CB(skb)->same_flow = 0;
3434 NAPI_GRO_CB(skb)->flush = 0;
3435 NAPI_GRO_CB(skb)->free = 0;
3437 pp = ptype->gro_receive(&napi->gro_list, skb);
3442 if (&ptype->list == head)
3445 same_flow = NAPI_GRO_CB(skb)->same_flow;
3446 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3449 struct sk_buff *nskb = *pp;
3453 napi_gro_complete(nskb);
3460 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3464 NAPI_GRO_CB(skb)->count = 1;
3465 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3466 skb->next = napi->gro_list;
3467 napi->gro_list = skb;
3471 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3472 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3474 BUG_ON(skb->end - skb->tail < grow);
3476 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3479 skb->data_len -= grow;
3481 skb_shinfo(skb)->frags[0].page_offset += grow;
3482 skb_shinfo(skb)->frags[0].size -= grow;
3484 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3485 put_page(skb_shinfo(skb)->frags[0].page);
3486 memmove(skb_shinfo(skb)->frags,
3487 skb_shinfo(skb)->frags + 1,
3488 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3499 EXPORT_SYMBOL(dev_gro_receive);
3501 static inline gro_result_t
3502 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3506 for (p = napi->gro_list; p; p = p->next) {
3507 unsigned long diffs;
3509 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3510 diffs |= p->vlan_tci ^ skb->vlan_tci;
3511 diffs |= compare_ether_header(skb_mac_header(p),
3512 skb_gro_mac_header(skb));
3513 NAPI_GRO_CB(p)->same_flow = !diffs;
3514 NAPI_GRO_CB(p)->flush = 0;
3517 return dev_gro_receive(napi, skb);
3520 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3524 if (netif_receive_skb(skb))
3529 case GRO_MERGED_FREE:
3540 EXPORT_SYMBOL(napi_skb_finish);
3542 void skb_gro_reset_offset(struct sk_buff *skb)
3544 NAPI_GRO_CB(skb)->data_offset = 0;
3545 NAPI_GRO_CB(skb)->frag0 = NULL;
3546 NAPI_GRO_CB(skb)->frag0_len = 0;
3548 if (skb->mac_header == skb->tail &&
3549 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3550 NAPI_GRO_CB(skb)->frag0 =
3551 page_address(skb_shinfo(skb)->frags[0].page) +
3552 skb_shinfo(skb)->frags[0].page_offset;
3553 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3556 EXPORT_SYMBOL(skb_gro_reset_offset);
3558 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3560 skb_gro_reset_offset(skb);
3562 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3564 EXPORT_SYMBOL(napi_gro_receive);
3566 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3568 __skb_pull(skb, skb_headlen(skb));
3569 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3571 skb->dev = napi->dev;
3577 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3579 struct sk_buff *skb = napi->skb;
3582 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3588 EXPORT_SYMBOL(napi_get_frags);
3590 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3596 skb->protocol = eth_type_trans(skb, skb->dev);
3598 if (ret == GRO_HELD)
3599 skb_gro_pull(skb, -ETH_HLEN);
3600 else if (netif_receive_skb(skb))
3605 case GRO_MERGED_FREE:
3606 napi_reuse_skb(napi, skb);
3615 EXPORT_SYMBOL(napi_frags_finish);
3617 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3619 struct sk_buff *skb = napi->skb;
3626 skb_reset_mac_header(skb);
3627 skb_gro_reset_offset(skb);
3629 off = skb_gro_offset(skb);
3630 hlen = off + sizeof(*eth);
3631 eth = skb_gro_header_fast(skb, off);
3632 if (skb_gro_header_hard(skb, hlen)) {
3633 eth = skb_gro_header_slow(skb, hlen, off);
3634 if (unlikely(!eth)) {
3635 napi_reuse_skb(napi, skb);
3641 skb_gro_pull(skb, sizeof(*eth));
3644 * This works because the only protocols we care about don't require
3645 * special handling. We'll fix it up properly at the end.
3647 skb->protocol = eth->h_proto;
3652 EXPORT_SYMBOL(napi_frags_skb);
3654 gro_result_t napi_gro_frags(struct napi_struct *napi)
3656 struct sk_buff *skb = napi_frags_skb(napi);
3661 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3663 EXPORT_SYMBOL(napi_gro_frags);
3666 * net_rps_action sends any pending IPI's for rps.
3667 * Note: called with local irq disabled, but exits with local irq enabled.
3669 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3672 struct softnet_data *remsd = sd->rps_ipi_list;
3675 sd->rps_ipi_list = NULL;
3679 /* Send pending IPI's to kick RPS processing on remote cpus. */
3681 struct softnet_data *next = remsd->rps_ipi_next;
3683 if (cpu_online(remsd->cpu))
3684 __smp_call_function_single(remsd->cpu,
3693 static int process_backlog(struct napi_struct *napi, int quota)
3696 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3699 /* Check if we have pending ipi, its better to send them now,
3700 * not waiting net_rx_action() end.
3702 if (sd->rps_ipi_list) {
3703 local_irq_disable();
3704 net_rps_action_and_irq_enable(sd);
3707 napi->weight = weight_p;
3708 local_irq_disable();
3709 while (work < quota) {
3710 struct sk_buff *skb;
3713 while ((skb = __skb_dequeue(&sd->process_queue))) {
3715 __netif_receive_skb(skb);
3716 local_irq_disable();
3717 input_queue_head_incr(sd);
3718 if (++work >= quota) {
3725 qlen = skb_queue_len(&sd->input_pkt_queue);
3727 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3728 &sd->process_queue);
3730 if (qlen < quota - work) {
3732 * Inline a custom version of __napi_complete().
3733 * only current cpu owns and manipulates this napi,
3734 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3735 * we can use a plain write instead of clear_bit(),
3736 * and we dont need an smp_mb() memory barrier.
3738 list_del(&napi->poll_list);
3741 quota = work + qlen;
3751 * __napi_schedule - schedule for receive
3752 * @n: entry to schedule
3754 * The entry's receive function will be scheduled to run
3756 void __napi_schedule(struct napi_struct *n)
3758 unsigned long flags;
3760 local_irq_save(flags);
3761 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3762 local_irq_restore(flags);
3764 EXPORT_SYMBOL(__napi_schedule);
3766 void __napi_complete(struct napi_struct *n)
3768 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3769 BUG_ON(n->gro_list);
3771 list_del(&n->poll_list);
3772 smp_mb__before_clear_bit();
3773 clear_bit(NAPI_STATE_SCHED, &n->state);
3775 EXPORT_SYMBOL(__napi_complete);
3777 void napi_complete(struct napi_struct *n)
3779 unsigned long flags;
3782 * don't let napi dequeue from the cpu poll list
3783 * just in case its running on a different cpu
3785 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3789 local_irq_save(flags);
3791 local_irq_restore(flags);
3793 EXPORT_SYMBOL(napi_complete);
3795 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3796 int (*poll)(struct napi_struct *, int), int weight)
3798 INIT_LIST_HEAD(&napi->poll_list);
3799 napi->gro_count = 0;
3800 napi->gro_list = NULL;
3803 napi->weight = weight;
3804 list_add(&napi->dev_list, &dev->napi_list);
3806 #ifdef CONFIG_NETPOLL
3807 spin_lock_init(&napi->poll_lock);
3808 napi->poll_owner = -1;
3810 set_bit(NAPI_STATE_SCHED, &napi->state);
3812 EXPORT_SYMBOL(netif_napi_add);
3814 void netif_napi_del(struct napi_struct *napi)
3816 struct sk_buff *skb, *next;
3818 list_del_init(&napi->dev_list);
3819 napi_free_frags(napi);
3821 for (skb = napi->gro_list; skb; skb = next) {
3827 napi->gro_list = NULL;
3828 napi->gro_count = 0;
3830 EXPORT_SYMBOL(netif_napi_del);
3832 static void net_rx_action(struct softirq_action *h)
3834 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3835 unsigned long time_limit = jiffies + 2;
3836 int budget = netdev_budget;
3839 local_irq_disable();
3841 while (!list_empty(&sd->poll_list)) {
3842 struct napi_struct *n;
3845 /* If softirq window is exhuasted then punt.
3846 * Allow this to run for 2 jiffies since which will allow
3847 * an average latency of 1.5/HZ.
3849 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3854 /* Even though interrupts have been re-enabled, this
3855 * access is safe because interrupts can only add new
3856 * entries to the tail of this list, and only ->poll()
3857 * calls can remove this head entry from the list.
3859 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3861 have = netpoll_poll_lock(n);
3865 /* This NAPI_STATE_SCHED test is for avoiding a race
3866 * with netpoll's poll_napi(). Only the entity which
3867 * obtains the lock and sees NAPI_STATE_SCHED set will
3868 * actually make the ->poll() call. Therefore we avoid
3869 * accidently calling ->poll() when NAPI is not scheduled.
3872 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3873 work = n->poll(n, weight);
3877 WARN_ON_ONCE(work > weight);
3881 local_irq_disable();
3883 /* Drivers must not modify the NAPI state if they
3884 * consume the entire weight. In such cases this code
3885 * still "owns" the NAPI instance and therefore can
3886 * move the instance around on the list at-will.
3888 if (unlikely(work == weight)) {
3889 if (unlikely(napi_disable_pending(n))) {
3892 local_irq_disable();
3894 list_move_tail(&n->poll_list, &sd->poll_list);
3897 netpoll_poll_unlock(have);
3900 net_rps_action_and_irq_enable(sd);
3902 #ifdef CONFIG_NET_DMA
3904 * There may not be any more sk_buffs coming right now, so push
3905 * any pending DMA copies to hardware
3907 dma_issue_pending_all();
3914 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3918 static gifconf_func_t *gifconf_list[NPROTO];
3921 * register_gifconf - register a SIOCGIF handler
3922 * @family: Address family
3923 * @gifconf: Function handler
3925 * Register protocol dependent address dumping routines. The handler
3926 * that is passed must not be freed or reused until it has been replaced
3927 * by another handler.
3929 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3931 if (family >= NPROTO)
3933 gifconf_list[family] = gifconf;
3936 EXPORT_SYMBOL(register_gifconf);
3940 * Map an interface index to its name (SIOCGIFNAME)
3944 * We need this ioctl for efficient implementation of the
3945 * if_indextoname() function required by the IPv6 API. Without
3946 * it, we would have to search all the interfaces to find a
3950 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3952 struct net_device *dev;
3956 * Fetch the caller's info block.
3959 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3963 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3969 strcpy(ifr.ifr_name, dev->name);
3972 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3978 * Perform a SIOCGIFCONF call. This structure will change
3979 * size eventually, and there is nothing I can do about it.
3980 * Thus we will need a 'compatibility mode'.
3983 static int dev_ifconf(struct net *net, char __user *arg)
3986 struct net_device *dev;
3993 * Fetch the caller's info block.
3996 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4003 * Loop over the interfaces, and write an info block for each.
4007 for_each_netdev(net, dev) {
4008 for (i = 0; i < NPROTO; i++) {
4009 if (gifconf_list[i]) {
4012 done = gifconf_list[i](dev, NULL, 0);
4014 done = gifconf_list[i](dev, pos + total,
4024 * All done. Write the updated control block back to the caller.
4026 ifc.ifc_len = total;
4029 * Both BSD and Solaris return 0 here, so we do too.
4031 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4034 #ifdef CONFIG_PROC_FS
4036 * This is invoked by the /proc filesystem handler to display a device
4039 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4042 struct net *net = seq_file_net(seq);
4044 struct net_device *dev;
4048 return SEQ_START_TOKEN;
4051 for_each_netdev_rcu(net, dev)
4058 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4060 struct net_device *dev = v;
4062 if (v == SEQ_START_TOKEN)
4063 dev = first_net_device_rcu(seq_file_net(seq));
4065 dev = next_net_device_rcu(dev);
4071 void dev_seq_stop(struct seq_file *seq, void *v)
4077 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4079 struct rtnl_link_stats64 temp;
4080 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4082 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4083 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4084 dev->name, stats->rx_bytes, stats->rx_packets,
4086 stats->rx_dropped + stats->rx_missed_errors,
4087 stats->rx_fifo_errors,
4088 stats->rx_length_errors + stats->rx_over_errors +
4089 stats->rx_crc_errors + stats->rx_frame_errors,
4090 stats->rx_compressed, stats->multicast,
4091 stats->tx_bytes, stats->tx_packets,
4092 stats->tx_errors, stats->tx_dropped,
4093 stats->tx_fifo_errors, stats->collisions,
4094 stats->tx_carrier_errors +
4095 stats->tx_aborted_errors +
4096 stats->tx_window_errors +
4097 stats->tx_heartbeat_errors,
4098 stats->tx_compressed);
4102 * Called from the PROCfs module. This now uses the new arbitrary sized
4103 * /proc/net interface to create /proc/net/dev
4105 static int dev_seq_show(struct seq_file *seq, void *v)
4107 if (v == SEQ_START_TOKEN)
4108 seq_puts(seq, "Inter-| Receive "
4110 " face |bytes packets errs drop fifo frame "
4111 "compressed multicast|bytes packets errs "
4112 "drop fifo colls carrier compressed\n");
4114 dev_seq_printf_stats(seq, v);
4118 static struct softnet_data *softnet_get_online(loff_t *pos)
4120 struct softnet_data *sd = NULL;
4122 while (*pos < nr_cpu_ids)
4123 if (cpu_online(*pos)) {
4124 sd = &per_cpu(softnet_data, *pos);
4131 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4133 return softnet_get_online(pos);
4136 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4139 return softnet_get_online(pos);
4142 static void softnet_seq_stop(struct seq_file *seq, void *v)
4146 static int softnet_seq_show(struct seq_file *seq, void *v)
4148 struct softnet_data *sd = v;
4150 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4151 sd->processed, sd->dropped, sd->time_squeeze, 0,
4152 0, 0, 0, 0, /* was fastroute */
4153 sd->cpu_collision, sd->received_rps);
4157 static const struct seq_operations dev_seq_ops = {
4158 .start = dev_seq_start,
4159 .next = dev_seq_next,
4160 .stop = dev_seq_stop,
4161 .show = dev_seq_show,
4164 static int dev_seq_open(struct inode *inode, struct file *file)
4166 return seq_open_net(inode, file, &dev_seq_ops,
4167 sizeof(struct seq_net_private));
4170 static const struct file_operations dev_seq_fops = {
4171 .owner = THIS_MODULE,
4172 .open = dev_seq_open,
4174 .llseek = seq_lseek,
4175 .release = seq_release_net,
4178 static const struct seq_operations softnet_seq_ops = {
4179 .start = softnet_seq_start,
4180 .next = softnet_seq_next,
4181 .stop = softnet_seq_stop,
4182 .show = softnet_seq_show,
4185 static int softnet_seq_open(struct inode *inode, struct file *file)
4187 return seq_open(file, &softnet_seq_ops);
4190 static const struct file_operations softnet_seq_fops = {
4191 .owner = THIS_MODULE,
4192 .open = softnet_seq_open,
4194 .llseek = seq_lseek,
4195 .release = seq_release,
4198 static void *ptype_get_idx(loff_t pos)
4200 struct packet_type *pt = NULL;
4204 list_for_each_entry_rcu(pt, &ptype_all, list) {
4210 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4211 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4220 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4224 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4227 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4229 struct packet_type *pt;
4230 struct list_head *nxt;
4234 if (v == SEQ_START_TOKEN)
4235 return ptype_get_idx(0);
4238 nxt = pt->list.next;
4239 if (pt->type == htons(ETH_P_ALL)) {
4240 if (nxt != &ptype_all)
4243 nxt = ptype_base[0].next;
4245 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4247 while (nxt == &ptype_base[hash]) {
4248 if (++hash >= PTYPE_HASH_SIZE)
4250 nxt = ptype_base[hash].next;
4253 return list_entry(nxt, struct packet_type, list);
4256 static void ptype_seq_stop(struct seq_file *seq, void *v)
4262 static int ptype_seq_show(struct seq_file *seq, void *v)
4264 struct packet_type *pt = v;
4266 if (v == SEQ_START_TOKEN)
4267 seq_puts(seq, "Type Device Function\n");
4268 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4269 if (pt->type == htons(ETH_P_ALL))
4270 seq_puts(seq, "ALL ");
4272 seq_printf(seq, "%04x", ntohs(pt->type));
4274 seq_printf(seq, " %-8s %pF\n",
4275 pt->dev ? pt->dev->name : "", pt->func);
4281 static const struct seq_operations ptype_seq_ops = {
4282 .start = ptype_seq_start,
4283 .next = ptype_seq_next,
4284 .stop = ptype_seq_stop,
4285 .show = ptype_seq_show,
4288 static int ptype_seq_open(struct inode *inode, struct file *file)
4290 return seq_open_net(inode, file, &ptype_seq_ops,
4291 sizeof(struct seq_net_private));
4294 static const struct file_operations ptype_seq_fops = {
4295 .owner = THIS_MODULE,
4296 .open = ptype_seq_open,
4298 .llseek = seq_lseek,
4299 .release = seq_release_net,
4303 static int __net_init dev_proc_net_init(struct net *net)
4307 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4309 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4311 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4314 if (wext_proc_init(net))
4320 proc_net_remove(net, "ptype");
4322 proc_net_remove(net, "softnet_stat");
4324 proc_net_remove(net, "dev");
4328 static void __net_exit dev_proc_net_exit(struct net *net)
4330 wext_proc_exit(net);
4332 proc_net_remove(net, "ptype");
4333 proc_net_remove(net, "softnet_stat");
4334 proc_net_remove(net, "dev");
4337 static struct pernet_operations __net_initdata dev_proc_ops = {
4338 .init = dev_proc_net_init,
4339 .exit = dev_proc_net_exit,
4342 static int __init dev_proc_init(void)
4344 return register_pernet_subsys(&dev_proc_ops);
4347 #define dev_proc_init() 0
4348 #endif /* CONFIG_PROC_FS */
4352 * netdev_set_master - set up master pointer
4353 * @slave: slave device
4354 * @master: new master device
4356 * Changes the master device of the slave. Pass %NULL to break the
4357 * bonding. The caller must hold the RTNL semaphore. On a failure
4358 * a negative errno code is returned. On success the reference counts
4359 * are adjusted and the function returns zero.
4361 int netdev_set_master(struct net_device *slave, struct net_device *master)
4363 struct net_device *old = slave->master;
4373 slave->master = master;
4381 EXPORT_SYMBOL(netdev_set_master);
4384 * netdev_set_bond_master - set up bonding master/slave pair
4385 * @slave: slave device
4386 * @master: new master device
4388 * Changes the master device of the slave. Pass %NULL to break the
4389 * bonding. The caller must hold the RTNL semaphore. On a failure
4390 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4391 * to the routing socket and the function returns zero.
4393 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4399 err = netdev_set_master(slave, master);
4403 slave->flags |= IFF_SLAVE;
4405 slave->flags &= ~IFF_SLAVE;
4407 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4410 EXPORT_SYMBOL(netdev_set_bond_master);
4412 static void dev_change_rx_flags(struct net_device *dev, int flags)
4414 const struct net_device_ops *ops = dev->netdev_ops;
4416 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4417 ops->ndo_change_rx_flags(dev, flags);
4420 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4422 unsigned short old_flags = dev->flags;
4428 dev->flags |= IFF_PROMISC;
4429 dev->promiscuity += inc;
4430 if (dev->promiscuity == 0) {
4433 * If inc causes overflow, untouch promisc and return error.
4436 dev->flags &= ~IFF_PROMISC;
4438 dev->promiscuity -= inc;
4439 printk(KERN_WARNING "%s: promiscuity touches roof, "
4440 "set promiscuity failed, promiscuity feature "
4441 "of device might be broken.\n", dev->name);
4445 if (dev->flags != old_flags) {
4446 printk(KERN_INFO "device %s %s promiscuous mode\n",
4447 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4449 if (audit_enabled) {
4450 current_uid_gid(&uid, &gid);
4451 audit_log(current->audit_context, GFP_ATOMIC,
4452 AUDIT_ANOM_PROMISCUOUS,
4453 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4454 dev->name, (dev->flags & IFF_PROMISC),
4455 (old_flags & IFF_PROMISC),
4456 audit_get_loginuid(current),
4458 audit_get_sessionid(current));
4461 dev_change_rx_flags(dev, IFF_PROMISC);
4467 * dev_set_promiscuity - update promiscuity count on a device
4471 * Add or remove promiscuity from a device. While the count in the device
4472 * remains above zero the interface remains promiscuous. Once it hits zero
4473 * the device reverts back to normal filtering operation. A negative inc
4474 * value is used to drop promiscuity on the device.
4475 * Return 0 if successful or a negative errno code on error.
4477 int dev_set_promiscuity(struct net_device *dev, int inc)
4479 unsigned short old_flags = dev->flags;
4482 err = __dev_set_promiscuity(dev, inc);
4485 if (dev->flags != old_flags)
4486 dev_set_rx_mode(dev);
4489 EXPORT_SYMBOL(dev_set_promiscuity);
4492 * dev_set_allmulti - update allmulti count on a device
4496 * Add or remove reception of all multicast frames to a device. While the
4497 * count in the device remains above zero the interface remains listening
4498 * to all interfaces. Once it hits zero the device reverts back to normal
4499 * filtering operation. A negative @inc value is used to drop the counter
4500 * when releasing a resource needing all multicasts.
4501 * Return 0 if successful or a negative errno code on error.
4504 int dev_set_allmulti(struct net_device *dev, int inc)
4506 unsigned short old_flags = dev->flags;
4510 dev->flags |= IFF_ALLMULTI;
4511 dev->allmulti += inc;
4512 if (dev->allmulti == 0) {
4515 * If inc causes overflow, untouch allmulti and return error.
4518 dev->flags &= ~IFF_ALLMULTI;
4520 dev->allmulti -= inc;
4521 printk(KERN_WARNING "%s: allmulti touches roof, "
4522 "set allmulti failed, allmulti feature of "
4523 "device might be broken.\n", dev->name);
4527 if (dev->flags ^ old_flags) {
4528 dev_change_rx_flags(dev, IFF_ALLMULTI);
4529 dev_set_rx_mode(dev);
4533 EXPORT_SYMBOL(dev_set_allmulti);
4536 * Upload unicast and multicast address lists to device and
4537 * configure RX filtering. When the device doesn't support unicast
4538 * filtering it is put in promiscuous mode while unicast addresses
4541 void __dev_set_rx_mode(struct net_device *dev)
4543 const struct net_device_ops *ops = dev->netdev_ops;
4545 /* dev_open will call this function so the list will stay sane. */
4546 if (!(dev->flags&IFF_UP))
4549 if (!netif_device_present(dev))
4552 if (ops->ndo_set_rx_mode)
4553 ops->ndo_set_rx_mode(dev);
4555 /* Unicast addresses changes may only happen under the rtnl,
4556 * therefore calling __dev_set_promiscuity here is safe.
4558 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4559 __dev_set_promiscuity(dev, 1);
4560 dev->uc_promisc = 1;
4561 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4562 __dev_set_promiscuity(dev, -1);
4563 dev->uc_promisc = 0;
4566 if (ops->ndo_set_multicast_list)
4567 ops->ndo_set_multicast_list(dev);
4571 void dev_set_rx_mode(struct net_device *dev)
4573 netif_addr_lock_bh(dev);
4574 __dev_set_rx_mode(dev);
4575 netif_addr_unlock_bh(dev);
4579 * dev_get_flags - get flags reported to userspace
4582 * Get the combination of flag bits exported through APIs to userspace.
4584 unsigned dev_get_flags(const struct net_device *dev)
4588 flags = (dev->flags & ~(IFF_PROMISC |
4593 (dev->gflags & (IFF_PROMISC |
4596 if (netif_running(dev)) {
4597 if (netif_oper_up(dev))
4598 flags |= IFF_RUNNING;
4599 if (netif_carrier_ok(dev))
4600 flags |= IFF_LOWER_UP;
4601 if (netif_dormant(dev))
4602 flags |= IFF_DORMANT;
4607 EXPORT_SYMBOL(dev_get_flags);
4609 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4611 int old_flags = dev->flags;
4617 * Set the flags on our device.
4620 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4621 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4623 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4627 * Load in the correct multicast list now the flags have changed.
4630 if ((old_flags ^ flags) & IFF_MULTICAST)
4631 dev_change_rx_flags(dev, IFF_MULTICAST);
4633 dev_set_rx_mode(dev);
4636 * Have we downed the interface. We handle IFF_UP ourselves
4637 * according to user attempts to set it, rather than blindly
4642 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4643 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4646 dev_set_rx_mode(dev);
4649 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4650 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4652 dev->gflags ^= IFF_PROMISC;
4653 dev_set_promiscuity(dev, inc);
4656 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4657 is important. Some (broken) drivers set IFF_PROMISC, when
4658 IFF_ALLMULTI is requested not asking us and not reporting.
4660 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4661 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4663 dev->gflags ^= IFF_ALLMULTI;
4664 dev_set_allmulti(dev, inc);
4670 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4672 unsigned int changes = dev->flags ^ old_flags;
4674 if (changes & IFF_UP) {
4675 if (dev->flags & IFF_UP)
4676 call_netdevice_notifiers(NETDEV_UP, dev);
4678 call_netdevice_notifiers(NETDEV_DOWN, dev);
4681 if (dev->flags & IFF_UP &&
4682 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4683 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4687 * dev_change_flags - change device settings
4689 * @flags: device state flags
4691 * Change settings on device based state flags. The flags are
4692 * in the userspace exported format.
4694 int dev_change_flags(struct net_device *dev, unsigned flags)
4697 int old_flags = dev->flags;
4699 ret = __dev_change_flags(dev, flags);
4703 changes = old_flags ^ dev->flags;
4705 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4707 __dev_notify_flags(dev, old_flags);
4710 EXPORT_SYMBOL(dev_change_flags);
4713 * dev_set_mtu - Change maximum transfer unit
4715 * @new_mtu: new transfer unit
4717 * Change the maximum transfer size of the network device.
4719 int dev_set_mtu(struct net_device *dev, int new_mtu)
4721 const struct net_device_ops *ops = dev->netdev_ops;
4724 if (new_mtu == dev->mtu)
4727 /* MTU must be positive. */
4731 if (!netif_device_present(dev))
4735 if (ops->ndo_change_mtu)
4736 err = ops->ndo_change_mtu(dev, new_mtu);
4740 if (!err && dev->flags & IFF_UP)
4741 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4744 EXPORT_SYMBOL(dev_set_mtu);
4747 * dev_set_group - Change group this device belongs to
4749 * @new_group: group this device should belong to
4751 void dev_set_group(struct net_device *dev, int new_group)
4753 dev->group = new_group;
4755 EXPORT_SYMBOL(dev_set_group);
4758 * dev_set_mac_address - Change Media Access Control Address
4762 * Change the hardware (MAC) address of the device
4764 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4766 const struct net_device_ops *ops = dev->netdev_ops;
4769 if (!ops->ndo_set_mac_address)
4771 if (sa->sa_family != dev->type)
4773 if (!netif_device_present(dev))
4775 err = ops->ndo_set_mac_address(dev, sa);
4777 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4780 EXPORT_SYMBOL(dev_set_mac_address);
4783 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4785 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4788 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4794 case SIOCGIFFLAGS: /* Get interface flags */
4795 ifr->ifr_flags = (short) dev_get_flags(dev);
4798 case SIOCGIFMETRIC: /* Get the metric on the interface
4799 (currently unused) */
4800 ifr->ifr_metric = 0;
4803 case SIOCGIFMTU: /* Get the MTU of a device */
4804 ifr->ifr_mtu = dev->mtu;
4809 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4811 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4812 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4813 ifr->ifr_hwaddr.sa_family = dev->type;
4821 ifr->ifr_map.mem_start = dev->mem_start;
4822 ifr->ifr_map.mem_end = dev->mem_end;
4823 ifr->ifr_map.base_addr = dev->base_addr;
4824 ifr->ifr_map.irq = dev->irq;
4825 ifr->ifr_map.dma = dev->dma;
4826 ifr->ifr_map.port = dev->if_port;
4830 ifr->ifr_ifindex = dev->ifindex;
4834 ifr->ifr_qlen = dev->tx_queue_len;
4838 /* dev_ioctl() should ensure this case
4850 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4852 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4855 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4856 const struct net_device_ops *ops;
4861 ops = dev->netdev_ops;
4864 case SIOCSIFFLAGS: /* Set interface flags */
4865 return dev_change_flags(dev, ifr->ifr_flags);
4867 case SIOCSIFMETRIC: /* Set the metric on the interface
4868 (currently unused) */
4871 case SIOCSIFMTU: /* Set the MTU of a device */
4872 return dev_set_mtu(dev, ifr->ifr_mtu);
4875 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4877 case SIOCSIFHWBROADCAST:
4878 if (ifr->ifr_hwaddr.sa_family != dev->type)
4880 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4881 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4882 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4886 if (ops->ndo_set_config) {
4887 if (!netif_device_present(dev))
4889 return ops->ndo_set_config(dev, &ifr->ifr_map);
4894 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4895 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4897 if (!netif_device_present(dev))
4899 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4902 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4903 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4905 if (!netif_device_present(dev))
4907 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4910 if (ifr->ifr_qlen < 0)
4912 dev->tx_queue_len = ifr->ifr_qlen;
4916 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4917 return dev_change_name(dev, ifr->ifr_newname);
4920 * Unknown or private ioctl
4923 if ((cmd >= SIOCDEVPRIVATE &&
4924 cmd <= SIOCDEVPRIVATE + 15) ||
4925 cmd == SIOCBONDENSLAVE ||
4926 cmd == SIOCBONDRELEASE ||
4927 cmd == SIOCBONDSETHWADDR ||
4928 cmd == SIOCBONDSLAVEINFOQUERY ||
4929 cmd == SIOCBONDINFOQUERY ||
4930 cmd == SIOCBONDCHANGEACTIVE ||
4931 cmd == SIOCGMIIPHY ||
4932 cmd == SIOCGMIIREG ||
4933 cmd == SIOCSMIIREG ||
4934 cmd == SIOCBRADDIF ||
4935 cmd == SIOCBRDELIF ||
4936 cmd == SIOCSHWTSTAMP ||
4937 cmd == SIOCWANDEV) {
4939 if (ops->ndo_do_ioctl) {
4940 if (netif_device_present(dev))
4941 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4953 * This function handles all "interface"-type I/O control requests. The actual
4954 * 'doing' part of this is dev_ifsioc above.
4958 * dev_ioctl - network device ioctl
4959 * @net: the applicable net namespace
4960 * @cmd: command to issue
4961 * @arg: pointer to a struct ifreq in user space
4963 * Issue ioctl functions to devices. This is normally called by the
4964 * user space syscall interfaces but can sometimes be useful for
4965 * other purposes. The return value is the return from the syscall if
4966 * positive or a negative errno code on error.
4969 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4975 /* One special case: SIOCGIFCONF takes ifconf argument
4976 and requires shared lock, because it sleeps writing
4980 if (cmd == SIOCGIFCONF) {
4982 ret = dev_ifconf(net, (char __user *) arg);
4986 if (cmd == SIOCGIFNAME)
4987 return dev_ifname(net, (struct ifreq __user *)arg);
4989 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4992 ifr.ifr_name[IFNAMSIZ-1] = 0;
4994 colon = strchr(ifr.ifr_name, ':');
4999 * See which interface the caller is talking about.
5004 * These ioctl calls:
5005 * - can be done by all.
5006 * - atomic and do not require locking.
5017 dev_load(net, ifr.ifr_name);
5019 ret = dev_ifsioc_locked(net, &ifr, cmd);
5024 if (copy_to_user(arg, &ifr,
5025 sizeof(struct ifreq)))
5031 dev_load(net, ifr.ifr_name);
5033 ret = dev_ethtool(net, &ifr);
5038 if (copy_to_user(arg, &ifr,
5039 sizeof(struct ifreq)))
5045 * These ioctl calls:
5046 * - require superuser power.
5047 * - require strict serialization.
5053 if (!capable(CAP_NET_ADMIN))
5055 dev_load(net, ifr.ifr_name);
5057 ret = dev_ifsioc(net, &ifr, cmd);
5062 if (copy_to_user(arg, &ifr,
5063 sizeof(struct ifreq)))
5069 * These ioctl calls:
5070 * - require superuser power.
5071 * - require strict serialization.
5072 * - do not return a value
5082 case SIOCSIFHWBROADCAST:
5085 case SIOCBONDENSLAVE:
5086 case SIOCBONDRELEASE:
5087 case SIOCBONDSETHWADDR:
5088 case SIOCBONDCHANGEACTIVE:
5092 if (!capable(CAP_NET_ADMIN))
5095 case SIOCBONDSLAVEINFOQUERY:
5096 case SIOCBONDINFOQUERY:
5097 dev_load(net, ifr.ifr_name);
5099 ret = dev_ifsioc(net, &ifr, cmd);
5104 /* Get the per device memory space. We can add this but
5105 * currently do not support it */
5107 /* Set the per device memory buffer space.
5108 * Not applicable in our case */
5113 * Unknown or private ioctl.
5116 if (cmd == SIOCWANDEV ||
5117 (cmd >= SIOCDEVPRIVATE &&
5118 cmd <= SIOCDEVPRIVATE + 15)) {
5119 dev_load(net, ifr.ifr_name);
5121 ret = dev_ifsioc(net, &ifr, cmd);
5123 if (!ret && copy_to_user(arg, &ifr,
5124 sizeof(struct ifreq)))
5128 /* Take care of Wireless Extensions */
5129 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5130 return wext_handle_ioctl(net, &ifr, cmd, arg);
5137 * dev_new_index - allocate an ifindex
5138 * @net: the applicable net namespace
5140 * Returns a suitable unique value for a new device interface
5141 * number. The caller must hold the rtnl semaphore or the
5142 * dev_base_lock to be sure it remains unique.
5144 static int dev_new_index(struct net *net)
5150 if (!__dev_get_by_index(net, ifindex))
5155 /* Delayed registration/unregisteration */
5156 static LIST_HEAD(net_todo_list);
5158 static void net_set_todo(struct net_device *dev)
5160 list_add_tail(&dev->todo_list, &net_todo_list);
5163 static void rollback_registered_many(struct list_head *head)
5165 struct net_device *dev, *tmp;
5167 BUG_ON(dev_boot_phase);
5170 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5171 /* Some devices call without registering
5172 * for initialization unwind. Remove those
5173 * devices and proceed with the remaining.
5175 if (dev->reg_state == NETREG_UNINITIALIZED) {
5176 pr_debug("unregister_netdevice: device %s/%p never "
5177 "was registered\n", dev->name, dev);
5180 list_del(&dev->unreg_list);
5184 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5187 /* If device is running, close it first. */
5188 dev_close_many(head);
5190 list_for_each_entry(dev, head, unreg_list) {
5191 /* And unlink it from device chain. */
5192 unlist_netdevice(dev);
5194 dev->reg_state = NETREG_UNREGISTERING;
5199 list_for_each_entry(dev, head, unreg_list) {
5200 /* Shutdown queueing discipline. */
5204 /* Notify protocols, that we are about to destroy
5205 this device. They should clean all the things.
5207 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5209 if (!dev->rtnl_link_ops ||
5210 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5211 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5214 * Flush the unicast and multicast chains
5219 if (dev->netdev_ops->ndo_uninit)
5220 dev->netdev_ops->ndo_uninit(dev);
5222 /* Notifier chain MUST detach us from master device. */
5223 WARN_ON(dev->master);
5225 /* Remove entries from kobject tree */
5226 netdev_unregister_kobject(dev);
5229 /* Process any work delayed until the end of the batch */
5230 dev = list_first_entry(head, struct net_device, unreg_list);
5231 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5235 list_for_each_entry(dev, head, unreg_list)
5239 static void rollback_registered(struct net_device *dev)
5243 list_add(&dev->unreg_list, &single);
5244 rollback_registered_many(&single);
5247 u32 netdev_fix_features(struct net_device *dev, u32 features)
5249 /* Fix illegal checksum combinations */
5250 if ((features & NETIF_F_HW_CSUM) &&
5251 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5252 netdev_info(dev, "mixed HW and IP checksum settings.\n");
5253 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5256 if ((features & NETIF_F_NO_CSUM) &&
5257 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5258 netdev_info(dev, "mixed no checksumming and other settings.\n");
5259 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5262 /* Fix illegal SG+CSUM combinations. */
5263 if ((features & NETIF_F_SG) &&
5264 !(features & NETIF_F_ALL_CSUM)) {
5266 "Dropping NETIF_F_SG since no checksum feature.\n");
5267 features &= ~NETIF_F_SG;
5270 /* TSO requires that SG is present as well. */
5271 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
5272 netdev_info(dev, "Dropping NETIF_F_TSO since no SG feature.\n");
5273 features &= ~NETIF_F_TSO;
5276 /* UFO needs SG and checksumming */
5277 if (features & NETIF_F_UFO) {
5278 /* maybe split UFO into V4 and V6? */
5279 if (!((features & NETIF_F_GEN_CSUM) ||
5280 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5281 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5283 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5284 features &= ~NETIF_F_UFO;
5287 if (!(features & NETIF_F_SG)) {
5289 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5290 features &= ~NETIF_F_UFO;
5296 EXPORT_SYMBOL(netdev_fix_features);
5299 * netif_stacked_transfer_operstate - transfer operstate
5300 * @rootdev: the root or lower level device to transfer state from
5301 * @dev: the device to transfer operstate to
5303 * Transfer operational state from root to device. This is normally
5304 * called when a stacking relationship exists between the root
5305 * device and the device(a leaf device).
5307 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5308 struct net_device *dev)
5310 if (rootdev->operstate == IF_OPER_DORMANT)
5311 netif_dormant_on(dev);
5313 netif_dormant_off(dev);
5315 if (netif_carrier_ok(rootdev)) {
5316 if (!netif_carrier_ok(dev))
5317 netif_carrier_on(dev);
5319 if (netif_carrier_ok(dev))
5320 netif_carrier_off(dev);
5323 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5326 static int netif_alloc_rx_queues(struct net_device *dev)
5328 unsigned int i, count = dev->num_rx_queues;
5329 struct netdev_rx_queue *rx;
5333 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5335 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5340 for (i = 0; i < count; i++)
5346 static void netdev_init_one_queue(struct net_device *dev,
5347 struct netdev_queue *queue, void *_unused)
5349 /* Initialize queue lock */
5350 spin_lock_init(&queue->_xmit_lock);
5351 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5352 queue->xmit_lock_owner = -1;
5353 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5357 static int netif_alloc_netdev_queues(struct net_device *dev)
5359 unsigned int count = dev->num_tx_queues;
5360 struct netdev_queue *tx;
5364 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5366 pr_err("netdev: Unable to allocate %u tx queues.\n",
5372 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5373 spin_lock_init(&dev->tx_global_lock);
5379 * register_netdevice - register a network device
5380 * @dev: device to register
5382 * Take a completed network device structure and add it to the kernel
5383 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5384 * chain. 0 is returned on success. A negative errno code is returned
5385 * on a failure to set up the device, or if the name is a duplicate.
5387 * Callers must hold the rtnl semaphore. You may want
5388 * register_netdev() instead of this.
5391 * The locking appears insufficient to guarantee two parallel registers
5392 * will not get the same name.
5395 int register_netdevice(struct net_device *dev)
5398 struct net *net = dev_net(dev);
5400 BUG_ON(dev_boot_phase);
5405 /* When net_device's are persistent, this will be fatal. */
5406 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5409 spin_lock_init(&dev->addr_list_lock);
5410 netdev_set_addr_lockdep_class(dev);
5414 /* Init, if this function is available */
5415 if (dev->netdev_ops->ndo_init) {
5416 ret = dev->netdev_ops->ndo_init(dev);
5424 ret = dev_get_valid_name(dev, dev->name, 0);
5428 dev->ifindex = dev_new_index(net);
5429 if (dev->iflink == -1)
5430 dev->iflink = dev->ifindex;
5432 dev->features = netdev_fix_features(dev, dev->features);
5434 /* Enable software GSO if SG is supported. */
5435 if (dev->features & NETIF_F_SG)
5436 dev->features |= NETIF_F_GSO;
5438 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5439 * vlan_dev_init() will do the dev->features check, so these features
5440 * are enabled only if supported by underlying device.
5442 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5444 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5445 ret = notifier_to_errno(ret);
5449 ret = netdev_register_kobject(dev);
5452 dev->reg_state = NETREG_REGISTERED;
5455 * Default initial state at registry is that the
5456 * device is present.
5459 set_bit(__LINK_STATE_PRESENT, &dev->state);
5461 dev_init_scheduler(dev);
5463 list_netdevice(dev);
5465 /* Notify protocols, that a new device appeared. */
5466 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5467 ret = notifier_to_errno(ret);
5469 rollback_registered(dev);
5470 dev->reg_state = NETREG_UNREGISTERED;
5473 * Prevent userspace races by waiting until the network
5474 * device is fully setup before sending notifications.
5476 if (!dev->rtnl_link_ops ||
5477 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5478 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5484 if (dev->netdev_ops->ndo_uninit)
5485 dev->netdev_ops->ndo_uninit(dev);
5488 EXPORT_SYMBOL(register_netdevice);
5491 * init_dummy_netdev - init a dummy network device for NAPI
5492 * @dev: device to init
5494 * This takes a network device structure and initialize the minimum
5495 * amount of fields so it can be used to schedule NAPI polls without
5496 * registering a full blown interface. This is to be used by drivers
5497 * that need to tie several hardware interfaces to a single NAPI
5498 * poll scheduler due to HW limitations.
5500 int init_dummy_netdev(struct net_device *dev)
5502 /* Clear everything. Note we don't initialize spinlocks
5503 * are they aren't supposed to be taken by any of the
5504 * NAPI code and this dummy netdev is supposed to be
5505 * only ever used for NAPI polls
5507 memset(dev, 0, sizeof(struct net_device));
5509 /* make sure we BUG if trying to hit standard
5510 * register/unregister code path
5512 dev->reg_state = NETREG_DUMMY;
5514 /* NAPI wants this */
5515 INIT_LIST_HEAD(&dev->napi_list);
5517 /* a dummy interface is started by default */
5518 set_bit(__LINK_STATE_PRESENT, &dev->state);
5519 set_bit(__LINK_STATE_START, &dev->state);
5521 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5522 * because users of this 'device' dont need to change
5528 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5532 * register_netdev - register a network device
5533 * @dev: device to register
5535 * Take a completed network device structure and add it to the kernel
5536 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5537 * chain. 0 is returned on success. A negative errno code is returned
5538 * on a failure to set up the device, or if the name is a duplicate.
5540 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5541 * and expands the device name if you passed a format string to
5544 int register_netdev(struct net_device *dev)
5551 * If the name is a format string the caller wants us to do a
5554 if (strchr(dev->name, '%')) {
5555 err = dev_alloc_name(dev, dev->name);
5560 err = register_netdevice(dev);
5565 EXPORT_SYMBOL(register_netdev);
5567 int netdev_refcnt_read(const struct net_device *dev)
5571 for_each_possible_cpu(i)
5572 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5575 EXPORT_SYMBOL(netdev_refcnt_read);
5578 * netdev_wait_allrefs - wait until all references are gone.
5580 * This is called when unregistering network devices.
5582 * Any protocol or device that holds a reference should register
5583 * for netdevice notification, and cleanup and put back the
5584 * reference if they receive an UNREGISTER event.
5585 * We can get stuck here if buggy protocols don't correctly
5588 static void netdev_wait_allrefs(struct net_device *dev)
5590 unsigned long rebroadcast_time, warning_time;
5593 linkwatch_forget_dev(dev);
5595 rebroadcast_time = warning_time = jiffies;
5596 refcnt = netdev_refcnt_read(dev);
5598 while (refcnt != 0) {
5599 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5602 /* Rebroadcast unregister notification */
5603 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5604 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5605 * should have already handle it the first time */
5607 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5609 /* We must not have linkwatch events
5610 * pending on unregister. If this
5611 * happens, we simply run the queue
5612 * unscheduled, resulting in a noop
5615 linkwatch_run_queue();
5620 rebroadcast_time = jiffies;
5625 refcnt = netdev_refcnt_read(dev);
5627 if (time_after(jiffies, warning_time + 10 * HZ)) {
5628 printk(KERN_EMERG "unregister_netdevice: "
5629 "waiting for %s to become free. Usage "
5632 warning_time = jiffies;
5641 * register_netdevice(x1);
5642 * register_netdevice(x2);
5644 * unregister_netdevice(y1);
5645 * unregister_netdevice(y2);
5651 * We are invoked by rtnl_unlock().
5652 * This allows us to deal with problems:
5653 * 1) We can delete sysfs objects which invoke hotplug
5654 * without deadlocking with linkwatch via keventd.
5655 * 2) Since we run with the RTNL semaphore not held, we can sleep
5656 * safely in order to wait for the netdev refcnt to drop to zero.
5658 * We must not return until all unregister events added during
5659 * the interval the lock was held have been completed.
5661 void netdev_run_todo(void)
5663 struct list_head list;
5665 /* Snapshot list, allow later requests */
5666 list_replace_init(&net_todo_list, &list);
5670 while (!list_empty(&list)) {
5671 struct net_device *dev
5672 = list_first_entry(&list, struct net_device, todo_list);
5673 list_del(&dev->todo_list);
5675 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5676 printk(KERN_ERR "network todo '%s' but state %d\n",
5677 dev->name, dev->reg_state);
5682 dev->reg_state = NETREG_UNREGISTERED;
5684 on_each_cpu(flush_backlog, dev, 1);
5686 netdev_wait_allrefs(dev);
5689 BUG_ON(netdev_refcnt_read(dev));
5690 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5691 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5692 WARN_ON(dev->dn_ptr);
5694 if (dev->destructor)
5695 dev->destructor(dev);
5697 /* Free network device */
5698 kobject_put(&dev->dev.kobj);
5702 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5703 * fields in the same order, with only the type differing.
5705 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5706 const struct net_device_stats *netdev_stats)
5708 #if BITS_PER_LONG == 64
5709 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5710 memcpy(stats64, netdev_stats, sizeof(*stats64));
5712 size_t i, n = sizeof(*stats64) / sizeof(u64);
5713 const unsigned long *src = (const unsigned long *)netdev_stats;
5714 u64 *dst = (u64 *)stats64;
5716 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5717 sizeof(*stats64) / sizeof(u64));
5718 for (i = 0; i < n; i++)
5724 * dev_get_stats - get network device statistics
5725 * @dev: device to get statistics from
5726 * @storage: place to store stats
5728 * Get network statistics from device. Return @storage.
5729 * The device driver may provide its own method by setting
5730 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5731 * otherwise the internal statistics structure is used.
5733 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5734 struct rtnl_link_stats64 *storage)
5736 const struct net_device_ops *ops = dev->netdev_ops;
5738 if (ops->ndo_get_stats64) {
5739 memset(storage, 0, sizeof(*storage));
5740 ops->ndo_get_stats64(dev, storage);
5741 } else if (ops->ndo_get_stats) {
5742 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5744 netdev_stats_to_stats64(storage, &dev->stats);
5746 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5749 EXPORT_SYMBOL(dev_get_stats);
5751 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5753 struct netdev_queue *queue = dev_ingress_queue(dev);
5755 #ifdef CONFIG_NET_CLS_ACT
5758 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5761 netdev_init_one_queue(dev, queue, NULL);
5762 queue->qdisc = &noop_qdisc;
5763 queue->qdisc_sleeping = &noop_qdisc;
5764 rcu_assign_pointer(dev->ingress_queue, queue);
5770 * alloc_netdev_mqs - allocate network device
5771 * @sizeof_priv: size of private data to allocate space for
5772 * @name: device name format string
5773 * @setup: callback to initialize device
5774 * @txqs: the number of TX subqueues to allocate
5775 * @rxqs: the number of RX subqueues to allocate
5777 * Allocates a struct net_device with private data area for driver use
5778 * and performs basic initialization. Also allocates subquue structs
5779 * for each queue on the device.
5781 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5782 void (*setup)(struct net_device *),
5783 unsigned int txqs, unsigned int rxqs)
5785 struct net_device *dev;
5787 struct net_device *p;
5789 BUG_ON(strlen(name) >= sizeof(dev->name));
5792 pr_err("alloc_netdev: Unable to allocate device "
5793 "with zero queues.\n");
5799 pr_err("alloc_netdev: Unable to allocate device "
5800 "with zero RX queues.\n");
5805 alloc_size = sizeof(struct net_device);
5807 /* ensure 32-byte alignment of private area */
5808 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5809 alloc_size += sizeof_priv;
5811 /* ensure 32-byte alignment of whole construct */
5812 alloc_size += NETDEV_ALIGN - 1;
5814 p = kzalloc(alloc_size, GFP_KERNEL);
5816 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5820 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5821 dev->padded = (char *)dev - (char *)p;
5823 dev->pcpu_refcnt = alloc_percpu(int);
5824 if (!dev->pcpu_refcnt)
5827 if (dev_addr_init(dev))
5833 dev_net_set(dev, &init_net);
5835 dev->gso_max_size = GSO_MAX_SIZE;
5837 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5838 dev->ethtool_ntuple_list.count = 0;
5839 INIT_LIST_HEAD(&dev->napi_list);
5840 INIT_LIST_HEAD(&dev->unreg_list);
5841 INIT_LIST_HEAD(&dev->link_watch_list);
5842 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5845 dev->num_tx_queues = txqs;
5846 dev->real_num_tx_queues = txqs;
5847 if (netif_alloc_netdev_queues(dev))
5851 dev->num_rx_queues = rxqs;
5852 dev->real_num_rx_queues = rxqs;
5853 if (netif_alloc_rx_queues(dev))
5857 strcpy(dev->name, name);
5858 dev->group = INIT_NETDEV_GROUP;
5866 free_percpu(dev->pcpu_refcnt);
5876 EXPORT_SYMBOL(alloc_netdev_mqs);
5879 * free_netdev - free network device
5882 * This function does the last stage of destroying an allocated device
5883 * interface. The reference to the device object is released.
5884 * If this is the last reference then it will be freed.
5886 void free_netdev(struct net_device *dev)
5888 struct napi_struct *p, *n;
5890 release_net(dev_net(dev));
5897 kfree(rcu_dereference_raw(dev->ingress_queue));
5899 /* Flush device addresses */
5900 dev_addr_flush(dev);
5902 /* Clear ethtool n-tuple list */
5903 ethtool_ntuple_flush(dev);
5905 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5908 free_percpu(dev->pcpu_refcnt);
5909 dev->pcpu_refcnt = NULL;
5911 /* Compatibility with error handling in drivers */
5912 if (dev->reg_state == NETREG_UNINITIALIZED) {
5913 kfree((char *)dev - dev->padded);
5917 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5918 dev->reg_state = NETREG_RELEASED;
5920 /* will free via device release */
5921 put_device(&dev->dev);
5923 EXPORT_SYMBOL(free_netdev);
5926 * synchronize_net - Synchronize with packet receive processing
5928 * Wait for packets currently being received to be done.
5929 * Does not block later packets from starting.
5931 void synchronize_net(void)
5936 EXPORT_SYMBOL(synchronize_net);
5939 * unregister_netdevice_queue - remove device from the kernel
5943 * This function shuts down a device interface and removes it
5944 * from the kernel tables.
5945 * If head not NULL, device is queued to be unregistered later.
5947 * Callers must hold the rtnl semaphore. You may want
5948 * unregister_netdev() instead of this.
5951 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5956 list_move_tail(&dev->unreg_list, head);
5958 rollback_registered(dev);
5959 /* Finish processing unregister after unlock */
5963 EXPORT_SYMBOL(unregister_netdevice_queue);
5966 * unregister_netdevice_many - unregister many devices
5967 * @head: list of devices
5969 void unregister_netdevice_many(struct list_head *head)
5971 struct net_device *dev;
5973 if (!list_empty(head)) {
5974 rollback_registered_many(head);
5975 list_for_each_entry(dev, head, unreg_list)
5979 EXPORT_SYMBOL(unregister_netdevice_many);
5982 * unregister_netdev - remove device from the kernel
5985 * This function shuts down a device interface and removes it
5986 * from the kernel tables.
5988 * This is just a wrapper for unregister_netdevice that takes
5989 * the rtnl semaphore. In general you want to use this and not
5990 * unregister_netdevice.
5992 void unregister_netdev(struct net_device *dev)
5995 unregister_netdevice(dev);
5998 EXPORT_SYMBOL(unregister_netdev);
6001 * dev_change_net_namespace - move device to different nethost namespace
6003 * @net: network namespace
6004 * @pat: If not NULL name pattern to try if the current device name
6005 * is already taken in the destination network namespace.
6007 * This function shuts down a device interface and moves it
6008 * to a new network namespace. On success 0 is returned, on
6009 * a failure a netagive errno code is returned.
6011 * Callers must hold the rtnl semaphore.
6014 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6020 /* Don't allow namespace local devices to be moved. */
6022 if (dev->features & NETIF_F_NETNS_LOCAL)
6025 /* Ensure the device has been registrered */
6027 if (dev->reg_state != NETREG_REGISTERED)
6030 /* Get out if there is nothing todo */
6032 if (net_eq(dev_net(dev), net))
6035 /* Pick the destination device name, and ensure
6036 * we can use it in the destination network namespace.
6039 if (__dev_get_by_name(net, dev->name)) {
6040 /* We get here if we can't use the current device name */
6043 if (dev_get_valid_name(dev, pat, 1))
6048 * And now a mini version of register_netdevice unregister_netdevice.
6051 /* If device is running close it first. */
6054 /* And unlink it from device chain */
6056 unlist_netdevice(dev);
6060 /* Shutdown queueing discipline. */
6063 /* Notify protocols, that we are about to destroy
6064 this device. They should clean all the things.
6066 Note that dev->reg_state stays at NETREG_REGISTERED.
6067 This is wanted because this way 8021q and macvlan know
6068 the device is just moving and can keep their slaves up.
6070 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6071 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6074 * Flush the unicast and multicast chains
6079 /* Actually switch the network namespace */
6080 dev_net_set(dev, net);
6082 /* If there is an ifindex conflict assign a new one */
6083 if (__dev_get_by_index(net, dev->ifindex)) {
6084 int iflink = (dev->iflink == dev->ifindex);
6085 dev->ifindex = dev_new_index(net);
6087 dev->iflink = dev->ifindex;
6090 /* Fixup kobjects */
6091 err = device_rename(&dev->dev, dev->name);
6094 /* Add the device back in the hashes */
6095 list_netdevice(dev);
6097 /* Notify protocols, that a new device appeared. */
6098 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6101 * Prevent userspace races by waiting until the network
6102 * device is fully setup before sending notifications.
6104 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6111 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6113 static int dev_cpu_callback(struct notifier_block *nfb,
6114 unsigned long action,
6117 struct sk_buff **list_skb;
6118 struct sk_buff *skb;
6119 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6120 struct softnet_data *sd, *oldsd;
6122 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6125 local_irq_disable();
6126 cpu = smp_processor_id();
6127 sd = &per_cpu(softnet_data, cpu);
6128 oldsd = &per_cpu(softnet_data, oldcpu);
6130 /* Find end of our completion_queue. */
6131 list_skb = &sd->completion_queue;
6133 list_skb = &(*list_skb)->next;
6134 /* Append completion queue from offline CPU. */
6135 *list_skb = oldsd->completion_queue;
6136 oldsd->completion_queue = NULL;
6138 /* Append output queue from offline CPU. */
6139 if (oldsd->output_queue) {
6140 *sd->output_queue_tailp = oldsd->output_queue;
6141 sd->output_queue_tailp = oldsd->output_queue_tailp;
6142 oldsd->output_queue = NULL;
6143 oldsd->output_queue_tailp = &oldsd->output_queue;
6146 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6149 /* Process offline CPU's input_pkt_queue */
6150 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6152 input_queue_head_incr(oldsd);
6154 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6156 input_queue_head_incr(oldsd);
6164 * netdev_increment_features - increment feature set by one
6165 * @all: current feature set
6166 * @one: new feature set
6167 * @mask: mask feature set
6169 * Computes a new feature set after adding a device with feature set
6170 * @one to the master device with current feature set @all. Will not
6171 * enable anything that is off in @mask. Returns the new feature set.
6173 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6175 /* If device needs checksumming, downgrade to it. */
6176 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
6177 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
6178 else if (mask & NETIF_F_ALL_CSUM) {
6179 /* If one device supports v4/v6 checksumming, set for all. */
6180 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
6181 !(all & NETIF_F_GEN_CSUM)) {
6182 all &= ~NETIF_F_ALL_CSUM;
6183 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
6186 /* If one device supports hw checksumming, set for all. */
6187 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
6188 all &= ~NETIF_F_ALL_CSUM;
6189 all |= NETIF_F_HW_CSUM;
6193 one |= NETIF_F_ALL_CSUM;
6195 one |= all & NETIF_F_ONE_FOR_ALL;
6196 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
6197 all |= one & mask & NETIF_F_ONE_FOR_ALL;
6201 EXPORT_SYMBOL(netdev_increment_features);
6203 static struct hlist_head *netdev_create_hash(void)
6206 struct hlist_head *hash;
6208 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6210 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6211 INIT_HLIST_HEAD(&hash[i]);
6216 /* Initialize per network namespace state */
6217 static int __net_init netdev_init(struct net *net)
6219 INIT_LIST_HEAD(&net->dev_base_head);
6221 net->dev_name_head = netdev_create_hash();
6222 if (net->dev_name_head == NULL)
6225 net->dev_index_head = netdev_create_hash();
6226 if (net->dev_index_head == NULL)
6232 kfree(net->dev_name_head);
6238 * netdev_drivername - network driver for the device
6239 * @dev: network device
6240 * @buffer: buffer for resulting name
6241 * @len: size of buffer
6243 * Determine network driver for device.
6245 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6247 const struct device_driver *driver;
6248 const struct device *parent;
6250 if (len <= 0 || !buffer)
6254 parent = dev->dev.parent;
6259 driver = parent->driver;
6260 if (driver && driver->name)
6261 strlcpy(buffer, driver->name, len);
6265 static int __netdev_printk(const char *level, const struct net_device *dev,
6266 struct va_format *vaf)
6270 if (dev && dev->dev.parent)
6271 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6272 netdev_name(dev), vaf);
6274 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6276 r = printk("%s(NULL net_device): %pV", level, vaf);
6281 int netdev_printk(const char *level, const struct net_device *dev,
6282 const char *format, ...)
6284 struct va_format vaf;
6288 va_start(args, format);
6293 r = __netdev_printk(level, dev, &vaf);
6298 EXPORT_SYMBOL(netdev_printk);
6300 #define define_netdev_printk_level(func, level) \
6301 int func(const struct net_device *dev, const char *fmt, ...) \
6304 struct va_format vaf; \
6307 va_start(args, fmt); \
6312 r = __netdev_printk(level, dev, &vaf); \
6317 EXPORT_SYMBOL(func);
6319 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6320 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6321 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6322 define_netdev_printk_level(netdev_err, KERN_ERR);
6323 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6324 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6325 define_netdev_printk_level(netdev_info, KERN_INFO);
6327 static void __net_exit netdev_exit(struct net *net)
6329 kfree(net->dev_name_head);
6330 kfree(net->dev_index_head);
6333 static struct pernet_operations __net_initdata netdev_net_ops = {
6334 .init = netdev_init,
6335 .exit = netdev_exit,
6338 static void __net_exit default_device_exit(struct net *net)
6340 struct net_device *dev, *aux;
6342 * Push all migratable network devices back to the
6343 * initial network namespace
6346 for_each_netdev_safe(net, dev, aux) {
6348 char fb_name[IFNAMSIZ];
6350 /* Ignore unmoveable devices (i.e. loopback) */
6351 if (dev->features & NETIF_F_NETNS_LOCAL)
6354 /* Leave virtual devices for the generic cleanup */
6355 if (dev->rtnl_link_ops)
6358 /* Push remaing network devices to init_net */
6359 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6360 err = dev_change_net_namespace(dev, &init_net, fb_name);
6362 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6363 __func__, dev->name, err);
6370 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6372 /* At exit all network devices most be removed from a network
6373 * namespace. Do this in the reverse order of registration.
6374 * Do this across as many network namespaces as possible to
6375 * improve batching efficiency.
6377 struct net_device *dev;
6379 LIST_HEAD(dev_kill_list);
6382 list_for_each_entry(net, net_list, exit_list) {
6383 for_each_netdev_reverse(net, dev) {
6384 if (dev->rtnl_link_ops)
6385 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6387 unregister_netdevice_queue(dev, &dev_kill_list);
6390 unregister_netdevice_many(&dev_kill_list);
6394 static struct pernet_operations __net_initdata default_device_ops = {
6395 .exit = default_device_exit,
6396 .exit_batch = default_device_exit_batch,
6400 * Initialize the DEV module. At boot time this walks the device list and
6401 * unhooks any devices that fail to initialise (normally hardware not
6402 * present) and leaves us with a valid list of present and active devices.
6407 * This is called single threaded during boot, so no need
6408 * to take the rtnl semaphore.
6410 static int __init net_dev_init(void)
6412 int i, rc = -ENOMEM;
6414 BUG_ON(!dev_boot_phase);
6416 if (dev_proc_init())
6419 if (netdev_kobject_init())
6422 INIT_LIST_HEAD(&ptype_all);
6423 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6424 INIT_LIST_HEAD(&ptype_base[i]);
6426 if (register_pernet_subsys(&netdev_net_ops))
6430 * Initialise the packet receive queues.
6433 for_each_possible_cpu(i) {
6434 struct softnet_data *sd = &per_cpu(softnet_data, i);
6436 memset(sd, 0, sizeof(*sd));
6437 skb_queue_head_init(&sd->input_pkt_queue);
6438 skb_queue_head_init(&sd->process_queue);
6439 sd->completion_queue = NULL;
6440 INIT_LIST_HEAD(&sd->poll_list);
6441 sd->output_queue = NULL;
6442 sd->output_queue_tailp = &sd->output_queue;
6444 sd->csd.func = rps_trigger_softirq;
6450 sd->backlog.poll = process_backlog;
6451 sd->backlog.weight = weight_p;
6452 sd->backlog.gro_list = NULL;
6453 sd->backlog.gro_count = 0;
6458 /* The loopback device is special if any other network devices
6459 * is present in a network namespace the loopback device must
6460 * be present. Since we now dynamically allocate and free the
6461 * loopback device ensure this invariant is maintained by
6462 * keeping the loopback device as the first device on the
6463 * list of network devices. Ensuring the loopback devices
6464 * is the first device that appears and the last network device
6467 if (register_pernet_device(&loopback_net_ops))
6470 if (register_pernet_device(&default_device_ops))
6473 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6474 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6476 hotcpu_notifier(dev_cpu_callback, 0);
6484 subsys_initcall(net_dev_init);
6486 static int __init initialize_hashrnd(void)
6488 get_random_bytes(&hashrnd, sizeof(hashrnd));
6492 late_initcall_sync(initialize_hashrnd);