2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
135 #include <linux/cpu_rmap.h>
137 #include "net-sysfs.h"
139 /* Instead of increasing this, you should create a hash table. */
140 #define MAX_GRO_SKBS 8
142 /* This should be increased if a protocol with a bigger head is added. */
143 #define GRO_MAX_HEAD (MAX_HEADER + 128)
146 * The list of packet types we will receive (as opposed to discard)
147 * and the routines to invoke.
149 * Why 16. Because with 16 the only overlap we get on a hash of the
150 * low nibble of the protocol value is RARP/SNAP/X.25.
152 * NOTE: That is no longer true with the addition of VLAN tags. Not
153 * sure which should go first, but I bet it won't make much
154 * difference if we are running VLANs. The good news is that
155 * this protocol won't be in the list unless compiled in, so
156 * the average user (w/out VLANs) will not be adversely affected.
173 #define PTYPE_HASH_SIZE (16)
174 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
176 static DEFINE_SPINLOCK(ptype_lock);
177 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
178 static struct list_head ptype_all __read_mostly; /* Taps */
181 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
184 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
186 * Writers must hold the rtnl semaphore while they loop through the
187 * dev_base_head list, and hold dev_base_lock for writing when they do the
188 * actual updates. This allows pure readers to access the list even
189 * while a writer is preparing to update it.
191 * To put it another way, dev_base_lock is held for writing only to
192 * protect against pure readers; the rtnl semaphore provides the
193 * protection against other writers.
195 * See, for example usages, register_netdevice() and
196 * unregister_netdevice(), which must be called with the rtnl
199 DEFINE_RWLOCK(dev_base_lock);
200 EXPORT_SYMBOL(dev_base_lock);
202 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
204 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
205 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
208 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
210 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
213 static inline void rps_lock(struct softnet_data *sd)
216 spin_lock(&sd->input_pkt_queue.lock);
220 static inline void rps_unlock(struct softnet_data *sd)
223 spin_unlock(&sd->input_pkt_queue.lock);
227 /* Device list insertion */
228 static int list_netdevice(struct net_device *dev)
230 struct net *net = dev_net(dev);
234 write_lock_bh(&dev_base_lock);
235 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
236 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
237 hlist_add_head_rcu(&dev->index_hlist,
238 dev_index_hash(net, dev->ifindex));
239 write_unlock_bh(&dev_base_lock);
243 /* Device list removal
244 * caller must respect a RCU grace period before freeing/reusing dev
246 static void unlist_netdevice(struct net_device *dev)
250 /* Unlink dev from the device chain */
251 write_lock_bh(&dev_base_lock);
252 list_del_rcu(&dev->dev_list);
253 hlist_del_rcu(&dev->name_hlist);
254 hlist_del_rcu(&dev->index_hlist);
255 write_unlock_bh(&dev_base_lock);
262 static RAW_NOTIFIER_HEAD(netdev_chain);
265 * Device drivers call our routines to queue packets here. We empty the
266 * queue in the local softnet handler.
269 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
270 EXPORT_PER_CPU_SYMBOL(softnet_data);
272 #ifdef CONFIG_LOCKDEP
274 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
275 * according to dev->type
277 static const unsigned short netdev_lock_type[] =
278 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
279 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
280 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
281 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
282 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
283 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
284 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
285 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
286 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
287 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
288 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
289 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
290 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
291 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
292 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
293 ARPHRD_VOID, ARPHRD_NONE};
295 static const char *const netdev_lock_name[] =
296 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
297 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
298 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
299 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
300 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
301 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
302 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
303 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
304 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
305 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
306 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
307 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
308 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
309 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
310 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
311 "_xmit_VOID", "_xmit_NONE"};
313 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
316 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
320 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
321 if (netdev_lock_type[i] == dev_type)
323 /* the last key is used by default */
324 return ARRAY_SIZE(netdev_lock_type) - 1;
327 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
328 unsigned short dev_type)
332 i = netdev_lock_pos(dev_type);
333 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
334 netdev_lock_name[i]);
337 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
341 i = netdev_lock_pos(dev->type);
342 lockdep_set_class_and_name(&dev->addr_list_lock,
343 &netdev_addr_lock_key[i],
344 netdev_lock_name[i]);
347 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
348 unsigned short dev_type)
351 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
356 /*******************************************************************************
358 Protocol management and registration routines
360 *******************************************************************************/
363 * Add a protocol ID to the list. Now that the input handler is
364 * smarter we can dispense with all the messy stuff that used to be
367 * BEWARE!!! Protocol handlers, mangling input packets,
368 * MUST BE last in hash buckets and checking protocol handlers
369 * MUST start from promiscuous ptype_all chain in net_bh.
370 * It is true now, do not change it.
371 * Explanation follows: if protocol handler, mangling packet, will
372 * be the first on list, it is not able to sense, that packet
373 * is cloned and should be copied-on-write, so that it will
374 * change it and subsequent readers will get broken packet.
378 static inline struct list_head *ptype_head(const struct packet_type *pt)
380 if (pt->type == htons(ETH_P_ALL))
383 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
387 * dev_add_pack - add packet handler
388 * @pt: packet type declaration
390 * Add a protocol handler to the networking stack. The passed &packet_type
391 * is linked into kernel lists and may not be freed until it has been
392 * removed from the kernel lists.
394 * This call does not sleep therefore it can not
395 * guarantee all CPU's that are in middle of receiving packets
396 * will see the new packet type (until the next received packet).
399 void dev_add_pack(struct packet_type *pt)
401 struct list_head *head = ptype_head(pt);
403 spin_lock(&ptype_lock);
404 list_add_rcu(&pt->list, head);
405 spin_unlock(&ptype_lock);
407 EXPORT_SYMBOL(dev_add_pack);
410 * __dev_remove_pack - remove packet handler
411 * @pt: packet type declaration
413 * Remove a protocol handler that was previously added to the kernel
414 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
415 * from the kernel lists and can be freed or reused once this function
418 * The packet type might still be in use by receivers
419 * and must not be freed until after all the CPU's have gone
420 * through a quiescent state.
422 void __dev_remove_pack(struct packet_type *pt)
424 struct list_head *head = ptype_head(pt);
425 struct packet_type *pt1;
427 spin_lock(&ptype_lock);
429 list_for_each_entry(pt1, head, list) {
431 list_del_rcu(&pt->list);
436 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
438 spin_unlock(&ptype_lock);
440 EXPORT_SYMBOL(__dev_remove_pack);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type *pt)
456 __dev_remove_pack(pt);
460 EXPORT_SYMBOL(dev_remove_pack);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name, struct ifmap *map)
482 struct netdev_boot_setup *s;
486 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
487 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
488 memset(s[i].name, 0, sizeof(s[i].name));
489 strlcpy(s[i].name, name, IFNAMSIZ);
490 memcpy(&s[i].map, map, sizeof(s[i].map));
495 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device *dev)
509 struct netdev_boot_setup *s = dev_boot_setup;
512 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
513 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
514 !strcmp(dev->name, s[i].name)) {
515 dev->irq = s[i].map.irq;
516 dev->base_addr = s[i].map.base_addr;
517 dev->mem_start = s[i].map.mem_start;
518 dev->mem_end = s[i].map.mem_end;
524 EXPORT_SYMBOL(netdev_boot_setup_check);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix, int unit)
539 const struct netdev_boot_setup *s = dev_boot_setup;
543 sprintf(name, "%s%d", prefix, unit);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net, name))
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
553 if (!strcmp(name, s[i].name))
554 return s[i].map.base_addr;
559 * Saves at boot time configured settings for any netdevice.
561 int __init netdev_boot_setup(char *str)
566 str = get_options(str, ARRAY_SIZE(ints), ints);
571 memset(&map, 0, sizeof(map));
575 map.base_addr = ints[2];
577 map.mem_start = ints[3];
579 map.mem_end = ints[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str, &map);
585 __setup("netdev=", netdev_boot_setup);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device *__dev_get_by_name(struct net *net, const char *name)
607 struct hlist_node *p;
608 struct net_device *dev;
609 struct hlist_head *head = dev_name_hash(net, name);
611 hlist_for_each_entry(dev, p, head, name_hlist)
612 if (!strncmp(dev->name, name, IFNAMSIZ))
617 EXPORT_SYMBOL(__dev_get_by_name);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
633 struct hlist_node *p;
634 struct net_device *dev;
635 struct hlist_head *head = dev_name_hash(net, name);
637 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
638 if (!strncmp(dev->name, name, IFNAMSIZ))
643 EXPORT_SYMBOL(dev_get_by_name_rcu);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device *dev_get_by_name(struct net *net, const char *name)
659 struct net_device *dev;
662 dev = dev_get_by_name_rcu(net, name);
668 EXPORT_SYMBOL(dev_get_by_name);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
684 struct hlist_node *p;
685 struct net_device *dev;
686 struct hlist_head *head = dev_index_hash(net, ifindex);
688 hlist_for_each_entry(dev, p, head, index_hlist)
689 if (dev->ifindex == ifindex)
694 EXPORT_SYMBOL(__dev_get_by_index);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
709 struct hlist_node *p;
710 struct net_device *dev;
711 struct hlist_head *head = dev_index_hash(net, ifindex);
713 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
714 if (dev->ifindex == ifindex)
719 EXPORT_SYMBOL(dev_get_by_index_rcu);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device *dev_get_by_index(struct net *net, int ifindex)
735 struct net_device *dev;
738 dev = dev_get_by_index_rcu(net, ifindex);
744 EXPORT_SYMBOL(dev_get_by_index);
747 * dev_getbyhwaddr_rcu - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device.
754 * The caller must hold RCU or RTNL.
755 * The returned device has not had its ref count increased
756 * and the caller must therefore be careful about locking
760 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
763 struct net_device *dev;
765 for_each_netdev_rcu(net, dev)
766 if (dev->type == type &&
767 !memcmp(dev->dev_addr, ha, dev->addr_len))
772 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
774 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
776 struct net_device *dev;
779 for_each_netdev(net, dev)
780 if (dev->type == type)
785 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
787 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
789 struct net_device *dev, *ret = NULL;
792 for_each_netdev_rcu(net, dev)
793 if (dev->type == type) {
801 EXPORT_SYMBOL(dev_getfirstbyhwtype);
804 * dev_get_by_flags_rcu - find any device with given flags
805 * @net: the applicable net namespace
806 * @if_flags: IFF_* values
807 * @mask: bitmask of bits in if_flags to check
809 * Search for any interface with the given flags. Returns NULL if a device
810 * is not found or a pointer to the device. Must be called inside
811 * rcu_read_lock(), and result refcount is unchanged.
814 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
817 struct net_device *dev, *ret;
820 for_each_netdev_rcu(net, dev) {
821 if (((dev->flags ^ if_flags) & mask) == 0) {
828 EXPORT_SYMBOL(dev_get_by_flags_rcu);
831 * dev_valid_name - check if name is okay for network device
834 * Network device names need to be valid file names to
835 * to allow sysfs to work. We also disallow any kind of
838 int dev_valid_name(const char *name)
842 if (strlen(name) >= IFNAMSIZ)
844 if (!strcmp(name, ".") || !strcmp(name, ".."))
848 if (*name == '/' || isspace(*name))
854 EXPORT_SYMBOL(dev_valid_name);
857 * __dev_alloc_name - allocate a name for a device
858 * @net: network namespace to allocate the device name in
859 * @name: name format string
860 * @buf: scratch buffer and result name string
862 * Passed a format string - eg "lt%d" it will try and find a suitable
863 * id. It scans list of devices to build up a free map, then chooses
864 * the first empty slot. The caller must hold the dev_base or rtnl lock
865 * while allocating the name and adding the device in order to avoid
867 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
868 * Returns the number of the unit assigned or a negative errno code.
871 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
875 const int max_netdevices = 8*PAGE_SIZE;
876 unsigned long *inuse;
877 struct net_device *d;
879 p = strnchr(name, IFNAMSIZ-1, '%');
882 * Verify the string as this thing may have come from
883 * the user. There must be either one "%d" and no other "%"
886 if (p[1] != 'd' || strchr(p + 2, '%'))
889 /* Use one page as a bit array of possible slots */
890 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
894 for_each_netdev(net, d) {
895 if (!sscanf(d->name, name, &i))
897 if (i < 0 || i >= max_netdevices)
900 /* avoid cases where sscanf is not exact inverse of printf */
901 snprintf(buf, IFNAMSIZ, name, i);
902 if (!strncmp(buf, d->name, IFNAMSIZ))
906 i = find_first_zero_bit(inuse, max_netdevices);
907 free_page((unsigned long) inuse);
911 snprintf(buf, IFNAMSIZ, name, i);
912 if (!__dev_get_by_name(net, buf))
915 /* It is possible to run out of possible slots
916 * when the name is long and there isn't enough space left
917 * for the digits, or if all bits are used.
923 * dev_alloc_name - allocate a name for a device
925 * @name: name format string
927 * Passed a format string - eg "lt%d" it will try and find a suitable
928 * id. It scans list of devices to build up a free map, then chooses
929 * the first empty slot. The caller must hold the dev_base or rtnl lock
930 * while allocating the name and adding the device in order to avoid
932 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
933 * Returns the number of the unit assigned or a negative errno code.
936 int dev_alloc_name(struct net_device *dev, const char *name)
942 BUG_ON(!dev_net(dev));
944 ret = __dev_alloc_name(net, name, buf);
946 strlcpy(dev->name, buf, IFNAMSIZ);
949 EXPORT_SYMBOL(dev_alloc_name);
951 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
955 BUG_ON(!dev_net(dev));
958 if (!dev_valid_name(name))
961 if (fmt && strchr(name, '%'))
962 return dev_alloc_name(dev, name);
963 else if (__dev_get_by_name(net, name))
965 else if (dev->name != name)
966 strlcpy(dev->name, name, IFNAMSIZ);
972 * dev_change_name - change name of a device
974 * @newname: name (or format string) must be at least IFNAMSIZ
976 * Change name of a device, can pass format strings "eth%d".
979 int dev_change_name(struct net_device *dev, const char *newname)
981 char oldname[IFNAMSIZ];
987 BUG_ON(!dev_net(dev));
990 if (dev->flags & IFF_UP)
993 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
996 memcpy(oldname, dev->name, IFNAMSIZ);
998 err = dev_get_valid_name(dev, newname, 1);
1003 ret = device_rename(&dev->dev, dev->name);
1005 memcpy(dev->name, oldname, IFNAMSIZ);
1009 write_lock_bh(&dev_base_lock);
1010 hlist_del(&dev->name_hlist);
1011 write_unlock_bh(&dev_base_lock);
1015 write_lock_bh(&dev_base_lock);
1016 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1017 write_unlock_bh(&dev_base_lock);
1019 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1020 ret = notifier_to_errno(ret);
1023 /* err >= 0 after dev_alloc_name() or stores the first errno */
1026 memcpy(dev->name, oldname, IFNAMSIZ);
1030 "%s: name change rollback failed: %d.\n",
1039 * dev_set_alias - change ifalias of a device
1041 * @alias: name up to IFALIASZ
1042 * @len: limit of bytes to copy from info
1044 * Set ifalias for a device,
1046 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1050 if (len >= IFALIASZ)
1055 kfree(dev->ifalias);
1056 dev->ifalias = NULL;
1061 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1065 strlcpy(dev->ifalias, alias, len+1);
1071 * netdev_features_change - device changes features
1072 * @dev: device to cause notification
1074 * Called to indicate a device has changed features.
1076 void netdev_features_change(struct net_device *dev)
1078 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1080 EXPORT_SYMBOL(netdev_features_change);
1083 * netdev_state_change - device changes state
1084 * @dev: device to cause notification
1086 * Called to indicate a device has changed state. This function calls
1087 * the notifier chains for netdev_chain and sends a NEWLINK message
1088 * to the routing socket.
1090 void netdev_state_change(struct net_device *dev)
1092 if (dev->flags & IFF_UP) {
1093 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1094 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1097 EXPORT_SYMBOL(netdev_state_change);
1099 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1101 return call_netdevice_notifiers(event, dev);
1103 EXPORT_SYMBOL(netdev_bonding_change);
1106 * dev_load - load a network module
1107 * @net: the applicable net namespace
1108 * @name: name of interface
1110 * If a network interface is not present and the process has suitable
1111 * privileges this function loads the module. If module loading is not
1112 * available in this kernel then it becomes a nop.
1115 void dev_load(struct net *net, const char *name)
1117 struct net_device *dev;
1120 dev = dev_get_by_name_rcu(net, name);
1123 if (!dev && capable(CAP_NET_ADMIN))
1124 request_module("%s", name);
1126 EXPORT_SYMBOL(dev_load);
1128 static int __dev_open(struct net_device *dev)
1130 const struct net_device_ops *ops = dev->netdev_ops;
1136 * Is it even present?
1138 if (!netif_device_present(dev))
1141 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1142 ret = notifier_to_errno(ret);
1147 * Call device private open method
1149 set_bit(__LINK_STATE_START, &dev->state);
1151 if (ops->ndo_validate_addr)
1152 ret = ops->ndo_validate_addr(dev);
1154 if (!ret && ops->ndo_open)
1155 ret = ops->ndo_open(dev);
1158 * If it went open OK then:
1162 clear_bit(__LINK_STATE_START, &dev->state);
1167 dev->flags |= IFF_UP;
1172 net_dmaengine_get();
1175 * Initialize multicasting status
1177 dev_set_rx_mode(dev);
1180 * Wakeup transmit queue engine
1189 * dev_open - prepare an interface for use.
1190 * @dev: device to open
1192 * Takes a device from down to up state. The device's private open
1193 * function is invoked and then the multicast lists are loaded. Finally
1194 * the device is moved into the up state and a %NETDEV_UP message is
1195 * sent to the netdev notifier chain.
1197 * Calling this function on an active interface is a nop. On a failure
1198 * a negative errno code is returned.
1200 int dev_open(struct net_device *dev)
1207 if (dev->flags & IFF_UP)
1213 ret = __dev_open(dev);
1218 * ... and announce new interface.
1220 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1221 call_netdevice_notifiers(NETDEV_UP, dev);
1225 EXPORT_SYMBOL(dev_open);
1227 static int __dev_close_many(struct list_head *head)
1229 struct net_device *dev;
1234 list_for_each_entry(dev, head, unreg_list) {
1236 * Tell people we are going down, so that they can
1237 * prepare to death, when device is still operating.
1239 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1241 clear_bit(__LINK_STATE_START, &dev->state);
1243 /* Synchronize to scheduled poll. We cannot touch poll list, it
1244 * can be even on different cpu. So just clear netif_running().
1246 * dev->stop() will invoke napi_disable() on all of it's
1247 * napi_struct instances on this device.
1249 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1252 dev_deactivate_many(head);
1254 list_for_each_entry(dev, head, unreg_list) {
1255 const struct net_device_ops *ops = dev->netdev_ops;
1258 * Call the device specific close. This cannot fail.
1259 * Only if device is UP
1261 * We allow it to be called even after a DETACH hot-plug
1268 * Device is now down.
1271 dev->flags &= ~IFF_UP;
1276 net_dmaengine_put();
1282 static int __dev_close(struct net_device *dev)
1287 list_add(&dev->unreg_list, &single);
1288 retval = __dev_close_many(&single);
1293 static int dev_close_many(struct list_head *head)
1295 struct net_device *dev, *tmp;
1296 LIST_HEAD(tmp_list);
1298 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1299 if (!(dev->flags & IFF_UP))
1300 list_move(&dev->unreg_list, &tmp_list);
1302 __dev_close_many(head);
1305 * Tell people we are down
1307 list_for_each_entry(dev, head, unreg_list) {
1308 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1309 call_netdevice_notifiers(NETDEV_DOWN, dev);
1312 /* rollback_registered_many needs the complete original list */
1313 list_splice(&tmp_list, head);
1318 * dev_close - shutdown an interface.
1319 * @dev: device to shutdown
1321 * This function moves an active device into down state. A
1322 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1323 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1326 int dev_close(struct net_device *dev)
1330 list_add(&dev->unreg_list, &single);
1331 dev_close_many(&single);
1335 EXPORT_SYMBOL(dev_close);
1339 * dev_disable_lro - disable Large Receive Offload on a device
1342 * Disable Large Receive Offload (LRO) on a net device. Must be
1343 * called under RTNL. This is needed if received packets may be
1344 * forwarded to another interface.
1346 void dev_disable_lro(struct net_device *dev)
1348 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1349 dev->ethtool_ops->set_flags) {
1350 u32 flags = dev->ethtool_ops->get_flags(dev);
1351 if (flags & ETH_FLAG_LRO) {
1352 flags &= ~ETH_FLAG_LRO;
1353 dev->ethtool_ops->set_flags(dev, flags);
1356 WARN_ON(dev->features & NETIF_F_LRO);
1358 EXPORT_SYMBOL(dev_disable_lro);
1361 static int dev_boot_phase = 1;
1364 * Device change register/unregister. These are not inline or static
1365 * as we export them to the world.
1369 * register_netdevice_notifier - register a network notifier block
1372 * Register a notifier to be called when network device events occur.
1373 * The notifier passed is linked into the kernel structures and must
1374 * not be reused until it has been unregistered. A negative errno code
1375 * is returned on a failure.
1377 * When registered all registration and up events are replayed
1378 * to the new notifier to allow device to have a race free
1379 * view of the network device list.
1382 int register_netdevice_notifier(struct notifier_block *nb)
1384 struct net_device *dev;
1385 struct net_device *last;
1390 err = raw_notifier_chain_register(&netdev_chain, nb);
1396 for_each_netdev(net, dev) {
1397 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1398 err = notifier_to_errno(err);
1402 if (!(dev->flags & IFF_UP))
1405 nb->notifier_call(nb, NETDEV_UP, dev);
1416 for_each_netdev(net, dev) {
1420 if (dev->flags & IFF_UP) {
1421 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1422 nb->notifier_call(nb, NETDEV_DOWN, dev);
1424 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1425 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1429 raw_notifier_chain_unregister(&netdev_chain, nb);
1432 EXPORT_SYMBOL(register_netdevice_notifier);
1435 * unregister_netdevice_notifier - unregister a network notifier block
1438 * Unregister a notifier previously registered by
1439 * register_netdevice_notifier(). The notifier is unlinked into the
1440 * kernel structures and may then be reused. A negative errno code
1441 * is returned on a failure.
1444 int unregister_netdevice_notifier(struct notifier_block *nb)
1449 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1453 EXPORT_SYMBOL(unregister_netdevice_notifier);
1456 * call_netdevice_notifiers - call all network notifier blocks
1457 * @val: value passed unmodified to notifier function
1458 * @dev: net_device pointer passed unmodified to notifier function
1460 * Call all network notifier blocks. Parameters and return value
1461 * are as for raw_notifier_call_chain().
1464 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1467 return raw_notifier_call_chain(&netdev_chain, val, dev);
1470 /* When > 0 there are consumers of rx skb time stamps */
1471 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1473 void net_enable_timestamp(void)
1475 atomic_inc(&netstamp_needed);
1477 EXPORT_SYMBOL(net_enable_timestamp);
1479 void net_disable_timestamp(void)
1481 atomic_dec(&netstamp_needed);
1483 EXPORT_SYMBOL(net_disable_timestamp);
1485 static inline void net_timestamp_set(struct sk_buff *skb)
1487 if (atomic_read(&netstamp_needed))
1488 __net_timestamp(skb);
1490 skb->tstamp.tv64 = 0;
1493 static inline void net_timestamp_check(struct sk_buff *skb)
1495 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1496 __net_timestamp(skb);
1500 * dev_forward_skb - loopback an skb to another netif
1502 * @dev: destination network device
1503 * @skb: buffer to forward
1506 * NET_RX_SUCCESS (no congestion)
1507 * NET_RX_DROP (packet was dropped, but freed)
1509 * dev_forward_skb can be used for injecting an skb from the
1510 * start_xmit function of one device into the receive queue
1511 * of another device.
1513 * The receiving device may be in another namespace, so
1514 * we have to clear all information in the skb that could
1515 * impact namespace isolation.
1517 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1522 if (unlikely(!(dev->flags & IFF_UP) ||
1523 (skb->len > (dev->mtu + dev->hard_header_len + VLAN_HLEN)))) {
1524 atomic_long_inc(&dev->rx_dropped);
1528 skb_set_dev(skb, dev);
1529 skb->tstamp.tv64 = 0;
1530 skb->pkt_type = PACKET_HOST;
1531 skb->protocol = eth_type_trans(skb, dev);
1532 return netif_rx(skb);
1534 EXPORT_SYMBOL_GPL(dev_forward_skb);
1536 static inline int deliver_skb(struct sk_buff *skb,
1537 struct packet_type *pt_prev,
1538 struct net_device *orig_dev)
1540 atomic_inc(&skb->users);
1541 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1545 * Support routine. Sends outgoing frames to any network
1546 * taps currently in use.
1549 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1551 struct packet_type *ptype;
1552 struct sk_buff *skb2 = NULL;
1553 struct packet_type *pt_prev = NULL;
1556 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1557 /* Never send packets back to the socket
1558 * they originated from - MvS (miquels@drinkel.ow.org)
1560 if ((ptype->dev == dev || !ptype->dev) &&
1561 (ptype->af_packet_priv == NULL ||
1562 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1564 deliver_skb(skb2, pt_prev, skb->dev);
1569 skb2 = skb_clone(skb, GFP_ATOMIC);
1573 net_timestamp_set(skb2);
1575 /* skb->nh should be correctly
1576 set by sender, so that the second statement is
1577 just protection against buggy protocols.
1579 skb_reset_mac_header(skb2);
1581 if (skb_network_header(skb2) < skb2->data ||
1582 skb2->network_header > skb2->tail) {
1583 if (net_ratelimit())
1584 printk(KERN_CRIT "protocol %04x is "
1586 ntohs(skb2->protocol),
1588 skb_reset_network_header(skb2);
1591 skb2->transport_header = skb2->network_header;
1592 skb2->pkt_type = PACKET_OUTGOING;
1597 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1601 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1602 * @dev: Network device
1603 * @txq: number of queues available
1605 * If real_num_tx_queues is changed the tc mappings may no longer be
1606 * valid. To resolve this verify the tc mapping remains valid and if
1607 * not NULL the mapping. With no priorities mapping to this
1608 * offset/count pair it will no longer be used. In the worst case TC0
1609 * is invalid nothing can be done so disable priority mappings. If is
1610 * expected that drivers will fix this mapping if they can before
1611 * calling netif_set_real_num_tx_queues.
1613 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1616 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1618 /* If TC0 is invalidated disable TC mapping */
1619 if (tc->offset + tc->count > txq) {
1620 pr_warning("Number of in use tx queues changed "
1621 "invalidating tc mappings. Priority "
1622 "traffic classification disabled!\n");
1627 /* Invalidated prio to tc mappings set to TC0 */
1628 for (i = 1; i < TC_BITMASK + 1; i++) {
1629 int q = netdev_get_prio_tc_map(dev, i);
1631 tc = &dev->tc_to_txq[q];
1632 if (tc->offset + tc->count > txq) {
1633 pr_warning("Number of in use tx queues "
1634 "changed. Priority %i to tc "
1635 "mapping %i is no longer valid "
1636 "setting map to 0\n",
1638 netdev_set_prio_tc_map(dev, i, 0);
1644 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1645 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1647 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1651 if (txq < 1 || txq > dev->num_tx_queues)
1654 if (dev->reg_state == NETREG_REGISTERED ||
1655 dev->reg_state == NETREG_UNREGISTERING) {
1658 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1664 netif_setup_tc(dev, txq);
1666 if (txq < dev->real_num_tx_queues)
1667 qdisc_reset_all_tx_gt(dev, txq);
1670 dev->real_num_tx_queues = txq;
1673 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1677 * netif_set_real_num_rx_queues - set actual number of RX queues used
1678 * @dev: Network device
1679 * @rxq: Actual number of RX queues
1681 * This must be called either with the rtnl_lock held or before
1682 * registration of the net device. Returns 0 on success, or a
1683 * negative error code. If called before registration, it always
1686 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1690 if (rxq < 1 || rxq > dev->num_rx_queues)
1693 if (dev->reg_state == NETREG_REGISTERED) {
1696 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1702 dev->real_num_rx_queues = rxq;
1705 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1708 static inline void __netif_reschedule(struct Qdisc *q)
1710 struct softnet_data *sd;
1711 unsigned long flags;
1713 local_irq_save(flags);
1714 sd = &__get_cpu_var(softnet_data);
1715 q->next_sched = NULL;
1716 *sd->output_queue_tailp = q;
1717 sd->output_queue_tailp = &q->next_sched;
1718 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1719 local_irq_restore(flags);
1722 void __netif_schedule(struct Qdisc *q)
1724 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1725 __netif_reschedule(q);
1727 EXPORT_SYMBOL(__netif_schedule);
1729 void dev_kfree_skb_irq(struct sk_buff *skb)
1731 if (atomic_dec_and_test(&skb->users)) {
1732 struct softnet_data *sd;
1733 unsigned long flags;
1735 local_irq_save(flags);
1736 sd = &__get_cpu_var(softnet_data);
1737 skb->next = sd->completion_queue;
1738 sd->completion_queue = skb;
1739 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1740 local_irq_restore(flags);
1743 EXPORT_SYMBOL(dev_kfree_skb_irq);
1745 void dev_kfree_skb_any(struct sk_buff *skb)
1747 if (in_irq() || irqs_disabled())
1748 dev_kfree_skb_irq(skb);
1752 EXPORT_SYMBOL(dev_kfree_skb_any);
1756 * netif_device_detach - mark device as removed
1757 * @dev: network device
1759 * Mark device as removed from system and therefore no longer available.
1761 void netif_device_detach(struct net_device *dev)
1763 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1764 netif_running(dev)) {
1765 netif_tx_stop_all_queues(dev);
1768 EXPORT_SYMBOL(netif_device_detach);
1771 * netif_device_attach - mark device as attached
1772 * @dev: network device
1774 * Mark device as attached from system and restart if needed.
1776 void netif_device_attach(struct net_device *dev)
1778 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1779 netif_running(dev)) {
1780 netif_tx_wake_all_queues(dev);
1781 __netdev_watchdog_up(dev);
1784 EXPORT_SYMBOL(netif_device_attach);
1787 * skb_dev_set -- assign a new device to a buffer
1788 * @skb: buffer for the new device
1789 * @dev: network device
1791 * If an skb is owned by a device already, we have to reset
1792 * all data private to the namespace a device belongs to
1793 * before assigning it a new device.
1795 #ifdef CONFIG_NET_NS
1796 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1799 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1802 skb_init_secmark(skb);
1806 skb->ipvs_property = 0;
1807 #ifdef CONFIG_NET_SCHED
1813 EXPORT_SYMBOL(skb_set_dev);
1814 #endif /* CONFIG_NET_NS */
1817 * Invalidate hardware checksum when packet is to be mangled, and
1818 * complete checksum manually on outgoing path.
1820 int skb_checksum_help(struct sk_buff *skb)
1823 int ret = 0, offset;
1825 if (skb->ip_summed == CHECKSUM_COMPLETE)
1826 goto out_set_summed;
1828 if (unlikely(skb_shinfo(skb)->gso_size)) {
1829 /* Let GSO fix up the checksum. */
1830 goto out_set_summed;
1833 offset = skb_checksum_start_offset(skb);
1834 BUG_ON(offset >= skb_headlen(skb));
1835 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1837 offset += skb->csum_offset;
1838 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1840 if (skb_cloned(skb) &&
1841 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1842 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1847 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1849 skb->ip_summed = CHECKSUM_NONE;
1853 EXPORT_SYMBOL(skb_checksum_help);
1856 * skb_gso_segment - Perform segmentation on skb.
1857 * @skb: buffer to segment
1858 * @features: features for the output path (see dev->features)
1860 * This function segments the given skb and returns a list of segments.
1862 * It may return NULL if the skb requires no segmentation. This is
1863 * only possible when GSO is used for verifying header integrity.
1865 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1867 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1868 struct packet_type *ptype;
1869 __be16 type = skb->protocol;
1870 int vlan_depth = ETH_HLEN;
1873 while (type == htons(ETH_P_8021Q)) {
1874 struct vlan_hdr *vh;
1876 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1877 return ERR_PTR(-EINVAL);
1879 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1880 type = vh->h_vlan_encapsulated_proto;
1881 vlan_depth += VLAN_HLEN;
1884 skb_reset_mac_header(skb);
1885 skb->mac_len = skb->network_header - skb->mac_header;
1886 __skb_pull(skb, skb->mac_len);
1888 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1889 struct net_device *dev = skb->dev;
1890 struct ethtool_drvinfo info = {};
1892 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1893 dev->ethtool_ops->get_drvinfo(dev, &info);
1895 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1896 info.driver, dev ? dev->features : 0L,
1897 skb->sk ? skb->sk->sk_route_caps : 0L,
1898 skb->len, skb->data_len, skb->ip_summed);
1900 if (skb_header_cloned(skb) &&
1901 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1902 return ERR_PTR(err);
1906 list_for_each_entry_rcu(ptype,
1907 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1908 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1909 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1910 err = ptype->gso_send_check(skb);
1911 segs = ERR_PTR(err);
1912 if (err || skb_gso_ok(skb, features))
1914 __skb_push(skb, (skb->data -
1915 skb_network_header(skb)));
1917 segs = ptype->gso_segment(skb, features);
1923 __skb_push(skb, skb->data - skb_mac_header(skb));
1927 EXPORT_SYMBOL(skb_gso_segment);
1929 /* Take action when hardware reception checksum errors are detected. */
1931 void netdev_rx_csum_fault(struct net_device *dev)
1933 if (net_ratelimit()) {
1934 printk(KERN_ERR "%s: hw csum failure.\n",
1935 dev ? dev->name : "<unknown>");
1939 EXPORT_SYMBOL(netdev_rx_csum_fault);
1942 /* Actually, we should eliminate this check as soon as we know, that:
1943 * 1. IOMMU is present and allows to map all the memory.
1944 * 2. No high memory really exists on this machine.
1947 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1949 #ifdef CONFIG_HIGHMEM
1951 if (!(dev->features & NETIF_F_HIGHDMA)) {
1952 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1953 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1957 if (PCI_DMA_BUS_IS_PHYS) {
1958 struct device *pdev = dev->dev.parent;
1962 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1963 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1964 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1973 void (*destructor)(struct sk_buff *skb);
1976 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1978 static void dev_gso_skb_destructor(struct sk_buff *skb)
1980 struct dev_gso_cb *cb;
1983 struct sk_buff *nskb = skb->next;
1985 skb->next = nskb->next;
1988 } while (skb->next);
1990 cb = DEV_GSO_CB(skb);
1992 cb->destructor(skb);
1996 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1997 * @skb: buffer to segment
1998 * @features: device features as applicable to this skb
2000 * This function segments the given skb and stores the list of segments
2003 static int dev_gso_segment(struct sk_buff *skb, int features)
2005 struct sk_buff *segs;
2007 segs = skb_gso_segment(skb, features);
2009 /* Verifying header integrity only. */
2014 return PTR_ERR(segs);
2017 DEV_GSO_CB(skb)->destructor = skb->destructor;
2018 skb->destructor = dev_gso_skb_destructor;
2024 * Try to orphan skb early, right before transmission by the device.
2025 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2026 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2028 static inline void skb_orphan_try(struct sk_buff *skb)
2030 struct sock *sk = skb->sk;
2032 if (sk && !skb_shinfo(skb)->tx_flags) {
2033 /* skb_tx_hash() wont be able to get sk.
2034 * We copy sk_hash into skb->rxhash
2037 skb->rxhash = sk->sk_hash;
2042 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2044 return ((features & NETIF_F_GEN_CSUM) ||
2045 ((features & NETIF_F_V4_CSUM) &&
2046 protocol == htons(ETH_P_IP)) ||
2047 ((features & NETIF_F_V6_CSUM) &&
2048 protocol == htons(ETH_P_IPV6)) ||
2049 ((features & NETIF_F_FCOE_CRC) &&
2050 protocol == htons(ETH_P_FCOE)));
2053 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2055 if (!can_checksum_protocol(features, protocol)) {
2056 features &= ~NETIF_F_ALL_CSUM;
2057 features &= ~NETIF_F_SG;
2058 } else if (illegal_highdma(skb->dev, skb)) {
2059 features &= ~NETIF_F_SG;
2065 u32 netif_skb_features(struct sk_buff *skb)
2067 __be16 protocol = skb->protocol;
2068 u32 features = skb->dev->features;
2070 if (protocol == htons(ETH_P_8021Q)) {
2071 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2072 protocol = veh->h_vlan_encapsulated_proto;
2073 } else if (!vlan_tx_tag_present(skb)) {
2074 return harmonize_features(skb, protocol, features);
2077 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2079 if (protocol != htons(ETH_P_8021Q)) {
2080 return harmonize_features(skb, protocol, features);
2082 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2083 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2084 return harmonize_features(skb, protocol, features);
2087 EXPORT_SYMBOL(netif_skb_features);
2090 * Returns true if either:
2091 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2092 * 2. skb is fragmented and the device does not support SG, or if
2093 * at least one of fragments is in highmem and device does not
2094 * support DMA from it.
2096 static inline int skb_needs_linearize(struct sk_buff *skb,
2099 return skb_is_nonlinear(skb) &&
2100 ((skb_has_frag_list(skb) &&
2101 !(features & NETIF_F_FRAGLIST)) ||
2102 (skb_shinfo(skb)->nr_frags &&
2103 !(features & NETIF_F_SG)));
2106 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2107 struct netdev_queue *txq)
2109 const struct net_device_ops *ops = dev->netdev_ops;
2110 int rc = NETDEV_TX_OK;
2112 if (likely(!skb->next)) {
2116 * If device doesnt need skb->dst, release it right now while
2117 * its hot in this cpu cache
2119 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2122 if (!list_empty(&ptype_all))
2123 dev_queue_xmit_nit(skb, dev);
2125 skb_orphan_try(skb);
2127 features = netif_skb_features(skb);
2129 if (vlan_tx_tag_present(skb) &&
2130 !(features & NETIF_F_HW_VLAN_TX)) {
2131 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2138 if (netif_needs_gso(skb, features)) {
2139 if (unlikely(dev_gso_segment(skb, features)))
2144 if (skb_needs_linearize(skb, features) &&
2145 __skb_linearize(skb))
2148 /* If packet is not checksummed and device does not
2149 * support checksumming for this protocol, complete
2150 * checksumming here.
2152 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2153 skb_set_transport_header(skb,
2154 skb_checksum_start_offset(skb));
2155 if (!(features & NETIF_F_ALL_CSUM) &&
2156 skb_checksum_help(skb))
2161 rc = ops->ndo_start_xmit(skb, dev);
2162 trace_net_dev_xmit(skb, rc);
2163 if (rc == NETDEV_TX_OK)
2164 txq_trans_update(txq);
2170 struct sk_buff *nskb = skb->next;
2172 skb->next = nskb->next;
2176 * If device doesnt need nskb->dst, release it right now while
2177 * its hot in this cpu cache
2179 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2182 rc = ops->ndo_start_xmit(nskb, dev);
2183 trace_net_dev_xmit(nskb, rc);
2184 if (unlikely(rc != NETDEV_TX_OK)) {
2185 if (rc & ~NETDEV_TX_MASK)
2186 goto out_kfree_gso_skb;
2187 nskb->next = skb->next;
2191 txq_trans_update(txq);
2192 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2193 return NETDEV_TX_BUSY;
2194 } while (skb->next);
2197 if (likely(skb->next == NULL))
2198 skb->destructor = DEV_GSO_CB(skb)->destructor;
2205 static u32 hashrnd __read_mostly;
2208 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2209 * to be used as a distribution range.
2211 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2212 unsigned int num_tx_queues)
2216 u16 qcount = num_tx_queues;
2218 if (skb_rx_queue_recorded(skb)) {
2219 hash = skb_get_rx_queue(skb);
2220 while (unlikely(hash >= num_tx_queues))
2221 hash -= num_tx_queues;
2226 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2227 qoffset = dev->tc_to_txq[tc].offset;
2228 qcount = dev->tc_to_txq[tc].count;
2231 if (skb->sk && skb->sk->sk_hash)
2232 hash = skb->sk->sk_hash;
2234 hash = (__force u16) skb->protocol ^ skb->rxhash;
2235 hash = jhash_1word(hash, hashrnd);
2237 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2239 EXPORT_SYMBOL(__skb_tx_hash);
2241 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2243 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2244 if (net_ratelimit()) {
2245 pr_warning("%s selects TX queue %d, but "
2246 "real number of TX queues is %d\n",
2247 dev->name, queue_index, dev->real_num_tx_queues);
2254 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2257 struct xps_dev_maps *dev_maps;
2258 struct xps_map *map;
2259 int queue_index = -1;
2262 dev_maps = rcu_dereference(dev->xps_maps);
2264 map = rcu_dereference(
2265 dev_maps->cpu_map[raw_smp_processor_id()]);
2268 queue_index = map->queues[0];
2271 if (skb->sk && skb->sk->sk_hash)
2272 hash = skb->sk->sk_hash;
2274 hash = (__force u16) skb->protocol ^
2276 hash = jhash_1word(hash, hashrnd);
2277 queue_index = map->queues[
2278 ((u64)hash * map->len) >> 32];
2280 if (unlikely(queue_index >= dev->real_num_tx_queues))
2292 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2293 struct sk_buff *skb)
2296 const struct net_device_ops *ops = dev->netdev_ops;
2298 if (dev->real_num_tx_queues == 1)
2300 else if (ops->ndo_select_queue) {
2301 queue_index = ops->ndo_select_queue(dev, skb);
2302 queue_index = dev_cap_txqueue(dev, queue_index);
2304 struct sock *sk = skb->sk;
2305 queue_index = sk_tx_queue_get(sk);
2307 if (queue_index < 0 || skb->ooo_okay ||
2308 queue_index >= dev->real_num_tx_queues) {
2309 int old_index = queue_index;
2311 queue_index = get_xps_queue(dev, skb);
2312 if (queue_index < 0)
2313 queue_index = skb_tx_hash(dev, skb);
2315 if (queue_index != old_index && sk) {
2316 struct dst_entry *dst =
2317 rcu_dereference_check(sk->sk_dst_cache, 1);
2319 if (dst && skb_dst(skb) == dst)
2320 sk_tx_queue_set(sk, queue_index);
2325 skb_set_queue_mapping(skb, queue_index);
2326 return netdev_get_tx_queue(dev, queue_index);
2329 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2330 struct net_device *dev,
2331 struct netdev_queue *txq)
2333 spinlock_t *root_lock = qdisc_lock(q);
2337 qdisc_skb_cb(skb)->pkt_len = skb->len;
2338 qdisc_calculate_pkt_len(skb, q);
2340 * Heuristic to force contended enqueues to serialize on a
2341 * separate lock before trying to get qdisc main lock.
2342 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2343 * and dequeue packets faster.
2345 contended = qdisc_is_running(q);
2346 if (unlikely(contended))
2347 spin_lock(&q->busylock);
2349 spin_lock(root_lock);
2350 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2353 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2354 qdisc_run_begin(q)) {
2356 * This is a work-conserving queue; there are no old skbs
2357 * waiting to be sent out; and the qdisc is not running -
2358 * xmit the skb directly.
2360 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2363 qdisc_bstats_update(q, skb);
2365 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2366 if (unlikely(contended)) {
2367 spin_unlock(&q->busylock);
2374 rc = NET_XMIT_SUCCESS;
2377 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2378 if (qdisc_run_begin(q)) {
2379 if (unlikely(contended)) {
2380 spin_unlock(&q->busylock);
2386 spin_unlock(root_lock);
2387 if (unlikely(contended))
2388 spin_unlock(&q->busylock);
2392 static DEFINE_PER_CPU(int, xmit_recursion);
2393 #define RECURSION_LIMIT 10
2396 * dev_queue_xmit - transmit a buffer
2397 * @skb: buffer to transmit
2399 * Queue a buffer for transmission to a network device. The caller must
2400 * have set the device and priority and built the buffer before calling
2401 * this function. The function can be called from an interrupt.
2403 * A negative errno code is returned on a failure. A success does not
2404 * guarantee the frame will be transmitted as it may be dropped due
2405 * to congestion or traffic shaping.
2407 * -----------------------------------------------------------------------------------
2408 * I notice this method can also return errors from the queue disciplines,
2409 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2412 * Regardless of the return value, the skb is consumed, so it is currently
2413 * difficult to retry a send to this method. (You can bump the ref count
2414 * before sending to hold a reference for retry if you are careful.)
2416 * When calling this method, interrupts MUST be enabled. This is because
2417 * the BH enable code must have IRQs enabled so that it will not deadlock.
2420 int dev_queue_xmit(struct sk_buff *skb)
2422 struct net_device *dev = skb->dev;
2423 struct netdev_queue *txq;
2427 /* Disable soft irqs for various locks below. Also
2428 * stops preemption for RCU.
2432 txq = dev_pick_tx(dev, skb);
2433 q = rcu_dereference_bh(txq->qdisc);
2435 #ifdef CONFIG_NET_CLS_ACT
2436 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2438 trace_net_dev_queue(skb);
2440 rc = __dev_xmit_skb(skb, q, dev, txq);
2444 /* The device has no queue. Common case for software devices:
2445 loopback, all the sorts of tunnels...
2447 Really, it is unlikely that netif_tx_lock protection is necessary
2448 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2450 However, it is possible, that they rely on protection
2453 Check this and shot the lock. It is not prone from deadlocks.
2454 Either shot noqueue qdisc, it is even simpler 8)
2456 if (dev->flags & IFF_UP) {
2457 int cpu = smp_processor_id(); /* ok because BHs are off */
2459 if (txq->xmit_lock_owner != cpu) {
2461 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2462 goto recursion_alert;
2464 HARD_TX_LOCK(dev, txq, cpu);
2466 if (!netif_tx_queue_stopped(txq)) {
2467 __this_cpu_inc(xmit_recursion);
2468 rc = dev_hard_start_xmit(skb, dev, txq);
2469 __this_cpu_dec(xmit_recursion);
2470 if (dev_xmit_complete(rc)) {
2471 HARD_TX_UNLOCK(dev, txq);
2475 HARD_TX_UNLOCK(dev, txq);
2476 if (net_ratelimit())
2477 printk(KERN_CRIT "Virtual device %s asks to "
2478 "queue packet!\n", dev->name);
2480 /* Recursion is detected! It is possible,
2484 if (net_ratelimit())
2485 printk(KERN_CRIT "Dead loop on virtual device "
2486 "%s, fix it urgently!\n", dev->name);
2491 rcu_read_unlock_bh();
2496 rcu_read_unlock_bh();
2499 EXPORT_SYMBOL(dev_queue_xmit);
2502 /*=======================================================================
2504 =======================================================================*/
2506 int netdev_max_backlog __read_mostly = 1000;
2507 int netdev_tstamp_prequeue __read_mostly = 1;
2508 int netdev_budget __read_mostly = 300;
2509 int weight_p __read_mostly = 64; /* old backlog weight */
2511 /* Called with irq disabled */
2512 static inline void ____napi_schedule(struct softnet_data *sd,
2513 struct napi_struct *napi)
2515 list_add_tail(&napi->poll_list, &sd->poll_list);
2516 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2520 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2521 * and src/dst port numbers. Returns a non-zero hash number on success
2524 __u32 __skb_get_rxhash(struct sk_buff *skb)
2526 int nhoff, hash = 0, poff;
2527 struct ipv6hdr *ip6;
2530 u32 addr1, addr2, ihl;
2536 nhoff = skb_network_offset(skb);
2538 switch (skb->protocol) {
2539 case __constant_htons(ETH_P_IP):
2540 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2543 ip = (struct iphdr *) (skb->data + nhoff);
2544 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2547 ip_proto = ip->protocol;
2548 addr1 = (__force u32) ip->saddr;
2549 addr2 = (__force u32) ip->daddr;
2552 case __constant_htons(ETH_P_IPV6):
2553 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2556 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2557 ip_proto = ip6->nexthdr;
2558 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2559 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2567 poff = proto_ports_offset(ip_proto);
2569 nhoff += ihl * 4 + poff;
2570 if (pskb_may_pull(skb, nhoff + 4)) {
2571 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2572 if (ports.v16[1] < ports.v16[0])
2573 swap(ports.v16[0], ports.v16[1]);
2577 /* get a consistent hash (same value on both flow directions) */
2581 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2588 EXPORT_SYMBOL(__skb_get_rxhash);
2592 /* One global table that all flow-based protocols share. */
2593 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2594 EXPORT_SYMBOL(rps_sock_flow_table);
2596 static struct rps_dev_flow *
2597 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2598 struct rps_dev_flow *rflow, u16 next_cpu)
2602 tcpu = rflow->cpu = next_cpu;
2603 if (tcpu != RPS_NO_CPU) {
2604 #ifdef CONFIG_RFS_ACCEL
2605 struct netdev_rx_queue *rxqueue;
2606 struct rps_dev_flow_table *flow_table;
2607 struct rps_dev_flow *old_rflow;
2612 /* Should we steer this flow to a different hardware queue? */
2613 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2614 !(dev->features & NETIF_F_NTUPLE))
2616 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2617 if (rxq_index == skb_get_rx_queue(skb))
2620 rxqueue = dev->_rx + rxq_index;
2621 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2624 flow_id = skb->rxhash & flow_table->mask;
2625 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2626 rxq_index, flow_id);
2630 rflow = &flow_table->flows[flow_id];
2631 rflow->cpu = next_cpu;
2633 if (old_rflow->filter == rflow->filter)
2634 old_rflow->filter = RPS_NO_FILTER;
2638 per_cpu(softnet_data, tcpu).input_queue_head;
2645 * get_rps_cpu is called from netif_receive_skb and returns the target
2646 * CPU from the RPS map of the receiving queue for a given skb.
2647 * rcu_read_lock must be held on entry.
2649 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2650 struct rps_dev_flow **rflowp)
2652 struct netdev_rx_queue *rxqueue;
2653 struct rps_map *map;
2654 struct rps_dev_flow_table *flow_table;
2655 struct rps_sock_flow_table *sock_flow_table;
2659 if (skb_rx_queue_recorded(skb)) {
2660 u16 index = skb_get_rx_queue(skb);
2661 if (unlikely(index >= dev->real_num_rx_queues)) {
2662 WARN_ONCE(dev->real_num_rx_queues > 1,
2663 "%s received packet on queue %u, but number "
2664 "of RX queues is %u\n",
2665 dev->name, index, dev->real_num_rx_queues);
2668 rxqueue = dev->_rx + index;
2672 map = rcu_dereference(rxqueue->rps_map);
2674 if (map->len == 1 &&
2675 !rcu_dereference_raw(rxqueue->rps_flow_table)) {
2676 tcpu = map->cpus[0];
2677 if (cpu_online(tcpu))
2681 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2685 skb_reset_network_header(skb);
2686 if (!skb_get_rxhash(skb))
2689 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2690 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2691 if (flow_table && sock_flow_table) {
2693 struct rps_dev_flow *rflow;
2695 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2698 next_cpu = sock_flow_table->ents[skb->rxhash &
2699 sock_flow_table->mask];
2702 * If the desired CPU (where last recvmsg was done) is
2703 * different from current CPU (one in the rx-queue flow
2704 * table entry), switch if one of the following holds:
2705 * - Current CPU is unset (equal to RPS_NO_CPU).
2706 * - Current CPU is offline.
2707 * - The current CPU's queue tail has advanced beyond the
2708 * last packet that was enqueued using this table entry.
2709 * This guarantees that all previous packets for the flow
2710 * have been dequeued, thus preserving in order delivery.
2712 if (unlikely(tcpu != next_cpu) &&
2713 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2714 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2715 rflow->last_qtail)) >= 0))
2716 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2718 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2726 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2728 if (cpu_online(tcpu)) {
2738 #ifdef CONFIG_RFS_ACCEL
2741 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2742 * @dev: Device on which the filter was set
2743 * @rxq_index: RX queue index
2744 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2745 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2747 * Drivers that implement ndo_rx_flow_steer() should periodically call
2748 * this function for each installed filter and remove the filters for
2749 * which it returns %true.
2751 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2752 u32 flow_id, u16 filter_id)
2754 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2755 struct rps_dev_flow_table *flow_table;
2756 struct rps_dev_flow *rflow;
2761 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2762 if (flow_table && flow_id <= flow_table->mask) {
2763 rflow = &flow_table->flows[flow_id];
2764 cpu = ACCESS_ONCE(rflow->cpu);
2765 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2766 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2767 rflow->last_qtail) <
2768 (int)(10 * flow_table->mask)))
2774 EXPORT_SYMBOL(rps_may_expire_flow);
2776 #endif /* CONFIG_RFS_ACCEL */
2778 /* Called from hardirq (IPI) context */
2779 static void rps_trigger_softirq(void *data)
2781 struct softnet_data *sd = data;
2783 ____napi_schedule(sd, &sd->backlog);
2787 #endif /* CONFIG_RPS */
2790 * Check if this softnet_data structure is another cpu one
2791 * If yes, queue it to our IPI list and return 1
2794 static int rps_ipi_queued(struct softnet_data *sd)
2797 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2800 sd->rps_ipi_next = mysd->rps_ipi_list;
2801 mysd->rps_ipi_list = sd;
2803 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2806 #endif /* CONFIG_RPS */
2811 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2812 * queue (may be a remote CPU queue).
2814 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2815 unsigned int *qtail)
2817 struct softnet_data *sd;
2818 unsigned long flags;
2820 sd = &per_cpu(softnet_data, cpu);
2822 local_irq_save(flags);
2825 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2826 if (skb_queue_len(&sd->input_pkt_queue)) {
2828 __skb_queue_tail(&sd->input_pkt_queue, skb);
2829 input_queue_tail_incr_save(sd, qtail);
2831 local_irq_restore(flags);
2832 return NET_RX_SUCCESS;
2835 /* Schedule NAPI for backlog device
2836 * We can use non atomic operation since we own the queue lock
2838 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2839 if (!rps_ipi_queued(sd))
2840 ____napi_schedule(sd, &sd->backlog);
2848 local_irq_restore(flags);
2850 atomic_long_inc(&skb->dev->rx_dropped);
2856 * netif_rx - post buffer to the network code
2857 * @skb: buffer to post
2859 * This function receives a packet from a device driver and queues it for
2860 * the upper (protocol) levels to process. It always succeeds. The buffer
2861 * may be dropped during processing for congestion control or by the
2865 * NET_RX_SUCCESS (no congestion)
2866 * NET_RX_DROP (packet was dropped)
2870 int netif_rx(struct sk_buff *skb)
2874 /* if netpoll wants it, pretend we never saw it */
2875 if (netpoll_rx(skb))
2878 if (netdev_tstamp_prequeue)
2879 net_timestamp_check(skb);
2881 trace_netif_rx(skb);
2884 struct rps_dev_flow voidflow, *rflow = &voidflow;
2890 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2892 cpu = smp_processor_id();
2894 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2902 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2908 EXPORT_SYMBOL(netif_rx);
2910 int netif_rx_ni(struct sk_buff *skb)
2915 err = netif_rx(skb);
2916 if (local_softirq_pending())
2922 EXPORT_SYMBOL(netif_rx_ni);
2924 static void net_tx_action(struct softirq_action *h)
2926 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2928 if (sd->completion_queue) {
2929 struct sk_buff *clist;
2931 local_irq_disable();
2932 clist = sd->completion_queue;
2933 sd->completion_queue = NULL;
2937 struct sk_buff *skb = clist;
2938 clist = clist->next;
2940 WARN_ON(atomic_read(&skb->users));
2941 trace_kfree_skb(skb, net_tx_action);
2946 if (sd->output_queue) {
2949 local_irq_disable();
2950 head = sd->output_queue;
2951 sd->output_queue = NULL;
2952 sd->output_queue_tailp = &sd->output_queue;
2956 struct Qdisc *q = head;
2957 spinlock_t *root_lock;
2959 head = head->next_sched;
2961 root_lock = qdisc_lock(q);
2962 if (spin_trylock(root_lock)) {
2963 smp_mb__before_clear_bit();
2964 clear_bit(__QDISC_STATE_SCHED,
2967 spin_unlock(root_lock);
2969 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2971 __netif_reschedule(q);
2973 smp_mb__before_clear_bit();
2974 clear_bit(__QDISC_STATE_SCHED,
2982 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2983 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2984 /* This hook is defined here for ATM LANE */
2985 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2986 unsigned char *addr) __read_mostly;
2987 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2990 #ifdef CONFIG_NET_CLS_ACT
2991 /* TODO: Maybe we should just force sch_ingress to be compiled in
2992 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2993 * a compare and 2 stores extra right now if we dont have it on
2994 * but have CONFIG_NET_CLS_ACT
2995 * NOTE: This doesnt stop any functionality; if you dont have
2996 * the ingress scheduler, you just cant add policies on ingress.
2999 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3001 struct net_device *dev = skb->dev;
3002 u32 ttl = G_TC_RTTL(skb->tc_verd);
3003 int result = TC_ACT_OK;
3006 if (unlikely(MAX_RED_LOOP < ttl++)) {
3007 if (net_ratelimit())
3008 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3009 skb->skb_iif, dev->ifindex);
3013 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3014 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3017 if (q != &noop_qdisc) {
3018 spin_lock(qdisc_lock(q));
3019 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3020 result = qdisc_enqueue_root(skb, q);
3021 spin_unlock(qdisc_lock(q));
3027 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3028 struct packet_type **pt_prev,
3029 int *ret, struct net_device *orig_dev)
3031 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3033 if (!rxq || rxq->qdisc == &noop_qdisc)
3037 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3041 switch (ing_filter(skb, rxq)) {
3055 * netdev_rx_handler_register - register receive handler
3056 * @dev: device to register a handler for
3057 * @rx_handler: receive handler to register
3058 * @rx_handler_data: data pointer that is used by rx handler
3060 * Register a receive hander for a device. This handler will then be
3061 * called from __netif_receive_skb. A negative errno code is returned
3064 * The caller must hold the rtnl_mutex.
3066 int netdev_rx_handler_register(struct net_device *dev,
3067 rx_handler_func_t *rx_handler,
3068 void *rx_handler_data)
3072 if (dev->rx_handler)
3075 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3076 rcu_assign_pointer(dev->rx_handler, rx_handler);
3080 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3083 * netdev_rx_handler_unregister - unregister receive handler
3084 * @dev: device to unregister a handler from
3086 * Unregister a receive hander from a device.
3088 * The caller must hold the rtnl_mutex.
3090 void netdev_rx_handler_unregister(struct net_device *dev)
3094 rcu_assign_pointer(dev->rx_handler, NULL);
3095 rcu_assign_pointer(dev->rx_handler_data, NULL);
3097 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3099 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
3100 struct net_device *master)
3102 if (skb->pkt_type == PACKET_HOST) {
3103 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
3105 memcpy(dest, master->dev_addr, ETH_ALEN);
3109 /* On bonding slaves other than the currently active slave, suppress
3110 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
3111 * ARP on active-backup slaves with arp_validate enabled.
3113 static int __skb_bond_should_drop(struct sk_buff *skb,
3114 struct net_device *master)
3116 struct net_device *dev = skb->dev;
3118 if (master->priv_flags & IFF_MASTER_ARPMON)
3119 dev->last_rx = jiffies;
3121 if ((master->priv_flags & IFF_MASTER_ALB) &&
3122 (master->priv_flags & IFF_BRIDGE_PORT)) {
3123 /* Do address unmangle. The local destination address
3124 * will be always the one master has. Provides the right
3125 * functionality in a bridge.
3127 skb_bond_set_mac_by_master(skb, master);
3130 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
3131 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
3132 skb->protocol == __cpu_to_be16(ETH_P_ARP))
3135 if (master->priv_flags & IFF_MASTER_ALB) {
3136 if (skb->pkt_type != PACKET_BROADCAST &&
3137 skb->pkt_type != PACKET_MULTICAST)
3140 if (master->priv_flags & IFF_MASTER_8023AD &&
3141 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
3149 static int __netif_receive_skb(struct sk_buff *skb)
3151 struct packet_type *ptype, *pt_prev;
3152 rx_handler_func_t *rx_handler;
3153 struct net_device *orig_dev;
3154 struct net_device *null_or_orig;
3155 struct net_device *orig_or_bond;
3156 int ret = NET_RX_DROP;
3159 if (!netdev_tstamp_prequeue)
3160 net_timestamp_check(skb);
3162 trace_netif_receive_skb(skb);
3164 /* if we've gotten here through NAPI, check netpoll */
3165 if (netpoll_receive_skb(skb))
3169 skb->skb_iif = skb->dev->ifindex;
3172 * bonding note: skbs received on inactive slaves should only
3173 * be delivered to pkt handlers that are exact matches. Also
3174 * the deliver_no_wcard flag will be set. If packet handlers
3175 * are sensitive to duplicate packets these skbs will need to
3176 * be dropped at the handler.
3178 null_or_orig = NULL;
3179 orig_dev = skb->dev;
3180 if (skb->deliver_no_wcard)
3181 null_or_orig = orig_dev;
3182 else if (netif_is_bond_slave(orig_dev)) {
3183 struct net_device *bond_master = ACCESS_ONCE(orig_dev->master);
3185 if (likely(bond_master)) {
3186 if (__skb_bond_should_drop(skb, bond_master)) {
3187 skb->deliver_no_wcard = 1;
3188 /* deliver only exact match */
3189 null_or_orig = orig_dev;
3191 skb->dev = bond_master;
3195 __this_cpu_inc(softnet_data.processed);
3196 skb_reset_network_header(skb);
3197 skb_reset_transport_header(skb);
3198 skb->mac_len = skb->network_header - skb->mac_header;
3204 #ifdef CONFIG_NET_CLS_ACT
3205 if (skb->tc_verd & TC_NCLS) {
3206 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3211 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3212 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
3213 ptype->dev == orig_dev) {
3215 ret = deliver_skb(skb, pt_prev, orig_dev);
3220 #ifdef CONFIG_NET_CLS_ACT
3221 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3227 /* Handle special case of bridge or macvlan */
3228 rx_handler = rcu_dereference(skb->dev->rx_handler);
3231 ret = deliver_skb(skb, pt_prev, orig_dev);
3234 skb = rx_handler(skb);
3239 if (vlan_tx_tag_present(skb)) {
3241 ret = deliver_skb(skb, pt_prev, orig_dev);
3244 if (vlan_hwaccel_do_receive(&skb)) {
3245 ret = __netif_receive_skb(skb);
3247 } else if (unlikely(!skb))
3252 * Make sure frames received on VLAN interfaces stacked on
3253 * bonding interfaces still make their way to any base bonding
3254 * device that may have registered for a specific ptype. The
3255 * handler may have to adjust skb->dev and orig_dev.
3257 orig_or_bond = orig_dev;
3258 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
3259 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
3260 orig_or_bond = vlan_dev_real_dev(skb->dev);
3263 type = skb->protocol;
3264 list_for_each_entry_rcu(ptype,
3265 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3266 if (ptype->type == type && (ptype->dev == null_or_orig ||
3267 ptype->dev == skb->dev || ptype->dev == orig_dev ||
3268 ptype->dev == orig_or_bond)) {
3270 ret = deliver_skb(skb, pt_prev, orig_dev);
3276 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3278 atomic_long_inc(&skb->dev->rx_dropped);
3280 /* Jamal, now you will not able to escape explaining
3281 * me how you were going to use this. :-)
3292 * netif_receive_skb - process receive buffer from network
3293 * @skb: buffer to process
3295 * netif_receive_skb() is the main receive data processing function.
3296 * It always succeeds. The buffer may be dropped during processing
3297 * for congestion control or by the protocol layers.
3299 * This function may only be called from softirq context and interrupts
3300 * should be enabled.
3302 * Return values (usually ignored):
3303 * NET_RX_SUCCESS: no congestion
3304 * NET_RX_DROP: packet was dropped
3306 int netif_receive_skb(struct sk_buff *skb)
3308 if (netdev_tstamp_prequeue)
3309 net_timestamp_check(skb);
3311 if (skb_defer_rx_timestamp(skb))
3312 return NET_RX_SUCCESS;
3316 struct rps_dev_flow voidflow, *rflow = &voidflow;
3321 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3324 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3328 ret = __netif_receive_skb(skb);
3334 return __netif_receive_skb(skb);
3337 EXPORT_SYMBOL(netif_receive_skb);
3339 /* Network device is going away, flush any packets still pending
3340 * Called with irqs disabled.
3342 static void flush_backlog(void *arg)
3344 struct net_device *dev = arg;
3345 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3346 struct sk_buff *skb, *tmp;
3349 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3350 if (skb->dev == dev) {
3351 __skb_unlink(skb, &sd->input_pkt_queue);
3353 input_queue_head_incr(sd);
3358 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3359 if (skb->dev == dev) {
3360 __skb_unlink(skb, &sd->process_queue);
3362 input_queue_head_incr(sd);
3367 static int napi_gro_complete(struct sk_buff *skb)
3369 struct packet_type *ptype;
3370 __be16 type = skb->protocol;
3371 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3374 if (NAPI_GRO_CB(skb)->count == 1) {
3375 skb_shinfo(skb)->gso_size = 0;
3380 list_for_each_entry_rcu(ptype, head, list) {
3381 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3384 err = ptype->gro_complete(skb);
3390 WARN_ON(&ptype->list == head);
3392 return NET_RX_SUCCESS;
3396 return netif_receive_skb(skb);
3399 inline void napi_gro_flush(struct napi_struct *napi)
3401 struct sk_buff *skb, *next;
3403 for (skb = napi->gro_list; skb; skb = next) {
3406 napi_gro_complete(skb);
3409 napi->gro_count = 0;
3410 napi->gro_list = NULL;
3412 EXPORT_SYMBOL(napi_gro_flush);
3414 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3416 struct sk_buff **pp = NULL;
3417 struct packet_type *ptype;
3418 __be16 type = skb->protocol;
3419 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3422 enum gro_result ret;
3424 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3427 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3431 list_for_each_entry_rcu(ptype, head, list) {
3432 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3435 skb_set_network_header(skb, skb_gro_offset(skb));
3436 mac_len = skb->network_header - skb->mac_header;
3437 skb->mac_len = mac_len;
3438 NAPI_GRO_CB(skb)->same_flow = 0;
3439 NAPI_GRO_CB(skb)->flush = 0;
3440 NAPI_GRO_CB(skb)->free = 0;
3442 pp = ptype->gro_receive(&napi->gro_list, skb);
3447 if (&ptype->list == head)
3450 same_flow = NAPI_GRO_CB(skb)->same_flow;
3451 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3454 struct sk_buff *nskb = *pp;
3458 napi_gro_complete(nskb);
3465 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3469 NAPI_GRO_CB(skb)->count = 1;
3470 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3471 skb->next = napi->gro_list;
3472 napi->gro_list = skb;
3476 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3477 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3479 BUG_ON(skb->end - skb->tail < grow);
3481 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3484 skb->data_len -= grow;
3486 skb_shinfo(skb)->frags[0].page_offset += grow;
3487 skb_shinfo(skb)->frags[0].size -= grow;
3489 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3490 put_page(skb_shinfo(skb)->frags[0].page);
3491 memmove(skb_shinfo(skb)->frags,
3492 skb_shinfo(skb)->frags + 1,
3493 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3504 EXPORT_SYMBOL(dev_gro_receive);
3506 static inline gro_result_t
3507 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3511 for (p = napi->gro_list; p; p = p->next) {
3512 unsigned long diffs;
3514 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3515 diffs |= p->vlan_tci ^ skb->vlan_tci;
3516 diffs |= compare_ether_header(skb_mac_header(p),
3517 skb_gro_mac_header(skb));
3518 NAPI_GRO_CB(p)->same_flow = !diffs;
3519 NAPI_GRO_CB(p)->flush = 0;
3522 return dev_gro_receive(napi, skb);
3525 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3529 if (netif_receive_skb(skb))
3534 case GRO_MERGED_FREE:
3545 EXPORT_SYMBOL(napi_skb_finish);
3547 void skb_gro_reset_offset(struct sk_buff *skb)
3549 NAPI_GRO_CB(skb)->data_offset = 0;
3550 NAPI_GRO_CB(skb)->frag0 = NULL;
3551 NAPI_GRO_CB(skb)->frag0_len = 0;
3553 if (skb->mac_header == skb->tail &&
3554 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3555 NAPI_GRO_CB(skb)->frag0 =
3556 page_address(skb_shinfo(skb)->frags[0].page) +
3557 skb_shinfo(skb)->frags[0].page_offset;
3558 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3561 EXPORT_SYMBOL(skb_gro_reset_offset);
3563 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3565 skb_gro_reset_offset(skb);
3567 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3569 EXPORT_SYMBOL(napi_gro_receive);
3571 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3573 __skb_pull(skb, skb_headlen(skb));
3574 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3576 skb->dev = napi->dev;
3582 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3584 struct sk_buff *skb = napi->skb;
3587 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3593 EXPORT_SYMBOL(napi_get_frags);
3595 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3601 skb->protocol = eth_type_trans(skb, skb->dev);
3603 if (ret == GRO_HELD)
3604 skb_gro_pull(skb, -ETH_HLEN);
3605 else if (netif_receive_skb(skb))
3610 case GRO_MERGED_FREE:
3611 napi_reuse_skb(napi, skb);
3620 EXPORT_SYMBOL(napi_frags_finish);
3622 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3624 struct sk_buff *skb = napi->skb;
3631 skb_reset_mac_header(skb);
3632 skb_gro_reset_offset(skb);
3634 off = skb_gro_offset(skb);
3635 hlen = off + sizeof(*eth);
3636 eth = skb_gro_header_fast(skb, off);
3637 if (skb_gro_header_hard(skb, hlen)) {
3638 eth = skb_gro_header_slow(skb, hlen, off);
3639 if (unlikely(!eth)) {
3640 napi_reuse_skb(napi, skb);
3646 skb_gro_pull(skb, sizeof(*eth));
3649 * This works because the only protocols we care about don't require
3650 * special handling. We'll fix it up properly at the end.
3652 skb->protocol = eth->h_proto;
3657 EXPORT_SYMBOL(napi_frags_skb);
3659 gro_result_t napi_gro_frags(struct napi_struct *napi)
3661 struct sk_buff *skb = napi_frags_skb(napi);
3666 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3668 EXPORT_SYMBOL(napi_gro_frags);
3671 * net_rps_action sends any pending IPI's for rps.
3672 * Note: called with local irq disabled, but exits with local irq enabled.
3674 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3677 struct softnet_data *remsd = sd->rps_ipi_list;
3680 sd->rps_ipi_list = NULL;
3684 /* Send pending IPI's to kick RPS processing on remote cpus. */
3686 struct softnet_data *next = remsd->rps_ipi_next;
3688 if (cpu_online(remsd->cpu))
3689 __smp_call_function_single(remsd->cpu,
3698 static int process_backlog(struct napi_struct *napi, int quota)
3701 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3704 /* Check if we have pending ipi, its better to send them now,
3705 * not waiting net_rx_action() end.
3707 if (sd->rps_ipi_list) {
3708 local_irq_disable();
3709 net_rps_action_and_irq_enable(sd);
3712 napi->weight = weight_p;
3713 local_irq_disable();
3714 while (work < quota) {
3715 struct sk_buff *skb;
3718 while ((skb = __skb_dequeue(&sd->process_queue))) {
3720 __netif_receive_skb(skb);
3721 local_irq_disable();
3722 input_queue_head_incr(sd);
3723 if (++work >= quota) {
3730 qlen = skb_queue_len(&sd->input_pkt_queue);
3732 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3733 &sd->process_queue);
3735 if (qlen < quota - work) {
3737 * Inline a custom version of __napi_complete().
3738 * only current cpu owns and manipulates this napi,
3739 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3740 * we can use a plain write instead of clear_bit(),
3741 * and we dont need an smp_mb() memory barrier.
3743 list_del(&napi->poll_list);
3746 quota = work + qlen;
3756 * __napi_schedule - schedule for receive
3757 * @n: entry to schedule
3759 * The entry's receive function will be scheduled to run
3761 void __napi_schedule(struct napi_struct *n)
3763 unsigned long flags;
3765 local_irq_save(flags);
3766 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3767 local_irq_restore(flags);
3769 EXPORT_SYMBOL(__napi_schedule);
3771 void __napi_complete(struct napi_struct *n)
3773 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3774 BUG_ON(n->gro_list);
3776 list_del(&n->poll_list);
3777 smp_mb__before_clear_bit();
3778 clear_bit(NAPI_STATE_SCHED, &n->state);
3780 EXPORT_SYMBOL(__napi_complete);
3782 void napi_complete(struct napi_struct *n)
3784 unsigned long flags;
3787 * don't let napi dequeue from the cpu poll list
3788 * just in case its running on a different cpu
3790 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3794 local_irq_save(flags);
3796 local_irq_restore(flags);
3798 EXPORT_SYMBOL(napi_complete);
3800 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3801 int (*poll)(struct napi_struct *, int), int weight)
3803 INIT_LIST_HEAD(&napi->poll_list);
3804 napi->gro_count = 0;
3805 napi->gro_list = NULL;
3808 napi->weight = weight;
3809 list_add(&napi->dev_list, &dev->napi_list);
3811 #ifdef CONFIG_NETPOLL
3812 spin_lock_init(&napi->poll_lock);
3813 napi->poll_owner = -1;
3815 set_bit(NAPI_STATE_SCHED, &napi->state);
3817 EXPORT_SYMBOL(netif_napi_add);
3819 void netif_napi_del(struct napi_struct *napi)
3821 struct sk_buff *skb, *next;
3823 list_del_init(&napi->dev_list);
3824 napi_free_frags(napi);
3826 for (skb = napi->gro_list; skb; skb = next) {
3832 napi->gro_list = NULL;
3833 napi->gro_count = 0;
3835 EXPORT_SYMBOL(netif_napi_del);
3837 static void net_rx_action(struct softirq_action *h)
3839 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3840 unsigned long time_limit = jiffies + 2;
3841 int budget = netdev_budget;
3844 local_irq_disable();
3846 while (!list_empty(&sd->poll_list)) {
3847 struct napi_struct *n;
3850 /* If softirq window is exhuasted then punt.
3851 * Allow this to run for 2 jiffies since which will allow
3852 * an average latency of 1.5/HZ.
3854 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3859 /* Even though interrupts have been re-enabled, this
3860 * access is safe because interrupts can only add new
3861 * entries to the tail of this list, and only ->poll()
3862 * calls can remove this head entry from the list.
3864 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3866 have = netpoll_poll_lock(n);
3870 /* This NAPI_STATE_SCHED test is for avoiding a race
3871 * with netpoll's poll_napi(). Only the entity which
3872 * obtains the lock and sees NAPI_STATE_SCHED set will
3873 * actually make the ->poll() call. Therefore we avoid
3874 * accidently calling ->poll() when NAPI is not scheduled.
3877 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3878 work = n->poll(n, weight);
3882 WARN_ON_ONCE(work > weight);
3886 local_irq_disable();
3888 /* Drivers must not modify the NAPI state if they
3889 * consume the entire weight. In such cases this code
3890 * still "owns" the NAPI instance and therefore can
3891 * move the instance around on the list at-will.
3893 if (unlikely(work == weight)) {
3894 if (unlikely(napi_disable_pending(n))) {
3897 local_irq_disable();
3899 list_move_tail(&n->poll_list, &sd->poll_list);
3902 netpoll_poll_unlock(have);
3905 net_rps_action_and_irq_enable(sd);
3907 #ifdef CONFIG_NET_DMA
3909 * There may not be any more sk_buffs coming right now, so push
3910 * any pending DMA copies to hardware
3912 dma_issue_pending_all();
3919 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3923 static gifconf_func_t *gifconf_list[NPROTO];
3926 * register_gifconf - register a SIOCGIF handler
3927 * @family: Address family
3928 * @gifconf: Function handler
3930 * Register protocol dependent address dumping routines. The handler
3931 * that is passed must not be freed or reused until it has been replaced
3932 * by another handler.
3934 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3936 if (family >= NPROTO)
3938 gifconf_list[family] = gifconf;
3941 EXPORT_SYMBOL(register_gifconf);
3945 * Map an interface index to its name (SIOCGIFNAME)
3949 * We need this ioctl for efficient implementation of the
3950 * if_indextoname() function required by the IPv6 API. Without
3951 * it, we would have to search all the interfaces to find a
3955 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3957 struct net_device *dev;
3961 * Fetch the caller's info block.
3964 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3968 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3974 strcpy(ifr.ifr_name, dev->name);
3977 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3983 * Perform a SIOCGIFCONF call. This structure will change
3984 * size eventually, and there is nothing I can do about it.
3985 * Thus we will need a 'compatibility mode'.
3988 static int dev_ifconf(struct net *net, char __user *arg)
3991 struct net_device *dev;
3998 * Fetch the caller's info block.
4001 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4008 * Loop over the interfaces, and write an info block for each.
4012 for_each_netdev(net, dev) {
4013 for (i = 0; i < NPROTO; i++) {
4014 if (gifconf_list[i]) {
4017 done = gifconf_list[i](dev, NULL, 0);
4019 done = gifconf_list[i](dev, pos + total,
4029 * All done. Write the updated control block back to the caller.
4031 ifc.ifc_len = total;
4034 * Both BSD and Solaris return 0 here, so we do too.
4036 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4039 #ifdef CONFIG_PROC_FS
4041 * This is invoked by the /proc filesystem handler to display a device
4044 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4047 struct net *net = seq_file_net(seq);
4049 struct net_device *dev;
4053 return SEQ_START_TOKEN;
4056 for_each_netdev_rcu(net, dev)
4063 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4065 struct net_device *dev = v;
4067 if (v == SEQ_START_TOKEN)
4068 dev = first_net_device_rcu(seq_file_net(seq));
4070 dev = next_net_device_rcu(dev);
4076 void dev_seq_stop(struct seq_file *seq, void *v)
4082 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4084 struct rtnl_link_stats64 temp;
4085 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4087 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4088 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4089 dev->name, stats->rx_bytes, stats->rx_packets,
4091 stats->rx_dropped + stats->rx_missed_errors,
4092 stats->rx_fifo_errors,
4093 stats->rx_length_errors + stats->rx_over_errors +
4094 stats->rx_crc_errors + stats->rx_frame_errors,
4095 stats->rx_compressed, stats->multicast,
4096 stats->tx_bytes, stats->tx_packets,
4097 stats->tx_errors, stats->tx_dropped,
4098 stats->tx_fifo_errors, stats->collisions,
4099 stats->tx_carrier_errors +
4100 stats->tx_aborted_errors +
4101 stats->tx_window_errors +
4102 stats->tx_heartbeat_errors,
4103 stats->tx_compressed);
4107 * Called from the PROCfs module. This now uses the new arbitrary sized
4108 * /proc/net interface to create /proc/net/dev
4110 static int dev_seq_show(struct seq_file *seq, void *v)
4112 if (v == SEQ_START_TOKEN)
4113 seq_puts(seq, "Inter-| Receive "
4115 " face |bytes packets errs drop fifo frame "
4116 "compressed multicast|bytes packets errs "
4117 "drop fifo colls carrier compressed\n");
4119 dev_seq_printf_stats(seq, v);
4123 static struct softnet_data *softnet_get_online(loff_t *pos)
4125 struct softnet_data *sd = NULL;
4127 while (*pos < nr_cpu_ids)
4128 if (cpu_online(*pos)) {
4129 sd = &per_cpu(softnet_data, *pos);
4136 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4138 return softnet_get_online(pos);
4141 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4144 return softnet_get_online(pos);
4147 static void softnet_seq_stop(struct seq_file *seq, void *v)
4151 static int softnet_seq_show(struct seq_file *seq, void *v)
4153 struct softnet_data *sd = v;
4155 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4156 sd->processed, sd->dropped, sd->time_squeeze, 0,
4157 0, 0, 0, 0, /* was fastroute */
4158 sd->cpu_collision, sd->received_rps);
4162 static const struct seq_operations dev_seq_ops = {
4163 .start = dev_seq_start,
4164 .next = dev_seq_next,
4165 .stop = dev_seq_stop,
4166 .show = dev_seq_show,
4169 static int dev_seq_open(struct inode *inode, struct file *file)
4171 return seq_open_net(inode, file, &dev_seq_ops,
4172 sizeof(struct seq_net_private));
4175 static const struct file_operations dev_seq_fops = {
4176 .owner = THIS_MODULE,
4177 .open = dev_seq_open,
4179 .llseek = seq_lseek,
4180 .release = seq_release_net,
4183 static const struct seq_operations softnet_seq_ops = {
4184 .start = softnet_seq_start,
4185 .next = softnet_seq_next,
4186 .stop = softnet_seq_stop,
4187 .show = softnet_seq_show,
4190 static int softnet_seq_open(struct inode *inode, struct file *file)
4192 return seq_open(file, &softnet_seq_ops);
4195 static const struct file_operations softnet_seq_fops = {
4196 .owner = THIS_MODULE,
4197 .open = softnet_seq_open,
4199 .llseek = seq_lseek,
4200 .release = seq_release,
4203 static void *ptype_get_idx(loff_t pos)
4205 struct packet_type *pt = NULL;
4209 list_for_each_entry_rcu(pt, &ptype_all, list) {
4215 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4216 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4225 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4229 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4232 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4234 struct packet_type *pt;
4235 struct list_head *nxt;
4239 if (v == SEQ_START_TOKEN)
4240 return ptype_get_idx(0);
4243 nxt = pt->list.next;
4244 if (pt->type == htons(ETH_P_ALL)) {
4245 if (nxt != &ptype_all)
4248 nxt = ptype_base[0].next;
4250 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4252 while (nxt == &ptype_base[hash]) {
4253 if (++hash >= PTYPE_HASH_SIZE)
4255 nxt = ptype_base[hash].next;
4258 return list_entry(nxt, struct packet_type, list);
4261 static void ptype_seq_stop(struct seq_file *seq, void *v)
4267 static int ptype_seq_show(struct seq_file *seq, void *v)
4269 struct packet_type *pt = v;
4271 if (v == SEQ_START_TOKEN)
4272 seq_puts(seq, "Type Device Function\n");
4273 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4274 if (pt->type == htons(ETH_P_ALL))
4275 seq_puts(seq, "ALL ");
4277 seq_printf(seq, "%04x", ntohs(pt->type));
4279 seq_printf(seq, " %-8s %pF\n",
4280 pt->dev ? pt->dev->name : "", pt->func);
4286 static const struct seq_operations ptype_seq_ops = {
4287 .start = ptype_seq_start,
4288 .next = ptype_seq_next,
4289 .stop = ptype_seq_stop,
4290 .show = ptype_seq_show,
4293 static int ptype_seq_open(struct inode *inode, struct file *file)
4295 return seq_open_net(inode, file, &ptype_seq_ops,
4296 sizeof(struct seq_net_private));
4299 static const struct file_operations ptype_seq_fops = {
4300 .owner = THIS_MODULE,
4301 .open = ptype_seq_open,
4303 .llseek = seq_lseek,
4304 .release = seq_release_net,
4308 static int __net_init dev_proc_net_init(struct net *net)
4312 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4314 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4316 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4319 if (wext_proc_init(net))
4325 proc_net_remove(net, "ptype");
4327 proc_net_remove(net, "softnet_stat");
4329 proc_net_remove(net, "dev");
4333 static void __net_exit dev_proc_net_exit(struct net *net)
4335 wext_proc_exit(net);
4337 proc_net_remove(net, "ptype");
4338 proc_net_remove(net, "softnet_stat");
4339 proc_net_remove(net, "dev");
4342 static struct pernet_operations __net_initdata dev_proc_ops = {
4343 .init = dev_proc_net_init,
4344 .exit = dev_proc_net_exit,
4347 static int __init dev_proc_init(void)
4349 return register_pernet_subsys(&dev_proc_ops);
4352 #define dev_proc_init() 0
4353 #endif /* CONFIG_PROC_FS */
4357 * netdev_set_master - set up master pointer
4358 * @slave: slave device
4359 * @master: new master device
4361 * Changes the master device of the slave. Pass %NULL to break the
4362 * bonding. The caller must hold the RTNL semaphore. On a failure
4363 * a negative errno code is returned. On success the reference counts
4364 * are adjusted and the function returns zero.
4366 int netdev_set_master(struct net_device *slave, struct net_device *master)
4368 struct net_device *old = slave->master;
4378 slave->master = master;
4386 EXPORT_SYMBOL(netdev_set_master);
4389 * netdev_set_bond_master - set up bonding master/slave pair
4390 * @slave: slave device
4391 * @master: new master device
4393 * Changes the master device of the slave. Pass %NULL to break the
4394 * bonding. The caller must hold the RTNL semaphore. On a failure
4395 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4396 * to the routing socket and the function returns zero.
4398 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4404 err = netdev_set_master(slave, master);
4408 slave->flags |= IFF_SLAVE;
4410 slave->flags &= ~IFF_SLAVE;
4412 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4415 EXPORT_SYMBOL(netdev_set_bond_master);
4417 static void dev_change_rx_flags(struct net_device *dev, int flags)
4419 const struct net_device_ops *ops = dev->netdev_ops;
4421 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4422 ops->ndo_change_rx_flags(dev, flags);
4425 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4427 unsigned short old_flags = dev->flags;
4433 dev->flags |= IFF_PROMISC;
4434 dev->promiscuity += inc;
4435 if (dev->promiscuity == 0) {
4438 * If inc causes overflow, untouch promisc and return error.
4441 dev->flags &= ~IFF_PROMISC;
4443 dev->promiscuity -= inc;
4444 printk(KERN_WARNING "%s: promiscuity touches roof, "
4445 "set promiscuity failed, promiscuity feature "
4446 "of device might be broken.\n", dev->name);
4450 if (dev->flags != old_flags) {
4451 printk(KERN_INFO "device %s %s promiscuous mode\n",
4452 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4454 if (audit_enabled) {
4455 current_uid_gid(&uid, &gid);
4456 audit_log(current->audit_context, GFP_ATOMIC,
4457 AUDIT_ANOM_PROMISCUOUS,
4458 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4459 dev->name, (dev->flags & IFF_PROMISC),
4460 (old_flags & IFF_PROMISC),
4461 audit_get_loginuid(current),
4463 audit_get_sessionid(current));
4466 dev_change_rx_flags(dev, IFF_PROMISC);
4472 * dev_set_promiscuity - update promiscuity count on a device
4476 * Add or remove promiscuity from a device. While the count in the device
4477 * remains above zero the interface remains promiscuous. Once it hits zero
4478 * the device reverts back to normal filtering operation. A negative inc
4479 * value is used to drop promiscuity on the device.
4480 * Return 0 if successful or a negative errno code on error.
4482 int dev_set_promiscuity(struct net_device *dev, int inc)
4484 unsigned short old_flags = dev->flags;
4487 err = __dev_set_promiscuity(dev, inc);
4490 if (dev->flags != old_flags)
4491 dev_set_rx_mode(dev);
4494 EXPORT_SYMBOL(dev_set_promiscuity);
4497 * dev_set_allmulti - update allmulti count on a device
4501 * Add or remove reception of all multicast frames to a device. While the
4502 * count in the device remains above zero the interface remains listening
4503 * to all interfaces. Once it hits zero the device reverts back to normal
4504 * filtering operation. A negative @inc value is used to drop the counter
4505 * when releasing a resource needing all multicasts.
4506 * Return 0 if successful or a negative errno code on error.
4509 int dev_set_allmulti(struct net_device *dev, int inc)
4511 unsigned short old_flags = dev->flags;
4515 dev->flags |= IFF_ALLMULTI;
4516 dev->allmulti += inc;
4517 if (dev->allmulti == 0) {
4520 * If inc causes overflow, untouch allmulti and return error.
4523 dev->flags &= ~IFF_ALLMULTI;
4525 dev->allmulti -= inc;
4526 printk(KERN_WARNING "%s: allmulti touches roof, "
4527 "set allmulti failed, allmulti feature of "
4528 "device might be broken.\n", dev->name);
4532 if (dev->flags ^ old_flags) {
4533 dev_change_rx_flags(dev, IFF_ALLMULTI);
4534 dev_set_rx_mode(dev);
4538 EXPORT_SYMBOL(dev_set_allmulti);
4541 * Upload unicast and multicast address lists to device and
4542 * configure RX filtering. When the device doesn't support unicast
4543 * filtering it is put in promiscuous mode while unicast addresses
4546 void __dev_set_rx_mode(struct net_device *dev)
4548 const struct net_device_ops *ops = dev->netdev_ops;
4550 /* dev_open will call this function so the list will stay sane. */
4551 if (!(dev->flags&IFF_UP))
4554 if (!netif_device_present(dev))
4557 if (ops->ndo_set_rx_mode)
4558 ops->ndo_set_rx_mode(dev);
4560 /* Unicast addresses changes may only happen under the rtnl,
4561 * therefore calling __dev_set_promiscuity here is safe.
4563 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4564 __dev_set_promiscuity(dev, 1);
4565 dev->uc_promisc = 1;
4566 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4567 __dev_set_promiscuity(dev, -1);
4568 dev->uc_promisc = 0;
4571 if (ops->ndo_set_multicast_list)
4572 ops->ndo_set_multicast_list(dev);
4576 void dev_set_rx_mode(struct net_device *dev)
4578 netif_addr_lock_bh(dev);
4579 __dev_set_rx_mode(dev);
4580 netif_addr_unlock_bh(dev);
4584 * dev_get_flags - get flags reported to userspace
4587 * Get the combination of flag bits exported through APIs to userspace.
4589 unsigned dev_get_flags(const struct net_device *dev)
4593 flags = (dev->flags & ~(IFF_PROMISC |
4598 (dev->gflags & (IFF_PROMISC |
4601 if (netif_running(dev)) {
4602 if (netif_oper_up(dev))
4603 flags |= IFF_RUNNING;
4604 if (netif_carrier_ok(dev))
4605 flags |= IFF_LOWER_UP;
4606 if (netif_dormant(dev))
4607 flags |= IFF_DORMANT;
4612 EXPORT_SYMBOL(dev_get_flags);
4614 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4616 int old_flags = dev->flags;
4622 * Set the flags on our device.
4625 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4626 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4628 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4632 * Load in the correct multicast list now the flags have changed.
4635 if ((old_flags ^ flags) & IFF_MULTICAST)
4636 dev_change_rx_flags(dev, IFF_MULTICAST);
4638 dev_set_rx_mode(dev);
4641 * Have we downed the interface. We handle IFF_UP ourselves
4642 * according to user attempts to set it, rather than blindly
4647 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4648 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4651 dev_set_rx_mode(dev);
4654 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4655 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4657 dev->gflags ^= IFF_PROMISC;
4658 dev_set_promiscuity(dev, inc);
4661 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4662 is important. Some (broken) drivers set IFF_PROMISC, when
4663 IFF_ALLMULTI is requested not asking us and not reporting.
4665 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4666 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4668 dev->gflags ^= IFF_ALLMULTI;
4669 dev_set_allmulti(dev, inc);
4675 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4677 unsigned int changes = dev->flags ^ old_flags;
4679 if (changes & IFF_UP) {
4680 if (dev->flags & IFF_UP)
4681 call_netdevice_notifiers(NETDEV_UP, dev);
4683 call_netdevice_notifiers(NETDEV_DOWN, dev);
4686 if (dev->flags & IFF_UP &&
4687 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4688 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4692 * dev_change_flags - change device settings
4694 * @flags: device state flags
4696 * Change settings on device based state flags. The flags are
4697 * in the userspace exported format.
4699 int dev_change_flags(struct net_device *dev, unsigned flags)
4702 int old_flags = dev->flags;
4704 ret = __dev_change_flags(dev, flags);
4708 changes = old_flags ^ dev->flags;
4710 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4712 __dev_notify_flags(dev, old_flags);
4715 EXPORT_SYMBOL(dev_change_flags);
4718 * dev_set_mtu - Change maximum transfer unit
4720 * @new_mtu: new transfer unit
4722 * Change the maximum transfer size of the network device.
4724 int dev_set_mtu(struct net_device *dev, int new_mtu)
4726 const struct net_device_ops *ops = dev->netdev_ops;
4729 if (new_mtu == dev->mtu)
4732 /* MTU must be positive. */
4736 if (!netif_device_present(dev))
4740 if (ops->ndo_change_mtu)
4741 err = ops->ndo_change_mtu(dev, new_mtu);
4745 if (!err && dev->flags & IFF_UP)
4746 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4749 EXPORT_SYMBOL(dev_set_mtu);
4752 * dev_set_group - Change group this device belongs to
4754 * @new_group: group this device should belong to
4756 void dev_set_group(struct net_device *dev, int new_group)
4758 dev->group = new_group;
4760 EXPORT_SYMBOL(dev_set_group);
4763 * dev_set_mac_address - Change Media Access Control Address
4767 * Change the hardware (MAC) address of the device
4769 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4771 const struct net_device_ops *ops = dev->netdev_ops;
4774 if (!ops->ndo_set_mac_address)
4776 if (sa->sa_family != dev->type)
4778 if (!netif_device_present(dev))
4780 err = ops->ndo_set_mac_address(dev, sa);
4782 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4785 EXPORT_SYMBOL(dev_set_mac_address);
4788 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4790 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4793 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4799 case SIOCGIFFLAGS: /* Get interface flags */
4800 ifr->ifr_flags = (short) dev_get_flags(dev);
4803 case SIOCGIFMETRIC: /* Get the metric on the interface
4804 (currently unused) */
4805 ifr->ifr_metric = 0;
4808 case SIOCGIFMTU: /* Get the MTU of a device */
4809 ifr->ifr_mtu = dev->mtu;
4814 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4816 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4817 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4818 ifr->ifr_hwaddr.sa_family = dev->type;
4826 ifr->ifr_map.mem_start = dev->mem_start;
4827 ifr->ifr_map.mem_end = dev->mem_end;
4828 ifr->ifr_map.base_addr = dev->base_addr;
4829 ifr->ifr_map.irq = dev->irq;
4830 ifr->ifr_map.dma = dev->dma;
4831 ifr->ifr_map.port = dev->if_port;
4835 ifr->ifr_ifindex = dev->ifindex;
4839 ifr->ifr_qlen = dev->tx_queue_len;
4843 /* dev_ioctl() should ensure this case
4855 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4857 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4860 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4861 const struct net_device_ops *ops;
4866 ops = dev->netdev_ops;
4869 case SIOCSIFFLAGS: /* Set interface flags */
4870 return dev_change_flags(dev, ifr->ifr_flags);
4872 case SIOCSIFMETRIC: /* Set the metric on the interface
4873 (currently unused) */
4876 case SIOCSIFMTU: /* Set the MTU of a device */
4877 return dev_set_mtu(dev, ifr->ifr_mtu);
4880 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4882 case SIOCSIFHWBROADCAST:
4883 if (ifr->ifr_hwaddr.sa_family != dev->type)
4885 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4886 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4887 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4891 if (ops->ndo_set_config) {
4892 if (!netif_device_present(dev))
4894 return ops->ndo_set_config(dev, &ifr->ifr_map);
4899 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4900 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4902 if (!netif_device_present(dev))
4904 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4907 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4908 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4910 if (!netif_device_present(dev))
4912 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4915 if (ifr->ifr_qlen < 0)
4917 dev->tx_queue_len = ifr->ifr_qlen;
4921 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4922 return dev_change_name(dev, ifr->ifr_newname);
4925 * Unknown or private ioctl
4928 if ((cmd >= SIOCDEVPRIVATE &&
4929 cmd <= SIOCDEVPRIVATE + 15) ||
4930 cmd == SIOCBONDENSLAVE ||
4931 cmd == SIOCBONDRELEASE ||
4932 cmd == SIOCBONDSETHWADDR ||
4933 cmd == SIOCBONDSLAVEINFOQUERY ||
4934 cmd == SIOCBONDINFOQUERY ||
4935 cmd == SIOCBONDCHANGEACTIVE ||
4936 cmd == SIOCGMIIPHY ||
4937 cmd == SIOCGMIIREG ||
4938 cmd == SIOCSMIIREG ||
4939 cmd == SIOCBRADDIF ||
4940 cmd == SIOCBRDELIF ||
4941 cmd == SIOCSHWTSTAMP ||
4942 cmd == SIOCWANDEV) {
4944 if (ops->ndo_do_ioctl) {
4945 if (netif_device_present(dev))
4946 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4958 * This function handles all "interface"-type I/O control requests. The actual
4959 * 'doing' part of this is dev_ifsioc above.
4963 * dev_ioctl - network device ioctl
4964 * @net: the applicable net namespace
4965 * @cmd: command to issue
4966 * @arg: pointer to a struct ifreq in user space
4968 * Issue ioctl functions to devices. This is normally called by the
4969 * user space syscall interfaces but can sometimes be useful for
4970 * other purposes. The return value is the return from the syscall if
4971 * positive or a negative errno code on error.
4974 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4980 /* One special case: SIOCGIFCONF takes ifconf argument
4981 and requires shared lock, because it sleeps writing
4985 if (cmd == SIOCGIFCONF) {
4987 ret = dev_ifconf(net, (char __user *) arg);
4991 if (cmd == SIOCGIFNAME)
4992 return dev_ifname(net, (struct ifreq __user *)arg);
4994 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4997 ifr.ifr_name[IFNAMSIZ-1] = 0;
4999 colon = strchr(ifr.ifr_name, ':');
5004 * See which interface the caller is talking about.
5009 * These ioctl calls:
5010 * - can be done by all.
5011 * - atomic and do not require locking.
5022 dev_load(net, ifr.ifr_name);
5024 ret = dev_ifsioc_locked(net, &ifr, cmd);
5029 if (copy_to_user(arg, &ifr,
5030 sizeof(struct ifreq)))
5036 dev_load(net, ifr.ifr_name);
5038 ret = dev_ethtool(net, &ifr);
5043 if (copy_to_user(arg, &ifr,
5044 sizeof(struct ifreq)))
5050 * These ioctl calls:
5051 * - require superuser power.
5052 * - require strict serialization.
5058 if (!capable(CAP_NET_ADMIN))
5060 dev_load(net, ifr.ifr_name);
5062 ret = dev_ifsioc(net, &ifr, cmd);
5067 if (copy_to_user(arg, &ifr,
5068 sizeof(struct ifreq)))
5074 * These ioctl calls:
5075 * - require superuser power.
5076 * - require strict serialization.
5077 * - do not return a value
5087 case SIOCSIFHWBROADCAST:
5090 case SIOCBONDENSLAVE:
5091 case SIOCBONDRELEASE:
5092 case SIOCBONDSETHWADDR:
5093 case SIOCBONDCHANGEACTIVE:
5097 if (!capable(CAP_NET_ADMIN))
5100 case SIOCBONDSLAVEINFOQUERY:
5101 case SIOCBONDINFOQUERY:
5102 dev_load(net, ifr.ifr_name);
5104 ret = dev_ifsioc(net, &ifr, cmd);
5109 /* Get the per device memory space. We can add this but
5110 * currently do not support it */
5112 /* Set the per device memory buffer space.
5113 * Not applicable in our case */
5118 * Unknown or private ioctl.
5121 if (cmd == SIOCWANDEV ||
5122 (cmd >= SIOCDEVPRIVATE &&
5123 cmd <= SIOCDEVPRIVATE + 15)) {
5124 dev_load(net, ifr.ifr_name);
5126 ret = dev_ifsioc(net, &ifr, cmd);
5128 if (!ret && copy_to_user(arg, &ifr,
5129 sizeof(struct ifreq)))
5133 /* Take care of Wireless Extensions */
5134 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5135 return wext_handle_ioctl(net, &ifr, cmd, arg);
5142 * dev_new_index - allocate an ifindex
5143 * @net: the applicable net namespace
5145 * Returns a suitable unique value for a new device interface
5146 * number. The caller must hold the rtnl semaphore or the
5147 * dev_base_lock to be sure it remains unique.
5149 static int dev_new_index(struct net *net)
5155 if (!__dev_get_by_index(net, ifindex))
5160 /* Delayed registration/unregisteration */
5161 static LIST_HEAD(net_todo_list);
5163 static void net_set_todo(struct net_device *dev)
5165 list_add_tail(&dev->todo_list, &net_todo_list);
5168 static void rollback_registered_many(struct list_head *head)
5170 struct net_device *dev, *tmp;
5172 BUG_ON(dev_boot_phase);
5175 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5176 /* Some devices call without registering
5177 * for initialization unwind. Remove those
5178 * devices and proceed with the remaining.
5180 if (dev->reg_state == NETREG_UNINITIALIZED) {
5181 pr_debug("unregister_netdevice: device %s/%p never "
5182 "was registered\n", dev->name, dev);
5185 list_del(&dev->unreg_list);
5189 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5192 /* If device is running, close it first. */
5193 dev_close_many(head);
5195 list_for_each_entry(dev, head, unreg_list) {
5196 /* And unlink it from device chain. */
5197 unlist_netdevice(dev);
5199 dev->reg_state = NETREG_UNREGISTERING;
5204 list_for_each_entry(dev, head, unreg_list) {
5205 /* Shutdown queueing discipline. */
5209 /* Notify protocols, that we are about to destroy
5210 this device. They should clean all the things.
5212 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5214 if (!dev->rtnl_link_ops ||
5215 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5216 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5219 * Flush the unicast and multicast chains
5224 if (dev->netdev_ops->ndo_uninit)
5225 dev->netdev_ops->ndo_uninit(dev);
5227 /* Notifier chain MUST detach us from master device. */
5228 WARN_ON(dev->master);
5230 /* Remove entries from kobject tree */
5231 netdev_unregister_kobject(dev);
5234 /* Process any work delayed until the end of the batch */
5235 dev = list_first_entry(head, struct net_device, unreg_list);
5236 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5240 list_for_each_entry(dev, head, unreg_list)
5244 static void rollback_registered(struct net_device *dev)
5248 list_add(&dev->unreg_list, &single);
5249 rollback_registered_many(&single);
5253 u32 netdev_fix_features(struct net_device *dev, u32 features)
5255 /* Fix illegal checksum combinations */
5256 if ((features & NETIF_F_HW_CSUM) &&
5257 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5258 netdev_info(dev, "mixed HW and IP checksum settings.\n");
5259 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5262 if ((features & NETIF_F_NO_CSUM) &&
5263 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5264 netdev_info(dev, "mixed no checksumming and other settings.\n");
5265 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5268 /* Fix illegal SG+CSUM combinations. */
5269 if ((features & NETIF_F_SG) &&
5270 !(features & NETIF_F_ALL_CSUM)) {
5272 "Dropping NETIF_F_SG since no checksum feature.\n");
5273 features &= ~NETIF_F_SG;
5276 /* TSO requires that SG is present as well. */
5277 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
5278 netdev_info(dev, "Dropping NETIF_F_TSO since no SG feature.\n");
5279 features &= ~NETIF_F_TSO;
5282 /* Software GSO depends on SG. */
5283 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5284 netdev_info(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5285 features &= ~NETIF_F_GSO;
5288 /* UFO needs SG and checksumming */
5289 if (features & NETIF_F_UFO) {
5290 /* maybe split UFO into V4 and V6? */
5291 if (!((features & NETIF_F_GEN_CSUM) ||
5292 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5293 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5295 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5296 features &= ~NETIF_F_UFO;
5299 if (!(features & NETIF_F_SG)) {
5301 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5302 features &= ~NETIF_F_UFO;
5308 EXPORT_SYMBOL(netdev_fix_features);
5310 void netdev_update_features(struct net_device *dev)
5315 features = netdev_get_wanted_features(dev);
5317 if (dev->netdev_ops->ndo_fix_features)
5318 features = dev->netdev_ops->ndo_fix_features(dev, features);
5320 /* driver might be less strict about feature dependencies */
5321 features = netdev_fix_features(dev, features);
5323 if (dev->features == features)
5326 netdev_info(dev, "Features changed: 0x%08x -> 0x%08x\n",
5327 dev->features, features);
5329 if (dev->netdev_ops->ndo_set_features)
5330 err = dev->netdev_ops->ndo_set_features(dev, features);
5333 dev->features = features;
5336 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5337 err, features, dev->features);
5339 EXPORT_SYMBOL(netdev_update_features);
5342 * netif_stacked_transfer_operstate - transfer operstate
5343 * @rootdev: the root or lower level device to transfer state from
5344 * @dev: the device to transfer operstate to
5346 * Transfer operational state from root to device. This is normally
5347 * called when a stacking relationship exists between the root
5348 * device and the device(a leaf device).
5350 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5351 struct net_device *dev)
5353 if (rootdev->operstate == IF_OPER_DORMANT)
5354 netif_dormant_on(dev);
5356 netif_dormant_off(dev);
5358 if (netif_carrier_ok(rootdev)) {
5359 if (!netif_carrier_ok(dev))
5360 netif_carrier_on(dev);
5362 if (netif_carrier_ok(dev))
5363 netif_carrier_off(dev);
5366 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5369 static int netif_alloc_rx_queues(struct net_device *dev)
5371 unsigned int i, count = dev->num_rx_queues;
5372 struct netdev_rx_queue *rx;
5376 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5378 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5383 for (i = 0; i < count; i++)
5389 static void netdev_init_one_queue(struct net_device *dev,
5390 struct netdev_queue *queue, void *_unused)
5392 /* Initialize queue lock */
5393 spin_lock_init(&queue->_xmit_lock);
5394 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5395 queue->xmit_lock_owner = -1;
5396 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5400 static int netif_alloc_netdev_queues(struct net_device *dev)
5402 unsigned int count = dev->num_tx_queues;
5403 struct netdev_queue *tx;
5407 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5409 pr_err("netdev: Unable to allocate %u tx queues.\n",
5415 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5416 spin_lock_init(&dev->tx_global_lock);
5422 * register_netdevice - register a network device
5423 * @dev: device to register
5425 * Take a completed network device structure and add it to the kernel
5426 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5427 * chain. 0 is returned on success. A negative errno code is returned
5428 * on a failure to set up the device, or if the name is a duplicate.
5430 * Callers must hold the rtnl semaphore. You may want
5431 * register_netdev() instead of this.
5434 * The locking appears insufficient to guarantee two parallel registers
5435 * will not get the same name.
5438 int register_netdevice(struct net_device *dev)
5441 struct net *net = dev_net(dev);
5443 BUG_ON(dev_boot_phase);
5448 /* When net_device's are persistent, this will be fatal. */
5449 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5452 spin_lock_init(&dev->addr_list_lock);
5453 netdev_set_addr_lockdep_class(dev);
5457 /* Init, if this function is available */
5458 if (dev->netdev_ops->ndo_init) {
5459 ret = dev->netdev_ops->ndo_init(dev);
5467 ret = dev_get_valid_name(dev, dev->name, 0);
5471 dev->ifindex = dev_new_index(net);
5472 if (dev->iflink == -1)
5473 dev->iflink = dev->ifindex;
5475 /* Transfer changeable features to wanted_features and enable
5476 * software offloads (GSO and GRO).
5478 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5479 dev->wanted_features = (dev->features & dev->hw_features)
5480 | NETIF_F_SOFT_FEATURES;
5482 /* Avoid warning from netdev_fix_features() for GSO without SG */
5483 if (!(dev->wanted_features & NETIF_F_SG))
5484 dev->wanted_features &= ~NETIF_F_GSO;
5486 netdev_update_features(dev);
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;
5505 * Default initial state at registry is that the
5506 * device is present.
5509 set_bit(__LINK_STATE_PRESENT, &dev->state);
5511 dev_init_scheduler(dev);
5513 list_netdevice(dev);
5515 /* Notify protocols, that a new device appeared. */
5516 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5517 ret = notifier_to_errno(ret);
5519 rollback_registered(dev);
5520 dev->reg_state = NETREG_UNREGISTERED;
5523 * Prevent userspace races by waiting until the network
5524 * device is fully setup before sending notifications.
5526 if (!dev->rtnl_link_ops ||
5527 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5528 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5534 if (dev->netdev_ops->ndo_uninit)
5535 dev->netdev_ops->ndo_uninit(dev);
5538 EXPORT_SYMBOL(register_netdevice);
5541 * init_dummy_netdev - init a dummy network device for NAPI
5542 * @dev: device to init
5544 * This takes a network device structure and initialize the minimum
5545 * amount of fields so it can be used to schedule NAPI polls without
5546 * registering a full blown interface. This is to be used by drivers
5547 * that need to tie several hardware interfaces to a single NAPI
5548 * poll scheduler due to HW limitations.
5550 int init_dummy_netdev(struct net_device *dev)
5552 /* Clear everything. Note we don't initialize spinlocks
5553 * are they aren't supposed to be taken by any of the
5554 * NAPI code and this dummy netdev is supposed to be
5555 * only ever used for NAPI polls
5557 memset(dev, 0, sizeof(struct net_device));
5559 /* make sure we BUG if trying to hit standard
5560 * register/unregister code path
5562 dev->reg_state = NETREG_DUMMY;
5564 /* NAPI wants this */
5565 INIT_LIST_HEAD(&dev->napi_list);
5567 /* a dummy interface is started by default */
5568 set_bit(__LINK_STATE_PRESENT, &dev->state);
5569 set_bit(__LINK_STATE_START, &dev->state);
5571 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5572 * because users of this 'device' dont need to change
5578 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5582 * register_netdev - register a network device
5583 * @dev: device to register
5585 * Take a completed network device structure and add it to the kernel
5586 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5587 * chain. 0 is returned on success. A negative errno code is returned
5588 * on a failure to set up the device, or if the name is a duplicate.
5590 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5591 * and expands the device name if you passed a format string to
5594 int register_netdev(struct net_device *dev)
5601 * If the name is a format string the caller wants us to do a
5604 if (strchr(dev->name, '%')) {
5605 err = dev_alloc_name(dev, dev->name);
5610 err = register_netdevice(dev);
5615 EXPORT_SYMBOL(register_netdev);
5617 int netdev_refcnt_read(const struct net_device *dev)
5621 for_each_possible_cpu(i)
5622 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5625 EXPORT_SYMBOL(netdev_refcnt_read);
5628 * netdev_wait_allrefs - wait until all references are gone.
5630 * This is called when unregistering network devices.
5632 * Any protocol or device that holds a reference should register
5633 * for netdevice notification, and cleanup and put back the
5634 * reference if they receive an UNREGISTER event.
5635 * We can get stuck here if buggy protocols don't correctly
5638 static void netdev_wait_allrefs(struct net_device *dev)
5640 unsigned long rebroadcast_time, warning_time;
5643 linkwatch_forget_dev(dev);
5645 rebroadcast_time = warning_time = jiffies;
5646 refcnt = netdev_refcnt_read(dev);
5648 while (refcnt != 0) {
5649 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5652 /* Rebroadcast unregister notification */
5653 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5654 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5655 * should have already handle it the first time */
5657 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5659 /* We must not have linkwatch events
5660 * pending on unregister. If this
5661 * happens, we simply run the queue
5662 * unscheduled, resulting in a noop
5665 linkwatch_run_queue();
5670 rebroadcast_time = jiffies;
5675 refcnt = netdev_refcnt_read(dev);
5677 if (time_after(jiffies, warning_time + 10 * HZ)) {
5678 printk(KERN_EMERG "unregister_netdevice: "
5679 "waiting for %s to become free. Usage "
5682 warning_time = jiffies;
5691 * register_netdevice(x1);
5692 * register_netdevice(x2);
5694 * unregister_netdevice(y1);
5695 * unregister_netdevice(y2);
5701 * We are invoked by rtnl_unlock().
5702 * This allows us to deal with problems:
5703 * 1) We can delete sysfs objects which invoke hotplug
5704 * without deadlocking with linkwatch via keventd.
5705 * 2) Since we run with the RTNL semaphore not held, we can sleep
5706 * safely in order to wait for the netdev refcnt to drop to zero.
5708 * We must not return until all unregister events added during
5709 * the interval the lock was held have been completed.
5711 void netdev_run_todo(void)
5713 struct list_head list;
5715 /* Snapshot list, allow later requests */
5716 list_replace_init(&net_todo_list, &list);
5720 while (!list_empty(&list)) {
5721 struct net_device *dev
5722 = list_first_entry(&list, struct net_device, todo_list);
5723 list_del(&dev->todo_list);
5725 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5726 printk(KERN_ERR "network todo '%s' but state %d\n",
5727 dev->name, dev->reg_state);
5732 dev->reg_state = NETREG_UNREGISTERED;
5734 on_each_cpu(flush_backlog, dev, 1);
5736 netdev_wait_allrefs(dev);
5739 BUG_ON(netdev_refcnt_read(dev));
5740 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5741 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5742 WARN_ON(dev->dn_ptr);
5744 if (dev->destructor)
5745 dev->destructor(dev);
5747 /* Free network device */
5748 kobject_put(&dev->dev.kobj);
5752 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5753 * fields in the same order, with only the type differing.
5755 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5756 const struct net_device_stats *netdev_stats)
5758 #if BITS_PER_LONG == 64
5759 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5760 memcpy(stats64, netdev_stats, sizeof(*stats64));
5762 size_t i, n = sizeof(*stats64) / sizeof(u64);
5763 const unsigned long *src = (const unsigned long *)netdev_stats;
5764 u64 *dst = (u64 *)stats64;
5766 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5767 sizeof(*stats64) / sizeof(u64));
5768 for (i = 0; i < n; i++)
5774 * dev_get_stats - get network device statistics
5775 * @dev: device to get statistics from
5776 * @storage: place to store stats
5778 * Get network statistics from device. Return @storage.
5779 * The device driver may provide its own method by setting
5780 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5781 * otherwise the internal statistics structure is used.
5783 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5784 struct rtnl_link_stats64 *storage)
5786 const struct net_device_ops *ops = dev->netdev_ops;
5788 if (ops->ndo_get_stats64) {
5789 memset(storage, 0, sizeof(*storage));
5790 ops->ndo_get_stats64(dev, storage);
5791 } else if (ops->ndo_get_stats) {
5792 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5794 netdev_stats_to_stats64(storage, &dev->stats);
5796 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5799 EXPORT_SYMBOL(dev_get_stats);
5801 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5803 struct netdev_queue *queue = dev_ingress_queue(dev);
5805 #ifdef CONFIG_NET_CLS_ACT
5808 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5811 netdev_init_one_queue(dev, queue, NULL);
5812 queue->qdisc = &noop_qdisc;
5813 queue->qdisc_sleeping = &noop_qdisc;
5814 rcu_assign_pointer(dev->ingress_queue, queue);
5820 * alloc_netdev_mqs - allocate network device
5821 * @sizeof_priv: size of private data to allocate space for
5822 * @name: device name format string
5823 * @setup: callback to initialize device
5824 * @txqs: the number of TX subqueues to allocate
5825 * @rxqs: the number of RX subqueues to allocate
5827 * Allocates a struct net_device with private data area for driver use
5828 * and performs basic initialization. Also allocates subquue structs
5829 * for each queue on the device.
5831 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5832 void (*setup)(struct net_device *),
5833 unsigned int txqs, unsigned int rxqs)
5835 struct net_device *dev;
5837 struct net_device *p;
5839 BUG_ON(strlen(name) >= sizeof(dev->name));
5842 pr_err("alloc_netdev: Unable to allocate device "
5843 "with zero queues.\n");
5849 pr_err("alloc_netdev: Unable to allocate device "
5850 "with zero RX queues.\n");
5855 alloc_size = sizeof(struct net_device);
5857 /* ensure 32-byte alignment of private area */
5858 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5859 alloc_size += sizeof_priv;
5861 /* ensure 32-byte alignment of whole construct */
5862 alloc_size += NETDEV_ALIGN - 1;
5864 p = kzalloc(alloc_size, GFP_KERNEL);
5866 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5870 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5871 dev->padded = (char *)dev - (char *)p;
5873 dev->pcpu_refcnt = alloc_percpu(int);
5874 if (!dev->pcpu_refcnt)
5877 if (dev_addr_init(dev))
5883 dev_net_set(dev, &init_net);
5885 dev->gso_max_size = GSO_MAX_SIZE;
5887 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5888 dev->ethtool_ntuple_list.count = 0;
5889 INIT_LIST_HEAD(&dev->napi_list);
5890 INIT_LIST_HEAD(&dev->unreg_list);
5891 INIT_LIST_HEAD(&dev->link_watch_list);
5892 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5895 dev->num_tx_queues = txqs;
5896 dev->real_num_tx_queues = txqs;
5897 if (netif_alloc_netdev_queues(dev))
5901 dev->num_rx_queues = rxqs;
5902 dev->real_num_rx_queues = rxqs;
5903 if (netif_alloc_rx_queues(dev))
5907 strcpy(dev->name, name);
5908 dev->group = INIT_NETDEV_GROUP;
5916 free_percpu(dev->pcpu_refcnt);
5926 EXPORT_SYMBOL(alloc_netdev_mqs);
5929 * free_netdev - free network device
5932 * This function does the last stage of destroying an allocated device
5933 * interface. The reference to the device object is released.
5934 * If this is the last reference then it will be freed.
5936 void free_netdev(struct net_device *dev)
5938 struct napi_struct *p, *n;
5940 release_net(dev_net(dev));
5947 kfree(rcu_dereference_raw(dev->ingress_queue));
5949 /* Flush device addresses */
5950 dev_addr_flush(dev);
5952 /* Clear ethtool n-tuple list */
5953 ethtool_ntuple_flush(dev);
5955 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5958 free_percpu(dev->pcpu_refcnt);
5959 dev->pcpu_refcnt = NULL;
5961 /* Compatibility with error handling in drivers */
5962 if (dev->reg_state == NETREG_UNINITIALIZED) {
5963 kfree((char *)dev - dev->padded);
5967 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5968 dev->reg_state = NETREG_RELEASED;
5970 /* will free via device release */
5971 put_device(&dev->dev);
5973 EXPORT_SYMBOL(free_netdev);
5976 * synchronize_net - Synchronize with packet receive processing
5978 * Wait for packets currently being received to be done.
5979 * Does not block later packets from starting.
5981 void synchronize_net(void)
5986 EXPORT_SYMBOL(synchronize_net);
5989 * unregister_netdevice_queue - remove device from the kernel
5993 * This function shuts down a device interface and removes it
5994 * from the kernel tables.
5995 * If head not NULL, device is queued to be unregistered later.
5997 * Callers must hold the rtnl semaphore. You may want
5998 * unregister_netdev() instead of this.
6001 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6006 list_move_tail(&dev->unreg_list, head);
6008 rollback_registered(dev);
6009 /* Finish processing unregister after unlock */
6013 EXPORT_SYMBOL(unregister_netdevice_queue);
6016 * unregister_netdevice_many - unregister many devices
6017 * @head: list of devices
6019 void unregister_netdevice_many(struct list_head *head)
6021 struct net_device *dev;
6023 if (!list_empty(head)) {
6024 rollback_registered_many(head);
6025 list_for_each_entry(dev, head, unreg_list)
6029 EXPORT_SYMBOL(unregister_netdevice_many);
6032 * unregister_netdev - remove device from the kernel
6035 * This function shuts down a device interface and removes it
6036 * from the kernel tables.
6038 * This is just a wrapper for unregister_netdevice that takes
6039 * the rtnl semaphore. In general you want to use this and not
6040 * unregister_netdevice.
6042 void unregister_netdev(struct net_device *dev)
6045 unregister_netdevice(dev);
6048 EXPORT_SYMBOL(unregister_netdev);
6051 * dev_change_net_namespace - move device to different nethost namespace
6053 * @net: network namespace
6054 * @pat: If not NULL name pattern to try if the current device name
6055 * is already taken in the destination network namespace.
6057 * This function shuts down a device interface and moves it
6058 * to a new network namespace. On success 0 is returned, on
6059 * a failure a netagive errno code is returned.
6061 * Callers must hold the rtnl semaphore.
6064 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6070 /* Don't allow namespace local devices to be moved. */
6072 if (dev->features & NETIF_F_NETNS_LOCAL)
6075 /* Ensure the device has been registrered */
6077 if (dev->reg_state != NETREG_REGISTERED)
6080 /* Get out if there is nothing todo */
6082 if (net_eq(dev_net(dev), net))
6085 /* Pick the destination device name, and ensure
6086 * we can use it in the destination network namespace.
6089 if (__dev_get_by_name(net, dev->name)) {
6090 /* We get here if we can't use the current device name */
6093 if (dev_get_valid_name(dev, pat, 1))
6098 * And now a mini version of register_netdevice unregister_netdevice.
6101 /* If device is running close it first. */
6104 /* And unlink it from device chain */
6106 unlist_netdevice(dev);
6110 /* Shutdown queueing discipline. */
6113 /* Notify protocols, that we are about to destroy
6114 this device. They should clean all the things.
6116 Note that dev->reg_state stays at NETREG_REGISTERED.
6117 This is wanted because this way 8021q and macvlan know
6118 the device is just moving and can keep their slaves up.
6120 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6121 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6124 * Flush the unicast and multicast chains
6129 /* Actually switch the network namespace */
6130 dev_net_set(dev, net);
6132 /* If there is an ifindex conflict assign a new one */
6133 if (__dev_get_by_index(net, dev->ifindex)) {
6134 int iflink = (dev->iflink == dev->ifindex);
6135 dev->ifindex = dev_new_index(net);
6137 dev->iflink = dev->ifindex;
6140 /* Fixup kobjects */
6141 err = device_rename(&dev->dev, dev->name);
6144 /* Add the device back in the hashes */
6145 list_netdevice(dev);
6147 /* Notify protocols, that a new device appeared. */
6148 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6151 * Prevent userspace races by waiting until the network
6152 * device is fully setup before sending notifications.
6154 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6161 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6163 static int dev_cpu_callback(struct notifier_block *nfb,
6164 unsigned long action,
6167 struct sk_buff **list_skb;
6168 struct sk_buff *skb;
6169 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6170 struct softnet_data *sd, *oldsd;
6172 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6175 local_irq_disable();
6176 cpu = smp_processor_id();
6177 sd = &per_cpu(softnet_data, cpu);
6178 oldsd = &per_cpu(softnet_data, oldcpu);
6180 /* Find end of our completion_queue. */
6181 list_skb = &sd->completion_queue;
6183 list_skb = &(*list_skb)->next;
6184 /* Append completion queue from offline CPU. */
6185 *list_skb = oldsd->completion_queue;
6186 oldsd->completion_queue = NULL;
6188 /* Append output queue from offline CPU. */
6189 if (oldsd->output_queue) {
6190 *sd->output_queue_tailp = oldsd->output_queue;
6191 sd->output_queue_tailp = oldsd->output_queue_tailp;
6192 oldsd->output_queue = NULL;
6193 oldsd->output_queue_tailp = &oldsd->output_queue;
6196 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6199 /* Process offline CPU's input_pkt_queue */
6200 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6202 input_queue_head_incr(oldsd);
6204 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6206 input_queue_head_incr(oldsd);
6214 * netdev_increment_features - increment feature set by one
6215 * @all: current feature set
6216 * @one: new feature set
6217 * @mask: mask feature set
6219 * Computes a new feature set after adding a device with feature set
6220 * @one to the master device with current feature set @all. Will not
6221 * enable anything that is off in @mask. Returns the new feature set.
6223 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6225 /* If device needs checksumming, downgrade to it. */
6226 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
6227 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
6228 else if (mask & NETIF_F_ALL_CSUM) {
6229 /* If one device supports v4/v6 checksumming, set for all. */
6230 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
6231 !(all & NETIF_F_GEN_CSUM)) {
6232 all &= ~NETIF_F_ALL_CSUM;
6233 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
6236 /* If one device supports hw checksumming, set for all. */
6237 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
6238 all &= ~NETIF_F_ALL_CSUM;
6239 all |= NETIF_F_HW_CSUM;
6243 one |= NETIF_F_ALL_CSUM;
6245 one |= all & NETIF_F_ONE_FOR_ALL;
6246 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
6247 all |= one & mask & NETIF_F_ONE_FOR_ALL;
6251 EXPORT_SYMBOL(netdev_increment_features);
6253 static struct hlist_head *netdev_create_hash(void)
6256 struct hlist_head *hash;
6258 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6260 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6261 INIT_HLIST_HEAD(&hash[i]);
6266 /* Initialize per network namespace state */
6267 static int __net_init netdev_init(struct net *net)
6269 INIT_LIST_HEAD(&net->dev_base_head);
6271 net->dev_name_head = netdev_create_hash();
6272 if (net->dev_name_head == NULL)
6275 net->dev_index_head = netdev_create_hash();
6276 if (net->dev_index_head == NULL)
6282 kfree(net->dev_name_head);
6288 * netdev_drivername - network driver for the device
6289 * @dev: network device
6290 * @buffer: buffer for resulting name
6291 * @len: size of buffer
6293 * Determine network driver for device.
6295 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6297 const struct device_driver *driver;
6298 const struct device *parent;
6300 if (len <= 0 || !buffer)
6304 parent = dev->dev.parent;
6309 driver = parent->driver;
6310 if (driver && driver->name)
6311 strlcpy(buffer, driver->name, len);
6315 static int __netdev_printk(const char *level, const struct net_device *dev,
6316 struct va_format *vaf)
6320 if (dev && dev->dev.parent)
6321 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6322 netdev_name(dev), vaf);
6324 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6326 r = printk("%s(NULL net_device): %pV", level, vaf);
6331 int netdev_printk(const char *level, const struct net_device *dev,
6332 const char *format, ...)
6334 struct va_format vaf;
6338 va_start(args, format);
6343 r = __netdev_printk(level, dev, &vaf);
6348 EXPORT_SYMBOL(netdev_printk);
6350 #define define_netdev_printk_level(func, level) \
6351 int func(const struct net_device *dev, const char *fmt, ...) \
6354 struct va_format vaf; \
6357 va_start(args, fmt); \
6362 r = __netdev_printk(level, dev, &vaf); \
6367 EXPORT_SYMBOL(func);
6369 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6370 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6371 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6372 define_netdev_printk_level(netdev_err, KERN_ERR);
6373 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6374 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6375 define_netdev_printk_level(netdev_info, KERN_INFO);
6377 static void __net_exit netdev_exit(struct net *net)
6379 kfree(net->dev_name_head);
6380 kfree(net->dev_index_head);
6383 static struct pernet_operations __net_initdata netdev_net_ops = {
6384 .init = netdev_init,
6385 .exit = netdev_exit,
6388 static void __net_exit default_device_exit(struct net *net)
6390 struct net_device *dev, *aux;
6392 * Push all migratable network devices back to the
6393 * initial network namespace
6396 for_each_netdev_safe(net, dev, aux) {
6398 char fb_name[IFNAMSIZ];
6400 /* Ignore unmoveable devices (i.e. loopback) */
6401 if (dev->features & NETIF_F_NETNS_LOCAL)
6404 /* Leave virtual devices for the generic cleanup */
6405 if (dev->rtnl_link_ops)
6408 /* Push remaing network devices to init_net */
6409 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6410 err = dev_change_net_namespace(dev, &init_net, fb_name);
6412 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6413 __func__, dev->name, err);
6420 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6422 /* At exit all network devices most be removed from a network
6423 * namespace. Do this in the reverse order of registration.
6424 * Do this across as many network namespaces as possible to
6425 * improve batching efficiency.
6427 struct net_device *dev;
6429 LIST_HEAD(dev_kill_list);
6432 list_for_each_entry(net, net_list, exit_list) {
6433 for_each_netdev_reverse(net, dev) {
6434 if (dev->rtnl_link_ops)
6435 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6437 unregister_netdevice_queue(dev, &dev_kill_list);
6440 unregister_netdevice_many(&dev_kill_list);
6441 list_del(&dev_kill_list);
6445 static struct pernet_operations __net_initdata default_device_ops = {
6446 .exit = default_device_exit,
6447 .exit_batch = default_device_exit_batch,
6451 * Initialize the DEV module. At boot time this walks the device list and
6452 * unhooks any devices that fail to initialise (normally hardware not
6453 * present) and leaves us with a valid list of present and active devices.
6458 * This is called single threaded during boot, so no need
6459 * to take the rtnl semaphore.
6461 static int __init net_dev_init(void)
6463 int i, rc = -ENOMEM;
6465 BUG_ON(!dev_boot_phase);
6467 if (dev_proc_init())
6470 if (netdev_kobject_init())
6473 INIT_LIST_HEAD(&ptype_all);
6474 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6475 INIT_LIST_HEAD(&ptype_base[i]);
6477 if (register_pernet_subsys(&netdev_net_ops))
6481 * Initialise the packet receive queues.
6484 for_each_possible_cpu(i) {
6485 struct softnet_data *sd = &per_cpu(softnet_data, i);
6487 memset(sd, 0, sizeof(*sd));
6488 skb_queue_head_init(&sd->input_pkt_queue);
6489 skb_queue_head_init(&sd->process_queue);
6490 sd->completion_queue = NULL;
6491 INIT_LIST_HEAD(&sd->poll_list);
6492 sd->output_queue = NULL;
6493 sd->output_queue_tailp = &sd->output_queue;
6495 sd->csd.func = rps_trigger_softirq;
6501 sd->backlog.poll = process_backlog;
6502 sd->backlog.weight = weight_p;
6503 sd->backlog.gro_list = NULL;
6504 sd->backlog.gro_count = 0;
6509 /* The loopback device is special if any other network devices
6510 * is present in a network namespace the loopback device must
6511 * be present. Since we now dynamically allocate and free the
6512 * loopback device ensure this invariant is maintained by
6513 * keeping the loopback device as the first device on the
6514 * list of network devices. Ensuring the loopback devices
6515 * is the first device that appears and the last network device
6518 if (register_pernet_device(&loopback_net_ops))
6521 if (register_pernet_device(&default_device_ops))
6524 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6525 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6527 hotcpu_notifier(dev_cpu_callback, 0);
6535 subsys_initcall(net_dev_init);
6537 static int __init initialize_hashrnd(void)
6539 get_random_bytes(&hashrnd, sizeof(hashrnd));
6543 late_initcall_sync(initialize_hashrnd);