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;
1121 dev = dev_get_by_name_rcu(net, name);
1125 if (no_module && capable(CAP_NET_ADMIN))
1126 no_module = request_module("netdev-%s", name);
1127 if (no_module && capable(CAP_SYS_MODULE)) {
1128 if (!request_module("%s", name))
1129 pr_err("Loading kernel module for a network device "
1130 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1134 EXPORT_SYMBOL(dev_load);
1136 static int __dev_open(struct net_device *dev)
1138 const struct net_device_ops *ops = dev->netdev_ops;
1143 if (!netif_device_present(dev))
1146 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1147 ret = notifier_to_errno(ret);
1151 set_bit(__LINK_STATE_START, &dev->state);
1153 if (ops->ndo_validate_addr)
1154 ret = ops->ndo_validate_addr(dev);
1156 if (!ret && ops->ndo_open)
1157 ret = ops->ndo_open(dev);
1160 clear_bit(__LINK_STATE_START, &dev->state);
1162 dev->flags |= IFF_UP;
1163 net_dmaengine_get();
1164 dev_set_rx_mode(dev);
1172 * dev_open - prepare an interface for use.
1173 * @dev: device to open
1175 * Takes a device from down to up state. The device's private open
1176 * function is invoked and then the multicast lists are loaded. Finally
1177 * the device is moved into the up state and a %NETDEV_UP message is
1178 * sent to the netdev notifier chain.
1180 * Calling this function on an active interface is a nop. On a failure
1181 * a negative errno code is returned.
1183 int dev_open(struct net_device *dev)
1187 if (dev->flags & IFF_UP)
1190 ret = __dev_open(dev);
1194 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1195 call_netdevice_notifiers(NETDEV_UP, dev);
1199 EXPORT_SYMBOL(dev_open);
1201 static int __dev_close_many(struct list_head *head)
1203 struct net_device *dev;
1208 list_for_each_entry(dev, head, unreg_list) {
1209 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1211 clear_bit(__LINK_STATE_START, &dev->state);
1213 /* Synchronize to scheduled poll. We cannot touch poll list, it
1214 * can be even on different cpu. So just clear netif_running().
1216 * dev->stop() will invoke napi_disable() on all of it's
1217 * napi_struct instances on this device.
1219 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1222 dev_deactivate_many(head);
1224 list_for_each_entry(dev, head, unreg_list) {
1225 const struct net_device_ops *ops = dev->netdev_ops;
1228 * Call the device specific close. This cannot fail.
1229 * Only if device is UP
1231 * We allow it to be called even after a DETACH hot-plug
1237 dev->flags &= ~IFF_UP;
1238 net_dmaengine_put();
1244 static int __dev_close(struct net_device *dev)
1249 list_add(&dev->unreg_list, &single);
1250 retval = __dev_close_many(&single);
1255 static int dev_close_many(struct list_head *head)
1257 struct net_device *dev, *tmp;
1258 LIST_HEAD(tmp_list);
1260 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1261 if (!(dev->flags & IFF_UP))
1262 list_move(&dev->unreg_list, &tmp_list);
1264 __dev_close_many(head);
1266 list_for_each_entry(dev, head, unreg_list) {
1267 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1268 call_netdevice_notifiers(NETDEV_DOWN, dev);
1271 /* rollback_registered_many needs the complete original list */
1272 list_splice(&tmp_list, head);
1277 * dev_close - shutdown an interface.
1278 * @dev: device to shutdown
1280 * This function moves an active device into down state. A
1281 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1282 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1285 int dev_close(struct net_device *dev)
1289 list_add(&dev->unreg_list, &single);
1290 dev_close_many(&single);
1294 EXPORT_SYMBOL(dev_close);
1298 * dev_disable_lro - disable Large Receive Offload on a device
1301 * Disable Large Receive Offload (LRO) on a net device. Must be
1302 * called under RTNL. This is needed if received packets may be
1303 * forwarded to another interface.
1305 void dev_disable_lro(struct net_device *dev)
1309 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1310 flags = dev->ethtool_ops->get_flags(dev);
1312 flags = ethtool_op_get_flags(dev);
1314 if (!(flags & ETH_FLAG_LRO))
1317 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1318 if (unlikely(dev->features & NETIF_F_LRO))
1319 netdev_WARN(dev, "failed to disable LRO!\n");
1321 EXPORT_SYMBOL(dev_disable_lro);
1324 static int dev_boot_phase = 1;
1327 * register_netdevice_notifier - register a network notifier block
1330 * Register a notifier to be called when network device events occur.
1331 * The notifier passed is linked into the kernel structures and must
1332 * not be reused until it has been unregistered. A negative errno code
1333 * is returned on a failure.
1335 * When registered all registration and up events are replayed
1336 * to the new notifier to allow device to have a race free
1337 * view of the network device list.
1340 int register_netdevice_notifier(struct notifier_block *nb)
1342 struct net_device *dev;
1343 struct net_device *last;
1348 err = raw_notifier_chain_register(&netdev_chain, nb);
1354 for_each_netdev(net, dev) {
1355 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1356 err = notifier_to_errno(err);
1360 if (!(dev->flags & IFF_UP))
1363 nb->notifier_call(nb, NETDEV_UP, dev);
1374 for_each_netdev(net, dev) {
1378 if (dev->flags & IFF_UP) {
1379 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1380 nb->notifier_call(nb, NETDEV_DOWN, dev);
1382 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1383 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1387 raw_notifier_chain_unregister(&netdev_chain, nb);
1390 EXPORT_SYMBOL(register_netdevice_notifier);
1393 * unregister_netdevice_notifier - unregister a network notifier block
1396 * Unregister a notifier previously registered by
1397 * register_netdevice_notifier(). The notifier is unlinked into the
1398 * kernel structures and may then be reused. A negative errno code
1399 * is returned on a failure.
1402 int unregister_netdevice_notifier(struct notifier_block *nb)
1407 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1411 EXPORT_SYMBOL(unregister_netdevice_notifier);
1414 * call_netdevice_notifiers - call all network notifier blocks
1415 * @val: value passed unmodified to notifier function
1416 * @dev: net_device pointer passed unmodified to notifier function
1418 * Call all network notifier blocks. Parameters and return value
1419 * are as for raw_notifier_call_chain().
1422 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1425 return raw_notifier_call_chain(&netdev_chain, val, dev);
1427 EXPORT_SYMBOL(call_netdevice_notifiers);
1429 /* When > 0 there are consumers of rx skb time stamps */
1430 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1432 void net_enable_timestamp(void)
1434 atomic_inc(&netstamp_needed);
1436 EXPORT_SYMBOL(net_enable_timestamp);
1438 void net_disable_timestamp(void)
1440 atomic_dec(&netstamp_needed);
1442 EXPORT_SYMBOL(net_disable_timestamp);
1444 static inline void net_timestamp_set(struct sk_buff *skb)
1446 if (atomic_read(&netstamp_needed))
1447 __net_timestamp(skb);
1449 skb->tstamp.tv64 = 0;
1452 static inline void net_timestamp_check(struct sk_buff *skb)
1454 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1455 __net_timestamp(skb);
1458 static inline bool is_skb_forwardable(struct net_device *dev,
1459 struct sk_buff *skb)
1463 if (!(dev->flags & IFF_UP))
1466 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1467 if (skb->len <= len)
1470 /* if TSO is enabled, we don't care about the length as the packet
1471 * could be forwarded without being segmented before
1473 if (skb_is_gso(skb))
1480 * dev_forward_skb - loopback an skb to another netif
1482 * @dev: destination network device
1483 * @skb: buffer to forward
1486 * NET_RX_SUCCESS (no congestion)
1487 * NET_RX_DROP (packet was dropped, but freed)
1489 * dev_forward_skb can be used for injecting an skb from the
1490 * start_xmit function of one device into the receive queue
1491 * of another device.
1493 * The receiving device may be in another namespace, so
1494 * we have to clear all information in the skb that could
1495 * impact namespace isolation.
1497 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1502 if (unlikely(!is_skb_forwardable(dev, skb))) {
1503 atomic_long_inc(&dev->rx_dropped);
1507 skb_set_dev(skb, dev);
1508 skb->tstamp.tv64 = 0;
1509 skb->pkt_type = PACKET_HOST;
1510 skb->protocol = eth_type_trans(skb, dev);
1511 return netif_rx(skb);
1513 EXPORT_SYMBOL_GPL(dev_forward_skb);
1515 static inline int deliver_skb(struct sk_buff *skb,
1516 struct packet_type *pt_prev,
1517 struct net_device *orig_dev)
1519 atomic_inc(&skb->users);
1520 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1524 * Support routine. Sends outgoing frames to any network
1525 * taps currently in use.
1528 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1530 struct packet_type *ptype;
1531 struct sk_buff *skb2 = NULL;
1532 struct packet_type *pt_prev = NULL;
1535 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1536 /* Never send packets back to the socket
1537 * they originated from - MvS (miquels@drinkel.ow.org)
1539 if ((ptype->dev == dev || !ptype->dev) &&
1540 (ptype->af_packet_priv == NULL ||
1541 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1543 deliver_skb(skb2, pt_prev, skb->dev);
1548 skb2 = skb_clone(skb, GFP_ATOMIC);
1552 net_timestamp_set(skb2);
1554 /* skb->nh should be correctly
1555 set by sender, so that the second statement is
1556 just protection against buggy protocols.
1558 skb_reset_mac_header(skb2);
1560 if (skb_network_header(skb2) < skb2->data ||
1561 skb2->network_header > skb2->tail) {
1562 if (net_ratelimit())
1563 printk(KERN_CRIT "protocol %04x is "
1565 ntohs(skb2->protocol),
1567 skb_reset_network_header(skb2);
1570 skb2->transport_header = skb2->network_header;
1571 skb2->pkt_type = PACKET_OUTGOING;
1576 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1580 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1581 * @dev: Network device
1582 * @txq: number of queues available
1584 * If real_num_tx_queues is changed the tc mappings may no longer be
1585 * valid. To resolve this verify the tc mapping remains valid and if
1586 * not NULL the mapping. With no priorities mapping to this
1587 * offset/count pair it will no longer be used. In the worst case TC0
1588 * is invalid nothing can be done so disable priority mappings. If is
1589 * expected that drivers will fix this mapping if they can before
1590 * calling netif_set_real_num_tx_queues.
1592 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1595 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1597 /* If TC0 is invalidated disable TC mapping */
1598 if (tc->offset + tc->count > txq) {
1599 pr_warning("Number of in use tx queues changed "
1600 "invalidating tc mappings. Priority "
1601 "traffic classification disabled!\n");
1606 /* Invalidated prio to tc mappings set to TC0 */
1607 for (i = 1; i < TC_BITMASK + 1; i++) {
1608 int q = netdev_get_prio_tc_map(dev, i);
1610 tc = &dev->tc_to_txq[q];
1611 if (tc->offset + tc->count > txq) {
1612 pr_warning("Number of in use tx queues "
1613 "changed. Priority %i to tc "
1614 "mapping %i is no longer valid "
1615 "setting map to 0\n",
1617 netdev_set_prio_tc_map(dev, i, 0);
1623 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1624 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1626 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1630 if (txq < 1 || txq > dev->num_tx_queues)
1633 if (dev->reg_state == NETREG_REGISTERED ||
1634 dev->reg_state == NETREG_UNREGISTERING) {
1637 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1643 netif_setup_tc(dev, txq);
1645 if (txq < dev->real_num_tx_queues)
1646 qdisc_reset_all_tx_gt(dev, txq);
1649 dev->real_num_tx_queues = txq;
1652 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1656 * netif_set_real_num_rx_queues - set actual number of RX queues used
1657 * @dev: Network device
1658 * @rxq: Actual number of RX queues
1660 * This must be called either with the rtnl_lock held or before
1661 * registration of the net device. Returns 0 on success, or a
1662 * negative error code. If called before registration, it always
1665 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1669 if (rxq < 1 || rxq > dev->num_rx_queues)
1672 if (dev->reg_state == NETREG_REGISTERED) {
1675 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1681 dev->real_num_rx_queues = rxq;
1684 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1687 static inline void __netif_reschedule(struct Qdisc *q)
1689 struct softnet_data *sd;
1690 unsigned long flags;
1692 local_irq_save(flags);
1693 sd = &__get_cpu_var(softnet_data);
1694 q->next_sched = NULL;
1695 *sd->output_queue_tailp = q;
1696 sd->output_queue_tailp = &q->next_sched;
1697 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1698 local_irq_restore(flags);
1701 void __netif_schedule(struct Qdisc *q)
1703 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1704 __netif_reschedule(q);
1706 EXPORT_SYMBOL(__netif_schedule);
1708 void dev_kfree_skb_irq(struct sk_buff *skb)
1710 if (atomic_dec_and_test(&skb->users)) {
1711 struct softnet_data *sd;
1712 unsigned long flags;
1714 local_irq_save(flags);
1715 sd = &__get_cpu_var(softnet_data);
1716 skb->next = sd->completion_queue;
1717 sd->completion_queue = skb;
1718 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1719 local_irq_restore(flags);
1722 EXPORT_SYMBOL(dev_kfree_skb_irq);
1724 void dev_kfree_skb_any(struct sk_buff *skb)
1726 if (in_irq() || irqs_disabled())
1727 dev_kfree_skb_irq(skb);
1731 EXPORT_SYMBOL(dev_kfree_skb_any);
1735 * netif_device_detach - mark device as removed
1736 * @dev: network device
1738 * Mark device as removed from system and therefore no longer available.
1740 void netif_device_detach(struct net_device *dev)
1742 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1743 netif_running(dev)) {
1744 netif_tx_stop_all_queues(dev);
1747 EXPORT_SYMBOL(netif_device_detach);
1750 * netif_device_attach - mark device as attached
1751 * @dev: network device
1753 * Mark device as attached from system and restart if needed.
1755 void netif_device_attach(struct net_device *dev)
1757 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1758 netif_running(dev)) {
1759 netif_tx_wake_all_queues(dev);
1760 __netdev_watchdog_up(dev);
1763 EXPORT_SYMBOL(netif_device_attach);
1766 * skb_dev_set -- assign a new device to a buffer
1767 * @skb: buffer for the new device
1768 * @dev: network device
1770 * If an skb is owned by a device already, we have to reset
1771 * all data private to the namespace a device belongs to
1772 * before assigning it a new device.
1774 #ifdef CONFIG_NET_NS
1775 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1778 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1781 skb_init_secmark(skb);
1785 skb->ipvs_property = 0;
1786 #ifdef CONFIG_NET_SCHED
1792 EXPORT_SYMBOL(skb_set_dev);
1793 #endif /* CONFIG_NET_NS */
1796 * Invalidate hardware checksum when packet is to be mangled, and
1797 * complete checksum manually on outgoing path.
1799 int skb_checksum_help(struct sk_buff *skb)
1802 int ret = 0, offset;
1804 if (skb->ip_summed == CHECKSUM_COMPLETE)
1805 goto out_set_summed;
1807 if (unlikely(skb_shinfo(skb)->gso_size)) {
1808 /* Let GSO fix up the checksum. */
1809 goto out_set_summed;
1812 offset = skb_checksum_start_offset(skb);
1813 BUG_ON(offset >= skb_headlen(skb));
1814 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1816 offset += skb->csum_offset;
1817 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1819 if (skb_cloned(skb) &&
1820 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1821 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1826 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1828 skb->ip_summed = CHECKSUM_NONE;
1832 EXPORT_SYMBOL(skb_checksum_help);
1835 * skb_gso_segment - Perform segmentation on skb.
1836 * @skb: buffer to segment
1837 * @features: features for the output path (see dev->features)
1839 * This function segments the given skb and returns a list of segments.
1841 * It may return NULL if the skb requires no segmentation. This is
1842 * only possible when GSO is used for verifying header integrity.
1844 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1846 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1847 struct packet_type *ptype;
1848 __be16 type = skb->protocol;
1849 int vlan_depth = ETH_HLEN;
1852 while (type == htons(ETH_P_8021Q)) {
1853 struct vlan_hdr *vh;
1855 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1856 return ERR_PTR(-EINVAL);
1858 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1859 type = vh->h_vlan_encapsulated_proto;
1860 vlan_depth += VLAN_HLEN;
1863 skb_reset_mac_header(skb);
1864 skb->mac_len = skb->network_header - skb->mac_header;
1865 __skb_pull(skb, skb->mac_len);
1867 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1868 struct net_device *dev = skb->dev;
1869 struct ethtool_drvinfo info = {};
1871 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1872 dev->ethtool_ops->get_drvinfo(dev, &info);
1874 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1875 info.driver, dev ? dev->features : 0L,
1876 skb->sk ? skb->sk->sk_route_caps : 0L,
1877 skb->len, skb->data_len, skb->ip_summed);
1879 if (skb_header_cloned(skb) &&
1880 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1881 return ERR_PTR(err);
1885 list_for_each_entry_rcu(ptype,
1886 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1887 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1888 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1889 err = ptype->gso_send_check(skb);
1890 segs = ERR_PTR(err);
1891 if (err || skb_gso_ok(skb, features))
1893 __skb_push(skb, (skb->data -
1894 skb_network_header(skb)));
1896 segs = ptype->gso_segment(skb, features);
1902 __skb_push(skb, skb->data - skb_mac_header(skb));
1906 EXPORT_SYMBOL(skb_gso_segment);
1908 /* Take action when hardware reception checksum errors are detected. */
1910 void netdev_rx_csum_fault(struct net_device *dev)
1912 if (net_ratelimit()) {
1913 printk(KERN_ERR "%s: hw csum failure.\n",
1914 dev ? dev->name : "<unknown>");
1918 EXPORT_SYMBOL(netdev_rx_csum_fault);
1921 /* Actually, we should eliminate this check as soon as we know, that:
1922 * 1. IOMMU is present and allows to map all the memory.
1923 * 2. No high memory really exists on this machine.
1926 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1928 #ifdef CONFIG_HIGHMEM
1930 if (!(dev->features & NETIF_F_HIGHDMA)) {
1931 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1932 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1936 if (PCI_DMA_BUS_IS_PHYS) {
1937 struct device *pdev = dev->dev.parent;
1941 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1942 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1943 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1952 void (*destructor)(struct sk_buff *skb);
1955 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1957 static void dev_gso_skb_destructor(struct sk_buff *skb)
1959 struct dev_gso_cb *cb;
1962 struct sk_buff *nskb = skb->next;
1964 skb->next = nskb->next;
1967 } while (skb->next);
1969 cb = DEV_GSO_CB(skb);
1971 cb->destructor(skb);
1975 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1976 * @skb: buffer to segment
1977 * @features: device features as applicable to this skb
1979 * This function segments the given skb and stores the list of segments
1982 static int dev_gso_segment(struct sk_buff *skb, int features)
1984 struct sk_buff *segs;
1986 segs = skb_gso_segment(skb, features);
1988 /* Verifying header integrity only. */
1993 return PTR_ERR(segs);
1996 DEV_GSO_CB(skb)->destructor = skb->destructor;
1997 skb->destructor = dev_gso_skb_destructor;
2003 * Try to orphan skb early, right before transmission by the device.
2004 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2005 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2007 static inline void skb_orphan_try(struct sk_buff *skb)
2009 struct sock *sk = skb->sk;
2011 if (sk && !skb_shinfo(skb)->tx_flags) {
2012 /* skb_tx_hash() wont be able to get sk.
2013 * We copy sk_hash into skb->rxhash
2016 skb->rxhash = sk->sk_hash;
2021 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2023 return ((features & NETIF_F_GEN_CSUM) ||
2024 ((features & NETIF_F_V4_CSUM) &&
2025 protocol == htons(ETH_P_IP)) ||
2026 ((features & NETIF_F_V6_CSUM) &&
2027 protocol == htons(ETH_P_IPV6)) ||
2028 ((features & NETIF_F_FCOE_CRC) &&
2029 protocol == htons(ETH_P_FCOE)));
2032 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2034 if (!can_checksum_protocol(features, protocol)) {
2035 features &= ~NETIF_F_ALL_CSUM;
2036 features &= ~NETIF_F_SG;
2037 } else if (illegal_highdma(skb->dev, skb)) {
2038 features &= ~NETIF_F_SG;
2044 u32 netif_skb_features(struct sk_buff *skb)
2046 __be16 protocol = skb->protocol;
2047 u32 features = skb->dev->features;
2049 if (protocol == htons(ETH_P_8021Q)) {
2050 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2051 protocol = veh->h_vlan_encapsulated_proto;
2052 } else if (!vlan_tx_tag_present(skb)) {
2053 return harmonize_features(skb, protocol, features);
2056 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2058 if (protocol != htons(ETH_P_8021Q)) {
2059 return harmonize_features(skb, protocol, features);
2061 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2062 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2063 return harmonize_features(skb, protocol, features);
2066 EXPORT_SYMBOL(netif_skb_features);
2069 * Returns true if either:
2070 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2071 * 2. skb is fragmented and the device does not support SG, or if
2072 * at least one of fragments is in highmem and device does not
2073 * support DMA from it.
2075 static inline int skb_needs_linearize(struct sk_buff *skb,
2078 return skb_is_nonlinear(skb) &&
2079 ((skb_has_frag_list(skb) &&
2080 !(features & NETIF_F_FRAGLIST)) ||
2081 (skb_shinfo(skb)->nr_frags &&
2082 !(features & NETIF_F_SG)));
2085 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2086 struct netdev_queue *txq)
2088 const struct net_device_ops *ops = dev->netdev_ops;
2089 int rc = NETDEV_TX_OK;
2091 if (likely(!skb->next)) {
2095 * If device doesn't need skb->dst, release it right now while
2096 * its hot in this cpu cache
2098 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2101 if (!list_empty(&ptype_all))
2102 dev_queue_xmit_nit(skb, dev);
2104 skb_orphan_try(skb);
2106 features = netif_skb_features(skb);
2108 if (vlan_tx_tag_present(skb) &&
2109 !(features & NETIF_F_HW_VLAN_TX)) {
2110 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2117 if (netif_needs_gso(skb, features)) {
2118 if (unlikely(dev_gso_segment(skb, features)))
2123 if (skb_needs_linearize(skb, features) &&
2124 __skb_linearize(skb))
2127 /* If packet is not checksummed and device does not
2128 * support checksumming for this protocol, complete
2129 * checksumming here.
2131 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2132 skb_set_transport_header(skb,
2133 skb_checksum_start_offset(skb));
2134 if (!(features & NETIF_F_ALL_CSUM) &&
2135 skb_checksum_help(skb))
2140 rc = ops->ndo_start_xmit(skb, dev);
2141 trace_net_dev_xmit(skb, rc);
2142 if (rc == NETDEV_TX_OK)
2143 txq_trans_update(txq);
2149 struct sk_buff *nskb = skb->next;
2151 skb->next = nskb->next;
2155 * If device doesn't need nskb->dst, release it right now while
2156 * its hot in this cpu cache
2158 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2161 rc = ops->ndo_start_xmit(nskb, dev);
2162 trace_net_dev_xmit(nskb, rc);
2163 if (unlikely(rc != NETDEV_TX_OK)) {
2164 if (rc & ~NETDEV_TX_MASK)
2165 goto out_kfree_gso_skb;
2166 nskb->next = skb->next;
2170 txq_trans_update(txq);
2171 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2172 return NETDEV_TX_BUSY;
2173 } while (skb->next);
2176 if (likely(skb->next == NULL))
2177 skb->destructor = DEV_GSO_CB(skb)->destructor;
2184 static u32 hashrnd __read_mostly;
2187 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2188 * to be used as a distribution range.
2190 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2191 unsigned int num_tx_queues)
2195 u16 qcount = num_tx_queues;
2197 if (skb_rx_queue_recorded(skb)) {
2198 hash = skb_get_rx_queue(skb);
2199 while (unlikely(hash >= num_tx_queues))
2200 hash -= num_tx_queues;
2205 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2206 qoffset = dev->tc_to_txq[tc].offset;
2207 qcount = dev->tc_to_txq[tc].count;
2210 if (skb->sk && skb->sk->sk_hash)
2211 hash = skb->sk->sk_hash;
2213 hash = (__force u16) skb->protocol ^ skb->rxhash;
2214 hash = jhash_1word(hash, hashrnd);
2216 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2218 EXPORT_SYMBOL(__skb_tx_hash);
2220 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2222 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2223 if (net_ratelimit()) {
2224 pr_warning("%s selects TX queue %d, but "
2225 "real number of TX queues is %d\n",
2226 dev->name, queue_index, dev->real_num_tx_queues);
2233 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2236 struct xps_dev_maps *dev_maps;
2237 struct xps_map *map;
2238 int queue_index = -1;
2241 dev_maps = rcu_dereference(dev->xps_maps);
2243 map = rcu_dereference(
2244 dev_maps->cpu_map[raw_smp_processor_id()]);
2247 queue_index = map->queues[0];
2250 if (skb->sk && skb->sk->sk_hash)
2251 hash = skb->sk->sk_hash;
2253 hash = (__force u16) skb->protocol ^
2255 hash = jhash_1word(hash, hashrnd);
2256 queue_index = map->queues[
2257 ((u64)hash * map->len) >> 32];
2259 if (unlikely(queue_index >= dev->real_num_tx_queues))
2271 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2272 struct sk_buff *skb)
2275 const struct net_device_ops *ops = dev->netdev_ops;
2277 if (dev->real_num_tx_queues == 1)
2279 else if (ops->ndo_select_queue) {
2280 queue_index = ops->ndo_select_queue(dev, skb);
2281 queue_index = dev_cap_txqueue(dev, queue_index);
2283 struct sock *sk = skb->sk;
2284 queue_index = sk_tx_queue_get(sk);
2286 if (queue_index < 0 || skb->ooo_okay ||
2287 queue_index >= dev->real_num_tx_queues) {
2288 int old_index = queue_index;
2290 queue_index = get_xps_queue(dev, skb);
2291 if (queue_index < 0)
2292 queue_index = skb_tx_hash(dev, skb);
2294 if (queue_index != old_index && sk) {
2295 struct dst_entry *dst =
2296 rcu_dereference_check(sk->sk_dst_cache, 1);
2298 if (dst && skb_dst(skb) == dst)
2299 sk_tx_queue_set(sk, queue_index);
2304 skb_set_queue_mapping(skb, queue_index);
2305 return netdev_get_tx_queue(dev, queue_index);
2308 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2309 struct net_device *dev,
2310 struct netdev_queue *txq)
2312 spinlock_t *root_lock = qdisc_lock(q);
2316 qdisc_skb_cb(skb)->pkt_len = skb->len;
2317 qdisc_calculate_pkt_len(skb, q);
2319 * Heuristic to force contended enqueues to serialize on a
2320 * separate lock before trying to get qdisc main lock.
2321 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2322 * and dequeue packets faster.
2324 contended = qdisc_is_running(q);
2325 if (unlikely(contended))
2326 spin_lock(&q->busylock);
2328 spin_lock(root_lock);
2329 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2332 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2333 qdisc_run_begin(q)) {
2335 * This is a work-conserving queue; there are no old skbs
2336 * waiting to be sent out; and the qdisc is not running -
2337 * xmit the skb directly.
2339 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2342 qdisc_bstats_update(q, skb);
2344 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2345 if (unlikely(contended)) {
2346 spin_unlock(&q->busylock);
2353 rc = NET_XMIT_SUCCESS;
2356 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2357 if (qdisc_run_begin(q)) {
2358 if (unlikely(contended)) {
2359 spin_unlock(&q->busylock);
2365 spin_unlock(root_lock);
2366 if (unlikely(contended))
2367 spin_unlock(&q->busylock);
2371 static DEFINE_PER_CPU(int, xmit_recursion);
2372 #define RECURSION_LIMIT 10
2375 * dev_queue_xmit - transmit a buffer
2376 * @skb: buffer to transmit
2378 * Queue a buffer for transmission to a network device. The caller must
2379 * have set the device and priority and built the buffer before calling
2380 * this function. The function can be called from an interrupt.
2382 * A negative errno code is returned on a failure. A success does not
2383 * guarantee the frame will be transmitted as it may be dropped due
2384 * to congestion or traffic shaping.
2386 * -----------------------------------------------------------------------------------
2387 * I notice this method can also return errors from the queue disciplines,
2388 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2391 * Regardless of the return value, the skb is consumed, so it is currently
2392 * difficult to retry a send to this method. (You can bump the ref count
2393 * before sending to hold a reference for retry if you are careful.)
2395 * When calling this method, interrupts MUST be enabled. This is because
2396 * the BH enable code must have IRQs enabled so that it will not deadlock.
2399 int dev_queue_xmit(struct sk_buff *skb)
2401 struct net_device *dev = skb->dev;
2402 struct netdev_queue *txq;
2406 /* Disable soft irqs for various locks below. Also
2407 * stops preemption for RCU.
2411 txq = dev_pick_tx(dev, skb);
2412 q = rcu_dereference_bh(txq->qdisc);
2414 #ifdef CONFIG_NET_CLS_ACT
2415 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2417 trace_net_dev_queue(skb);
2419 rc = __dev_xmit_skb(skb, q, dev, txq);
2423 /* The device has no queue. Common case for software devices:
2424 loopback, all the sorts of tunnels...
2426 Really, it is unlikely that netif_tx_lock protection is necessary
2427 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2429 However, it is possible, that they rely on protection
2432 Check this and shot the lock. It is not prone from deadlocks.
2433 Either shot noqueue qdisc, it is even simpler 8)
2435 if (dev->flags & IFF_UP) {
2436 int cpu = smp_processor_id(); /* ok because BHs are off */
2438 if (txq->xmit_lock_owner != cpu) {
2440 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2441 goto recursion_alert;
2443 HARD_TX_LOCK(dev, txq, cpu);
2445 if (!netif_tx_queue_stopped(txq)) {
2446 __this_cpu_inc(xmit_recursion);
2447 rc = dev_hard_start_xmit(skb, dev, txq);
2448 __this_cpu_dec(xmit_recursion);
2449 if (dev_xmit_complete(rc)) {
2450 HARD_TX_UNLOCK(dev, txq);
2454 HARD_TX_UNLOCK(dev, txq);
2455 if (net_ratelimit())
2456 printk(KERN_CRIT "Virtual device %s asks to "
2457 "queue packet!\n", dev->name);
2459 /* Recursion is detected! It is possible,
2463 if (net_ratelimit())
2464 printk(KERN_CRIT "Dead loop on virtual device "
2465 "%s, fix it urgently!\n", dev->name);
2470 rcu_read_unlock_bh();
2475 rcu_read_unlock_bh();
2478 EXPORT_SYMBOL(dev_queue_xmit);
2481 /*=======================================================================
2483 =======================================================================*/
2485 int netdev_max_backlog __read_mostly = 1000;
2486 int netdev_tstamp_prequeue __read_mostly = 1;
2487 int netdev_budget __read_mostly = 300;
2488 int weight_p __read_mostly = 64; /* old backlog weight */
2490 /* Called with irq disabled */
2491 static inline void ____napi_schedule(struct softnet_data *sd,
2492 struct napi_struct *napi)
2494 list_add_tail(&napi->poll_list, &sd->poll_list);
2495 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2499 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2500 * and src/dst port numbers. Returns a non-zero hash number on success
2503 __u32 __skb_get_rxhash(struct sk_buff *skb)
2505 int nhoff, hash = 0, poff;
2506 const struct ipv6hdr *ip6;
2507 const struct iphdr *ip;
2509 u32 addr1, addr2, ihl;
2515 nhoff = skb_network_offset(skb);
2517 switch (skb->protocol) {
2518 case __constant_htons(ETH_P_IP):
2519 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2522 ip = (const struct iphdr *) (skb->data + nhoff);
2523 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2526 ip_proto = ip->protocol;
2527 addr1 = (__force u32) ip->saddr;
2528 addr2 = (__force u32) ip->daddr;
2531 case __constant_htons(ETH_P_IPV6):
2532 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2535 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2536 ip_proto = ip6->nexthdr;
2537 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2538 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2546 poff = proto_ports_offset(ip_proto);
2548 nhoff += ihl * 4 + poff;
2549 if (pskb_may_pull(skb, nhoff + 4)) {
2550 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2551 if (ports.v16[1] < ports.v16[0])
2552 swap(ports.v16[0], ports.v16[1]);
2556 /* get a consistent hash (same value on both flow directions) */
2560 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2567 EXPORT_SYMBOL(__skb_get_rxhash);
2571 /* One global table that all flow-based protocols share. */
2572 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2573 EXPORT_SYMBOL(rps_sock_flow_table);
2575 static struct rps_dev_flow *
2576 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2577 struct rps_dev_flow *rflow, u16 next_cpu)
2581 tcpu = rflow->cpu = next_cpu;
2582 if (tcpu != RPS_NO_CPU) {
2583 #ifdef CONFIG_RFS_ACCEL
2584 struct netdev_rx_queue *rxqueue;
2585 struct rps_dev_flow_table *flow_table;
2586 struct rps_dev_flow *old_rflow;
2591 /* Should we steer this flow to a different hardware queue? */
2592 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2593 !(dev->features & NETIF_F_NTUPLE))
2595 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2596 if (rxq_index == skb_get_rx_queue(skb))
2599 rxqueue = dev->_rx + rxq_index;
2600 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2603 flow_id = skb->rxhash & flow_table->mask;
2604 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2605 rxq_index, flow_id);
2609 rflow = &flow_table->flows[flow_id];
2610 rflow->cpu = next_cpu;
2612 if (old_rflow->filter == rflow->filter)
2613 old_rflow->filter = RPS_NO_FILTER;
2617 per_cpu(softnet_data, tcpu).input_queue_head;
2624 * get_rps_cpu is called from netif_receive_skb and returns the target
2625 * CPU from the RPS map of the receiving queue for a given skb.
2626 * rcu_read_lock must be held on entry.
2628 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2629 struct rps_dev_flow **rflowp)
2631 struct netdev_rx_queue *rxqueue;
2632 struct rps_map *map;
2633 struct rps_dev_flow_table *flow_table;
2634 struct rps_sock_flow_table *sock_flow_table;
2638 if (skb_rx_queue_recorded(skb)) {
2639 u16 index = skb_get_rx_queue(skb);
2640 if (unlikely(index >= dev->real_num_rx_queues)) {
2641 WARN_ONCE(dev->real_num_rx_queues > 1,
2642 "%s received packet on queue %u, but number "
2643 "of RX queues is %u\n",
2644 dev->name, index, dev->real_num_rx_queues);
2647 rxqueue = dev->_rx + index;
2651 map = rcu_dereference(rxqueue->rps_map);
2653 if (map->len == 1 &&
2654 !rcu_dereference_raw(rxqueue->rps_flow_table)) {
2655 tcpu = map->cpus[0];
2656 if (cpu_online(tcpu))
2660 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2664 skb_reset_network_header(skb);
2665 if (!skb_get_rxhash(skb))
2668 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2669 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2670 if (flow_table && sock_flow_table) {
2672 struct rps_dev_flow *rflow;
2674 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2677 next_cpu = sock_flow_table->ents[skb->rxhash &
2678 sock_flow_table->mask];
2681 * If the desired CPU (where last recvmsg was done) is
2682 * different from current CPU (one in the rx-queue flow
2683 * table entry), switch if one of the following holds:
2684 * - Current CPU is unset (equal to RPS_NO_CPU).
2685 * - Current CPU is offline.
2686 * - The current CPU's queue tail has advanced beyond the
2687 * last packet that was enqueued using this table entry.
2688 * This guarantees that all previous packets for the flow
2689 * have been dequeued, thus preserving in order delivery.
2691 if (unlikely(tcpu != next_cpu) &&
2692 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2693 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2694 rflow->last_qtail)) >= 0))
2695 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2697 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2705 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2707 if (cpu_online(tcpu)) {
2717 #ifdef CONFIG_RFS_ACCEL
2720 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2721 * @dev: Device on which the filter was set
2722 * @rxq_index: RX queue index
2723 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2724 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2726 * Drivers that implement ndo_rx_flow_steer() should periodically call
2727 * this function for each installed filter and remove the filters for
2728 * which it returns %true.
2730 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2731 u32 flow_id, u16 filter_id)
2733 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2734 struct rps_dev_flow_table *flow_table;
2735 struct rps_dev_flow *rflow;
2740 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2741 if (flow_table && flow_id <= flow_table->mask) {
2742 rflow = &flow_table->flows[flow_id];
2743 cpu = ACCESS_ONCE(rflow->cpu);
2744 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2745 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2746 rflow->last_qtail) <
2747 (int)(10 * flow_table->mask)))
2753 EXPORT_SYMBOL(rps_may_expire_flow);
2755 #endif /* CONFIG_RFS_ACCEL */
2757 /* Called from hardirq (IPI) context */
2758 static void rps_trigger_softirq(void *data)
2760 struct softnet_data *sd = data;
2762 ____napi_schedule(sd, &sd->backlog);
2766 #endif /* CONFIG_RPS */
2769 * Check if this softnet_data structure is another cpu one
2770 * If yes, queue it to our IPI list and return 1
2773 static int rps_ipi_queued(struct softnet_data *sd)
2776 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2779 sd->rps_ipi_next = mysd->rps_ipi_list;
2780 mysd->rps_ipi_list = sd;
2782 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2785 #endif /* CONFIG_RPS */
2790 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2791 * queue (may be a remote CPU queue).
2793 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2794 unsigned int *qtail)
2796 struct softnet_data *sd;
2797 unsigned long flags;
2799 sd = &per_cpu(softnet_data, cpu);
2801 local_irq_save(flags);
2804 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2805 if (skb_queue_len(&sd->input_pkt_queue)) {
2807 __skb_queue_tail(&sd->input_pkt_queue, skb);
2808 input_queue_tail_incr_save(sd, qtail);
2810 local_irq_restore(flags);
2811 return NET_RX_SUCCESS;
2814 /* Schedule NAPI for backlog device
2815 * We can use non atomic operation since we own the queue lock
2817 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2818 if (!rps_ipi_queued(sd))
2819 ____napi_schedule(sd, &sd->backlog);
2827 local_irq_restore(flags);
2829 atomic_long_inc(&skb->dev->rx_dropped);
2835 * netif_rx - post buffer to the network code
2836 * @skb: buffer to post
2838 * This function receives a packet from a device driver and queues it for
2839 * the upper (protocol) levels to process. It always succeeds. The buffer
2840 * may be dropped during processing for congestion control or by the
2844 * NET_RX_SUCCESS (no congestion)
2845 * NET_RX_DROP (packet was dropped)
2849 int netif_rx(struct sk_buff *skb)
2853 /* if netpoll wants it, pretend we never saw it */
2854 if (netpoll_rx(skb))
2857 if (netdev_tstamp_prequeue)
2858 net_timestamp_check(skb);
2860 trace_netif_rx(skb);
2863 struct rps_dev_flow voidflow, *rflow = &voidflow;
2869 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2871 cpu = smp_processor_id();
2873 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2881 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2887 EXPORT_SYMBOL(netif_rx);
2889 int netif_rx_ni(struct sk_buff *skb)
2894 err = netif_rx(skb);
2895 if (local_softirq_pending())
2901 EXPORT_SYMBOL(netif_rx_ni);
2903 static void net_tx_action(struct softirq_action *h)
2905 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2907 if (sd->completion_queue) {
2908 struct sk_buff *clist;
2910 local_irq_disable();
2911 clist = sd->completion_queue;
2912 sd->completion_queue = NULL;
2916 struct sk_buff *skb = clist;
2917 clist = clist->next;
2919 WARN_ON(atomic_read(&skb->users));
2920 trace_kfree_skb(skb, net_tx_action);
2925 if (sd->output_queue) {
2928 local_irq_disable();
2929 head = sd->output_queue;
2930 sd->output_queue = NULL;
2931 sd->output_queue_tailp = &sd->output_queue;
2935 struct Qdisc *q = head;
2936 spinlock_t *root_lock;
2938 head = head->next_sched;
2940 root_lock = qdisc_lock(q);
2941 if (spin_trylock(root_lock)) {
2942 smp_mb__before_clear_bit();
2943 clear_bit(__QDISC_STATE_SCHED,
2946 spin_unlock(root_lock);
2948 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2950 __netif_reschedule(q);
2952 smp_mb__before_clear_bit();
2953 clear_bit(__QDISC_STATE_SCHED,
2961 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2962 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2963 /* This hook is defined here for ATM LANE */
2964 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2965 unsigned char *addr) __read_mostly;
2966 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2969 #ifdef CONFIG_NET_CLS_ACT
2970 /* TODO: Maybe we should just force sch_ingress to be compiled in
2971 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2972 * a compare and 2 stores extra right now if we dont have it on
2973 * but have CONFIG_NET_CLS_ACT
2974 * NOTE: This doesn't stop any functionality; if you dont have
2975 * the ingress scheduler, you just can't add policies on ingress.
2978 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2980 struct net_device *dev = skb->dev;
2981 u32 ttl = G_TC_RTTL(skb->tc_verd);
2982 int result = TC_ACT_OK;
2985 if (unlikely(MAX_RED_LOOP < ttl++)) {
2986 if (net_ratelimit())
2987 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2988 skb->skb_iif, dev->ifindex);
2992 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2993 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2996 if (q != &noop_qdisc) {
2997 spin_lock(qdisc_lock(q));
2998 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2999 result = qdisc_enqueue_root(skb, q);
3000 spin_unlock(qdisc_lock(q));
3006 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3007 struct packet_type **pt_prev,
3008 int *ret, struct net_device *orig_dev)
3010 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3012 if (!rxq || rxq->qdisc == &noop_qdisc)
3016 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3020 switch (ing_filter(skb, rxq)) {
3034 * netdev_rx_handler_register - register receive handler
3035 * @dev: device to register a handler for
3036 * @rx_handler: receive handler to register
3037 * @rx_handler_data: data pointer that is used by rx handler
3039 * Register a receive hander for a device. This handler will then be
3040 * called from __netif_receive_skb. A negative errno code is returned
3043 * The caller must hold the rtnl_mutex.
3045 * For a general description of rx_handler, see enum rx_handler_result.
3047 int netdev_rx_handler_register(struct net_device *dev,
3048 rx_handler_func_t *rx_handler,
3049 void *rx_handler_data)
3053 if (dev->rx_handler)
3056 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3057 rcu_assign_pointer(dev->rx_handler, rx_handler);
3061 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3064 * netdev_rx_handler_unregister - unregister receive handler
3065 * @dev: device to unregister a handler from
3067 * Unregister a receive hander from a device.
3069 * The caller must hold the rtnl_mutex.
3071 void netdev_rx_handler_unregister(struct net_device *dev)
3075 rcu_assign_pointer(dev->rx_handler, NULL);
3076 rcu_assign_pointer(dev->rx_handler_data, NULL);
3078 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3080 static void vlan_on_bond_hook(struct sk_buff *skb)
3083 * Make sure ARP frames received on VLAN interfaces stacked on
3084 * bonding interfaces still make their way to any base bonding
3085 * device that may have registered for a specific ptype.
3087 if (skb->dev->priv_flags & IFF_802_1Q_VLAN &&
3088 vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING &&
3089 skb->protocol == htons(ETH_P_ARP)) {
3090 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3094 skb2->dev = vlan_dev_real_dev(skb->dev);
3099 static int __netif_receive_skb(struct sk_buff *skb)
3101 struct packet_type *ptype, *pt_prev;
3102 rx_handler_func_t *rx_handler;
3103 struct net_device *orig_dev;
3104 struct net_device *null_or_dev;
3105 bool deliver_exact = false;
3106 int ret = NET_RX_DROP;
3109 if (!netdev_tstamp_prequeue)
3110 net_timestamp_check(skb);
3112 trace_netif_receive_skb(skb);
3114 /* if we've gotten here through NAPI, check netpoll */
3115 if (netpoll_receive_skb(skb))
3119 skb->skb_iif = skb->dev->ifindex;
3120 orig_dev = skb->dev;
3122 skb_reset_network_header(skb);
3123 skb_reset_transport_header(skb);
3124 skb->mac_len = skb->network_header - skb->mac_header;
3132 __this_cpu_inc(softnet_data.processed);
3134 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3135 skb = vlan_untag(skb);
3140 #ifdef CONFIG_NET_CLS_ACT
3141 if (skb->tc_verd & TC_NCLS) {
3142 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3147 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3148 if (!ptype->dev || ptype->dev == skb->dev) {
3150 ret = deliver_skb(skb, pt_prev, orig_dev);
3155 #ifdef CONFIG_NET_CLS_ACT
3156 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3162 rx_handler = rcu_dereference(skb->dev->rx_handler);
3165 ret = deliver_skb(skb, pt_prev, orig_dev);
3168 switch (rx_handler(&skb)) {
3169 case RX_HANDLER_CONSUMED:
3171 case RX_HANDLER_ANOTHER:
3173 case RX_HANDLER_EXACT:
3174 deliver_exact = true;
3175 case RX_HANDLER_PASS:
3182 if (vlan_tx_tag_present(skb)) {
3184 ret = deliver_skb(skb, pt_prev, orig_dev);
3187 if (vlan_do_receive(&skb)) {
3188 ret = __netif_receive_skb(skb);
3190 } else if (unlikely(!skb))
3194 vlan_on_bond_hook(skb);
3196 /* deliver only exact match when indicated */
3197 null_or_dev = deliver_exact ? skb->dev : NULL;
3199 type = skb->protocol;
3200 list_for_each_entry_rcu(ptype,
3201 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3202 if (ptype->type == type &&
3203 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3204 ptype->dev == orig_dev)) {
3206 ret = deliver_skb(skb, pt_prev, orig_dev);
3212 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3214 atomic_long_inc(&skb->dev->rx_dropped);
3216 /* Jamal, now you will not able to escape explaining
3217 * me how you were going to use this. :-)
3228 * netif_receive_skb - process receive buffer from network
3229 * @skb: buffer to process
3231 * netif_receive_skb() is the main receive data processing function.
3232 * It always succeeds. The buffer may be dropped during processing
3233 * for congestion control or by the protocol layers.
3235 * This function may only be called from softirq context and interrupts
3236 * should be enabled.
3238 * Return values (usually ignored):
3239 * NET_RX_SUCCESS: no congestion
3240 * NET_RX_DROP: packet was dropped
3242 int netif_receive_skb(struct sk_buff *skb)
3244 if (netdev_tstamp_prequeue)
3245 net_timestamp_check(skb);
3247 if (skb_defer_rx_timestamp(skb))
3248 return NET_RX_SUCCESS;
3252 struct rps_dev_flow voidflow, *rflow = &voidflow;
3257 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3260 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3264 ret = __netif_receive_skb(skb);
3270 return __netif_receive_skb(skb);
3273 EXPORT_SYMBOL(netif_receive_skb);
3275 /* Network device is going away, flush any packets still pending
3276 * Called with irqs disabled.
3278 static void flush_backlog(void *arg)
3280 struct net_device *dev = arg;
3281 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3282 struct sk_buff *skb, *tmp;
3285 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3286 if (skb->dev == dev) {
3287 __skb_unlink(skb, &sd->input_pkt_queue);
3289 input_queue_head_incr(sd);
3294 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3295 if (skb->dev == dev) {
3296 __skb_unlink(skb, &sd->process_queue);
3298 input_queue_head_incr(sd);
3303 static int napi_gro_complete(struct sk_buff *skb)
3305 struct packet_type *ptype;
3306 __be16 type = skb->protocol;
3307 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3310 if (NAPI_GRO_CB(skb)->count == 1) {
3311 skb_shinfo(skb)->gso_size = 0;
3316 list_for_each_entry_rcu(ptype, head, list) {
3317 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3320 err = ptype->gro_complete(skb);
3326 WARN_ON(&ptype->list == head);
3328 return NET_RX_SUCCESS;
3332 return netif_receive_skb(skb);
3335 inline void napi_gro_flush(struct napi_struct *napi)
3337 struct sk_buff *skb, *next;
3339 for (skb = napi->gro_list; skb; skb = next) {
3342 napi_gro_complete(skb);
3345 napi->gro_count = 0;
3346 napi->gro_list = NULL;
3348 EXPORT_SYMBOL(napi_gro_flush);
3350 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3352 struct sk_buff **pp = NULL;
3353 struct packet_type *ptype;
3354 __be16 type = skb->protocol;
3355 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3358 enum gro_result ret;
3360 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3363 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3367 list_for_each_entry_rcu(ptype, head, list) {
3368 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3371 skb_set_network_header(skb, skb_gro_offset(skb));
3372 mac_len = skb->network_header - skb->mac_header;
3373 skb->mac_len = mac_len;
3374 NAPI_GRO_CB(skb)->same_flow = 0;
3375 NAPI_GRO_CB(skb)->flush = 0;
3376 NAPI_GRO_CB(skb)->free = 0;
3378 pp = ptype->gro_receive(&napi->gro_list, skb);
3383 if (&ptype->list == head)
3386 same_flow = NAPI_GRO_CB(skb)->same_flow;
3387 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3390 struct sk_buff *nskb = *pp;
3394 napi_gro_complete(nskb);
3401 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3405 NAPI_GRO_CB(skb)->count = 1;
3406 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3407 skb->next = napi->gro_list;
3408 napi->gro_list = skb;
3412 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3413 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3415 BUG_ON(skb->end - skb->tail < grow);
3417 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3420 skb->data_len -= grow;
3422 skb_shinfo(skb)->frags[0].page_offset += grow;
3423 skb_shinfo(skb)->frags[0].size -= grow;
3425 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3426 put_page(skb_shinfo(skb)->frags[0].page);
3427 memmove(skb_shinfo(skb)->frags,
3428 skb_shinfo(skb)->frags + 1,
3429 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3440 EXPORT_SYMBOL(dev_gro_receive);
3442 static inline gro_result_t
3443 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3447 for (p = napi->gro_list; p; p = p->next) {
3448 unsigned long diffs;
3450 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3451 diffs |= p->vlan_tci ^ skb->vlan_tci;
3452 diffs |= compare_ether_header(skb_mac_header(p),
3453 skb_gro_mac_header(skb));
3454 NAPI_GRO_CB(p)->same_flow = !diffs;
3455 NAPI_GRO_CB(p)->flush = 0;
3458 return dev_gro_receive(napi, skb);
3461 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3465 if (netif_receive_skb(skb))
3470 case GRO_MERGED_FREE:
3481 EXPORT_SYMBOL(napi_skb_finish);
3483 void skb_gro_reset_offset(struct sk_buff *skb)
3485 NAPI_GRO_CB(skb)->data_offset = 0;
3486 NAPI_GRO_CB(skb)->frag0 = NULL;
3487 NAPI_GRO_CB(skb)->frag0_len = 0;
3489 if (skb->mac_header == skb->tail &&
3490 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3491 NAPI_GRO_CB(skb)->frag0 =
3492 page_address(skb_shinfo(skb)->frags[0].page) +
3493 skb_shinfo(skb)->frags[0].page_offset;
3494 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3497 EXPORT_SYMBOL(skb_gro_reset_offset);
3499 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3501 skb_gro_reset_offset(skb);
3503 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3505 EXPORT_SYMBOL(napi_gro_receive);
3507 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3509 __skb_pull(skb, skb_headlen(skb));
3510 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3512 skb->dev = napi->dev;
3518 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3520 struct sk_buff *skb = napi->skb;
3523 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3529 EXPORT_SYMBOL(napi_get_frags);
3531 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3537 skb->protocol = eth_type_trans(skb, skb->dev);
3539 if (ret == GRO_HELD)
3540 skb_gro_pull(skb, -ETH_HLEN);
3541 else if (netif_receive_skb(skb))
3546 case GRO_MERGED_FREE:
3547 napi_reuse_skb(napi, skb);
3556 EXPORT_SYMBOL(napi_frags_finish);
3558 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3560 struct sk_buff *skb = napi->skb;
3567 skb_reset_mac_header(skb);
3568 skb_gro_reset_offset(skb);
3570 off = skb_gro_offset(skb);
3571 hlen = off + sizeof(*eth);
3572 eth = skb_gro_header_fast(skb, off);
3573 if (skb_gro_header_hard(skb, hlen)) {
3574 eth = skb_gro_header_slow(skb, hlen, off);
3575 if (unlikely(!eth)) {
3576 napi_reuse_skb(napi, skb);
3582 skb_gro_pull(skb, sizeof(*eth));
3585 * This works because the only protocols we care about don't require
3586 * special handling. We'll fix it up properly at the end.
3588 skb->protocol = eth->h_proto;
3593 EXPORT_SYMBOL(napi_frags_skb);
3595 gro_result_t napi_gro_frags(struct napi_struct *napi)
3597 struct sk_buff *skb = napi_frags_skb(napi);
3602 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3604 EXPORT_SYMBOL(napi_gro_frags);
3607 * net_rps_action sends any pending IPI's for rps.
3608 * Note: called with local irq disabled, but exits with local irq enabled.
3610 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3613 struct softnet_data *remsd = sd->rps_ipi_list;
3616 sd->rps_ipi_list = NULL;
3620 /* Send pending IPI's to kick RPS processing on remote cpus. */
3622 struct softnet_data *next = remsd->rps_ipi_next;
3624 if (cpu_online(remsd->cpu))
3625 __smp_call_function_single(remsd->cpu,
3634 static int process_backlog(struct napi_struct *napi, int quota)
3637 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3640 /* Check if we have pending ipi, its better to send them now,
3641 * not waiting net_rx_action() end.
3643 if (sd->rps_ipi_list) {
3644 local_irq_disable();
3645 net_rps_action_and_irq_enable(sd);
3648 napi->weight = weight_p;
3649 local_irq_disable();
3650 while (work < quota) {
3651 struct sk_buff *skb;
3654 while ((skb = __skb_dequeue(&sd->process_queue))) {
3656 __netif_receive_skb(skb);
3657 local_irq_disable();
3658 input_queue_head_incr(sd);
3659 if (++work >= quota) {
3666 qlen = skb_queue_len(&sd->input_pkt_queue);
3668 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3669 &sd->process_queue);
3671 if (qlen < quota - work) {
3673 * Inline a custom version of __napi_complete().
3674 * only current cpu owns and manipulates this napi,
3675 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3676 * we can use a plain write instead of clear_bit(),
3677 * and we dont need an smp_mb() memory barrier.
3679 list_del(&napi->poll_list);
3682 quota = work + qlen;
3692 * __napi_schedule - schedule for receive
3693 * @n: entry to schedule
3695 * The entry's receive function will be scheduled to run
3697 void __napi_schedule(struct napi_struct *n)
3699 unsigned long flags;
3701 local_irq_save(flags);
3702 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3703 local_irq_restore(flags);
3705 EXPORT_SYMBOL(__napi_schedule);
3707 void __napi_complete(struct napi_struct *n)
3709 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3710 BUG_ON(n->gro_list);
3712 list_del(&n->poll_list);
3713 smp_mb__before_clear_bit();
3714 clear_bit(NAPI_STATE_SCHED, &n->state);
3716 EXPORT_SYMBOL(__napi_complete);
3718 void napi_complete(struct napi_struct *n)
3720 unsigned long flags;
3723 * don't let napi dequeue from the cpu poll list
3724 * just in case its running on a different cpu
3726 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3730 local_irq_save(flags);
3732 local_irq_restore(flags);
3734 EXPORT_SYMBOL(napi_complete);
3736 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3737 int (*poll)(struct napi_struct *, int), int weight)
3739 INIT_LIST_HEAD(&napi->poll_list);
3740 napi->gro_count = 0;
3741 napi->gro_list = NULL;
3744 napi->weight = weight;
3745 list_add(&napi->dev_list, &dev->napi_list);
3747 #ifdef CONFIG_NETPOLL
3748 spin_lock_init(&napi->poll_lock);
3749 napi->poll_owner = -1;
3751 set_bit(NAPI_STATE_SCHED, &napi->state);
3753 EXPORT_SYMBOL(netif_napi_add);
3755 void netif_napi_del(struct napi_struct *napi)
3757 struct sk_buff *skb, *next;
3759 list_del_init(&napi->dev_list);
3760 napi_free_frags(napi);
3762 for (skb = napi->gro_list; skb; skb = next) {
3768 napi->gro_list = NULL;
3769 napi->gro_count = 0;
3771 EXPORT_SYMBOL(netif_napi_del);
3773 static void net_rx_action(struct softirq_action *h)
3775 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3776 unsigned long time_limit = jiffies + 2;
3777 int budget = netdev_budget;
3780 local_irq_disable();
3782 while (!list_empty(&sd->poll_list)) {
3783 struct napi_struct *n;
3786 /* If softirq window is exhuasted then punt.
3787 * Allow this to run for 2 jiffies since which will allow
3788 * an average latency of 1.5/HZ.
3790 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3795 /* Even though interrupts have been re-enabled, this
3796 * access is safe because interrupts can only add new
3797 * entries to the tail of this list, and only ->poll()
3798 * calls can remove this head entry from the list.
3800 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3802 have = netpoll_poll_lock(n);
3806 /* This NAPI_STATE_SCHED test is for avoiding a race
3807 * with netpoll's poll_napi(). Only the entity which
3808 * obtains the lock and sees NAPI_STATE_SCHED set will
3809 * actually make the ->poll() call. Therefore we avoid
3810 * accidentally calling ->poll() when NAPI is not scheduled.
3813 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3814 work = n->poll(n, weight);
3818 WARN_ON_ONCE(work > weight);
3822 local_irq_disable();
3824 /* Drivers must not modify the NAPI state if they
3825 * consume the entire weight. In such cases this code
3826 * still "owns" the NAPI instance and therefore can
3827 * move the instance around on the list at-will.
3829 if (unlikely(work == weight)) {
3830 if (unlikely(napi_disable_pending(n))) {
3833 local_irq_disable();
3835 list_move_tail(&n->poll_list, &sd->poll_list);
3838 netpoll_poll_unlock(have);
3841 net_rps_action_and_irq_enable(sd);
3843 #ifdef CONFIG_NET_DMA
3845 * There may not be any more sk_buffs coming right now, so push
3846 * any pending DMA copies to hardware
3848 dma_issue_pending_all();
3855 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3859 static gifconf_func_t *gifconf_list[NPROTO];
3862 * register_gifconf - register a SIOCGIF handler
3863 * @family: Address family
3864 * @gifconf: Function handler
3866 * Register protocol dependent address dumping routines. The handler
3867 * that is passed must not be freed or reused until it has been replaced
3868 * by another handler.
3870 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3872 if (family >= NPROTO)
3874 gifconf_list[family] = gifconf;
3877 EXPORT_SYMBOL(register_gifconf);
3881 * Map an interface index to its name (SIOCGIFNAME)
3885 * We need this ioctl for efficient implementation of the
3886 * if_indextoname() function required by the IPv6 API. Without
3887 * it, we would have to search all the interfaces to find a
3891 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3893 struct net_device *dev;
3897 * Fetch the caller's info block.
3900 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3904 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3910 strcpy(ifr.ifr_name, dev->name);
3913 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3919 * Perform a SIOCGIFCONF call. This structure will change
3920 * size eventually, and there is nothing I can do about it.
3921 * Thus we will need a 'compatibility mode'.
3924 static int dev_ifconf(struct net *net, char __user *arg)
3927 struct net_device *dev;
3934 * Fetch the caller's info block.
3937 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3944 * Loop over the interfaces, and write an info block for each.
3948 for_each_netdev(net, dev) {
3949 for (i = 0; i < NPROTO; i++) {
3950 if (gifconf_list[i]) {
3953 done = gifconf_list[i](dev, NULL, 0);
3955 done = gifconf_list[i](dev, pos + total,
3965 * All done. Write the updated control block back to the caller.
3967 ifc.ifc_len = total;
3970 * Both BSD and Solaris return 0 here, so we do too.
3972 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3975 #ifdef CONFIG_PROC_FS
3977 * This is invoked by the /proc filesystem handler to display a device
3980 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3983 struct net *net = seq_file_net(seq);
3985 struct net_device *dev;
3989 return SEQ_START_TOKEN;
3992 for_each_netdev_rcu(net, dev)
3999 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4001 struct net_device *dev = v;
4003 if (v == SEQ_START_TOKEN)
4004 dev = first_net_device_rcu(seq_file_net(seq));
4006 dev = next_net_device_rcu(dev);
4012 void dev_seq_stop(struct seq_file *seq, void *v)
4018 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4020 struct rtnl_link_stats64 temp;
4021 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4023 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4024 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4025 dev->name, stats->rx_bytes, stats->rx_packets,
4027 stats->rx_dropped + stats->rx_missed_errors,
4028 stats->rx_fifo_errors,
4029 stats->rx_length_errors + stats->rx_over_errors +
4030 stats->rx_crc_errors + stats->rx_frame_errors,
4031 stats->rx_compressed, stats->multicast,
4032 stats->tx_bytes, stats->tx_packets,
4033 stats->tx_errors, stats->tx_dropped,
4034 stats->tx_fifo_errors, stats->collisions,
4035 stats->tx_carrier_errors +
4036 stats->tx_aborted_errors +
4037 stats->tx_window_errors +
4038 stats->tx_heartbeat_errors,
4039 stats->tx_compressed);
4043 * Called from the PROCfs module. This now uses the new arbitrary sized
4044 * /proc/net interface to create /proc/net/dev
4046 static int dev_seq_show(struct seq_file *seq, void *v)
4048 if (v == SEQ_START_TOKEN)
4049 seq_puts(seq, "Inter-| Receive "
4051 " face |bytes packets errs drop fifo frame "
4052 "compressed multicast|bytes packets errs "
4053 "drop fifo colls carrier compressed\n");
4055 dev_seq_printf_stats(seq, v);
4059 static struct softnet_data *softnet_get_online(loff_t *pos)
4061 struct softnet_data *sd = NULL;
4063 while (*pos < nr_cpu_ids)
4064 if (cpu_online(*pos)) {
4065 sd = &per_cpu(softnet_data, *pos);
4072 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4074 return softnet_get_online(pos);
4077 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4080 return softnet_get_online(pos);
4083 static void softnet_seq_stop(struct seq_file *seq, void *v)
4087 static int softnet_seq_show(struct seq_file *seq, void *v)
4089 struct softnet_data *sd = v;
4091 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4092 sd->processed, sd->dropped, sd->time_squeeze, 0,
4093 0, 0, 0, 0, /* was fastroute */
4094 sd->cpu_collision, sd->received_rps);
4098 static const struct seq_operations dev_seq_ops = {
4099 .start = dev_seq_start,
4100 .next = dev_seq_next,
4101 .stop = dev_seq_stop,
4102 .show = dev_seq_show,
4105 static int dev_seq_open(struct inode *inode, struct file *file)
4107 return seq_open_net(inode, file, &dev_seq_ops,
4108 sizeof(struct seq_net_private));
4111 static const struct file_operations dev_seq_fops = {
4112 .owner = THIS_MODULE,
4113 .open = dev_seq_open,
4115 .llseek = seq_lseek,
4116 .release = seq_release_net,
4119 static const struct seq_operations softnet_seq_ops = {
4120 .start = softnet_seq_start,
4121 .next = softnet_seq_next,
4122 .stop = softnet_seq_stop,
4123 .show = softnet_seq_show,
4126 static int softnet_seq_open(struct inode *inode, struct file *file)
4128 return seq_open(file, &softnet_seq_ops);
4131 static const struct file_operations softnet_seq_fops = {
4132 .owner = THIS_MODULE,
4133 .open = softnet_seq_open,
4135 .llseek = seq_lseek,
4136 .release = seq_release,
4139 static void *ptype_get_idx(loff_t pos)
4141 struct packet_type *pt = NULL;
4145 list_for_each_entry_rcu(pt, &ptype_all, list) {
4151 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4152 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4161 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4165 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4168 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4170 struct packet_type *pt;
4171 struct list_head *nxt;
4175 if (v == SEQ_START_TOKEN)
4176 return ptype_get_idx(0);
4179 nxt = pt->list.next;
4180 if (pt->type == htons(ETH_P_ALL)) {
4181 if (nxt != &ptype_all)
4184 nxt = ptype_base[0].next;
4186 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4188 while (nxt == &ptype_base[hash]) {
4189 if (++hash >= PTYPE_HASH_SIZE)
4191 nxt = ptype_base[hash].next;
4194 return list_entry(nxt, struct packet_type, list);
4197 static void ptype_seq_stop(struct seq_file *seq, void *v)
4203 static int ptype_seq_show(struct seq_file *seq, void *v)
4205 struct packet_type *pt = v;
4207 if (v == SEQ_START_TOKEN)
4208 seq_puts(seq, "Type Device Function\n");
4209 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4210 if (pt->type == htons(ETH_P_ALL))
4211 seq_puts(seq, "ALL ");
4213 seq_printf(seq, "%04x", ntohs(pt->type));
4215 seq_printf(seq, " %-8s %pF\n",
4216 pt->dev ? pt->dev->name : "", pt->func);
4222 static const struct seq_operations ptype_seq_ops = {
4223 .start = ptype_seq_start,
4224 .next = ptype_seq_next,
4225 .stop = ptype_seq_stop,
4226 .show = ptype_seq_show,
4229 static int ptype_seq_open(struct inode *inode, struct file *file)
4231 return seq_open_net(inode, file, &ptype_seq_ops,
4232 sizeof(struct seq_net_private));
4235 static const struct file_operations ptype_seq_fops = {
4236 .owner = THIS_MODULE,
4237 .open = ptype_seq_open,
4239 .llseek = seq_lseek,
4240 .release = seq_release_net,
4244 static int __net_init dev_proc_net_init(struct net *net)
4248 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4250 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4252 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4255 if (wext_proc_init(net))
4261 proc_net_remove(net, "ptype");
4263 proc_net_remove(net, "softnet_stat");
4265 proc_net_remove(net, "dev");
4269 static void __net_exit dev_proc_net_exit(struct net *net)
4271 wext_proc_exit(net);
4273 proc_net_remove(net, "ptype");
4274 proc_net_remove(net, "softnet_stat");
4275 proc_net_remove(net, "dev");
4278 static struct pernet_operations __net_initdata dev_proc_ops = {
4279 .init = dev_proc_net_init,
4280 .exit = dev_proc_net_exit,
4283 static int __init dev_proc_init(void)
4285 return register_pernet_subsys(&dev_proc_ops);
4288 #define dev_proc_init() 0
4289 #endif /* CONFIG_PROC_FS */
4293 * netdev_set_master - set up master pointer
4294 * @slave: slave device
4295 * @master: new master device
4297 * Changes the master device of the slave. Pass %NULL to break the
4298 * bonding. The caller must hold the RTNL semaphore. On a failure
4299 * a negative errno code is returned. On success the reference counts
4300 * are adjusted and the function returns zero.
4302 int netdev_set_master(struct net_device *slave, struct net_device *master)
4304 struct net_device *old = slave->master;
4314 slave->master = master;
4322 EXPORT_SYMBOL(netdev_set_master);
4325 * netdev_set_bond_master - set up bonding master/slave pair
4326 * @slave: slave device
4327 * @master: new master device
4329 * Changes the master device of the slave. Pass %NULL to break the
4330 * bonding. The caller must hold the RTNL semaphore. On a failure
4331 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4332 * to the routing socket and the function returns zero.
4334 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4340 err = netdev_set_master(slave, master);
4344 slave->flags |= IFF_SLAVE;
4346 slave->flags &= ~IFF_SLAVE;
4348 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4351 EXPORT_SYMBOL(netdev_set_bond_master);
4353 static void dev_change_rx_flags(struct net_device *dev, int flags)
4355 const struct net_device_ops *ops = dev->netdev_ops;
4357 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4358 ops->ndo_change_rx_flags(dev, flags);
4361 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4363 unsigned short old_flags = dev->flags;
4369 dev->flags |= IFF_PROMISC;
4370 dev->promiscuity += inc;
4371 if (dev->promiscuity == 0) {
4374 * If inc causes overflow, untouch promisc and return error.
4377 dev->flags &= ~IFF_PROMISC;
4379 dev->promiscuity -= inc;
4380 printk(KERN_WARNING "%s: promiscuity touches roof, "
4381 "set promiscuity failed, promiscuity feature "
4382 "of device might be broken.\n", dev->name);
4386 if (dev->flags != old_flags) {
4387 printk(KERN_INFO "device %s %s promiscuous mode\n",
4388 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4390 if (audit_enabled) {
4391 current_uid_gid(&uid, &gid);
4392 audit_log(current->audit_context, GFP_ATOMIC,
4393 AUDIT_ANOM_PROMISCUOUS,
4394 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4395 dev->name, (dev->flags & IFF_PROMISC),
4396 (old_flags & IFF_PROMISC),
4397 audit_get_loginuid(current),
4399 audit_get_sessionid(current));
4402 dev_change_rx_flags(dev, IFF_PROMISC);
4408 * dev_set_promiscuity - update promiscuity count on a device
4412 * Add or remove promiscuity from a device. While the count in the device
4413 * remains above zero the interface remains promiscuous. Once it hits zero
4414 * the device reverts back to normal filtering operation. A negative inc
4415 * value is used to drop promiscuity on the device.
4416 * Return 0 if successful or a negative errno code on error.
4418 int dev_set_promiscuity(struct net_device *dev, int inc)
4420 unsigned short old_flags = dev->flags;
4423 err = __dev_set_promiscuity(dev, inc);
4426 if (dev->flags != old_flags)
4427 dev_set_rx_mode(dev);
4430 EXPORT_SYMBOL(dev_set_promiscuity);
4433 * dev_set_allmulti - update allmulti count on a device
4437 * Add or remove reception of all multicast frames to a device. While the
4438 * count in the device remains above zero the interface remains listening
4439 * to all interfaces. Once it hits zero the device reverts back to normal
4440 * filtering operation. A negative @inc value is used to drop the counter
4441 * when releasing a resource needing all multicasts.
4442 * Return 0 if successful or a negative errno code on error.
4445 int dev_set_allmulti(struct net_device *dev, int inc)
4447 unsigned short old_flags = dev->flags;
4451 dev->flags |= IFF_ALLMULTI;
4452 dev->allmulti += inc;
4453 if (dev->allmulti == 0) {
4456 * If inc causes overflow, untouch allmulti and return error.
4459 dev->flags &= ~IFF_ALLMULTI;
4461 dev->allmulti -= inc;
4462 printk(KERN_WARNING "%s: allmulti touches roof, "
4463 "set allmulti failed, allmulti feature of "
4464 "device might be broken.\n", dev->name);
4468 if (dev->flags ^ old_flags) {
4469 dev_change_rx_flags(dev, IFF_ALLMULTI);
4470 dev_set_rx_mode(dev);
4474 EXPORT_SYMBOL(dev_set_allmulti);
4477 * Upload unicast and multicast address lists to device and
4478 * configure RX filtering. When the device doesn't support unicast
4479 * filtering it is put in promiscuous mode while unicast addresses
4482 void __dev_set_rx_mode(struct net_device *dev)
4484 const struct net_device_ops *ops = dev->netdev_ops;
4486 /* dev_open will call this function so the list will stay sane. */
4487 if (!(dev->flags&IFF_UP))
4490 if (!netif_device_present(dev))
4493 if (ops->ndo_set_rx_mode)
4494 ops->ndo_set_rx_mode(dev);
4496 /* Unicast addresses changes may only happen under the rtnl,
4497 * therefore calling __dev_set_promiscuity here is safe.
4499 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4500 __dev_set_promiscuity(dev, 1);
4501 dev->uc_promisc = 1;
4502 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4503 __dev_set_promiscuity(dev, -1);
4504 dev->uc_promisc = 0;
4507 if (ops->ndo_set_multicast_list)
4508 ops->ndo_set_multicast_list(dev);
4512 void dev_set_rx_mode(struct net_device *dev)
4514 netif_addr_lock_bh(dev);
4515 __dev_set_rx_mode(dev);
4516 netif_addr_unlock_bh(dev);
4520 * dev_get_flags - get flags reported to userspace
4523 * Get the combination of flag bits exported through APIs to userspace.
4525 unsigned dev_get_flags(const struct net_device *dev)
4529 flags = (dev->flags & ~(IFF_PROMISC |
4534 (dev->gflags & (IFF_PROMISC |
4537 if (netif_running(dev)) {
4538 if (netif_oper_up(dev))
4539 flags |= IFF_RUNNING;
4540 if (netif_carrier_ok(dev))
4541 flags |= IFF_LOWER_UP;
4542 if (netif_dormant(dev))
4543 flags |= IFF_DORMANT;
4548 EXPORT_SYMBOL(dev_get_flags);
4550 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4552 int old_flags = dev->flags;
4558 * Set the flags on our device.
4561 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4562 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4564 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4568 * Load in the correct multicast list now the flags have changed.
4571 if ((old_flags ^ flags) & IFF_MULTICAST)
4572 dev_change_rx_flags(dev, IFF_MULTICAST);
4574 dev_set_rx_mode(dev);
4577 * Have we downed the interface. We handle IFF_UP ourselves
4578 * according to user attempts to set it, rather than blindly
4583 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4584 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4587 dev_set_rx_mode(dev);
4590 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4591 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4593 dev->gflags ^= IFF_PROMISC;
4594 dev_set_promiscuity(dev, inc);
4597 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4598 is important. Some (broken) drivers set IFF_PROMISC, when
4599 IFF_ALLMULTI is requested not asking us and not reporting.
4601 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4602 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4604 dev->gflags ^= IFF_ALLMULTI;
4605 dev_set_allmulti(dev, inc);
4611 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4613 unsigned int changes = dev->flags ^ old_flags;
4615 if (changes & IFF_UP) {
4616 if (dev->flags & IFF_UP)
4617 call_netdevice_notifiers(NETDEV_UP, dev);
4619 call_netdevice_notifiers(NETDEV_DOWN, dev);
4622 if (dev->flags & IFF_UP &&
4623 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4624 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4628 * dev_change_flags - change device settings
4630 * @flags: device state flags
4632 * Change settings on device based state flags. The flags are
4633 * in the userspace exported format.
4635 int dev_change_flags(struct net_device *dev, unsigned flags)
4638 int old_flags = dev->flags;
4640 ret = __dev_change_flags(dev, flags);
4644 changes = old_flags ^ dev->flags;
4646 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4648 __dev_notify_flags(dev, old_flags);
4651 EXPORT_SYMBOL(dev_change_flags);
4654 * dev_set_mtu - Change maximum transfer unit
4656 * @new_mtu: new transfer unit
4658 * Change the maximum transfer size of the network device.
4660 int dev_set_mtu(struct net_device *dev, int new_mtu)
4662 const struct net_device_ops *ops = dev->netdev_ops;
4665 if (new_mtu == dev->mtu)
4668 /* MTU must be positive. */
4672 if (!netif_device_present(dev))
4676 if (ops->ndo_change_mtu)
4677 err = ops->ndo_change_mtu(dev, new_mtu);
4681 if (!err && dev->flags & IFF_UP)
4682 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4685 EXPORT_SYMBOL(dev_set_mtu);
4688 * dev_set_group - Change group this device belongs to
4690 * @new_group: group this device should belong to
4692 void dev_set_group(struct net_device *dev, int new_group)
4694 dev->group = new_group;
4696 EXPORT_SYMBOL(dev_set_group);
4699 * dev_set_mac_address - Change Media Access Control Address
4703 * Change the hardware (MAC) address of the device
4705 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4707 const struct net_device_ops *ops = dev->netdev_ops;
4710 if (!ops->ndo_set_mac_address)
4712 if (sa->sa_family != dev->type)
4714 if (!netif_device_present(dev))
4716 err = ops->ndo_set_mac_address(dev, sa);
4718 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4721 EXPORT_SYMBOL(dev_set_mac_address);
4724 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4726 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4729 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4735 case SIOCGIFFLAGS: /* Get interface flags */
4736 ifr->ifr_flags = (short) dev_get_flags(dev);
4739 case SIOCGIFMETRIC: /* Get the metric on the interface
4740 (currently unused) */
4741 ifr->ifr_metric = 0;
4744 case SIOCGIFMTU: /* Get the MTU of a device */
4745 ifr->ifr_mtu = dev->mtu;
4750 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4752 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4753 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4754 ifr->ifr_hwaddr.sa_family = dev->type;
4762 ifr->ifr_map.mem_start = dev->mem_start;
4763 ifr->ifr_map.mem_end = dev->mem_end;
4764 ifr->ifr_map.base_addr = dev->base_addr;
4765 ifr->ifr_map.irq = dev->irq;
4766 ifr->ifr_map.dma = dev->dma;
4767 ifr->ifr_map.port = dev->if_port;
4771 ifr->ifr_ifindex = dev->ifindex;
4775 ifr->ifr_qlen = dev->tx_queue_len;
4779 /* dev_ioctl() should ensure this case
4791 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4793 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4796 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4797 const struct net_device_ops *ops;
4802 ops = dev->netdev_ops;
4805 case SIOCSIFFLAGS: /* Set interface flags */
4806 return dev_change_flags(dev, ifr->ifr_flags);
4808 case SIOCSIFMETRIC: /* Set the metric on the interface
4809 (currently unused) */
4812 case SIOCSIFMTU: /* Set the MTU of a device */
4813 return dev_set_mtu(dev, ifr->ifr_mtu);
4816 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4818 case SIOCSIFHWBROADCAST:
4819 if (ifr->ifr_hwaddr.sa_family != dev->type)
4821 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4822 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4823 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4827 if (ops->ndo_set_config) {
4828 if (!netif_device_present(dev))
4830 return ops->ndo_set_config(dev, &ifr->ifr_map);
4835 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4836 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4838 if (!netif_device_present(dev))
4840 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4843 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4844 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4846 if (!netif_device_present(dev))
4848 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4851 if (ifr->ifr_qlen < 0)
4853 dev->tx_queue_len = ifr->ifr_qlen;
4857 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4858 return dev_change_name(dev, ifr->ifr_newname);
4861 * Unknown or private ioctl
4864 if ((cmd >= SIOCDEVPRIVATE &&
4865 cmd <= SIOCDEVPRIVATE + 15) ||
4866 cmd == SIOCBONDENSLAVE ||
4867 cmd == SIOCBONDRELEASE ||
4868 cmd == SIOCBONDSETHWADDR ||
4869 cmd == SIOCBONDSLAVEINFOQUERY ||
4870 cmd == SIOCBONDINFOQUERY ||
4871 cmd == SIOCBONDCHANGEACTIVE ||
4872 cmd == SIOCGMIIPHY ||
4873 cmd == SIOCGMIIREG ||
4874 cmd == SIOCSMIIREG ||
4875 cmd == SIOCBRADDIF ||
4876 cmd == SIOCBRDELIF ||
4877 cmd == SIOCSHWTSTAMP ||
4878 cmd == SIOCWANDEV) {
4880 if (ops->ndo_do_ioctl) {
4881 if (netif_device_present(dev))
4882 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4894 * This function handles all "interface"-type I/O control requests. The actual
4895 * 'doing' part of this is dev_ifsioc above.
4899 * dev_ioctl - network device ioctl
4900 * @net: the applicable net namespace
4901 * @cmd: command to issue
4902 * @arg: pointer to a struct ifreq in user space
4904 * Issue ioctl functions to devices. This is normally called by the
4905 * user space syscall interfaces but can sometimes be useful for
4906 * other purposes. The return value is the return from the syscall if
4907 * positive or a negative errno code on error.
4910 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4916 /* One special case: SIOCGIFCONF takes ifconf argument
4917 and requires shared lock, because it sleeps writing
4921 if (cmd == SIOCGIFCONF) {
4923 ret = dev_ifconf(net, (char __user *) arg);
4927 if (cmd == SIOCGIFNAME)
4928 return dev_ifname(net, (struct ifreq __user *)arg);
4930 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4933 ifr.ifr_name[IFNAMSIZ-1] = 0;
4935 colon = strchr(ifr.ifr_name, ':');
4940 * See which interface the caller is talking about.
4945 * These ioctl calls:
4946 * - can be done by all.
4947 * - atomic and do not require locking.
4958 dev_load(net, ifr.ifr_name);
4960 ret = dev_ifsioc_locked(net, &ifr, cmd);
4965 if (copy_to_user(arg, &ifr,
4966 sizeof(struct ifreq)))
4972 dev_load(net, ifr.ifr_name);
4974 ret = dev_ethtool(net, &ifr);
4979 if (copy_to_user(arg, &ifr,
4980 sizeof(struct ifreq)))
4986 * These ioctl calls:
4987 * - require superuser power.
4988 * - require strict serialization.
4994 if (!capable(CAP_NET_ADMIN))
4996 dev_load(net, ifr.ifr_name);
4998 ret = dev_ifsioc(net, &ifr, cmd);
5003 if (copy_to_user(arg, &ifr,
5004 sizeof(struct ifreq)))
5010 * These ioctl calls:
5011 * - require superuser power.
5012 * - require strict serialization.
5013 * - do not return a value
5023 case SIOCSIFHWBROADCAST:
5026 case SIOCBONDENSLAVE:
5027 case SIOCBONDRELEASE:
5028 case SIOCBONDSETHWADDR:
5029 case SIOCBONDCHANGEACTIVE:
5033 if (!capable(CAP_NET_ADMIN))
5036 case SIOCBONDSLAVEINFOQUERY:
5037 case SIOCBONDINFOQUERY:
5038 dev_load(net, ifr.ifr_name);
5040 ret = dev_ifsioc(net, &ifr, cmd);
5045 /* Get the per device memory space. We can add this but
5046 * currently do not support it */
5048 /* Set the per device memory buffer space.
5049 * Not applicable in our case */
5054 * Unknown or private ioctl.
5057 if (cmd == SIOCWANDEV ||
5058 (cmd >= SIOCDEVPRIVATE &&
5059 cmd <= SIOCDEVPRIVATE + 15)) {
5060 dev_load(net, ifr.ifr_name);
5062 ret = dev_ifsioc(net, &ifr, cmd);
5064 if (!ret && copy_to_user(arg, &ifr,
5065 sizeof(struct ifreq)))
5069 /* Take care of Wireless Extensions */
5070 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5071 return wext_handle_ioctl(net, &ifr, cmd, arg);
5078 * dev_new_index - allocate an ifindex
5079 * @net: the applicable net namespace
5081 * Returns a suitable unique value for a new device interface
5082 * number. The caller must hold the rtnl semaphore or the
5083 * dev_base_lock to be sure it remains unique.
5085 static int dev_new_index(struct net *net)
5091 if (!__dev_get_by_index(net, ifindex))
5096 /* Delayed registration/unregisteration */
5097 static LIST_HEAD(net_todo_list);
5099 static void net_set_todo(struct net_device *dev)
5101 list_add_tail(&dev->todo_list, &net_todo_list);
5104 static void rollback_registered_many(struct list_head *head)
5106 struct net_device *dev, *tmp;
5108 BUG_ON(dev_boot_phase);
5111 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5112 /* Some devices call without registering
5113 * for initialization unwind. Remove those
5114 * devices and proceed with the remaining.
5116 if (dev->reg_state == NETREG_UNINITIALIZED) {
5117 pr_debug("unregister_netdevice: device %s/%p never "
5118 "was registered\n", dev->name, dev);
5121 list_del(&dev->unreg_list);
5125 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5128 /* If device is running, close it first. */
5129 dev_close_many(head);
5131 list_for_each_entry(dev, head, unreg_list) {
5132 /* And unlink it from device chain. */
5133 unlist_netdevice(dev);
5135 dev->reg_state = NETREG_UNREGISTERING;
5140 list_for_each_entry(dev, head, unreg_list) {
5141 /* Shutdown queueing discipline. */
5145 /* Notify protocols, that we are about to destroy
5146 this device. They should clean all the things.
5148 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5150 if (!dev->rtnl_link_ops ||
5151 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5152 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5155 * Flush the unicast and multicast chains
5160 if (dev->netdev_ops->ndo_uninit)
5161 dev->netdev_ops->ndo_uninit(dev);
5163 /* Notifier chain MUST detach us from master device. */
5164 WARN_ON(dev->master);
5166 /* Remove entries from kobject tree */
5167 netdev_unregister_kobject(dev);
5170 /* Process any work delayed until the end of the batch */
5171 dev = list_first_entry(head, struct net_device, unreg_list);
5172 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5176 list_for_each_entry(dev, head, unreg_list)
5180 static void rollback_registered(struct net_device *dev)
5184 list_add(&dev->unreg_list, &single);
5185 rollback_registered_many(&single);
5189 u32 netdev_fix_features(struct net_device *dev, u32 features)
5191 /* Fix illegal checksum combinations */
5192 if ((features & NETIF_F_HW_CSUM) &&
5193 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5194 netdev_info(dev, "mixed HW and IP checksum settings.\n");
5195 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5198 if ((features & NETIF_F_NO_CSUM) &&
5199 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5200 netdev_info(dev, "mixed no checksumming and other settings.\n");
5201 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5204 /* Fix illegal SG+CSUM combinations. */
5205 if ((features & NETIF_F_SG) &&
5206 !(features & NETIF_F_ALL_CSUM)) {
5208 "Dropping NETIF_F_SG since no checksum feature.\n");
5209 features &= ~NETIF_F_SG;
5212 /* TSO requires that SG is present as well. */
5213 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5214 netdev_info(dev, "Dropping TSO features since no SG feature.\n");
5215 features &= ~NETIF_F_ALL_TSO;
5218 /* TSO ECN requires that TSO is present as well. */
5219 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5220 features &= ~NETIF_F_TSO_ECN;
5222 /* Software GSO depends on SG. */
5223 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5224 netdev_info(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5225 features &= ~NETIF_F_GSO;
5228 /* UFO needs SG and checksumming */
5229 if (features & NETIF_F_UFO) {
5230 /* maybe split UFO into V4 and V6? */
5231 if (!((features & NETIF_F_GEN_CSUM) ||
5232 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5233 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5235 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5236 features &= ~NETIF_F_UFO;
5239 if (!(features & NETIF_F_SG)) {
5241 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5242 features &= ~NETIF_F_UFO;
5248 EXPORT_SYMBOL(netdev_fix_features);
5250 int __netdev_update_features(struct net_device *dev)
5257 features = netdev_get_wanted_features(dev);
5259 if (dev->netdev_ops->ndo_fix_features)
5260 features = dev->netdev_ops->ndo_fix_features(dev, features);
5262 /* driver might be less strict about feature dependencies */
5263 features = netdev_fix_features(dev, features);
5265 if (dev->features == features)
5268 netdev_info(dev, "Features changed: 0x%08x -> 0x%08x\n",
5269 dev->features, features);
5271 if (dev->netdev_ops->ndo_set_features)
5272 err = dev->netdev_ops->ndo_set_features(dev, features);
5274 if (unlikely(err < 0)) {
5276 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5277 err, features, dev->features);
5282 dev->features = features;
5287 void netdev_update_features(struct net_device *dev)
5289 if (__netdev_update_features(dev))
5290 netdev_features_change(dev);
5292 EXPORT_SYMBOL(netdev_update_features);
5295 * netif_stacked_transfer_operstate - transfer operstate
5296 * @rootdev: the root or lower level device to transfer state from
5297 * @dev: the device to transfer operstate to
5299 * Transfer operational state from root to device. This is normally
5300 * called when a stacking relationship exists between the root
5301 * device and the device(a leaf device).
5303 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5304 struct net_device *dev)
5306 if (rootdev->operstate == IF_OPER_DORMANT)
5307 netif_dormant_on(dev);
5309 netif_dormant_off(dev);
5311 if (netif_carrier_ok(rootdev)) {
5312 if (!netif_carrier_ok(dev))
5313 netif_carrier_on(dev);
5315 if (netif_carrier_ok(dev))
5316 netif_carrier_off(dev);
5319 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5322 static int netif_alloc_rx_queues(struct net_device *dev)
5324 unsigned int i, count = dev->num_rx_queues;
5325 struct netdev_rx_queue *rx;
5329 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5331 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5336 for (i = 0; i < count; i++)
5342 static void netdev_init_one_queue(struct net_device *dev,
5343 struct netdev_queue *queue, void *_unused)
5345 /* Initialize queue lock */
5346 spin_lock_init(&queue->_xmit_lock);
5347 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5348 queue->xmit_lock_owner = -1;
5349 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5353 static int netif_alloc_netdev_queues(struct net_device *dev)
5355 unsigned int count = dev->num_tx_queues;
5356 struct netdev_queue *tx;
5360 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5362 pr_err("netdev: Unable to allocate %u tx queues.\n",
5368 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5369 spin_lock_init(&dev->tx_global_lock);
5375 * register_netdevice - register a network device
5376 * @dev: device to register
5378 * Take a completed network device structure and add it to the kernel
5379 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5380 * chain. 0 is returned on success. A negative errno code is returned
5381 * on a failure to set up the device, or if the name is a duplicate.
5383 * Callers must hold the rtnl semaphore. You may want
5384 * register_netdev() instead of this.
5387 * The locking appears insufficient to guarantee two parallel registers
5388 * will not get the same name.
5391 int register_netdevice(struct net_device *dev)
5394 struct net *net = dev_net(dev);
5396 BUG_ON(dev_boot_phase);
5401 /* When net_device's are persistent, this will be fatal. */
5402 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5405 spin_lock_init(&dev->addr_list_lock);
5406 netdev_set_addr_lockdep_class(dev);
5410 /* Init, if this function is available */
5411 if (dev->netdev_ops->ndo_init) {
5412 ret = dev->netdev_ops->ndo_init(dev);
5420 ret = dev_get_valid_name(dev, dev->name, 0);
5424 dev->ifindex = dev_new_index(net);
5425 if (dev->iflink == -1)
5426 dev->iflink = dev->ifindex;
5428 /* Transfer changeable features to wanted_features and enable
5429 * software offloads (GSO and GRO).
5431 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5432 dev->features |= NETIF_F_SOFT_FEATURES;
5433 dev->wanted_features = dev->features & dev->hw_features;
5435 /* Avoid warning from netdev_fix_features() for GSO without SG */
5436 if (!(dev->wanted_features & NETIF_F_SG)) {
5437 dev->wanted_features &= ~NETIF_F_GSO;
5438 dev->features &= ~NETIF_F_GSO;
5441 /* Turn on no cache copy if HW is doing checksum */
5442 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5443 if ((dev->features & NETIF_F_ALL_CSUM) &&
5444 !(dev->features & NETIF_F_NO_CSUM)) {
5445 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5446 dev->features |= NETIF_F_NOCACHE_COPY;
5449 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5450 * vlan_dev_init() will do the dev->features check, so these features
5451 * are enabled only if supported by underlying device.
5453 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5455 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5456 ret = notifier_to_errno(ret);
5460 ret = netdev_register_kobject(dev);
5463 dev->reg_state = NETREG_REGISTERED;
5465 __netdev_update_features(dev);
5468 * Default initial state at registry is that the
5469 * device is present.
5472 set_bit(__LINK_STATE_PRESENT, &dev->state);
5474 dev_init_scheduler(dev);
5476 list_netdevice(dev);
5478 /* Notify protocols, that a new device appeared. */
5479 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5480 ret = notifier_to_errno(ret);
5482 rollback_registered(dev);
5483 dev->reg_state = NETREG_UNREGISTERED;
5486 * Prevent userspace races by waiting until the network
5487 * device is fully setup before sending notifications.
5489 if (!dev->rtnl_link_ops ||
5490 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5491 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5497 if (dev->netdev_ops->ndo_uninit)
5498 dev->netdev_ops->ndo_uninit(dev);
5501 EXPORT_SYMBOL(register_netdevice);
5504 * init_dummy_netdev - init a dummy network device for NAPI
5505 * @dev: device to init
5507 * This takes a network device structure and initialize the minimum
5508 * amount of fields so it can be used to schedule NAPI polls without
5509 * registering a full blown interface. This is to be used by drivers
5510 * that need to tie several hardware interfaces to a single NAPI
5511 * poll scheduler due to HW limitations.
5513 int init_dummy_netdev(struct net_device *dev)
5515 /* Clear everything. Note we don't initialize spinlocks
5516 * are they aren't supposed to be taken by any of the
5517 * NAPI code and this dummy netdev is supposed to be
5518 * only ever used for NAPI polls
5520 memset(dev, 0, sizeof(struct net_device));
5522 /* make sure we BUG if trying to hit standard
5523 * register/unregister code path
5525 dev->reg_state = NETREG_DUMMY;
5527 /* NAPI wants this */
5528 INIT_LIST_HEAD(&dev->napi_list);
5530 /* a dummy interface is started by default */
5531 set_bit(__LINK_STATE_PRESENT, &dev->state);
5532 set_bit(__LINK_STATE_START, &dev->state);
5534 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5535 * because users of this 'device' dont need to change
5541 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5545 * register_netdev - register a network device
5546 * @dev: device to register
5548 * Take a completed network device structure and add it to the kernel
5549 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5550 * chain. 0 is returned on success. A negative errno code is returned
5551 * on a failure to set up the device, or if the name is a duplicate.
5553 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5554 * and expands the device name if you passed a format string to
5557 int register_netdev(struct net_device *dev)
5564 * If the name is a format string the caller wants us to do a
5567 if (strchr(dev->name, '%')) {
5568 err = dev_alloc_name(dev, dev->name);
5573 err = register_netdevice(dev);
5578 EXPORT_SYMBOL(register_netdev);
5580 int netdev_refcnt_read(const struct net_device *dev)
5584 for_each_possible_cpu(i)
5585 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5588 EXPORT_SYMBOL(netdev_refcnt_read);
5591 * netdev_wait_allrefs - wait until all references are gone.
5593 * This is called when unregistering network devices.
5595 * Any protocol or device that holds a reference should register
5596 * for netdevice notification, and cleanup and put back the
5597 * reference if they receive an UNREGISTER event.
5598 * We can get stuck here if buggy protocols don't correctly
5601 static void netdev_wait_allrefs(struct net_device *dev)
5603 unsigned long rebroadcast_time, warning_time;
5606 linkwatch_forget_dev(dev);
5608 rebroadcast_time = warning_time = jiffies;
5609 refcnt = netdev_refcnt_read(dev);
5611 while (refcnt != 0) {
5612 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5615 /* Rebroadcast unregister notification */
5616 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5617 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5618 * should have already handle it the first time */
5620 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5622 /* We must not have linkwatch events
5623 * pending on unregister. If this
5624 * happens, we simply run the queue
5625 * unscheduled, resulting in a noop
5628 linkwatch_run_queue();
5633 rebroadcast_time = jiffies;
5638 refcnt = netdev_refcnt_read(dev);
5640 if (time_after(jiffies, warning_time + 10 * HZ)) {
5641 printk(KERN_EMERG "unregister_netdevice: "
5642 "waiting for %s to become free. Usage "
5645 warning_time = jiffies;
5654 * register_netdevice(x1);
5655 * register_netdevice(x2);
5657 * unregister_netdevice(y1);
5658 * unregister_netdevice(y2);
5664 * We are invoked by rtnl_unlock().
5665 * This allows us to deal with problems:
5666 * 1) We can delete sysfs objects which invoke hotplug
5667 * without deadlocking with linkwatch via keventd.
5668 * 2) Since we run with the RTNL semaphore not held, we can sleep
5669 * safely in order to wait for the netdev refcnt to drop to zero.
5671 * We must not return until all unregister events added during
5672 * the interval the lock was held have been completed.
5674 void netdev_run_todo(void)
5676 struct list_head list;
5678 /* Snapshot list, allow later requests */
5679 list_replace_init(&net_todo_list, &list);
5683 while (!list_empty(&list)) {
5684 struct net_device *dev
5685 = list_first_entry(&list, struct net_device, todo_list);
5686 list_del(&dev->todo_list);
5688 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5689 printk(KERN_ERR "network todo '%s' but state %d\n",
5690 dev->name, dev->reg_state);
5695 dev->reg_state = NETREG_UNREGISTERED;
5697 on_each_cpu(flush_backlog, dev, 1);
5699 netdev_wait_allrefs(dev);
5702 BUG_ON(netdev_refcnt_read(dev));
5703 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5704 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5705 WARN_ON(dev->dn_ptr);
5707 if (dev->destructor)
5708 dev->destructor(dev);
5710 /* Free network device */
5711 kobject_put(&dev->dev.kobj);
5715 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5716 * fields in the same order, with only the type differing.
5718 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5719 const struct net_device_stats *netdev_stats)
5721 #if BITS_PER_LONG == 64
5722 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5723 memcpy(stats64, netdev_stats, sizeof(*stats64));
5725 size_t i, n = sizeof(*stats64) / sizeof(u64);
5726 const unsigned long *src = (const unsigned long *)netdev_stats;
5727 u64 *dst = (u64 *)stats64;
5729 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5730 sizeof(*stats64) / sizeof(u64));
5731 for (i = 0; i < n; i++)
5737 * dev_get_stats - get network device statistics
5738 * @dev: device to get statistics from
5739 * @storage: place to store stats
5741 * Get network statistics from device. Return @storage.
5742 * The device driver may provide its own method by setting
5743 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5744 * otherwise the internal statistics structure is used.
5746 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5747 struct rtnl_link_stats64 *storage)
5749 const struct net_device_ops *ops = dev->netdev_ops;
5751 if (ops->ndo_get_stats64) {
5752 memset(storage, 0, sizeof(*storage));
5753 ops->ndo_get_stats64(dev, storage);
5754 } else if (ops->ndo_get_stats) {
5755 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5757 netdev_stats_to_stats64(storage, &dev->stats);
5759 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5762 EXPORT_SYMBOL(dev_get_stats);
5764 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5766 struct netdev_queue *queue = dev_ingress_queue(dev);
5768 #ifdef CONFIG_NET_CLS_ACT
5771 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5774 netdev_init_one_queue(dev, queue, NULL);
5775 queue->qdisc = &noop_qdisc;
5776 queue->qdisc_sleeping = &noop_qdisc;
5777 rcu_assign_pointer(dev->ingress_queue, queue);
5783 * alloc_netdev_mqs - allocate network device
5784 * @sizeof_priv: size of private data to allocate space for
5785 * @name: device name format string
5786 * @setup: callback to initialize device
5787 * @txqs: the number of TX subqueues to allocate
5788 * @rxqs: the number of RX subqueues to allocate
5790 * Allocates a struct net_device with private data area for driver use
5791 * and performs basic initialization. Also allocates subquue structs
5792 * for each queue on the device.
5794 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5795 void (*setup)(struct net_device *),
5796 unsigned int txqs, unsigned int rxqs)
5798 struct net_device *dev;
5800 struct net_device *p;
5802 BUG_ON(strlen(name) >= sizeof(dev->name));
5805 pr_err("alloc_netdev: Unable to allocate device "
5806 "with zero queues.\n");
5812 pr_err("alloc_netdev: Unable to allocate device "
5813 "with zero RX queues.\n");
5818 alloc_size = sizeof(struct net_device);
5820 /* ensure 32-byte alignment of private area */
5821 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5822 alloc_size += sizeof_priv;
5824 /* ensure 32-byte alignment of whole construct */
5825 alloc_size += NETDEV_ALIGN - 1;
5827 p = kzalloc(alloc_size, GFP_KERNEL);
5829 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5833 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5834 dev->padded = (char *)dev - (char *)p;
5836 dev->pcpu_refcnt = alloc_percpu(int);
5837 if (!dev->pcpu_refcnt)
5840 if (dev_addr_init(dev))
5846 dev_net_set(dev, &init_net);
5848 dev->gso_max_size = GSO_MAX_SIZE;
5850 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5851 dev->ethtool_ntuple_list.count = 0;
5852 INIT_LIST_HEAD(&dev->napi_list);
5853 INIT_LIST_HEAD(&dev->unreg_list);
5854 INIT_LIST_HEAD(&dev->link_watch_list);
5855 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5858 dev->num_tx_queues = txqs;
5859 dev->real_num_tx_queues = txqs;
5860 if (netif_alloc_netdev_queues(dev))
5864 dev->num_rx_queues = rxqs;
5865 dev->real_num_rx_queues = rxqs;
5866 if (netif_alloc_rx_queues(dev))
5870 strcpy(dev->name, name);
5871 dev->group = INIT_NETDEV_GROUP;
5879 free_percpu(dev->pcpu_refcnt);
5889 EXPORT_SYMBOL(alloc_netdev_mqs);
5892 * free_netdev - free network device
5895 * This function does the last stage of destroying an allocated device
5896 * interface. The reference to the device object is released.
5897 * If this is the last reference then it will be freed.
5899 void free_netdev(struct net_device *dev)
5901 struct napi_struct *p, *n;
5903 release_net(dev_net(dev));
5910 kfree(rcu_dereference_raw(dev->ingress_queue));
5912 /* Flush device addresses */
5913 dev_addr_flush(dev);
5915 /* Clear ethtool n-tuple list */
5916 ethtool_ntuple_flush(dev);
5918 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5921 free_percpu(dev->pcpu_refcnt);
5922 dev->pcpu_refcnt = NULL;
5924 /* Compatibility with error handling in drivers */
5925 if (dev->reg_state == NETREG_UNINITIALIZED) {
5926 kfree((char *)dev - dev->padded);
5930 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5931 dev->reg_state = NETREG_RELEASED;
5933 /* will free via device release */
5934 put_device(&dev->dev);
5936 EXPORT_SYMBOL(free_netdev);
5939 * synchronize_net - Synchronize with packet receive processing
5941 * Wait for packets currently being received to be done.
5942 * Does not block later packets from starting.
5944 void synchronize_net(void)
5949 EXPORT_SYMBOL(synchronize_net);
5952 * unregister_netdevice_queue - remove device from the kernel
5956 * This function shuts down a device interface and removes it
5957 * from the kernel tables.
5958 * If head not NULL, device is queued to be unregistered later.
5960 * Callers must hold the rtnl semaphore. You may want
5961 * unregister_netdev() instead of this.
5964 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5969 list_move_tail(&dev->unreg_list, head);
5971 rollback_registered(dev);
5972 /* Finish processing unregister after unlock */
5976 EXPORT_SYMBOL(unregister_netdevice_queue);
5979 * unregister_netdevice_many - unregister many devices
5980 * @head: list of devices
5982 void unregister_netdevice_many(struct list_head *head)
5984 struct net_device *dev;
5986 if (!list_empty(head)) {
5987 rollback_registered_many(head);
5988 list_for_each_entry(dev, head, unreg_list)
5992 EXPORT_SYMBOL(unregister_netdevice_many);
5995 * unregister_netdev - remove device from the kernel
5998 * This function shuts down a device interface and removes it
5999 * from the kernel tables.
6001 * This is just a wrapper for unregister_netdevice that takes
6002 * the rtnl semaphore. In general you want to use this and not
6003 * unregister_netdevice.
6005 void unregister_netdev(struct net_device *dev)
6008 unregister_netdevice(dev);
6011 EXPORT_SYMBOL(unregister_netdev);
6014 * dev_change_net_namespace - move device to different nethost namespace
6016 * @net: network namespace
6017 * @pat: If not NULL name pattern to try if the current device name
6018 * is already taken in the destination network namespace.
6020 * This function shuts down a device interface and moves it
6021 * to a new network namespace. On success 0 is returned, on
6022 * a failure a netagive errno code is returned.
6024 * Callers must hold the rtnl semaphore.
6027 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6033 /* Don't allow namespace local devices to be moved. */
6035 if (dev->features & NETIF_F_NETNS_LOCAL)
6038 /* Ensure the device has been registrered */
6040 if (dev->reg_state != NETREG_REGISTERED)
6043 /* Get out if there is nothing todo */
6045 if (net_eq(dev_net(dev), net))
6048 /* Pick the destination device name, and ensure
6049 * we can use it in the destination network namespace.
6052 if (__dev_get_by_name(net, dev->name)) {
6053 /* We get here if we can't use the current device name */
6056 if (dev_get_valid_name(dev, pat, 1))
6061 * And now a mini version of register_netdevice unregister_netdevice.
6064 /* If device is running close it first. */
6067 /* And unlink it from device chain */
6069 unlist_netdevice(dev);
6073 /* Shutdown queueing discipline. */
6076 /* Notify protocols, that we are about to destroy
6077 this device. They should clean all the things.
6079 Note that dev->reg_state stays at NETREG_REGISTERED.
6080 This is wanted because this way 8021q and macvlan know
6081 the device is just moving and can keep their slaves up.
6083 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6084 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6087 * Flush the unicast and multicast chains
6092 /* Actually switch the network namespace */
6093 dev_net_set(dev, net);
6095 /* If there is an ifindex conflict assign a new one */
6096 if (__dev_get_by_index(net, dev->ifindex)) {
6097 int iflink = (dev->iflink == dev->ifindex);
6098 dev->ifindex = dev_new_index(net);
6100 dev->iflink = dev->ifindex;
6103 /* Fixup kobjects */
6104 err = device_rename(&dev->dev, dev->name);
6107 /* Add the device back in the hashes */
6108 list_netdevice(dev);
6110 /* Notify protocols, that a new device appeared. */
6111 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6114 * Prevent userspace races by waiting until the network
6115 * device is fully setup before sending notifications.
6117 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6124 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6126 static int dev_cpu_callback(struct notifier_block *nfb,
6127 unsigned long action,
6130 struct sk_buff **list_skb;
6131 struct sk_buff *skb;
6132 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6133 struct softnet_data *sd, *oldsd;
6135 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6138 local_irq_disable();
6139 cpu = smp_processor_id();
6140 sd = &per_cpu(softnet_data, cpu);
6141 oldsd = &per_cpu(softnet_data, oldcpu);
6143 /* Find end of our completion_queue. */
6144 list_skb = &sd->completion_queue;
6146 list_skb = &(*list_skb)->next;
6147 /* Append completion queue from offline CPU. */
6148 *list_skb = oldsd->completion_queue;
6149 oldsd->completion_queue = NULL;
6151 /* Append output queue from offline CPU. */
6152 if (oldsd->output_queue) {
6153 *sd->output_queue_tailp = oldsd->output_queue;
6154 sd->output_queue_tailp = oldsd->output_queue_tailp;
6155 oldsd->output_queue = NULL;
6156 oldsd->output_queue_tailp = &oldsd->output_queue;
6159 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6162 /* Process offline CPU's input_pkt_queue */
6163 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6165 input_queue_head_incr(oldsd);
6167 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6169 input_queue_head_incr(oldsd);
6177 * netdev_increment_features - increment feature set by one
6178 * @all: current feature set
6179 * @one: new feature set
6180 * @mask: mask feature set
6182 * Computes a new feature set after adding a device with feature set
6183 * @one to the master device with current feature set @all. Will not
6184 * enable anything that is off in @mask. Returns the new feature set.
6186 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6188 /* If device needs checksumming, downgrade to it. */
6189 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
6190 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
6191 else if (mask & NETIF_F_ALL_CSUM) {
6192 /* If one device supports v4/v6 checksumming, set for all. */
6193 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
6194 !(all & NETIF_F_GEN_CSUM)) {
6195 all &= ~NETIF_F_ALL_CSUM;
6196 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
6199 /* If one device supports hw checksumming, set for all. */
6200 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
6201 all &= ~NETIF_F_ALL_CSUM;
6202 all |= NETIF_F_HW_CSUM;
6206 /* If device can't no cache copy, don't do for all */
6207 if (!(one & NETIF_F_NOCACHE_COPY))
6208 all &= ~NETIF_F_NOCACHE_COPY;
6210 one |= NETIF_F_ALL_CSUM;
6212 one |= all & NETIF_F_ONE_FOR_ALL;
6213 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
6214 all |= one & mask & NETIF_F_ONE_FOR_ALL;
6218 EXPORT_SYMBOL(netdev_increment_features);
6220 static struct hlist_head *netdev_create_hash(void)
6223 struct hlist_head *hash;
6225 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6227 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6228 INIT_HLIST_HEAD(&hash[i]);
6233 /* Initialize per network namespace state */
6234 static int __net_init netdev_init(struct net *net)
6236 INIT_LIST_HEAD(&net->dev_base_head);
6238 net->dev_name_head = netdev_create_hash();
6239 if (net->dev_name_head == NULL)
6242 net->dev_index_head = netdev_create_hash();
6243 if (net->dev_index_head == NULL)
6249 kfree(net->dev_name_head);
6255 * netdev_drivername - network driver for the device
6256 * @dev: network device
6257 * @buffer: buffer for resulting name
6258 * @len: size of buffer
6260 * Determine network driver for device.
6262 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6264 const struct device_driver *driver;
6265 const struct device *parent;
6267 if (len <= 0 || !buffer)
6271 parent = dev->dev.parent;
6276 driver = parent->driver;
6277 if (driver && driver->name)
6278 strlcpy(buffer, driver->name, len);
6282 static int __netdev_printk(const char *level, const struct net_device *dev,
6283 struct va_format *vaf)
6287 if (dev && dev->dev.parent)
6288 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6289 netdev_name(dev), vaf);
6291 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6293 r = printk("%s(NULL net_device): %pV", level, vaf);
6298 int netdev_printk(const char *level, const struct net_device *dev,
6299 const char *format, ...)
6301 struct va_format vaf;
6305 va_start(args, format);
6310 r = __netdev_printk(level, dev, &vaf);
6315 EXPORT_SYMBOL(netdev_printk);
6317 #define define_netdev_printk_level(func, level) \
6318 int func(const struct net_device *dev, const char *fmt, ...) \
6321 struct va_format vaf; \
6324 va_start(args, fmt); \
6329 r = __netdev_printk(level, dev, &vaf); \
6334 EXPORT_SYMBOL(func);
6336 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6337 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6338 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6339 define_netdev_printk_level(netdev_err, KERN_ERR);
6340 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6341 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6342 define_netdev_printk_level(netdev_info, KERN_INFO);
6344 static void __net_exit netdev_exit(struct net *net)
6346 kfree(net->dev_name_head);
6347 kfree(net->dev_index_head);
6350 static struct pernet_operations __net_initdata netdev_net_ops = {
6351 .init = netdev_init,
6352 .exit = netdev_exit,
6355 static void __net_exit default_device_exit(struct net *net)
6357 struct net_device *dev, *aux;
6359 * Push all migratable network devices back to the
6360 * initial network namespace
6363 for_each_netdev_safe(net, dev, aux) {
6365 char fb_name[IFNAMSIZ];
6367 /* Ignore unmoveable devices (i.e. loopback) */
6368 if (dev->features & NETIF_F_NETNS_LOCAL)
6371 /* Leave virtual devices for the generic cleanup */
6372 if (dev->rtnl_link_ops)
6375 /* Push remaining network devices to init_net */
6376 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6377 err = dev_change_net_namespace(dev, &init_net, fb_name);
6379 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6380 __func__, dev->name, err);
6387 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6389 /* At exit all network devices most be removed from a network
6390 * namespace. Do this in the reverse order of registration.
6391 * Do this across as many network namespaces as possible to
6392 * improve batching efficiency.
6394 struct net_device *dev;
6396 LIST_HEAD(dev_kill_list);
6399 list_for_each_entry(net, net_list, exit_list) {
6400 for_each_netdev_reverse(net, dev) {
6401 if (dev->rtnl_link_ops)
6402 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6404 unregister_netdevice_queue(dev, &dev_kill_list);
6407 unregister_netdevice_many(&dev_kill_list);
6408 list_del(&dev_kill_list);
6412 static struct pernet_operations __net_initdata default_device_ops = {
6413 .exit = default_device_exit,
6414 .exit_batch = default_device_exit_batch,
6418 * Initialize the DEV module. At boot time this walks the device list and
6419 * unhooks any devices that fail to initialise (normally hardware not
6420 * present) and leaves us with a valid list of present and active devices.
6425 * This is called single threaded during boot, so no need
6426 * to take the rtnl semaphore.
6428 static int __init net_dev_init(void)
6430 int i, rc = -ENOMEM;
6432 BUG_ON(!dev_boot_phase);
6434 if (dev_proc_init())
6437 if (netdev_kobject_init())
6440 INIT_LIST_HEAD(&ptype_all);
6441 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6442 INIT_LIST_HEAD(&ptype_base[i]);
6444 if (register_pernet_subsys(&netdev_net_ops))
6448 * Initialise the packet receive queues.
6451 for_each_possible_cpu(i) {
6452 struct softnet_data *sd = &per_cpu(softnet_data, i);
6454 memset(sd, 0, sizeof(*sd));
6455 skb_queue_head_init(&sd->input_pkt_queue);
6456 skb_queue_head_init(&sd->process_queue);
6457 sd->completion_queue = NULL;
6458 INIT_LIST_HEAD(&sd->poll_list);
6459 sd->output_queue = NULL;
6460 sd->output_queue_tailp = &sd->output_queue;
6462 sd->csd.func = rps_trigger_softirq;
6468 sd->backlog.poll = process_backlog;
6469 sd->backlog.weight = weight_p;
6470 sd->backlog.gro_list = NULL;
6471 sd->backlog.gro_count = 0;
6476 /* The loopback device is special if any other network devices
6477 * is present in a network namespace the loopback device must
6478 * be present. Since we now dynamically allocate and free the
6479 * loopback device ensure this invariant is maintained by
6480 * keeping the loopback device as the first device on the
6481 * list of network devices. Ensuring the loopback devices
6482 * is the first device that appears and the last network device
6485 if (register_pernet_device(&loopback_net_ops))
6488 if (register_pernet_device(&default_device_ops))
6491 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6492 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6494 hotcpu_notifier(dev_cpu_callback, 0);
6502 subsys_initcall(net_dev_init);
6504 static int __init initialize_hashrnd(void)
6506 get_random_bytes(&hashrnd, sizeof(hashrnd));
6510 late_initcall_sync(initialize_hashrnd);