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)
955 BUG_ON(!dev_net(dev));
958 if (!dev_valid_name(name))
961 if (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);
1003 ret = device_rename(&dev->dev, dev->name);
1005 memcpy(dev->name, oldname, IFNAMSIZ);
1009 write_lock_bh(&dev_base_lock);
1010 hlist_del_rcu(&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)
1287 if (dev->flags & IFF_UP) {
1290 list_add(&dev->unreg_list, &single);
1291 dev_close_many(&single);
1296 EXPORT_SYMBOL(dev_close);
1300 * dev_disable_lro - disable Large Receive Offload on a device
1303 * Disable Large Receive Offload (LRO) on a net device. Must be
1304 * called under RTNL. This is needed if received packets may be
1305 * forwarded to another interface.
1307 void dev_disable_lro(struct net_device *dev)
1312 * If we're trying to disable lro on a vlan device
1313 * use the underlying physical device instead
1315 if (is_vlan_dev(dev))
1316 dev = vlan_dev_real_dev(dev);
1318 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1319 flags = dev->ethtool_ops->get_flags(dev);
1321 flags = ethtool_op_get_flags(dev);
1323 if (!(flags & ETH_FLAG_LRO))
1326 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1327 if (unlikely(dev->features & NETIF_F_LRO))
1328 netdev_WARN(dev, "failed to disable LRO!\n");
1330 EXPORT_SYMBOL(dev_disable_lro);
1333 static int dev_boot_phase = 1;
1336 * register_netdevice_notifier - register a network notifier block
1339 * Register a notifier to be called when network device events occur.
1340 * The notifier passed is linked into the kernel structures and must
1341 * not be reused until it has been unregistered. A negative errno code
1342 * is returned on a failure.
1344 * When registered all registration and up events are replayed
1345 * to the new notifier to allow device to have a race free
1346 * view of the network device list.
1349 int register_netdevice_notifier(struct notifier_block *nb)
1351 struct net_device *dev;
1352 struct net_device *last;
1357 err = raw_notifier_chain_register(&netdev_chain, nb);
1363 for_each_netdev(net, dev) {
1364 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1365 err = notifier_to_errno(err);
1369 if (!(dev->flags & IFF_UP))
1372 nb->notifier_call(nb, NETDEV_UP, dev);
1383 for_each_netdev(net, dev) {
1387 if (dev->flags & IFF_UP) {
1388 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1389 nb->notifier_call(nb, NETDEV_DOWN, dev);
1391 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1392 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1396 raw_notifier_chain_unregister(&netdev_chain, nb);
1399 EXPORT_SYMBOL(register_netdevice_notifier);
1402 * unregister_netdevice_notifier - unregister a network notifier block
1405 * Unregister a notifier previously registered by
1406 * register_netdevice_notifier(). The notifier is unlinked into the
1407 * kernel structures and may then be reused. A negative errno code
1408 * is returned on a failure.
1411 int unregister_netdevice_notifier(struct notifier_block *nb)
1416 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1420 EXPORT_SYMBOL(unregister_netdevice_notifier);
1423 * call_netdevice_notifiers - call all network notifier blocks
1424 * @val: value passed unmodified to notifier function
1425 * @dev: net_device pointer passed unmodified to notifier function
1427 * Call all network notifier blocks. Parameters and return value
1428 * are as for raw_notifier_call_chain().
1431 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1434 return raw_notifier_call_chain(&netdev_chain, val, dev);
1436 EXPORT_SYMBOL(call_netdevice_notifiers);
1438 /* When > 0 there are consumers of rx skb time stamps */
1439 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1441 void net_enable_timestamp(void)
1443 atomic_inc(&netstamp_needed);
1445 EXPORT_SYMBOL(net_enable_timestamp);
1447 void net_disable_timestamp(void)
1449 atomic_dec(&netstamp_needed);
1451 EXPORT_SYMBOL(net_disable_timestamp);
1453 static inline void net_timestamp_set(struct sk_buff *skb)
1455 if (atomic_read(&netstamp_needed))
1456 __net_timestamp(skb);
1458 skb->tstamp.tv64 = 0;
1461 static inline void net_timestamp_check(struct sk_buff *skb)
1463 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1464 __net_timestamp(skb);
1467 static inline bool is_skb_forwardable(struct net_device *dev,
1468 struct sk_buff *skb)
1472 if (!(dev->flags & IFF_UP))
1475 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1476 if (skb->len <= len)
1479 /* if TSO is enabled, we don't care about the length as the packet
1480 * could be forwarded without being segmented before
1482 if (skb_is_gso(skb))
1489 * dev_forward_skb - loopback an skb to another netif
1491 * @dev: destination network device
1492 * @skb: buffer to forward
1495 * NET_RX_SUCCESS (no congestion)
1496 * NET_RX_DROP (packet was dropped, but freed)
1498 * dev_forward_skb can be used for injecting an skb from the
1499 * start_xmit function of one device into the receive queue
1500 * of another device.
1502 * The receiving device may be in another namespace, so
1503 * we have to clear all information in the skb that could
1504 * impact namespace isolation.
1506 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1511 if (unlikely(!is_skb_forwardable(dev, skb))) {
1512 atomic_long_inc(&dev->rx_dropped);
1516 skb_set_dev(skb, dev);
1517 skb->tstamp.tv64 = 0;
1518 skb->pkt_type = PACKET_HOST;
1519 skb->protocol = eth_type_trans(skb, dev);
1520 return netif_rx(skb);
1522 EXPORT_SYMBOL_GPL(dev_forward_skb);
1524 static inline int deliver_skb(struct sk_buff *skb,
1525 struct packet_type *pt_prev,
1526 struct net_device *orig_dev)
1528 atomic_inc(&skb->users);
1529 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1533 * Support routine. Sends outgoing frames to any network
1534 * taps currently in use.
1537 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1539 struct packet_type *ptype;
1540 struct sk_buff *skb2 = NULL;
1541 struct packet_type *pt_prev = NULL;
1544 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1545 /* Never send packets back to the socket
1546 * they originated from - MvS (miquels@drinkel.ow.org)
1548 if ((ptype->dev == dev || !ptype->dev) &&
1549 (ptype->af_packet_priv == NULL ||
1550 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1552 deliver_skb(skb2, pt_prev, skb->dev);
1557 skb2 = skb_clone(skb, GFP_ATOMIC);
1561 net_timestamp_set(skb2);
1563 /* skb->nh should be correctly
1564 set by sender, so that the second statement is
1565 just protection against buggy protocols.
1567 skb_reset_mac_header(skb2);
1569 if (skb_network_header(skb2) < skb2->data ||
1570 skb2->network_header > skb2->tail) {
1571 if (net_ratelimit())
1572 printk(KERN_CRIT "protocol %04x is "
1574 ntohs(skb2->protocol),
1576 skb_reset_network_header(skb2);
1579 skb2->transport_header = skb2->network_header;
1580 skb2->pkt_type = PACKET_OUTGOING;
1585 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1589 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1590 * @dev: Network device
1591 * @txq: number of queues available
1593 * If real_num_tx_queues is changed the tc mappings may no longer be
1594 * valid. To resolve this verify the tc mapping remains valid and if
1595 * not NULL the mapping. With no priorities mapping to this
1596 * offset/count pair it will no longer be used. In the worst case TC0
1597 * is invalid nothing can be done so disable priority mappings. If is
1598 * expected that drivers will fix this mapping if they can before
1599 * calling netif_set_real_num_tx_queues.
1601 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1604 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1606 /* If TC0 is invalidated disable TC mapping */
1607 if (tc->offset + tc->count > txq) {
1608 pr_warning("Number of in use tx queues changed "
1609 "invalidating tc mappings. Priority "
1610 "traffic classification disabled!\n");
1615 /* Invalidated prio to tc mappings set to TC0 */
1616 for (i = 1; i < TC_BITMASK + 1; i++) {
1617 int q = netdev_get_prio_tc_map(dev, i);
1619 tc = &dev->tc_to_txq[q];
1620 if (tc->offset + tc->count > txq) {
1621 pr_warning("Number of in use tx queues "
1622 "changed. Priority %i to tc "
1623 "mapping %i is no longer valid "
1624 "setting map to 0\n",
1626 netdev_set_prio_tc_map(dev, i, 0);
1632 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1633 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1635 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1639 if (txq < 1 || txq > dev->num_tx_queues)
1642 if (dev->reg_state == NETREG_REGISTERED ||
1643 dev->reg_state == NETREG_UNREGISTERING) {
1646 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1652 netif_setup_tc(dev, txq);
1654 if (txq < dev->real_num_tx_queues)
1655 qdisc_reset_all_tx_gt(dev, txq);
1658 dev->real_num_tx_queues = txq;
1661 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1665 * netif_set_real_num_rx_queues - set actual number of RX queues used
1666 * @dev: Network device
1667 * @rxq: Actual number of RX queues
1669 * This must be called either with the rtnl_lock held or before
1670 * registration of the net device. Returns 0 on success, or a
1671 * negative error code. If called before registration, it always
1674 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1678 if (rxq < 1 || rxq > dev->num_rx_queues)
1681 if (dev->reg_state == NETREG_REGISTERED) {
1684 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1690 dev->real_num_rx_queues = rxq;
1693 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1696 static inline void __netif_reschedule(struct Qdisc *q)
1698 struct softnet_data *sd;
1699 unsigned long flags;
1701 local_irq_save(flags);
1702 sd = &__get_cpu_var(softnet_data);
1703 q->next_sched = NULL;
1704 *sd->output_queue_tailp = q;
1705 sd->output_queue_tailp = &q->next_sched;
1706 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1707 local_irq_restore(flags);
1710 void __netif_schedule(struct Qdisc *q)
1712 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1713 __netif_reschedule(q);
1715 EXPORT_SYMBOL(__netif_schedule);
1717 void dev_kfree_skb_irq(struct sk_buff *skb)
1719 if (atomic_dec_and_test(&skb->users)) {
1720 struct softnet_data *sd;
1721 unsigned long flags;
1723 local_irq_save(flags);
1724 sd = &__get_cpu_var(softnet_data);
1725 skb->next = sd->completion_queue;
1726 sd->completion_queue = skb;
1727 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1728 local_irq_restore(flags);
1731 EXPORT_SYMBOL(dev_kfree_skb_irq);
1733 void dev_kfree_skb_any(struct sk_buff *skb)
1735 if (in_irq() || irqs_disabled())
1736 dev_kfree_skb_irq(skb);
1740 EXPORT_SYMBOL(dev_kfree_skb_any);
1744 * netif_device_detach - mark device as removed
1745 * @dev: network device
1747 * Mark device as removed from system and therefore no longer available.
1749 void netif_device_detach(struct net_device *dev)
1751 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1752 netif_running(dev)) {
1753 netif_tx_stop_all_queues(dev);
1756 EXPORT_SYMBOL(netif_device_detach);
1759 * netif_device_attach - mark device as attached
1760 * @dev: network device
1762 * Mark device as attached from system and restart if needed.
1764 void netif_device_attach(struct net_device *dev)
1766 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1767 netif_running(dev)) {
1768 netif_tx_wake_all_queues(dev);
1769 __netdev_watchdog_up(dev);
1772 EXPORT_SYMBOL(netif_device_attach);
1775 * skb_dev_set -- assign a new device to a buffer
1776 * @skb: buffer for the new device
1777 * @dev: network device
1779 * If an skb is owned by a device already, we have to reset
1780 * all data private to the namespace a device belongs to
1781 * before assigning it a new device.
1783 #ifdef CONFIG_NET_NS
1784 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1787 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1790 skb_init_secmark(skb);
1794 skb->ipvs_property = 0;
1795 #ifdef CONFIG_NET_SCHED
1801 EXPORT_SYMBOL(skb_set_dev);
1802 #endif /* CONFIG_NET_NS */
1805 * Invalidate hardware checksum when packet is to be mangled, and
1806 * complete checksum manually on outgoing path.
1808 int skb_checksum_help(struct sk_buff *skb)
1811 int ret = 0, offset;
1813 if (skb->ip_summed == CHECKSUM_COMPLETE)
1814 goto out_set_summed;
1816 if (unlikely(skb_shinfo(skb)->gso_size)) {
1817 /* Let GSO fix up the checksum. */
1818 goto out_set_summed;
1821 offset = skb_checksum_start_offset(skb);
1822 BUG_ON(offset >= skb_headlen(skb));
1823 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1825 offset += skb->csum_offset;
1826 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1828 if (skb_cloned(skb) &&
1829 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1830 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1835 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1837 skb->ip_summed = CHECKSUM_NONE;
1841 EXPORT_SYMBOL(skb_checksum_help);
1844 * skb_gso_segment - Perform segmentation on skb.
1845 * @skb: buffer to segment
1846 * @features: features for the output path (see dev->features)
1848 * This function segments the given skb and returns a list of segments.
1850 * It may return NULL if the skb requires no segmentation. This is
1851 * only possible when GSO is used for verifying header integrity.
1853 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1855 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1856 struct packet_type *ptype;
1857 __be16 type = skb->protocol;
1858 int vlan_depth = ETH_HLEN;
1861 while (type == htons(ETH_P_8021Q)) {
1862 struct vlan_hdr *vh;
1864 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1865 return ERR_PTR(-EINVAL);
1867 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1868 type = vh->h_vlan_encapsulated_proto;
1869 vlan_depth += VLAN_HLEN;
1872 skb_reset_mac_header(skb);
1873 skb->mac_len = skb->network_header - skb->mac_header;
1874 __skb_pull(skb, skb->mac_len);
1876 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1877 struct net_device *dev = skb->dev;
1878 struct ethtool_drvinfo info = {};
1880 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1881 dev->ethtool_ops->get_drvinfo(dev, &info);
1883 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1884 info.driver, dev ? dev->features : 0L,
1885 skb->sk ? skb->sk->sk_route_caps : 0L,
1886 skb->len, skb->data_len, skb->ip_summed);
1888 if (skb_header_cloned(skb) &&
1889 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1890 return ERR_PTR(err);
1894 list_for_each_entry_rcu(ptype,
1895 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1896 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1897 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1898 err = ptype->gso_send_check(skb);
1899 segs = ERR_PTR(err);
1900 if (err || skb_gso_ok(skb, features))
1902 __skb_push(skb, (skb->data -
1903 skb_network_header(skb)));
1905 segs = ptype->gso_segment(skb, features);
1911 __skb_push(skb, skb->data - skb_mac_header(skb));
1915 EXPORT_SYMBOL(skb_gso_segment);
1917 /* Take action when hardware reception checksum errors are detected. */
1919 void netdev_rx_csum_fault(struct net_device *dev)
1921 if (net_ratelimit()) {
1922 printk(KERN_ERR "%s: hw csum failure.\n",
1923 dev ? dev->name : "<unknown>");
1927 EXPORT_SYMBOL(netdev_rx_csum_fault);
1930 /* Actually, we should eliminate this check as soon as we know, that:
1931 * 1. IOMMU is present and allows to map all the memory.
1932 * 2. No high memory really exists on this machine.
1935 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1937 #ifdef CONFIG_HIGHMEM
1939 if (!(dev->features & NETIF_F_HIGHDMA)) {
1940 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1941 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1945 if (PCI_DMA_BUS_IS_PHYS) {
1946 struct device *pdev = dev->dev.parent;
1950 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1951 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1952 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1961 void (*destructor)(struct sk_buff *skb);
1964 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1966 static void dev_gso_skb_destructor(struct sk_buff *skb)
1968 struct dev_gso_cb *cb;
1971 struct sk_buff *nskb = skb->next;
1973 skb->next = nskb->next;
1976 } while (skb->next);
1978 cb = DEV_GSO_CB(skb);
1980 cb->destructor(skb);
1984 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1985 * @skb: buffer to segment
1986 * @features: device features as applicable to this skb
1988 * This function segments the given skb and stores the list of segments
1991 static int dev_gso_segment(struct sk_buff *skb, int features)
1993 struct sk_buff *segs;
1995 segs = skb_gso_segment(skb, features);
1997 /* Verifying header integrity only. */
2002 return PTR_ERR(segs);
2005 DEV_GSO_CB(skb)->destructor = skb->destructor;
2006 skb->destructor = dev_gso_skb_destructor;
2012 * Try to orphan skb early, right before transmission by the device.
2013 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2014 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2016 static inline void skb_orphan_try(struct sk_buff *skb)
2018 struct sock *sk = skb->sk;
2020 if (sk && !skb_shinfo(skb)->tx_flags) {
2021 /* skb_tx_hash() wont be able to get sk.
2022 * We copy sk_hash into skb->rxhash
2025 skb->rxhash = sk->sk_hash;
2030 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2032 return ((features & NETIF_F_GEN_CSUM) ||
2033 ((features & NETIF_F_V4_CSUM) &&
2034 protocol == htons(ETH_P_IP)) ||
2035 ((features & NETIF_F_V6_CSUM) &&
2036 protocol == htons(ETH_P_IPV6)) ||
2037 ((features & NETIF_F_FCOE_CRC) &&
2038 protocol == htons(ETH_P_FCOE)));
2041 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2043 if (!can_checksum_protocol(features, protocol)) {
2044 features &= ~NETIF_F_ALL_CSUM;
2045 features &= ~NETIF_F_SG;
2046 } else if (illegal_highdma(skb->dev, skb)) {
2047 features &= ~NETIF_F_SG;
2053 u32 netif_skb_features(struct sk_buff *skb)
2055 __be16 protocol = skb->protocol;
2056 u32 features = skb->dev->features;
2058 if (protocol == htons(ETH_P_8021Q)) {
2059 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2060 protocol = veh->h_vlan_encapsulated_proto;
2061 } else if (!vlan_tx_tag_present(skb)) {
2062 return harmonize_features(skb, protocol, features);
2065 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2067 if (protocol != htons(ETH_P_8021Q)) {
2068 return harmonize_features(skb, protocol, features);
2070 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2071 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2072 return harmonize_features(skb, protocol, features);
2075 EXPORT_SYMBOL(netif_skb_features);
2078 * Returns true if either:
2079 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2080 * 2. skb is fragmented and the device does not support SG, or if
2081 * at least one of fragments is in highmem and device does not
2082 * support DMA from it.
2084 static inline int skb_needs_linearize(struct sk_buff *skb,
2087 return skb_is_nonlinear(skb) &&
2088 ((skb_has_frag_list(skb) &&
2089 !(features & NETIF_F_FRAGLIST)) ||
2090 (skb_shinfo(skb)->nr_frags &&
2091 !(features & NETIF_F_SG)));
2094 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2095 struct netdev_queue *txq)
2097 const struct net_device_ops *ops = dev->netdev_ops;
2098 int rc = NETDEV_TX_OK;
2099 unsigned int skb_len;
2101 if (likely(!skb->next)) {
2105 * If device doesn't need skb->dst, release it right now while
2106 * its hot in this cpu cache
2108 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2111 if (!list_empty(&ptype_all))
2112 dev_queue_xmit_nit(skb, dev);
2114 skb_orphan_try(skb);
2116 features = netif_skb_features(skb);
2118 if (vlan_tx_tag_present(skb) &&
2119 !(features & NETIF_F_HW_VLAN_TX)) {
2120 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2127 if (netif_needs_gso(skb, features)) {
2128 if (unlikely(dev_gso_segment(skb, features)))
2133 if (skb_needs_linearize(skb, features) &&
2134 __skb_linearize(skb))
2137 /* If packet is not checksummed and device does not
2138 * support checksumming for this protocol, complete
2139 * checksumming here.
2141 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2142 skb_set_transport_header(skb,
2143 skb_checksum_start_offset(skb));
2144 if (!(features & NETIF_F_ALL_CSUM) &&
2145 skb_checksum_help(skb))
2151 rc = ops->ndo_start_xmit(skb, dev);
2152 trace_net_dev_xmit(skb, rc, dev, skb_len);
2153 if (rc == NETDEV_TX_OK)
2154 txq_trans_update(txq);
2160 struct sk_buff *nskb = skb->next;
2162 skb->next = nskb->next;
2166 * If device doesn't need nskb->dst, release it right now while
2167 * its hot in this cpu cache
2169 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2172 skb_len = nskb->len;
2173 rc = ops->ndo_start_xmit(nskb, dev);
2174 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2175 if (unlikely(rc != NETDEV_TX_OK)) {
2176 if (rc & ~NETDEV_TX_MASK)
2177 goto out_kfree_gso_skb;
2178 nskb->next = skb->next;
2182 txq_trans_update(txq);
2183 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2184 return NETDEV_TX_BUSY;
2185 } while (skb->next);
2188 if (likely(skb->next == NULL))
2189 skb->destructor = DEV_GSO_CB(skb)->destructor;
2196 static u32 hashrnd __read_mostly;
2199 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2200 * to be used as a distribution range.
2202 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2203 unsigned int num_tx_queues)
2207 u16 qcount = num_tx_queues;
2209 if (skb_rx_queue_recorded(skb)) {
2210 hash = skb_get_rx_queue(skb);
2211 while (unlikely(hash >= num_tx_queues))
2212 hash -= num_tx_queues;
2217 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2218 qoffset = dev->tc_to_txq[tc].offset;
2219 qcount = dev->tc_to_txq[tc].count;
2222 if (skb->sk && skb->sk->sk_hash)
2223 hash = skb->sk->sk_hash;
2225 hash = (__force u16) skb->protocol ^ skb->rxhash;
2226 hash = jhash_1word(hash, hashrnd);
2228 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2230 EXPORT_SYMBOL(__skb_tx_hash);
2232 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2234 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2235 if (net_ratelimit()) {
2236 pr_warning("%s selects TX queue %d, but "
2237 "real number of TX queues is %d\n",
2238 dev->name, queue_index, dev->real_num_tx_queues);
2245 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2248 struct xps_dev_maps *dev_maps;
2249 struct xps_map *map;
2250 int queue_index = -1;
2253 dev_maps = rcu_dereference(dev->xps_maps);
2255 map = rcu_dereference(
2256 dev_maps->cpu_map[raw_smp_processor_id()]);
2259 queue_index = map->queues[0];
2262 if (skb->sk && skb->sk->sk_hash)
2263 hash = skb->sk->sk_hash;
2265 hash = (__force u16) skb->protocol ^
2267 hash = jhash_1word(hash, hashrnd);
2268 queue_index = map->queues[
2269 ((u64)hash * map->len) >> 32];
2271 if (unlikely(queue_index >= dev->real_num_tx_queues))
2283 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2284 struct sk_buff *skb)
2287 const struct net_device_ops *ops = dev->netdev_ops;
2289 if (dev->real_num_tx_queues == 1)
2291 else if (ops->ndo_select_queue) {
2292 queue_index = ops->ndo_select_queue(dev, skb);
2293 queue_index = dev_cap_txqueue(dev, queue_index);
2295 struct sock *sk = skb->sk;
2296 queue_index = sk_tx_queue_get(sk);
2298 if (queue_index < 0 || skb->ooo_okay ||
2299 queue_index >= dev->real_num_tx_queues) {
2300 int old_index = queue_index;
2302 queue_index = get_xps_queue(dev, skb);
2303 if (queue_index < 0)
2304 queue_index = skb_tx_hash(dev, skb);
2306 if (queue_index != old_index && sk) {
2307 struct dst_entry *dst =
2308 rcu_dereference_check(sk->sk_dst_cache, 1);
2310 if (dst && skb_dst(skb) == dst)
2311 sk_tx_queue_set(sk, queue_index);
2316 skb_set_queue_mapping(skb, queue_index);
2317 return netdev_get_tx_queue(dev, queue_index);
2320 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2321 struct net_device *dev,
2322 struct netdev_queue *txq)
2324 spinlock_t *root_lock = qdisc_lock(q);
2328 qdisc_skb_cb(skb)->pkt_len = skb->len;
2329 qdisc_calculate_pkt_len(skb, q);
2331 * Heuristic to force contended enqueues to serialize on a
2332 * separate lock before trying to get qdisc main lock.
2333 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2334 * and dequeue packets faster.
2336 contended = qdisc_is_running(q);
2337 if (unlikely(contended))
2338 spin_lock(&q->busylock);
2340 spin_lock(root_lock);
2341 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2344 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2345 qdisc_run_begin(q)) {
2347 * This is a work-conserving queue; there are no old skbs
2348 * waiting to be sent out; and the qdisc is not running -
2349 * xmit the skb directly.
2351 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2354 qdisc_bstats_update(q, skb);
2356 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2357 if (unlikely(contended)) {
2358 spin_unlock(&q->busylock);
2365 rc = NET_XMIT_SUCCESS;
2368 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2369 if (qdisc_run_begin(q)) {
2370 if (unlikely(contended)) {
2371 spin_unlock(&q->busylock);
2377 spin_unlock(root_lock);
2378 if (unlikely(contended))
2379 spin_unlock(&q->busylock);
2383 static DEFINE_PER_CPU(int, xmit_recursion);
2384 #define RECURSION_LIMIT 10
2387 * dev_queue_xmit - transmit a buffer
2388 * @skb: buffer to transmit
2390 * Queue a buffer for transmission to a network device. The caller must
2391 * have set the device and priority and built the buffer before calling
2392 * this function. The function can be called from an interrupt.
2394 * A negative errno code is returned on a failure. A success does not
2395 * guarantee the frame will be transmitted as it may be dropped due
2396 * to congestion or traffic shaping.
2398 * -----------------------------------------------------------------------------------
2399 * I notice this method can also return errors from the queue disciplines,
2400 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2403 * Regardless of the return value, the skb is consumed, so it is currently
2404 * difficult to retry a send to this method. (You can bump the ref count
2405 * before sending to hold a reference for retry if you are careful.)
2407 * When calling this method, interrupts MUST be enabled. This is because
2408 * the BH enable code must have IRQs enabled so that it will not deadlock.
2411 int dev_queue_xmit(struct sk_buff *skb)
2413 struct net_device *dev = skb->dev;
2414 struct netdev_queue *txq;
2418 /* Disable soft irqs for various locks below. Also
2419 * stops preemption for RCU.
2423 txq = dev_pick_tx(dev, skb);
2424 q = rcu_dereference_bh(txq->qdisc);
2426 #ifdef CONFIG_NET_CLS_ACT
2427 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2429 trace_net_dev_queue(skb);
2431 rc = __dev_xmit_skb(skb, q, dev, txq);
2435 /* The device has no queue. Common case for software devices:
2436 loopback, all the sorts of tunnels...
2438 Really, it is unlikely that netif_tx_lock protection is necessary
2439 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2441 However, it is possible, that they rely on protection
2444 Check this and shot the lock. It is not prone from deadlocks.
2445 Either shot noqueue qdisc, it is even simpler 8)
2447 if (dev->flags & IFF_UP) {
2448 int cpu = smp_processor_id(); /* ok because BHs are off */
2450 if (txq->xmit_lock_owner != cpu) {
2452 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2453 goto recursion_alert;
2455 HARD_TX_LOCK(dev, txq, cpu);
2457 if (!netif_tx_queue_stopped(txq)) {
2458 __this_cpu_inc(xmit_recursion);
2459 rc = dev_hard_start_xmit(skb, dev, txq);
2460 __this_cpu_dec(xmit_recursion);
2461 if (dev_xmit_complete(rc)) {
2462 HARD_TX_UNLOCK(dev, txq);
2466 HARD_TX_UNLOCK(dev, txq);
2467 if (net_ratelimit())
2468 printk(KERN_CRIT "Virtual device %s asks to "
2469 "queue packet!\n", dev->name);
2471 /* Recursion is detected! It is possible,
2475 if (net_ratelimit())
2476 printk(KERN_CRIT "Dead loop on virtual device "
2477 "%s, fix it urgently!\n", dev->name);
2482 rcu_read_unlock_bh();
2487 rcu_read_unlock_bh();
2490 EXPORT_SYMBOL(dev_queue_xmit);
2493 /*=======================================================================
2495 =======================================================================*/
2497 int netdev_max_backlog __read_mostly = 1000;
2498 int netdev_tstamp_prequeue __read_mostly = 1;
2499 int netdev_budget __read_mostly = 300;
2500 int weight_p __read_mostly = 64; /* old backlog weight */
2502 /* Called with irq disabled */
2503 static inline void ____napi_schedule(struct softnet_data *sd,
2504 struct napi_struct *napi)
2506 list_add_tail(&napi->poll_list, &sd->poll_list);
2507 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2511 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2512 * and src/dst port numbers. Returns a non-zero hash number on success
2515 __u32 __skb_get_rxhash(struct sk_buff *skb)
2517 int nhoff, hash = 0, poff;
2518 const struct ipv6hdr *ip6;
2519 const struct iphdr *ip;
2521 u32 addr1, addr2, ihl;
2527 nhoff = skb_network_offset(skb);
2529 switch (skb->protocol) {
2530 case __constant_htons(ETH_P_IP):
2531 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2534 ip = (const struct iphdr *) (skb->data + nhoff);
2535 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2538 ip_proto = ip->protocol;
2539 addr1 = (__force u32) ip->saddr;
2540 addr2 = (__force u32) ip->daddr;
2543 case __constant_htons(ETH_P_IPV6):
2544 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2547 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2548 ip_proto = ip6->nexthdr;
2549 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2550 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2558 poff = proto_ports_offset(ip_proto);
2560 nhoff += ihl * 4 + poff;
2561 if (pskb_may_pull(skb, nhoff + 4)) {
2562 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2563 if (ports.v16[1] < ports.v16[0])
2564 swap(ports.v16[0], ports.v16[1]);
2568 /* get a consistent hash (same value on both flow directions) */
2572 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2579 EXPORT_SYMBOL(__skb_get_rxhash);
2583 /* One global table that all flow-based protocols share. */
2584 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2585 EXPORT_SYMBOL(rps_sock_flow_table);
2587 static struct rps_dev_flow *
2588 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2589 struct rps_dev_flow *rflow, u16 next_cpu)
2593 tcpu = rflow->cpu = next_cpu;
2594 if (tcpu != RPS_NO_CPU) {
2595 #ifdef CONFIG_RFS_ACCEL
2596 struct netdev_rx_queue *rxqueue;
2597 struct rps_dev_flow_table *flow_table;
2598 struct rps_dev_flow *old_rflow;
2603 /* Should we steer this flow to a different hardware queue? */
2604 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2605 !(dev->features & NETIF_F_NTUPLE))
2607 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2608 if (rxq_index == skb_get_rx_queue(skb))
2611 rxqueue = dev->_rx + rxq_index;
2612 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2615 flow_id = skb->rxhash & flow_table->mask;
2616 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2617 rxq_index, flow_id);
2621 rflow = &flow_table->flows[flow_id];
2622 rflow->cpu = next_cpu;
2624 if (old_rflow->filter == rflow->filter)
2625 old_rflow->filter = RPS_NO_FILTER;
2629 per_cpu(softnet_data, tcpu).input_queue_head;
2636 * get_rps_cpu is called from netif_receive_skb and returns the target
2637 * CPU from the RPS map of the receiving queue for a given skb.
2638 * rcu_read_lock must be held on entry.
2640 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2641 struct rps_dev_flow **rflowp)
2643 struct netdev_rx_queue *rxqueue;
2644 struct rps_map *map;
2645 struct rps_dev_flow_table *flow_table;
2646 struct rps_sock_flow_table *sock_flow_table;
2650 if (skb_rx_queue_recorded(skb)) {
2651 u16 index = skb_get_rx_queue(skb);
2652 if (unlikely(index >= dev->real_num_rx_queues)) {
2653 WARN_ONCE(dev->real_num_rx_queues > 1,
2654 "%s received packet on queue %u, but number "
2655 "of RX queues is %u\n",
2656 dev->name, index, dev->real_num_rx_queues);
2659 rxqueue = dev->_rx + index;
2663 map = rcu_dereference(rxqueue->rps_map);
2665 if (map->len == 1 &&
2666 !rcu_dereference_raw(rxqueue->rps_flow_table)) {
2667 tcpu = map->cpus[0];
2668 if (cpu_online(tcpu))
2672 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2676 skb_reset_network_header(skb);
2677 if (!skb_get_rxhash(skb))
2680 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2681 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2682 if (flow_table && sock_flow_table) {
2684 struct rps_dev_flow *rflow;
2686 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2689 next_cpu = sock_flow_table->ents[skb->rxhash &
2690 sock_flow_table->mask];
2693 * If the desired CPU (where last recvmsg was done) is
2694 * different from current CPU (one in the rx-queue flow
2695 * table entry), switch if one of the following holds:
2696 * - Current CPU is unset (equal to RPS_NO_CPU).
2697 * - Current CPU is offline.
2698 * - The current CPU's queue tail has advanced beyond the
2699 * last packet that was enqueued using this table entry.
2700 * This guarantees that all previous packets for the flow
2701 * have been dequeued, thus preserving in order delivery.
2703 if (unlikely(tcpu != next_cpu) &&
2704 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2705 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2706 rflow->last_qtail)) >= 0))
2707 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2709 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2717 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2719 if (cpu_online(tcpu)) {
2729 #ifdef CONFIG_RFS_ACCEL
2732 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2733 * @dev: Device on which the filter was set
2734 * @rxq_index: RX queue index
2735 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2736 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2738 * Drivers that implement ndo_rx_flow_steer() should periodically call
2739 * this function for each installed filter and remove the filters for
2740 * which it returns %true.
2742 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2743 u32 flow_id, u16 filter_id)
2745 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2746 struct rps_dev_flow_table *flow_table;
2747 struct rps_dev_flow *rflow;
2752 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2753 if (flow_table && flow_id <= flow_table->mask) {
2754 rflow = &flow_table->flows[flow_id];
2755 cpu = ACCESS_ONCE(rflow->cpu);
2756 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2757 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2758 rflow->last_qtail) <
2759 (int)(10 * flow_table->mask)))
2765 EXPORT_SYMBOL(rps_may_expire_flow);
2767 #endif /* CONFIG_RFS_ACCEL */
2769 /* Called from hardirq (IPI) context */
2770 static void rps_trigger_softirq(void *data)
2772 struct softnet_data *sd = data;
2774 ____napi_schedule(sd, &sd->backlog);
2778 #endif /* CONFIG_RPS */
2781 * Check if this softnet_data structure is another cpu one
2782 * If yes, queue it to our IPI list and return 1
2785 static int rps_ipi_queued(struct softnet_data *sd)
2788 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2791 sd->rps_ipi_next = mysd->rps_ipi_list;
2792 mysd->rps_ipi_list = sd;
2794 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2797 #endif /* CONFIG_RPS */
2802 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2803 * queue (may be a remote CPU queue).
2805 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2806 unsigned int *qtail)
2808 struct softnet_data *sd;
2809 unsigned long flags;
2811 sd = &per_cpu(softnet_data, cpu);
2813 local_irq_save(flags);
2816 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2817 if (skb_queue_len(&sd->input_pkt_queue)) {
2819 __skb_queue_tail(&sd->input_pkt_queue, skb);
2820 input_queue_tail_incr_save(sd, qtail);
2822 local_irq_restore(flags);
2823 return NET_RX_SUCCESS;
2826 /* Schedule NAPI for backlog device
2827 * We can use non atomic operation since we own the queue lock
2829 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2830 if (!rps_ipi_queued(sd))
2831 ____napi_schedule(sd, &sd->backlog);
2839 local_irq_restore(flags);
2841 atomic_long_inc(&skb->dev->rx_dropped);
2847 * netif_rx - post buffer to the network code
2848 * @skb: buffer to post
2850 * This function receives a packet from a device driver and queues it for
2851 * the upper (protocol) levels to process. It always succeeds. The buffer
2852 * may be dropped during processing for congestion control or by the
2856 * NET_RX_SUCCESS (no congestion)
2857 * NET_RX_DROP (packet was dropped)
2861 int netif_rx(struct sk_buff *skb)
2865 /* if netpoll wants it, pretend we never saw it */
2866 if (netpoll_rx(skb))
2869 if (netdev_tstamp_prequeue)
2870 net_timestamp_check(skb);
2872 trace_netif_rx(skb);
2875 struct rps_dev_flow voidflow, *rflow = &voidflow;
2881 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2883 cpu = smp_processor_id();
2885 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2893 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2899 EXPORT_SYMBOL(netif_rx);
2901 int netif_rx_ni(struct sk_buff *skb)
2906 err = netif_rx(skb);
2907 if (local_softirq_pending())
2913 EXPORT_SYMBOL(netif_rx_ni);
2915 static void net_tx_action(struct softirq_action *h)
2917 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2919 if (sd->completion_queue) {
2920 struct sk_buff *clist;
2922 local_irq_disable();
2923 clist = sd->completion_queue;
2924 sd->completion_queue = NULL;
2928 struct sk_buff *skb = clist;
2929 clist = clist->next;
2931 WARN_ON(atomic_read(&skb->users));
2932 trace_kfree_skb(skb, net_tx_action);
2937 if (sd->output_queue) {
2940 local_irq_disable();
2941 head = sd->output_queue;
2942 sd->output_queue = NULL;
2943 sd->output_queue_tailp = &sd->output_queue;
2947 struct Qdisc *q = head;
2948 spinlock_t *root_lock;
2950 head = head->next_sched;
2952 root_lock = qdisc_lock(q);
2953 if (spin_trylock(root_lock)) {
2954 smp_mb__before_clear_bit();
2955 clear_bit(__QDISC_STATE_SCHED,
2958 spin_unlock(root_lock);
2960 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2962 __netif_reschedule(q);
2964 smp_mb__before_clear_bit();
2965 clear_bit(__QDISC_STATE_SCHED,
2973 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2974 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2975 /* This hook is defined here for ATM LANE */
2976 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2977 unsigned char *addr) __read_mostly;
2978 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2981 #ifdef CONFIG_NET_CLS_ACT
2982 /* TODO: Maybe we should just force sch_ingress to be compiled in
2983 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2984 * a compare and 2 stores extra right now if we dont have it on
2985 * but have CONFIG_NET_CLS_ACT
2986 * NOTE: This doesn't stop any functionality; if you dont have
2987 * the ingress scheduler, you just can't add policies on ingress.
2990 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2992 struct net_device *dev = skb->dev;
2993 u32 ttl = G_TC_RTTL(skb->tc_verd);
2994 int result = TC_ACT_OK;
2997 if (unlikely(MAX_RED_LOOP < ttl++)) {
2998 if (net_ratelimit())
2999 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3000 skb->skb_iif, dev->ifindex);
3004 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3005 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3008 if (q != &noop_qdisc) {
3009 spin_lock(qdisc_lock(q));
3010 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3011 result = qdisc_enqueue_root(skb, q);
3012 spin_unlock(qdisc_lock(q));
3018 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3019 struct packet_type **pt_prev,
3020 int *ret, struct net_device *orig_dev)
3022 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3024 if (!rxq || rxq->qdisc == &noop_qdisc)
3028 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3032 switch (ing_filter(skb, rxq)) {
3046 * netdev_rx_handler_register - register receive handler
3047 * @dev: device to register a handler for
3048 * @rx_handler: receive handler to register
3049 * @rx_handler_data: data pointer that is used by rx handler
3051 * Register a receive hander for a device. This handler will then be
3052 * called from __netif_receive_skb. A negative errno code is returned
3055 * The caller must hold the rtnl_mutex.
3057 * For a general description of rx_handler, see enum rx_handler_result.
3059 int netdev_rx_handler_register(struct net_device *dev,
3060 rx_handler_func_t *rx_handler,
3061 void *rx_handler_data)
3065 if (dev->rx_handler)
3068 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3069 rcu_assign_pointer(dev->rx_handler, rx_handler);
3073 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3076 * netdev_rx_handler_unregister - unregister receive handler
3077 * @dev: device to unregister a handler from
3079 * Unregister a receive hander from a device.
3081 * The caller must hold the rtnl_mutex.
3083 void netdev_rx_handler_unregister(struct net_device *dev)
3087 rcu_assign_pointer(dev->rx_handler, NULL);
3088 rcu_assign_pointer(dev->rx_handler_data, NULL);
3090 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3092 static int __netif_receive_skb(struct sk_buff *skb)
3094 struct packet_type *ptype, *pt_prev;
3095 rx_handler_func_t *rx_handler;
3096 struct net_device *orig_dev;
3097 struct net_device *null_or_dev;
3098 bool deliver_exact = false;
3099 int ret = NET_RX_DROP;
3102 if (!netdev_tstamp_prequeue)
3103 net_timestamp_check(skb);
3105 trace_netif_receive_skb(skb);
3107 /* if we've gotten here through NAPI, check netpoll */
3108 if (netpoll_receive_skb(skb))
3112 skb->skb_iif = skb->dev->ifindex;
3113 orig_dev = skb->dev;
3115 skb_reset_network_header(skb);
3116 skb_reset_transport_header(skb);
3117 skb_reset_mac_len(skb);
3125 __this_cpu_inc(softnet_data.processed);
3127 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3128 skb = vlan_untag(skb);
3133 #ifdef CONFIG_NET_CLS_ACT
3134 if (skb->tc_verd & TC_NCLS) {
3135 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3140 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3141 if (!ptype->dev || ptype->dev == skb->dev) {
3143 ret = deliver_skb(skb, pt_prev, orig_dev);
3148 #ifdef CONFIG_NET_CLS_ACT
3149 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3155 rx_handler = rcu_dereference(skb->dev->rx_handler);
3158 ret = deliver_skb(skb, pt_prev, orig_dev);
3161 switch (rx_handler(&skb)) {
3162 case RX_HANDLER_CONSUMED:
3164 case RX_HANDLER_ANOTHER:
3166 case RX_HANDLER_EXACT:
3167 deliver_exact = true;
3168 case RX_HANDLER_PASS:
3175 if (vlan_tx_tag_present(skb)) {
3177 ret = deliver_skb(skb, pt_prev, orig_dev);
3180 if (vlan_do_receive(&skb)) {
3181 ret = __netif_receive_skb(skb);
3183 } else if (unlikely(!skb))
3187 /* deliver only exact match when indicated */
3188 null_or_dev = deliver_exact ? skb->dev : NULL;
3190 type = skb->protocol;
3191 list_for_each_entry_rcu(ptype,
3192 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3193 if (ptype->type == type &&
3194 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3195 ptype->dev == orig_dev)) {
3197 ret = deliver_skb(skb, pt_prev, orig_dev);
3203 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3205 atomic_long_inc(&skb->dev->rx_dropped);
3207 /* Jamal, now you will not able to escape explaining
3208 * me how you were going to use this. :-)
3219 * netif_receive_skb - process receive buffer from network
3220 * @skb: buffer to process
3222 * netif_receive_skb() is the main receive data processing function.
3223 * It always succeeds. The buffer may be dropped during processing
3224 * for congestion control or by the protocol layers.
3226 * This function may only be called from softirq context and interrupts
3227 * should be enabled.
3229 * Return values (usually ignored):
3230 * NET_RX_SUCCESS: no congestion
3231 * NET_RX_DROP: packet was dropped
3233 int netif_receive_skb(struct sk_buff *skb)
3235 if (netdev_tstamp_prequeue)
3236 net_timestamp_check(skb);
3238 if (skb_defer_rx_timestamp(skb))
3239 return NET_RX_SUCCESS;
3243 struct rps_dev_flow voidflow, *rflow = &voidflow;
3248 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3251 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3255 ret = __netif_receive_skb(skb);
3261 return __netif_receive_skb(skb);
3264 EXPORT_SYMBOL(netif_receive_skb);
3266 /* Network device is going away, flush any packets still pending
3267 * Called with irqs disabled.
3269 static void flush_backlog(void *arg)
3271 struct net_device *dev = arg;
3272 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3273 struct sk_buff *skb, *tmp;
3276 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3277 if (skb->dev == dev) {
3278 __skb_unlink(skb, &sd->input_pkt_queue);
3280 input_queue_head_incr(sd);
3285 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3286 if (skb->dev == dev) {
3287 __skb_unlink(skb, &sd->process_queue);
3289 input_queue_head_incr(sd);
3294 static int napi_gro_complete(struct sk_buff *skb)
3296 struct packet_type *ptype;
3297 __be16 type = skb->protocol;
3298 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3301 if (NAPI_GRO_CB(skb)->count == 1) {
3302 skb_shinfo(skb)->gso_size = 0;
3307 list_for_each_entry_rcu(ptype, head, list) {
3308 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3311 err = ptype->gro_complete(skb);
3317 WARN_ON(&ptype->list == head);
3319 return NET_RX_SUCCESS;
3323 return netif_receive_skb(skb);
3326 inline void napi_gro_flush(struct napi_struct *napi)
3328 struct sk_buff *skb, *next;
3330 for (skb = napi->gro_list; skb; skb = next) {
3333 napi_gro_complete(skb);
3336 napi->gro_count = 0;
3337 napi->gro_list = NULL;
3339 EXPORT_SYMBOL(napi_gro_flush);
3341 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3343 struct sk_buff **pp = NULL;
3344 struct packet_type *ptype;
3345 __be16 type = skb->protocol;
3346 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3349 enum gro_result ret;
3351 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3354 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3358 list_for_each_entry_rcu(ptype, head, list) {
3359 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3362 skb_set_network_header(skb, skb_gro_offset(skb));
3363 mac_len = skb->network_header - skb->mac_header;
3364 skb->mac_len = mac_len;
3365 NAPI_GRO_CB(skb)->same_flow = 0;
3366 NAPI_GRO_CB(skb)->flush = 0;
3367 NAPI_GRO_CB(skb)->free = 0;
3369 pp = ptype->gro_receive(&napi->gro_list, skb);
3374 if (&ptype->list == head)
3377 same_flow = NAPI_GRO_CB(skb)->same_flow;
3378 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3381 struct sk_buff *nskb = *pp;
3385 napi_gro_complete(nskb);
3392 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3396 NAPI_GRO_CB(skb)->count = 1;
3397 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3398 skb->next = napi->gro_list;
3399 napi->gro_list = skb;
3403 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3404 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3406 BUG_ON(skb->end - skb->tail < grow);
3408 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3411 skb->data_len -= grow;
3413 skb_shinfo(skb)->frags[0].page_offset += grow;
3414 skb_shinfo(skb)->frags[0].size -= grow;
3416 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3417 put_page(skb_shinfo(skb)->frags[0].page);
3418 memmove(skb_shinfo(skb)->frags,
3419 skb_shinfo(skb)->frags + 1,
3420 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3431 EXPORT_SYMBOL(dev_gro_receive);
3433 static inline gro_result_t
3434 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3437 unsigned int maclen = skb->dev->hard_header_len;
3439 for (p = napi->gro_list; p; p = p->next) {
3440 unsigned long diffs;
3442 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3443 diffs |= p->vlan_tci ^ skb->vlan_tci;
3444 if (maclen == ETH_HLEN)
3445 diffs |= compare_ether_header(skb_mac_header(p),
3446 skb_gro_mac_header(skb));
3448 diffs = memcmp(skb_mac_header(p),
3449 skb_gro_mac_header(skb),
3451 NAPI_GRO_CB(p)->same_flow = !diffs;
3452 NAPI_GRO_CB(p)->flush = 0;
3455 return dev_gro_receive(napi, skb);
3458 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3462 if (netif_receive_skb(skb))
3467 case GRO_MERGED_FREE:
3478 EXPORT_SYMBOL(napi_skb_finish);
3480 void skb_gro_reset_offset(struct sk_buff *skb)
3482 NAPI_GRO_CB(skb)->data_offset = 0;
3483 NAPI_GRO_CB(skb)->frag0 = NULL;
3484 NAPI_GRO_CB(skb)->frag0_len = 0;
3486 if (skb->mac_header == skb->tail &&
3487 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3488 NAPI_GRO_CB(skb)->frag0 =
3489 page_address(skb_shinfo(skb)->frags[0].page) +
3490 skb_shinfo(skb)->frags[0].page_offset;
3491 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3494 EXPORT_SYMBOL(skb_gro_reset_offset);
3496 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3498 skb_gro_reset_offset(skb);
3500 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3502 EXPORT_SYMBOL(napi_gro_receive);
3504 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3506 __skb_pull(skb, skb_headlen(skb));
3507 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3508 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3510 skb->dev = napi->dev;
3516 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3518 struct sk_buff *skb = napi->skb;
3521 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3527 EXPORT_SYMBOL(napi_get_frags);
3529 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3535 skb->protocol = eth_type_trans(skb, skb->dev);
3537 if (ret == GRO_HELD)
3538 skb_gro_pull(skb, -ETH_HLEN);
3539 else if (netif_receive_skb(skb))
3544 case GRO_MERGED_FREE:
3545 napi_reuse_skb(napi, skb);
3554 EXPORT_SYMBOL(napi_frags_finish);
3556 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3558 struct sk_buff *skb = napi->skb;
3565 skb_reset_mac_header(skb);
3566 skb_gro_reset_offset(skb);
3568 off = skb_gro_offset(skb);
3569 hlen = off + sizeof(*eth);
3570 eth = skb_gro_header_fast(skb, off);
3571 if (skb_gro_header_hard(skb, hlen)) {
3572 eth = skb_gro_header_slow(skb, hlen, off);
3573 if (unlikely(!eth)) {
3574 napi_reuse_skb(napi, skb);
3580 skb_gro_pull(skb, sizeof(*eth));
3583 * This works because the only protocols we care about don't require
3584 * special handling. We'll fix it up properly at the end.
3586 skb->protocol = eth->h_proto;
3591 EXPORT_SYMBOL(napi_frags_skb);
3593 gro_result_t napi_gro_frags(struct napi_struct *napi)
3595 struct sk_buff *skb = napi_frags_skb(napi);
3600 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3602 EXPORT_SYMBOL(napi_gro_frags);
3605 * net_rps_action sends any pending IPI's for rps.
3606 * Note: called with local irq disabled, but exits with local irq enabled.
3608 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3611 struct softnet_data *remsd = sd->rps_ipi_list;
3614 sd->rps_ipi_list = NULL;
3618 /* Send pending IPI's to kick RPS processing on remote cpus. */
3620 struct softnet_data *next = remsd->rps_ipi_next;
3622 if (cpu_online(remsd->cpu))
3623 __smp_call_function_single(remsd->cpu,
3632 static int process_backlog(struct napi_struct *napi, int quota)
3635 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3638 /* Check if we have pending ipi, its better to send them now,
3639 * not waiting net_rx_action() end.
3641 if (sd->rps_ipi_list) {
3642 local_irq_disable();
3643 net_rps_action_and_irq_enable(sd);
3646 napi->weight = weight_p;
3647 local_irq_disable();
3648 while (work < quota) {
3649 struct sk_buff *skb;
3652 while ((skb = __skb_dequeue(&sd->process_queue))) {
3654 __netif_receive_skb(skb);
3655 local_irq_disable();
3656 input_queue_head_incr(sd);
3657 if (++work >= quota) {
3664 qlen = skb_queue_len(&sd->input_pkt_queue);
3666 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3667 &sd->process_queue);
3669 if (qlen < quota - work) {
3671 * Inline a custom version of __napi_complete().
3672 * only current cpu owns and manipulates this napi,
3673 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3674 * we can use a plain write instead of clear_bit(),
3675 * and we dont need an smp_mb() memory barrier.
3677 list_del(&napi->poll_list);
3680 quota = work + qlen;
3690 * __napi_schedule - schedule for receive
3691 * @n: entry to schedule
3693 * The entry's receive function will be scheduled to run
3695 void __napi_schedule(struct napi_struct *n)
3697 unsigned long flags;
3699 local_irq_save(flags);
3700 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3701 local_irq_restore(flags);
3703 EXPORT_SYMBOL(__napi_schedule);
3705 void __napi_complete(struct napi_struct *n)
3707 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3708 BUG_ON(n->gro_list);
3710 list_del(&n->poll_list);
3711 smp_mb__before_clear_bit();
3712 clear_bit(NAPI_STATE_SCHED, &n->state);
3714 EXPORT_SYMBOL(__napi_complete);
3716 void napi_complete(struct napi_struct *n)
3718 unsigned long flags;
3721 * don't let napi dequeue from the cpu poll list
3722 * just in case its running on a different cpu
3724 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3728 local_irq_save(flags);
3730 local_irq_restore(flags);
3732 EXPORT_SYMBOL(napi_complete);
3734 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3735 int (*poll)(struct napi_struct *, int), int weight)
3737 INIT_LIST_HEAD(&napi->poll_list);
3738 napi->gro_count = 0;
3739 napi->gro_list = NULL;
3742 napi->weight = weight;
3743 list_add(&napi->dev_list, &dev->napi_list);
3745 #ifdef CONFIG_NETPOLL
3746 spin_lock_init(&napi->poll_lock);
3747 napi->poll_owner = -1;
3749 set_bit(NAPI_STATE_SCHED, &napi->state);
3751 EXPORT_SYMBOL(netif_napi_add);
3753 void netif_napi_del(struct napi_struct *napi)
3755 struct sk_buff *skb, *next;
3757 list_del_init(&napi->dev_list);
3758 napi_free_frags(napi);
3760 for (skb = napi->gro_list; skb; skb = next) {
3766 napi->gro_list = NULL;
3767 napi->gro_count = 0;
3769 EXPORT_SYMBOL(netif_napi_del);
3771 static void net_rx_action(struct softirq_action *h)
3773 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3774 unsigned long time_limit = jiffies + 2;
3775 int budget = netdev_budget;
3778 local_irq_disable();
3780 while (!list_empty(&sd->poll_list)) {
3781 struct napi_struct *n;
3784 /* If softirq window is exhuasted then punt.
3785 * Allow this to run for 2 jiffies since which will allow
3786 * an average latency of 1.5/HZ.
3788 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3793 /* Even though interrupts have been re-enabled, this
3794 * access is safe because interrupts can only add new
3795 * entries to the tail of this list, and only ->poll()
3796 * calls can remove this head entry from the list.
3798 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3800 have = netpoll_poll_lock(n);
3804 /* This NAPI_STATE_SCHED test is for avoiding a race
3805 * with netpoll's poll_napi(). Only the entity which
3806 * obtains the lock and sees NAPI_STATE_SCHED set will
3807 * actually make the ->poll() call. Therefore we avoid
3808 * accidentally calling ->poll() when NAPI is not scheduled.
3811 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3812 work = n->poll(n, weight);
3816 WARN_ON_ONCE(work > weight);
3820 local_irq_disable();
3822 /* Drivers must not modify the NAPI state if they
3823 * consume the entire weight. In such cases this code
3824 * still "owns" the NAPI instance and therefore can
3825 * move the instance around on the list at-will.
3827 if (unlikely(work == weight)) {
3828 if (unlikely(napi_disable_pending(n))) {
3831 local_irq_disable();
3833 list_move_tail(&n->poll_list, &sd->poll_list);
3836 netpoll_poll_unlock(have);
3839 net_rps_action_and_irq_enable(sd);
3841 #ifdef CONFIG_NET_DMA
3843 * There may not be any more sk_buffs coming right now, so push
3844 * any pending DMA copies to hardware
3846 dma_issue_pending_all();
3853 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3857 static gifconf_func_t *gifconf_list[NPROTO];
3860 * register_gifconf - register a SIOCGIF handler
3861 * @family: Address family
3862 * @gifconf: Function handler
3864 * Register protocol dependent address dumping routines. The handler
3865 * that is passed must not be freed or reused until it has been replaced
3866 * by another handler.
3868 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3870 if (family >= NPROTO)
3872 gifconf_list[family] = gifconf;
3875 EXPORT_SYMBOL(register_gifconf);
3879 * Map an interface index to its name (SIOCGIFNAME)
3883 * We need this ioctl for efficient implementation of the
3884 * if_indextoname() function required by the IPv6 API. Without
3885 * it, we would have to search all the interfaces to find a
3889 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3891 struct net_device *dev;
3895 * Fetch the caller's info block.
3898 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3902 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3908 strcpy(ifr.ifr_name, dev->name);
3911 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3917 * Perform a SIOCGIFCONF call. This structure will change
3918 * size eventually, and there is nothing I can do about it.
3919 * Thus we will need a 'compatibility mode'.
3922 static int dev_ifconf(struct net *net, char __user *arg)
3925 struct net_device *dev;
3932 * Fetch the caller's info block.
3935 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3942 * Loop over the interfaces, and write an info block for each.
3946 for_each_netdev(net, dev) {
3947 for (i = 0; i < NPROTO; i++) {
3948 if (gifconf_list[i]) {
3951 done = gifconf_list[i](dev, NULL, 0);
3953 done = gifconf_list[i](dev, pos + total,
3963 * All done. Write the updated control block back to the caller.
3965 ifc.ifc_len = total;
3968 * Both BSD and Solaris return 0 here, so we do too.
3970 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3973 #ifdef CONFIG_PROC_FS
3975 * This is invoked by the /proc filesystem handler to display a device
3978 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3981 struct net *net = seq_file_net(seq);
3983 struct net_device *dev;
3987 return SEQ_START_TOKEN;
3990 for_each_netdev_rcu(net, dev)
3997 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3999 struct net_device *dev = v;
4001 if (v == SEQ_START_TOKEN)
4002 dev = first_net_device_rcu(seq_file_net(seq));
4004 dev = next_net_device_rcu(dev);
4010 void dev_seq_stop(struct seq_file *seq, void *v)
4016 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4018 struct rtnl_link_stats64 temp;
4019 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4021 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4022 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4023 dev->name, stats->rx_bytes, stats->rx_packets,
4025 stats->rx_dropped + stats->rx_missed_errors,
4026 stats->rx_fifo_errors,
4027 stats->rx_length_errors + stats->rx_over_errors +
4028 stats->rx_crc_errors + stats->rx_frame_errors,
4029 stats->rx_compressed, stats->multicast,
4030 stats->tx_bytes, stats->tx_packets,
4031 stats->tx_errors, stats->tx_dropped,
4032 stats->tx_fifo_errors, stats->collisions,
4033 stats->tx_carrier_errors +
4034 stats->tx_aborted_errors +
4035 stats->tx_window_errors +
4036 stats->tx_heartbeat_errors,
4037 stats->tx_compressed);
4041 * Called from the PROCfs module. This now uses the new arbitrary sized
4042 * /proc/net interface to create /proc/net/dev
4044 static int dev_seq_show(struct seq_file *seq, void *v)
4046 if (v == SEQ_START_TOKEN)
4047 seq_puts(seq, "Inter-| Receive "
4049 " face |bytes packets errs drop fifo frame "
4050 "compressed multicast|bytes packets errs "
4051 "drop fifo colls carrier compressed\n");
4053 dev_seq_printf_stats(seq, v);
4057 static struct softnet_data *softnet_get_online(loff_t *pos)
4059 struct softnet_data *sd = NULL;
4061 while (*pos < nr_cpu_ids)
4062 if (cpu_online(*pos)) {
4063 sd = &per_cpu(softnet_data, *pos);
4070 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4072 return softnet_get_online(pos);
4075 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4078 return softnet_get_online(pos);
4081 static void softnet_seq_stop(struct seq_file *seq, void *v)
4085 static int softnet_seq_show(struct seq_file *seq, void *v)
4087 struct softnet_data *sd = v;
4089 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4090 sd->processed, sd->dropped, sd->time_squeeze, 0,
4091 0, 0, 0, 0, /* was fastroute */
4092 sd->cpu_collision, sd->received_rps);
4096 static const struct seq_operations dev_seq_ops = {
4097 .start = dev_seq_start,
4098 .next = dev_seq_next,
4099 .stop = dev_seq_stop,
4100 .show = dev_seq_show,
4103 static int dev_seq_open(struct inode *inode, struct file *file)
4105 return seq_open_net(inode, file, &dev_seq_ops,
4106 sizeof(struct seq_net_private));
4109 static const struct file_operations dev_seq_fops = {
4110 .owner = THIS_MODULE,
4111 .open = dev_seq_open,
4113 .llseek = seq_lseek,
4114 .release = seq_release_net,
4117 static const struct seq_operations softnet_seq_ops = {
4118 .start = softnet_seq_start,
4119 .next = softnet_seq_next,
4120 .stop = softnet_seq_stop,
4121 .show = softnet_seq_show,
4124 static int softnet_seq_open(struct inode *inode, struct file *file)
4126 return seq_open(file, &softnet_seq_ops);
4129 static const struct file_operations softnet_seq_fops = {
4130 .owner = THIS_MODULE,
4131 .open = softnet_seq_open,
4133 .llseek = seq_lseek,
4134 .release = seq_release,
4137 static void *ptype_get_idx(loff_t pos)
4139 struct packet_type *pt = NULL;
4143 list_for_each_entry_rcu(pt, &ptype_all, list) {
4149 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4150 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4159 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4163 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4166 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4168 struct packet_type *pt;
4169 struct list_head *nxt;
4173 if (v == SEQ_START_TOKEN)
4174 return ptype_get_idx(0);
4177 nxt = pt->list.next;
4178 if (pt->type == htons(ETH_P_ALL)) {
4179 if (nxt != &ptype_all)
4182 nxt = ptype_base[0].next;
4184 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4186 while (nxt == &ptype_base[hash]) {
4187 if (++hash >= PTYPE_HASH_SIZE)
4189 nxt = ptype_base[hash].next;
4192 return list_entry(nxt, struct packet_type, list);
4195 static void ptype_seq_stop(struct seq_file *seq, void *v)
4201 static int ptype_seq_show(struct seq_file *seq, void *v)
4203 struct packet_type *pt = v;
4205 if (v == SEQ_START_TOKEN)
4206 seq_puts(seq, "Type Device Function\n");
4207 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4208 if (pt->type == htons(ETH_P_ALL))
4209 seq_puts(seq, "ALL ");
4211 seq_printf(seq, "%04x", ntohs(pt->type));
4213 seq_printf(seq, " %-8s %pF\n",
4214 pt->dev ? pt->dev->name : "", pt->func);
4220 static const struct seq_operations ptype_seq_ops = {
4221 .start = ptype_seq_start,
4222 .next = ptype_seq_next,
4223 .stop = ptype_seq_stop,
4224 .show = ptype_seq_show,
4227 static int ptype_seq_open(struct inode *inode, struct file *file)
4229 return seq_open_net(inode, file, &ptype_seq_ops,
4230 sizeof(struct seq_net_private));
4233 static const struct file_operations ptype_seq_fops = {
4234 .owner = THIS_MODULE,
4235 .open = ptype_seq_open,
4237 .llseek = seq_lseek,
4238 .release = seq_release_net,
4242 static int __net_init dev_proc_net_init(struct net *net)
4246 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4248 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4250 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4253 if (wext_proc_init(net))
4259 proc_net_remove(net, "ptype");
4261 proc_net_remove(net, "softnet_stat");
4263 proc_net_remove(net, "dev");
4267 static void __net_exit dev_proc_net_exit(struct net *net)
4269 wext_proc_exit(net);
4271 proc_net_remove(net, "ptype");
4272 proc_net_remove(net, "softnet_stat");
4273 proc_net_remove(net, "dev");
4276 static struct pernet_operations __net_initdata dev_proc_ops = {
4277 .init = dev_proc_net_init,
4278 .exit = dev_proc_net_exit,
4281 static int __init dev_proc_init(void)
4283 return register_pernet_subsys(&dev_proc_ops);
4286 #define dev_proc_init() 0
4287 #endif /* CONFIG_PROC_FS */
4291 * netdev_set_master - set up master pointer
4292 * @slave: slave device
4293 * @master: new master device
4295 * Changes the master device of the slave. Pass %NULL to break the
4296 * bonding. The caller must hold the RTNL semaphore. On a failure
4297 * a negative errno code is returned. On success the reference counts
4298 * are adjusted and the function returns zero.
4300 int netdev_set_master(struct net_device *slave, struct net_device *master)
4302 struct net_device *old = slave->master;
4312 slave->master = master;
4318 EXPORT_SYMBOL(netdev_set_master);
4321 * netdev_set_bond_master - set up bonding master/slave pair
4322 * @slave: slave device
4323 * @master: new master device
4325 * Changes the master device of the slave. Pass %NULL to break the
4326 * bonding. The caller must hold the RTNL semaphore. On a failure
4327 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4328 * to the routing socket and the function returns zero.
4330 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4336 err = netdev_set_master(slave, master);
4340 slave->flags |= IFF_SLAVE;
4342 slave->flags &= ~IFF_SLAVE;
4344 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4347 EXPORT_SYMBOL(netdev_set_bond_master);
4349 static void dev_change_rx_flags(struct net_device *dev, int flags)
4351 const struct net_device_ops *ops = dev->netdev_ops;
4353 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4354 ops->ndo_change_rx_flags(dev, flags);
4357 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4359 unsigned short old_flags = dev->flags;
4365 dev->flags |= IFF_PROMISC;
4366 dev->promiscuity += inc;
4367 if (dev->promiscuity == 0) {
4370 * If inc causes overflow, untouch promisc and return error.
4373 dev->flags &= ~IFF_PROMISC;
4375 dev->promiscuity -= inc;
4376 printk(KERN_WARNING "%s: promiscuity touches roof, "
4377 "set promiscuity failed, promiscuity feature "
4378 "of device might be broken.\n", dev->name);
4382 if (dev->flags != old_flags) {
4383 printk(KERN_INFO "device %s %s promiscuous mode\n",
4384 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4386 if (audit_enabled) {
4387 current_uid_gid(&uid, &gid);
4388 audit_log(current->audit_context, GFP_ATOMIC,
4389 AUDIT_ANOM_PROMISCUOUS,
4390 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4391 dev->name, (dev->flags & IFF_PROMISC),
4392 (old_flags & IFF_PROMISC),
4393 audit_get_loginuid(current),
4395 audit_get_sessionid(current));
4398 dev_change_rx_flags(dev, IFF_PROMISC);
4404 * dev_set_promiscuity - update promiscuity count on a device
4408 * Add or remove promiscuity from a device. While the count in the device
4409 * remains above zero the interface remains promiscuous. Once it hits zero
4410 * the device reverts back to normal filtering operation. A negative inc
4411 * value is used to drop promiscuity on the device.
4412 * Return 0 if successful or a negative errno code on error.
4414 int dev_set_promiscuity(struct net_device *dev, int inc)
4416 unsigned short old_flags = dev->flags;
4419 err = __dev_set_promiscuity(dev, inc);
4422 if (dev->flags != old_flags)
4423 dev_set_rx_mode(dev);
4426 EXPORT_SYMBOL(dev_set_promiscuity);
4429 * dev_set_allmulti - update allmulti count on a device
4433 * Add or remove reception of all multicast frames to a device. While the
4434 * count in the device remains above zero the interface remains listening
4435 * to all interfaces. Once it hits zero the device reverts back to normal
4436 * filtering operation. A negative @inc value is used to drop the counter
4437 * when releasing a resource needing all multicasts.
4438 * Return 0 if successful or a negative errno code on error.
4441 int dev_set_allmulti(struct net_device *dev, int inc)
4443 unsigned short old_flags = dev->flags;
4447 dev->flags |= IFF_ALLMULTI;
4448 dev->allmulti += inc;
4449 if (dev->allmulti == 0) {
4452 * If inc causes overflow, untouch allmulti and return error.
4455 dev->flags &= ~IFF_ALLMULTI;
4457 dev->allmulti -= inc;
4458 printk(KERN_WARNING "%s: allmulti touches roof, "
4459 "set allmulti failed, allmulti feature of "
4460 "device might be broken.\n", dev->name);
4464 if (dev->flags ^ old_flags) {
4465 dev_change_rx_flags(dev, IFF_ALLMULTI);
4466 dev_set_rx_mode(dev);
4470 EXPORT_SYMBOL(dev_set_allmulti);
4473 * Upload unicast and multicast address lists to device and
4474 * configure RX filtering. When the device doesn't support unicast
4475 * filtering it is put in promiscuous mode while unicast addresses
4478 void __dev_set_rx_mode(struct net_device *dev)
4480 const struct net_device_ops *ops = dev->netdev_ops;
4482 /* dev_open will call this function so the list will stay sane. */
4483 if (!(dev->flags&IFF_UP))
4486 if (!netif_device_present(dev))
4489 if (ops->ndo_set_rx_mode)
4490 ops->ndo_set_rx_mode(dev);
4492 /* Unicast addresses changes may only happen under the rtnl,
4493 * therefore calling __dev_set_promiscuity here is safe.
4495 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4496 __dev_set_promiscuity(dev, 1);
4497 dev->uc_promisc = 1;
4498 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4499 __dev_set_promiscuity(dev, -1);
4500 dev->uc_promisc = 0;
4503 if (ops->ndo_set_multicast_list)
4504 ops->ndo_set_multicast_list(dev);
4508 void dev_set_rx_mode(struct net_device *dev)
4510 netif_addr_lock_bh(dev);
4511 __dev_set_rx_mode(dev);
4512 netif_addr_unlock_bh(dev);
4516 * dev_ethtool_get_settings - call device's ethtool_ops::get_settings()
4518 * @cmd: memory area for ethtool_ops::get_settings() result
4520 * The cmd arg is initialized properly (cleared and
4521 * ethtool_cmd::cmd field set to ETHTOOL_GSET).
4523 * Return device's ethtool_ops::get_settings() result value or
4524 * -EOPNOTSUPP when device doesn't expose
4525 * ethtool_ops::get_settings() operation.
4527 int dev_ethtool_get_settings(struct net_device *dev,
4528 struct ethtool_cmd *cmd)
4530 if (!dev->ethtool_ops || !dev->ethtool_ops->get_settings)
4533 memset(cmd, 0, sizeof(struct ethtool_cmd));
4534 cmd->cmd = ETHTOOL_GSET;
4535 return dev->ethtool_ops->get_settings(dev, cmd);
4537 EXPORT_SYMBOL(dev_ethtool_get_settings);
4540 * dev_get_flags - get flags reported to userspace
4543 * Get the combination of flag bits exported through APIs to userspace.
4545 unsigned dev_get_flags(const struct net_device *dev)
4549 flags = (dev->flags & ~(IFF_PROMISC |
4554 (dev->gflags & (IFF_PROMISC |
4557 if (netif_running(dev)) {
4558 if (netif_oper_up(dev))
4559 flags |= IFF_RUNNING;
4560 if (netif_carrier_ok(dev))
4561 flags |= IFF_LOWER_UP;
4562 if (netif_dormant(dev))
4563 flags |= IFF_DORMANT;
4568 EXPORT_SYMBOL(dev_get_flags);
4570 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4572 int old_flags = dev->flags;
4578 * Set the flags on our device.
4581 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4582 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4584 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4588 * Load in the correct multicast list now the flags have changed.
4591 if ((old_flags ^ flags) & IFF_MULTICAST)
4592 dev_change_rx_flags(dev, IFF_MULTICAST);
4594 dev_set_rx_mode(dev);
4597 * Have we downed the interface. We handle IFF_UP ourselves
4598 * according to user attempts to set it, rather than blindly
4603 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4604 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4607 dev_set_rx_mode(dev);
4610 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4611 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4613 dev->gflags ^= IFF_PROMISC;
4614 dev_set_promiscuity(dev, inc);
4617 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4618 is important. Some (broken) drivers set IFF_PROMISC, when
4619 IFF_ALLMULTI is requested not asking us and not reporting.
4621 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4622 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4624 dev->gflags ^= IFF_ALLMULTI;
4625 dev_set_allmulti(dev, inc);
4631 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4633 unsigned int changes = dev->flags ^ old_flags;
4635 if (changes & IFF_UP) {
4636 if (dev->flags & IFF_UP)
4637 call_netdevice_notifiers(NETDEV_UP, dev);
4639 call_netdevice_notifiers(NETDEV_DOWN, dev);
4642 if (dev->flags & IFF_UP &&
4643 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4644 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4648 * dev_change_flags - change device settings
4650 * @flags: device state flags
4652 * Change settings on device based state flags. The flags are
4653 * in the userspace exported format.
4655 int dev_change_flags(struct net_device *dev, unsigned flags)
4658 int old_flags = dev->flags;
4660 ret = __dev_change_flags(dev, flags);
4664 changes = old_flags ^ dev->flags;
4666 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4668 __dev_notify_flags(dev, old_flags);
4671 EXPORT_SYMBOL(dev_change_flags);
4674 * dev_set_mtu - Change maximum transfer unit
4676 * @new_mtu: new transfer unit
4678 * Change the maximum transfer size of the network device.
4680 int dev_set_mtu(struct net_device *dev, int new_mtu)
4682 const struct net_device_ops *ops = dev->netdev_ops;
4685 if (new_mtu == dev->mtu)
4688 /* MTU must be positive. */
4692 if (!netif_device_present(dev))
4696 if (ops->ndo_change_mtu)
4697 err = ops->ndo_change_mtu(dev, new_mtu);
4701 if (!err && dev->flags & IFF_UP)
4702 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4705 EXPORT_SYMBOL(dev_set_mtu);
4708 * dev_set_group - Change group this device belongs to
4710 * @new_group: group this device should belong to
4712 void dev_set_group(struct net_device *dev, int new_group)
4714 dev->group = new_group;
4716 EXPORT_SYMBOL(dev_set_group);
4719 * dev_set_mac_address - Change Media Access Control Address
4723 * Change the hardware (MAC) address of the device
4725 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4727 const struct net_device_ops *ops = dev->netdev_ops;
4730 if (!ops->ndo_set_mac_address)
4732 if (sa->sa_family != dev->type)
4734 if (!netif_device_present(dev))
4736 err = ops->ndo_set_mac_address(dev, sa);
4738 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4741 EXPORT_SYMBOL(dev_set_mac_address);
4744 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4746 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4749 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4755 case SIOCGIFFLAGS: /* Get interface flags */
4756 ifr->ifr_flags = (short) dev_get_flags(dev);
4759 case SIOCGIFMETRIC: /* Get the metric on the interface
4760 (currently unused) */
4761 ifr->ifr_metric = 0;
4764 case SIOCGIFMTU: /* Get the MTU of a device */
4765 ifr->ifr_mtu = dev->mtu;
4770 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4772 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4773 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4774 ifr->ifr_hwaddr.sa_family = dev->type;
4782 ifr->ifr_map.mem_start = dev->mem_start;
4783 ifr->ifr_map.mem_end = dev->mem_end;
4784 ifr->ifr_map.base_addr = dev->base_addr;
4785 ifr->ifr_map.irq = dev->irq;
4786 ifr->ifr_map.dma = dev->dma;
4787 ifr->ifr_map.port = dev->if_port;
4791 ifr->ifr_ifindex = dev->ifindex;
4795 ifr->ifr_qlen = dev->tx_queue_len;
4799 /* dev_ioctl() should ensure this case
4811 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4813 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4816 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4817 const struct net_device_ops *ops;
4822 ops = dev->netdev_ops;
4825 case SIOCSIFFLAGS: /* Set interface flags */
4826 return dev_change_flags(dev, ifr->ifr_flags);
4828 case SIOCSIFMETRIC: /* Set the metric on the interface
4829 (currently unused) */
4832 case SIOCSIFMTU: /* Set the MTU of a device */
4833 return dev_set_mtu(dev, ifr->ifr_mtu);
4836 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4838 case SIOCSIFHWBROADCAST:
4839 if (ifr->ifr_hwaddr.sa_family != dev->type)
4841 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4842 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4843 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4847 if (ops->ndo_set_config) {
4848 if (!netif_device_present(dev))
4850 return ops->ndo_set_config(dev, &ifr->ifr_map);
4855 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4856 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4858 if (!netif_device_present(dev))
4860 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4863 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4864 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4866 if (!netif_device_present(dev))
4868 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4871 if (ifr->ifr_qlen < 0)
4873 dev->tx_queue_len = ifr->ifr_qlen;
4877 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4878 return dev_change_name(dev, ifr->ifr_newname);
4881 * Unknown or private ioctl
4884 if ((cmd >= SIOCDEVPRIVATE &&
4885 cmd <= SIOCDEVPRIVATE + 15) ||
4886 cmd == SIOCBONDENSLAVE ||
4887 cmd == SIOCBONDRELEASE ||
4888 cmd == SIOCBONDSETHWADDR ||
4889 cmd == SIOCBONDSLAVEINFOQUERY ||
4890 cmd == SIOCBONDINFOQUERY ||
4891 cmd == SIOCBONDCHANGEACTIVE ||
4892 cmd == SIOCGMIIPHY ||
4893 cmd == SIOCGMIIREG ||
4894 cmd == SIOCSMIIREG ||
4895 cmd == SIOCBRADDIF ||
4896 cmd == SIOCBRDELIF ||
4897 cmd == SIOCSHWTSTAMP ||
4898 cmd == SIOCWANDEV) {
4900 if (ops->ndo_do_ioctl) {
4901 if (netif_device_present(dev))
4902 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4914 * This function handles all "interface"-type I/O control requests. The actual
4915 * 'doing' part of this is dev_ifsioc above.
4919 * dev_ioctl - network device ioctl
4920 * @net: the applicable net namespace
4921 * @cmd: command to issue
4922 * @arg: pointer to a struct ifreq in user space
4924 * Issue ioctl functions to devices. This is normally called by the
4925 * user space syscall interfaces but can sometimes be useful for
4926 * other purposes. The return value is the return from the syscall if
4927 * positive or a negative errno code on error.
4930 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4936 /* One special case: SIOCGIFCONF takes ifconf argument
4937 and requires shared lock, because it sleeps writing
4941 if (cmd == SIOCGIFCONF) {
4943 ret = dev_ifconf(net, (char __user *) arg);
4947 if (cmd == SIOCGIFNAME)
4948 return dev_ifname(net, (struct ifreq __user *)arg);
4950 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4953 ifr.ifr_name[IFNAMSIZ-1] = 0;
4955 colon = strchr(ifr.ifr_name, ':');
4960 * See which interface the caller is talking about.
4965 * These ioctl calls:
4966 * - can be done by all.
4967 * - atomic and do not require locking.
4978 dev_load(net, ifr.ifr_name);
4980 ret = dev_ifsioc_locked(net, &ifr, cmd);
4985 if (copy_to_user(arg, &ifr,
4986 sizeof(struct ifreq)))
4992 dev_load(net, ifr.ifr_name);
4994 ret = dev_ethtool(net, &ifr);
4999 if (copy_to_user(arg, &ifr,
5000 sizeof(struct ifreq)))
5006 * These ioctl calls:
5007 * - require superuser power.
5008 * - require strict serialization.
5014 if (!capable(CAP_NET_ADMIN))
5016 dev_load(net, ifr.ifr_name);
5018 ret = dev_ifsioc(net, &ifr, cmd);
5023 if (copy_to_user(arg, &ifr,
5024 sizeof(struct ifreq)))
5030 * These ioctl calls:
5031 * - require superuser power.
5032 * - require strict serialization.
5033 * - do not return a value
5043 case SIOCSIFHWBROADCAST:
5046 case SIOCBONDENSLAVE:
5047 case SIOCBONDRELEASE:
5048 case SIOCBONDSETHWADDR:
5049 case SIOCBONDCHANGEACTIVE:
5053 if (!capable(CAP_NET_ADMIN))
5056 case SIOCBONDSLAVEINFOQUERY:
5057 case SIOCBONDINFOQUERY:
5058 dev_load(net, ifr.ifr_name);
5060 ret = dev_ifsioc(net, &ifr, cmd);
5065 /* Get the per device memory space. We can add this but
5066 * currently do not support it */
5068 /* Set the per device memory buffer space.
5069 * Not applicable in our case */
5074 * Unknown or private ioctl.
5077 if (cmd == SIOCWANDEV ||
5078 (cmd >= SIOCDEVPRIVATE &&
5079 cmd <= SIOCDEVPRIVATE + 15)) {
5080 dev_load(net, ifr.ifr_name);
5082 ret = dev_ifsioc(net, &ifr, cmd);
5084 if (!ret && copy_to_user(arg, &ifr,
5085 sizeof(struct ifreq)))
5089 /* Take care of Wireless Extensions */
5090 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5091 return wext_handle_ioctl(net, &ifr, cmd, arg);
5098 * dev_new_index - allocate an ifindex
5099 * @net: the applicable net namespace
5101 * Returns a suitable unique value for a new device interface
5102 * number. The caller must hold the rtnl semaphore or the
5103 * dev_base_lock to be sure it remains unique.
5105 static int dev_new_index(struct net *net)
5111 if (!__dev_get_by_index(net, ifindex))
5116 /* Delayed registration/unregisteration */
5117 static LIST_HEAD(net_todo_list);
5119 static void net_set_todo(struct net_device *dev)
5121 list_add_tail(&dev->todo_list, &net_todo_list);
5124 static void rollback_registered_many(struct list_head *head)
5126 struct net_device *dev, *tmp;
5128 BUG_ON(dev_boot_phase);
5131 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5132 /* Some devices call without registering
5133 * for initialization unwind. Remove those
5134 * devices and proceed with the remaining.
5136 if (dev->reg_state == NETREG_UNINITIALIZED) {
5137 pr_debug("unregister_netdevice: device %s/%p never "
5138 "was registered\n", dev->name, dev);
5141 list_del(&dev->unreg_list);
5144 dev->dismantle = true;
5145 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5148 /* If device is running, close it first. */
5149 dev_close_many(head);
5151 list_for_each_entry(dev, head, unreg_list) {
5152 /* And unlink it from device chain. */
5153 unlist_netdevice(dev);
5155 dev->reg_state = NETREG_UNREGISTERING;
5160 list_for_each_entry(dev, head, unreg_list) {
5161 /* Shutdown queueing discipline. */
5165 /* Notify protocols, that we are about to destroy
5166 this device. They should clean all the things.
5168 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5170 if (!dev->rtnl_link_ops ||
5171 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5172 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5175 * Flush the unicast and multicast chains
5180 if (dev->netdev_ops->ndo_uninit)
5181 dev->netdev_ops->ndo_uninit(dev);
5183 /* Notifier chain MUST detach us from master device. */
5184 WARN_ON(dev->master);
5186 /* Remove entries from kobject tree */
5187 netdev_unregister_kobject(dev);
5190 /* Process any work delayed until the end of the batch */
5191 dev = list_first_entry(head, struct net_device, unreg_list);
5192 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5196 list_for_each_entry(dev, head, unreg_list)
5200 static void rollback_registered(struct net_device *dev)
5204 list_add(&dev->unreg_list, &single);
5205 rollback_registered_many(&single);
5209 u32 netdev_fix_features(struct net_device *dev, u32 features)
5211 /* Fix illegal checksum combinations */
5212 if ((features & NETIF_F_HW_CSUM) &&
5213 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5214 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5215 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5218 if ((features & NETIF_F_NO_CSUM) &&
5219 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5220 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5221 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5224 /* Fix illegal SG+CSUM combinations. */
5225 if ((features & NETIF_F_SG) &&
5226 !(features & NETIF_F_ALL_CSUM)) {
5228 "Dropping NETIF_F_SG since no checksum feature.\n");
5229 features &= ~NETIF_F_SG;
5232 /* TSO requires that SG is present as well. */
5233 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5234 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5235 features &= ~NETIF_F_ALL_TSO;
5238 /* TSO ECN requires that TSO is present as well. */
5239 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5240 features &= ~NETIF_F_TSO_ECN;
5242 /* Software GSO depends on SG. */
5243 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5244 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5245 features &= ~NETIF_F_GSO;
5248 /* UFO needs SG and checksumming */
5249 if (features & NETIF_F_UFO) {
5250 /* maybe split UFO into V4 and V6? */
5251 if (!((features & NETIF_F_GEN_CSUM) ||
5252 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5253 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5255 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5256 features &= ~NETIF_F_UFO;
5259 if (!(features & NETIF_F_SG)) {
5261 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5262 features &= ~NETIF_F_UFO;
5268 EXPORT_SYMBOL(netdev_fix_features);
5270 int __netdev_update_features(struct net_device *dev)
5277 features = netdev_get_wanted_features(dev);
5279 if (dev->netdev_ops->ndo_fix_features)
5280 features = dev->netdev_ops->ndo_fix_features(dev, features);
5282 /* driver might be less strict about feature dependencies */
5283 features = netdev_fix_features(dev, features);
5285 if (dev->features == features)
5288 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5289 dev->features, features);
5291 if (dev->netdev_ops->ndo_set_features)
5292 err = dev->netdev_ops->ndo_set_features(dev, features);
5294 if (unlikely(err < 0)) {
5296 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5297 err, features, dev->features);
5302 dev->features = features;
5308 * netdev_update_features - recalculate device features
5309 * @dev: the device to check
5311 * Recalculate dev->features set and send notifications if it
5312 * has changed. Should be called after driver or hardware dependent
5313 * conditions might have changed that influence the features.
5315 void netdev_update_features(struct net_device *dev)
5317 if (__netdev_update_features(dev))
5318 netdev_features_change(dev);
5320 EXPORT_SYMBOL(netdev_update_features);
5323 * netdev_change_features - recalculate device features
5324 * @dev: the device to check
5326 * Recalculate dev->features set and send notifications even
5327 * if they have not changed. Should be called instead of
5328 * netdev_update_features() if also dev->vlan_features might
5329 * have changed to allow the changes to be propagated to stacked
5332 void netdev_change_features(struct net_device *dev)
5334 __netdev_update_features(dev);
5335 netdev_features_change(dev);
5337 EXPORT_SYMBOL(netdev_change_features);
5340 * netif_stacked_transfer_operstate - transfer operstate
5341 * @rootdev: the root or lower level device to transfer state from
5342 * @dev: the device to transfer operstate to
5344 * Transfer operational state from root to device. This is normally
5345 * called when a stacking relationship exists between the root
5346 * device and the device(a leaf device).
5348 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5349 struct net_device *dev)
5351 if (rootdev->operstate == IF_OPER_DORMANT)
5352 netif_dormant_on(dev);
5354 netif_dormant_off(dev);
5356 if (netif_carrier_ok(rootdev)) {
5357 if (!netif_carrier_ok(dev))
5358 netif_carrier_on(dev);
5360 if (netif_carrier_ok(dev))
5361 netif_carrier_off(dev);
5364 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5367 static int netif_alloc_rx_queues(struct net_device *dev)
5369 unsigned int i, count = dev->num_rx_queues;
5370 struct netdev_rx_queue *rx;
5374 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5376 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5381 for (i = 0; i < count; i++)
5387 static void netdev_init_one_queue(struct net_device *dev,
5388 struct netdev_queue *queue, void *_unused)
5390 /* Initialize queue lock */
5391 spin_lock_init(&queue->_xmit_lock);
5392 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5393 queue->xmit_lock_owner = -1;
5394 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5398 static int netif_alloc_netdev_queues(struct net_device *dev)
5400 unsigned int count = dev->num_tx_queues;
5401 struct netdev_queue *tx;
5405 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5407 pr_err("netdev: Unable to allocate %u tx queues.\n",
5413 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5414 spin_lock_init(&dev->tx_global_lock);
5420 * register_netdevice - register a network device
5421 * @dev: device to register
5423 * Take a completed network device structure and add it to the kernel
5424 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5425 * chain. 0 is returned on success. A negative errno code is returned
5426 * on a failure to set up the device, or if the name is a duplicate.
5428 * Callers must hold the rtnl semaphore. You may want
5429 * register_netdev() instead of this.
5432 * The locking appears insufficient to guarantee two parallel registers
5433 * will not get the same name.
5436 int register_netdevice(struct net_device *dev)
5439 struct net *net = dev_net(dev);
5441 BUG_ON(dev_boot_phase);
5446 /* When net_device's are persistent, this will be fatal. */
5447 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5450 spin_lock_init(&dev->addr_list_lock);
5451 netdev_set_addr_lockdep_class(dev);
5455 ret = dev_get_valid_name(dev, dev->name);
5459 /* Init, if this function is available */
5460 if (dev->netdev_ops->ndo_init) {
5461 ret = dev->netdev_ops->ndo_init(dev);
5469 dev->ifindex = dev_new_index(net);
5470 if (dev->iflink == -1)
5471 dev->iflink = dev->ifindex;
5473 /* Transfer changeable features to wanted_features and enable
5474 * software offloads (GSO and GRO).
5476 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5477 dev->features |= NETIF_F_SOFT_FEATURES;
5478 dev->wanted_features = dev->features & dev->hw_features;
5480 /* Turn on no cache copy if HW is doing checksum */
5481 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5482 if ((dev->features & NETIF_F_ALL_CSUM) &&
5483 !(dev->features & NETIF_F_NO_CSUM)) {
5484 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5485 dev->features |= NETIF_F_NOCACHE_COPY;
5488 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5489 * vlan_dev_init() will do the dev->features check, so these features
5490 * are enabled only if supported by underlying device.
5492 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5494 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5495 ret = notifier_to_errno(ret);
5499 ret = netdev_register_kobject(dev);
5502 dev->reg_state = NETREG_REGISTERED;
5504 __netdev_update_features(dev);
5507 * Default initial state at registry is that the
5508 * device is present.
5511 set_bit(__LINK_STATE_PRESENT, &dev->state);
5513 dev_init_scheduler(dev);
5515 list_netdevice(dev);
5517 /* Notify protocols, that a new device appeared. */
5518 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5519 ret = notifier_to_errno(ret);
5521 rollback_registered(dev);
5522 dev->reg_state = NETREG_UNREGISTERED;
5525 * Prevent userspace races by waiting until the network
5526 * device is fully setup before sending notifications.
5528 if (!dev->rtnl_link_ops ||
5529 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5530 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5536 if (dev->netdev_ops->ndo_uninit)
5537 dev->netdev_ops->ndo_uninit(dev);
5540 EXPORT_SYMBOL(register_netdevice);
5543 * init_dummy_netdev - init a dummy network device for NAPI
5544 * @dev: device to init
5546 * This takes a network device structure and initialize the minimum
5547 * amount of fields so it can be used to schedule NAPI polls without
5548 * registering a full blown interface. This is to be used by drivers
5549 * that need to tie several hardware interfaces to a single NAPI
5550 * poll scheduler due to HW limitations.
5552 int init_dummy_netdev(struct net_device *dev)
5554 /* Clear everything. Note we don't initialize spinlocks
5555 * are they aren't supposed to be taken by any of the
5556 * NAPI code and this dummy netdev is supposed to be
5557 * only ever used for NAPI polls
5559 memset(dev, 0, sizeof(struct net_device));
5561 /* make sure we BUG if trying to hit standard
5562 * register/unregister code path
5564 dev->reg_state = NETREG_DUMMY;
5566 /* NAPI wants this */
5567 INIT_LIST_HEAD(&dev->napi_list);
5569 /* a dummy interface is started by default */
5570 set_bit(__LINK_STATE_PRESENT, &dev->state);
5571 set_bit(__LINK_STATE_START, &dev->state);
5573 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5574 * because users of this 'device' dont need to change
5580 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5584 * register_netdev - register a network device
5585 * @dev: device to register
5587 * Take a completed network device structure and add it to the kernel
5588 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5589 * chain. 0 is returned on success. A negative errno code is returned
5590 * on a failure to set up the device, or if the name is a duplicate.
5592 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5593 * and expands the device name if you passed a format string to
5596 int register_netdev(struct net_device *dev)
5601 err = register_netdevice(dev);
5605 EXPORT_SYMBOL(register_netdev);
5607 int netdev_refcnt_read(const struct net_device *dev)
5611 for_each_possible_cpu(i)
5612 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5615 EXPORT_SYMBOL(netdev_refcnt_read);
5618 * netdev_wait_allrefs - wait until all references are gone.
5620 * This is called when unregistering network devices.
5622 * Any protocol or device that holds a reference should register
5623 * for netdevice notification, and cleanup and put back the
5624 * reference if they receive an UNREGISTER event.
5625 * We can get stuck here if buggy protocols don't correctly
5628 static void netdev_wait_allrefs(struct net_device *dev)
5630 unsigned long rebroadcast_time, warning_time;
5633 linkwatch_forget_dev(dev);
5635 rebroadcast_time = warning_time = jiffies;
5636 refcnt = netdev_refcnt_read(dev);
5638 while (refcnt != 0) {
5639 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5642 /* Rebroadcast unregister notification */
5643 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5644 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5645 * should have already handle it the first time */
5647 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5649 /* We must not have linkwatch events
5650 * pending on unregister. If this
5651 * happens, we simply run the queue
5652 * unscheduled, resulting in a noop
5655 linkwatch_run_queue();
5660 rebroadcast_time = jiffies;
5665 refcnt = netdev_refcnt_read(dev);
5667 if (time_after(jiffies, warning_time + 10 * HZ)) {
5668 printk(KERN_EMERG "unregister_netdevice: "
5669 "waiting for %s to become free. Usage "
5672 warning_time = jiffies;
5681 * register_netdevice(x1);
5682 * register_netdevice(x2);
5684 * unregister_netdevice(y1);
5685 * unregister_netdevice(y2);
5691 * We are invoked by rtnl_unlock().
5692 * This allows us to deal with problems:
5693 * 1) We can delete sysfs objects which invoke hotplug
5694 * without deadlocking with linkwatch via keventd.
5695 * 2) Since we run with the RTNL semaphore not held, we can sleep
5696 * safely in order to wait for the netdev refcnt to drop to zero.
5698 * We must not return until all unregister events added during
5699 * the interval the lock was held have been completed.
5701 void netdev_run_todo(void)
5703 struct list_head list;
5705 /* Snapshot list, allow later requests */
5706 list_replace_init(&net_todo_list, &list);
5710 while (!list_empty(&list)) {
5711 struct net_device *dev
5712 = list_first_entry(&list, struct net_device, todo_list);
5713 list_del(&dev->todo_list);
5715 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5716 printk(KERN_ERR "network todo '%s' but state %d\n",
5717 dev->name, dev->reg_state);
5722 dev->reg_state = NETREG_UNREGISTERED;
5724 on_each_cpu(flush_backlog, dev, 1);
5726 netdev_wait_allrefs(dev);
5729 BUG_ON(netdev_refcnt_read(dev));
5730 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5731 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5732 WARN_ON(dev->dn_ptr);
5734 if (dev->destructor)
5735 dev->destructor(dev);
5737 /* Free network device */
5738 kobject_put(&dev->dev.kobj);
5742 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5743 * fields in the same order, with only the type differing.
5745 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5746 const struct net_device_stats *netdev_stats)
5748 #if BITS_PER_LONG == 64
5749 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5750 memcpy(stats64, netdev_stats, sizeof(*stats64));
5752 size_t i, n = sizeof(*stats64) / sizeof(u64);
5753 const unsigned long *src = (const unsigned long *)netdev_stats;
5754 u64 *dst = (u64 *)stats64;
5756 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5757 sizeof(*stats64) / sizeof(u64));
5758 for (i = 0; i < n; i++)
5764 * dev_get_stats - get network device statistics
5765 * @dev: device to get statistics from
5766 * @storage: place to store stats
5768 * Get network statistics from device. Return @storage.
5769 * The device driver may provide its own method by setting
5770 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5771 * otherwise the internal statistics structure is used.
5773 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5774 struct rtnl_link_stats64 *storage)
5776 const struct net_device_ops *ops = dev->netdev_ops;
5778 if (ops->ndo_get_stats64) {
5779 memset(storage, 0, sizeof(*storage));
5780 ops->ndo_get_stats64(dev, storage);
5781 } else if (ops->ndo_get_stats) {
5782 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5784 netdev_stats_to_stats64(storage, &dev->stats);
5786 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5789 EXPORT_SYMBOL(dev_get_stats);
5791 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5793 struct netdev_queue *queue = dev_ingress_queue(dev);
5795 #ifdef CONFIG_NET_CLS_ACT
5798 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5801 netdev_init_one_queue(dev, queue, NULL);
5802 queue->qdisc = &noop_qdisc;
5803 queue->qdisc_sleeping = &noop_qdisc;
5804 rcu_assign_pointer(dev->ingress_queue, queue);
5810 * alloc_netdev_mqs - allocate network device
5811 * @sizeof_priv: size of private data to allocate space for
5812 * @name: device name format string
5813 * @setup: callback to initialize device
5814 * @txqs: the number of TX subqueues to allocate
5815 * @rxqs: the number of RX subqueues to allocate
5817 * Allocates a struct net_device with private data area for driver use
5818 * and performs basic initialization. Also allocates subquue structs
5819 * for each queue on the device.
5821 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5822 void (*setup)(struct net_device *),
5823 unsigned int txqs, unsigned int rxqs)
5825 struct net_device *dev;
5827 struct net_device *p;
5829 BUG_ON(strlen(name) >= sizeof(dev->name));
5832 pr_err("alloc_netdev: Unable to allocate device "
5833 "with zero queues.\n");
5839 pr_err("alloc_netdev: Unable to allocate device "
5840 "with zero RX queues.\n");
5845 alloc_size = sizeof(struct net_device);
5847 /* ensure 32-byte alignment of private area */
5848 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5849 alloc_size += sizeof_priv;
5851 /* ensure 32-byte alignment of whole construct */
5852 alloc_size += NETDEV_ALIGN - 1;
5854 p = kzalloc(alloc_size, GFP_KERNEL);
5856 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5860 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5861 dev->padded = (char *)dev - (char *)p;
5863 dev->pcpu_refcnt = alloc_percpu(int);
5864 if (!dev->pcpu_refcnt)
5867 if (dev_addr_init(dev))
5873 dev_net_set(dev, &init_net);
5875 dev->gso_max_size = GSO_MAX_SIZE;
5877 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5878 dev->ethtool_ntuple_list.count = 0;
5879 INIT_LIST_HEAD(&dev->napi_list);
5880 INIT_LIST_HEAD(&dev->unreg_list);
5881 INIT_LIST_HEAD(&dev->link_watch_list);
5882 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5885 dev->num_tx_queues = txqs;
5886 dev->real_num_tx_queues = txqs;
5887 if (netif_alloc_netdev_queues(dev))
5891 dev->num_rx_queues = rxqs;
5892 dev->real_num_rx_queues = rxqs;
5893 if (netif_alloc_rx_queues(dev))
5897 strcpy(dev->name, name);
5898 dev->group = INIT_NETDEV_GROUP;
5906 free_percpu(dev->pcpu_refcnt);
5916 EXPORT_SYMBOL(alloc_netdev_mqs);
5919 * free_netdev - free network device
5922 * This function does the last stage of destroying an allocated device
5923 * interface. The reference to the device object is released.
5924 * If this is the last reference then it will be freed.
5926 void free_netdev(struct net_device *dev)
5928 struct napi_struct *p, *n;
5930 release_net(dev_net(dev));
5937 kfree(rcu_dereference_raw(dev->ingress_queue));
5939 /* Flush device addresses */
5940 dev_addr_flush(dev);
5942 /* Clear ethtool n-tuple list */
5943 ethtool_ntuple_flush(dev);
5945 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5948 free_percpu(dev->pcpu_refcnt);
5949 dev->pcpu_refcnt = NULL;
5951 /* Compatibility with error handling in drivers */
5952 if (dev->reg_state == NETREG_UNINITIALIZED) {
5953 kfree((char *)dev - dev->padded);
5957 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5958 dev->reg_state = NETREG_RELEASED;
5960 /* will free via device release */
5961 put_device(&dev->dev);
5963 EXPORT_SYMBOL(free_netdev);
5966 * synchronize_net - Synchronize with packet receive processing
5968 * Wait for packets currently being received to be done.
5969 * Does not block later packets from starting.
5971 void synchronize_net(void)
5974 if (rtnl_is_locked())
5975 synchronize_rcu_expedited();
5979 EXPORT_SYMBOL(synchronize_net);
5982 * unregister_netdevice_queue - remove device from the kernel
5986 * This function shuts down a device interface and removes it
5987 * from the kernel tables.
5988 * If head not NULL, device is queued to be unregistered later.
5990 * Callers must hold the rtnl semaphore. You may want
5991 * unregister_netdev() instead of this.
5994 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5999 list_move_tail(&dev->unreg_list, head);
6001 rollback_registered(dev);
6002 /* Finish processing unregister after unlock */
6006 EXPORT_SYMBOL(unregister_netdevice_queue);
6009 * unregister_netdevice_many - unregister many devices
6010 * @head: list of devices
6012 void unregister_netdevice_many(struct list_head *head)
6014 struct net_device *dev;
6016 if (!list_empty(head)) {
6017 rollback_registered_many(head);
6018 list_for_each_entry(dev, head, unreg_list)
6022 EXPORT_SYMBOL(unregister_netdevice_many);
6025 * unregister_netdev - remove device from the kernel
6028 * This function shuts down a device interface and removes it
6029 * from the kernel tables.
6031 * This is just a wrapper for unregister_netdevice that takes
6032 * the rtnl semaphore. In general you want to use this and not
6033 * unregister_netdevice.
6035 void unregister_netdev(struct net_device *dev)
6038 unregister_netdevice(dev);
6041 EXPORT_SYMBOL(unregister_netdev);
6044 * dev_change_net_namespace - move device to different nethost namespace
6046 * @net: network namespace
6047 * @pat: If not NULL name pattern to try if the current device name
6048 * is already taken in the destination network namespace.
6050 * This function shuts down a device interface and moves it
6051 * to a new network namespace. On success 0 is returned, on
6052 * a failure a netagive errno code is returned.
6054 * Callers must hold the rtnl semaphore.
6057 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6063 /* Don't allow namespace local devices to be moved. */
6065 if (dev->features & NETIF_F_NETNS_LOCAL)
6068 /* Ensure the device has been registrered */
6070 if (dev->reg_state != NETREG_REGISTERED)
6073 /* Get out if there is nothing todo */
6075 if (net_eq(dev_net(dev), net))
6078 /* Pick the destination device name, and ensure
6079 * we can use it in the destination network namespace.
6082 if (__dev_get_by_name(net, dev->name)) {
6083 /* We get here if we can't use the current device name */
6086 if (dev_get_valid_name(dev, pat) < 0)
6091 * And now a mini version of register_netdevice unregister_netdevice.
6094 /* If device is running close it first. */
6097 /* And unlink it from device chain */
6099 unlist_netdevice(dev);
6103 /* Shutdown queueing discipline. */
6106 /* Notify protocols, that we are about to destroy
6107 this device. They should clean all the things.
6109 Note that dev->reg_state stays at NETREG_REGISTERED.
6110 This is wanted because this way 8021q and macvlan know
6111 the device is just moving and can keep their slaves up.
6113 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6114 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6115 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6118 * Flush the unicast and multicast chains
6123 /* Actually switch the network namespace */
6124 dev_net_set(dev, net);
6126 /* If there is an ifindex conflict assign a new one */
6127 if (__dev_get_by_index(net, dev->ifindex)) {
6128 int iflink = (dev->iflink == dev->ifindex);
6129 dev->ifindex = dev_new_index(net);
6131 dev->iflink = dev->ifindex;
6134 /* Fixup kobjects */
6135 err = device_rename(&dev->dev, dev->name);
6138 /* Add the device back in the hashes */
6139 list_netdevice(dev);
6141 /* Notify protocols, that a new device appeared. */
6142 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6145 * Prevent userspace races by waiting until the network
6146 * device is fully setup before sending notifications.
6148 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6155 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6157 static int dev_cpu_callback(struct notifier_block *nfb,
6158 unsigned long action,
6161 struct sk_buff **list_skb;
6162 struct sk_buff *skb;
6163 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6164 struct softnet_data *sd, *oldsd;
6166 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6169 local_irq_disable();
6170 cpu = smp_processor_id();
6171 sd = &per_cpu(softnet_data, cpu);
6172 oldsd = &per_cpu(softnet_data, oldcpu);
6174 /* Find end of our completion_queue. */
6175 list_skb = &sd->completion_queue;
6177 list_skb = &(*list_skb)->next;
6178 /* Append completion queue from offline CPU. */
6179 *list_skb = oldsd->completion_queue;
6180 oldsd->completion_queue = NULL;
6182 /* Append output queue from offline CPU. */
6183 if (oldsd->output_queue) {
6184 *sd->output_queue_tailp = oldsd->output_queue;
6185 sd->output_queue_tailp = oldsd->output_queue_tailp;
6186 oldsd->output_queue = NULL;
6187 oldsd->output_queue_tailp = &oldsd->output_queue;
6189 /* Append NAPI poll list from offline CPU. */
6190 if (!list_empty(&oldsd->poll_list)) {
6191 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6192 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6195 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6198 /* Process offline CPU's input_pkt_queue */
6199 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6201 input_queue_head_incr(oldsd);
6203 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6205 input_queue_head_incr(oldsd);
6213 * netdev_increment_features - increment feature set by one
6214 * @all: current feature set
6215 * @one: new feature set
6216 * @mask: mask feature set
6218 * Computes a new feature set after adding a device with feature set
6219 * @one to the master device with current feature set @all. Will not
6220 * enable anything that is off in @mask. Returns the new feature set.
6222 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6224 if (mask & NETIF_F_GEN_CSUM)
6225 mask |= NETIF_F_ALL_CSUM;
6226 mask |= NETIF_F_VLAN_CHALLENGED;
6228 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6229 all &= one | ~NETIF_F_ALL_FOR_ALL;
6231 /* If device needs checksumming, downgrade to it. */
6232 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6233 all &= ~NETIF_F_NO_CSUM;
6235 /* If one device supports hw checksumming, set for all. */
6236 if (all & NETIF_F_GEN_CSUM)
6237 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6241 EXPORT_SYMBOL(netdev_increment_features);
6243 static struct hlist_head *netdev_create_hash(void)
6246 struct hlist_head *hash;
6248 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6250 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6251 INIT_HLIST_HEAD(&hash[i]);
6256 /* Initialize per network namespace state */
6257 static int __net_init netdev_init(struct net *net)
6259 INIT_LIST_HEAD(&net->dev_base_head);
6261 net->dev_name_head = netdev_create_hash();
6262 if (net->dev_name_head == NULL)
6265 net->dev_index_head = netdev_create_hash();
6266 if (net->dev_index_head == NULL)
6272 kfree(net->dev_name_head);
6278 * netdev_drivername - network driver for the device
6279 * @dev: network device
6281 * Determine network driver for device.
6283 const char *netdev_drivername(const struct net_device *dev)
6285 const struct device_driver *driver;
6286 const struct device *parent;
6287 const char *empty = "";
6289 parent = dev->dev.parent;
6293 driver = parent->driver;
6294 if (driver && driver->name)
6295 return driver->name;
6299 static int __netdev_printk(const char *level, const struct net_device *dev,
6300 struct va_format *vaf)
6304 if (dev && dev->dev.parent)
6305 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6306 netdev_name(dev), vaf);
6308 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6310 r = printk("%s(NULL net_device): %pV", level, vaf);
6315 int netdev_printk(const char *level, const struct net_device *dev,
6316 const char *format, ...)
6318 struct va_format vaf;
6322 va_start(args, format);
6327 r = __netdev_printk(level, dev, &vaf);
6332 EXPORT_SYMBOL(netdev_printk);
6334 #define define_netdev_printk_level(func, level) \
6335 int func(const struct net_device *dev, const char *fmt, ...) \
6338 struct va_format vaf; \
6341 va_start(args, fmt); \
6346 r = __netdev_printk(level, dev, &vaf); \
6351 EXPORT_SYMBOL(func);
6353 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6354 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6355 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6356 define_netdev_printk_level(netdev_err, KERN_ERR);
6357 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6358 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6359 define_netdev_printk_level(netdev_info, KERN_INFO);
6361 static void __net_exit netdev_exit(struct net *net)
6363 kfree(net->dev_name_head);
6364 kfree(net->dev_index_head);
6367 static struct pernet_operations __net_initdata netdev_net_ops = {
6368 .init = netdev_init,
6369 .exit = netdev_exit,
6372 static void __net_exit default_device_exit(struct net *net)
6374 struct net_device *dev, *aux;
6376 * Push all migratable network devices back to the
6377 * initial network namespace
6380 for_each_netdev_safe(net, dev, aux) {
6382 char fb_name[IFNAMSIZ];
6384 /* Ignore unmoveable devices (i.e. loopback) */
6385 if (dev->features & NETIF_F_NETNS_LOCAL)
6388 /* Leave virtual devices for the generic cleanup */
6389 if (dev->rtnl_link_ops)
6392 /* Push remaining network devices to init_net */
6393 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6394 err = dev_change_net_namespace(dev, &init_net, fb_name);
6396 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6397 __func__, dev->name, err);
6404 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6406 /* At exit all network devices most be removed from a network
6407 * namespace. Do this in the reverse order of registration.
6408 * Do this across as many network namespaces as possible to
6409 * improve batching efficiency.
6411 struct net_device *dev;
6413 LIST_HEAD(dev_kill_list);
6416 list_for_each_entry(net, net_list, exit_list) {
6417 for_each_netdev_reverse(net, dev) {
6418 if (dev->rtnl_link_ops)
6419 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6421 unregister_netdevice_queue(dev, &dev_kill_list);
6424 unregister_netdevice_many(&dev_kill_list);
6425 list_del(&dev_kill_list);
6429 static struct pernet_operations __net_initdata default_device_ops = {
6430 .exit = default_device_exit,
6431 .exit_batch = default_device_exit_batch,
6435 * Initialize the DEV module. At boot time this walks the device list and
6436 * unhooks any devices that fail to initialise (normally hardware not
6437 * present) and leaves us with a valid list of present and active devices.
6442 * This is called single threaded during boot, so no need
6443 * to take the rtnl semaphore.
6445 static int __init net_dev_init(void)
6447 int i, rc = -ENOMEM;
6449 BUG_ON(!dev_boot_phase);
6451 if (dev_proc_init())
6454 if (netdev_kobject_init())
6457 INIT_LIST_HEAD(&ptype_all);
6458 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6459 INIT_LIST_HEAD(&ptype_base[i]);
6461 if (register_pernet_subsys(&netdev_net_ops))
6465 * Initialise the packet receive queues.
6468 for_each_possible_cpu(i) {
6469 struct softnet_data *sd = &per_cpu(softnet_data, i);
6471 memset(sd, 0, sizeof(*sd));
6472 skb_queue_head_init(&sd->input_pkt_queue);
6473 skb_queue_head_init(&sd->process_queue);
6474 sd->completion_queue = NULL;
6475 INIT_LIST_HEAD(&sd->poll_list);
6476 sd->output_queue = NULL;
6477 sd->output_queue_tailp = &sd->output_queue;
6479 sd->csd.func = rps_trigger_softirq;
6485 sd->backlog.poll = process_backlog;
6486 sd->backlog.weight = weight_p;
6487 sd->backlog.gro_list = NULL;
6488 sd->backlog.gro_count = 0;
6493 /* The loopback device is special if any other network devices
6494 * is present in a network namespace the loopback device must
6495 * be present. Since we now dynamically allocate and free the
6496 * loopback device ensure this invariant is maintained by
6497 * keeping the loopback device as the first device on the
6498 * list of network devices. Ensuring the loopback devices
6499 * is the first device that appears and the last network device
6502 if (register_pernet_device(&loopback_net_ops))
6505 if (register_pernet_device(&default_device_ops))
6508 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6509 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6511 hotcpu_notifier(dev_cpu_callback, 0);
6519 subsys_initcall(net_dev_init);
6521 static int __init initialize_hashrnd(void)
6523 get_random_bytes(&hashrnd, sizeof(hashrnd));
6527 late_initcall_sync(initialize_hashrnd);