2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
135 #include <linux/cpu_rmap.h>
137 #include "net-sysfs.h"
139 /* Instead of increasing this, you should create a hash table. */
140 #define MAX_GRO_SKBS 8
142 /* This should be increased if a protocol with a bigger head is added. */
143 #define GRO_MAX_HEAD (MAX_HEADER + 128)
146 * The list of packet types we will receive (as opposed to discard)
147 * and the routines to invoke.
149 * Why 16. Because with 16 the only overlap we get on a hash of the
150 * low nibble of the protocol value is RARP/SNAP/X.25.
152 * NOTE: That is no longer true with the addition of VLAN tags. Not
153 * sure which should go first, but I bet it won't make much
154 * difference if we are running VLANs. The good news is that
155 * this protocol won't be in the list unless compiled in, so
156 * the average user (w/out VLANs) will not be adversely affected.
173 #define PTYPE_HASH_SIZE (16)
174 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
176 static DEFINE_SPINLOCK(ptype_lock);
177 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
178 static struct list_head ptype_all __read_mostly; /* Taps */
181 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
184 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
186 * Writers must hold the rtnl semaphore while they loop through the
187 * dev_base_head list, and hold dev_base_lock for writing when they do the
188 * actual updates. This allows pure readers to access the list even
189 * while a writer is preparing to update it.
191 * To put it another way, dev_base_lock is held for writing only to
192 * protect against pure readers; the rtnl semaphore provides the
193 * protection against other writers.
195 * See, for example usages, register_netdevice() and
196 * unregister_netdevice(), which must be called with the rtnl
199 DEFINE_RWLOCK(dev_base_lock);
200 EXPORT_SYMBOL(dev_base_lock);
202 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
204 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
205 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
208 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
210 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
213 static inline void rps_lock(struct softnet_data *sd)
216 spin_lock(&sd->input_pkt_queue.lock);
220 static inline void rps_unlock(struct softnet_data *sd)
223 spin_unlock(&sd->input_pkt_queue.lock);
227 /* Device list insertion */
228 static int list_netdevice(struct net_device *dev)
230 struct net *net = dev_net(dev);
234 write_lock_bh(&dev_base_lock);
235 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
236 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
237 hlist_add_head_rcu(&dev->index_hlist,
238 dev_index_hash(net, dev->ifindex));
239 write_unlock_bh(&dev_base_lock);
243 /* Device list removal
244 * caller must respect a RCU grace period before freeing/reusing dev
246 static void unlist_netdevice(struct net_device *dev)
250 /* Unlink dev from the device chain */
251 write_lock_bh(&dev_base_lock);
252 list_del_rcu(&dev->dev_list);
253 hlist_del_rcu(&dev->name_hlist);
254 hlist_del_rcu(&dev->index_hlist);
255 write_unlock_bh(&dev_base_lock);
262 static RAW_NOTIFIER_HEAD(netdev_chain);
265 * Device drivers call our routines to queue packets here. We empty the
266 * queue in the local softnet handler.
269 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
270 EXPORT_PER_CPU_SYMBOL(softnet_data);
272 #ifdef CONFIG_LOCKDEP
274 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
275 * according to dev->type
277 static const unsigned short netdev_lock_type[] =
278 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
279 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
280 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
281 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
282 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
283 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
284 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
285 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
286 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
287 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
288 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
289 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
290 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
291 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
292 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
293 ARPHRD_VOID, ARPHRD_NONE};
295 static const char *const netdev_lock_name[] =
296 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
297 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
298 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
299 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
300 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
301 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
302 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
303 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
304 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
305 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
306 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
307 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
308 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
309 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
310 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
311 "_xmit_VOID", "_xmit_NONE"};
313 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
316 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
320 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
321 if (netdev_lock_type[i] == dev_type)
323 /* the last key is used by default */
324 return ARRAY_SIZE(netdev_lock_type) - 1;
327 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
328 unsigned short dev_type)
332 i = netdev_lock_pos(dev_type);
333 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
334 netdev_lock_name[i]);
337 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
341 i = netdev_lock_pos(dev->type);
342 lockdep_set_class_and_name(&dev->addr_list_lock,
343 &netdev_addr_lock_key[i],
344 netdev_lock_name[i]);
347 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
348 unsigned short dev_type)
351 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
356 /*******************************************************************************
358 Protocol management and registration routines
360 *******************************************************************************/
363 * Add a protocol ID to the list. Now that the input handler is
364 * smarter we can dispense with all the messy stuff that used to be
367 * BEWARE!!! Protocol handlers, mangling input packets,
368 * MUST BE last in hash buckets and checking protocol handlers
369 * MUST start from promiscuous ptype_all chain in net_bh.
370 * It is true now, do not change it.
371 * Explanation follows: if protocol handler, mangling packet, will
372 * be the first on list, it is not able to sense, that packet
373 * is cloned and should be copied-on-write, so that it will
374 * change it and subsequent readers will get broken packet.
378 static inline struct list_head *ptype_head(const struct packet_type *pt)
380 if (pt->type == htons(ETH_P_ALL))
383 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
387 * dev_add_pack - add packet handler
388 * @pt: packet type declaration
390 * Add a protocol handler to the networking stack. The passed &packet_type
391 * is linked into kernel lists and may not be freed until it has been
392 * removed from the kernel lists.
394 * This call does not sleep therefore it can not
395 * guarantee all CPU's that are in middle of receiving packets
396 * will see the new packet type (until the next received packet).
399 void dev_add_pack(struct packet_type *pt)
401 struct list_head *head = ptype_head(pt);
403 spin_lock(&ptype_lock);
404 list_add_rcu(&pt->list, head);
405 spin_unlock(&ptype_lock);
407 EXPORT_SYMBOL(dev_add_pack);
410 * __dev_remove_pack - remove packet handler
411 * @pt: packet type declaration
413 * Remove a protocol handler that was previously added to the kernel
414 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
415 * from the kernel lists and can be freed or reused once this function
418 * The packet type might still be in use by receivers
419 * and must not be freed until after all the CPU's have gone
420 * through a quiescent state.
422 void __dev_remove_pack(struct packet_type *pt)
424 struct list_head *head = ptype_head(pt);
425 struct packet_type *pt1;
427 spin_lock(&ptype_lock);
429 list_for_each_entry(pt1, head, list) {
431 list_del_rcu(&pt->list);
436 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
438 spin_unlock(&ptype_lock);
440 EXPORT_SYMBOL(__dev_remove_pack);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type *pt)
456 __dev_remove_pack(pt);
460 EXPORT_SYMBOL(dev_remove_pack);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name, struct ifmap *map)
482 struct netdev_boot_setup *s;
486 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
487 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
488 memset(s[i].name, 0, sizeof(s[i].name));
489 strlcpy(s[i].name, name, IFNAMSIZ);
490 memcpy(&s[i].map, map, sizeof(s[i].map));
495 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device *dev)
509 struct netdev_boot_setup *s = dev_boot_setup;
512 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
513 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
514 !strcmp(dev->name, s[i].name)) {
515 dev->irq = s[i].map.irq;
516 dev->base_addr = s[i].map.base_addr;
517 dev->mem_start = s[i].map.mem_start;
518 dev->mem_end = s[i].map.mem_end;
524 EXPORT_SYMBOL(netdev_boot_setup_check);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix, int unit)
539 const struct netdev_boot_setup *s = dev_boot_setup;
543 sprintf(name, "%s%d", prefix, unit);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net, name))
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
553 if (!strcmp(name, s[i].name))
554 return s[i].map.base_addr;
559 * Saves at boot time configured settings for any netdevice.
561 int __init netdev_boot_setup(char *str)
566 str = get_options(str, ARRAY_SIZE(ints), ints);
571 memset(&map, 0, sizeof(map));
575 map.base_addr = ints[2];
577 map.mem_start = ints[3];
579 map.mem_end = ints[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str, &map);
585 __setup("netdev=", netdev_boot_setup);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device *__dev_get_by_name(struct net *net, const char *name)
607 struct hlist_node *p;
608 struct net_device *dev;
609 struct hlist_head *head = dev_name_hash(net, name);
611 hlist_for_each_entry(dev, p, head, name_hlist)
612 if (!strncmp(dev->name, name, IFNAMSIZ))
617 EXPORT_SYMBOL(__dev_get_by_name);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
633 struct hlist_node *p;
634 struct net_device *dev;
635 struct hlist_head *head = dev_name_hash(net, name);
637 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
638 if (!strncmp(dev->name, name, IFNAMSIZ))
643 EXPORT_SYMBOL(dev_get_by_name_rcu);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device *dev_get_by_name(struct net *net, const char *name)
659 struct net_device *dev;
662 dev = dev_get_by_name_rcu(net, name);
668 EXPORT_SYMBOL(dev_get_by_name);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
684 struct hlist_node *p;
685 struct net_device *dev;
686 struct hlist_head *head = dev_index_hash(net, ifindex);
688 hlist_for_each_entry(dev, p, head, index_hlist)
689 if (dev->ifindex == ifindex)
694 EXPORT_SYMBOL(__dev_get_by_index);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
709 struct hlist_node *p;
710 struct net_device *dev;
711 struct hlist_head *head = dev_index_hash(net, ifindex);
713 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
714 if (dev->ifindex == ifindex)
719 EXPORT_SYMBOL(dev_get_by_index_rcu);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device *dev_get_by_index(struct net *net, int ifindex)
735 struct net_device *dev;
738 dev = dev_get_by_index_rcu(net, ifindex);
744 EXPORT_SYMBOL(dev_get_by_index);
747 * dev_getbyhwaddr_rcu - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device.
754 * The caller must hold RCU or RTNL.
755 * The returned device has not had its ref count increased
756 * and the caller must therefore be careful about locking
760 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
763 struct net_device *dev;
765 for_each_netdev_rcu(net, dev)
766 if (dev->type == type &&
767 !memcmp(dev->dev_addr, ha, dev->addr_len))
772 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
774 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
776 struct net_device *dev;
779 for_each_netdev(net, dev)
780 if (dev->type == type)
785 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
787 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
789 struct net_device *dev, *ret = NULL;
792 for_each_netdev_rcu(net, dev)
793 if (dev->type == type) {
801 EXPORT_SYMBOL(dev_getfirstbyhwtype);
804 * dev_get_by_flags_rcu - find any device with given flags
805 * @net: the applicable net namespace
806 * @if_flags: IFF_* values
807 * @mask: bitmask of bits in if_flags to check
809 * Search for any interface with the given flags. Returns NULL if a device
810 * is not found or a pointer to the device. Must be called inside
811 * rcu_read_lock(), and result refcount is unchanged.
814 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
817 struct net_device *dev, *ret;
820 for_each_netdev_rcu(net, dev) {
821 if (((dev->flags ^ if_flags) & mask) == 0) {
828 EXPORT_SYMBOL(dev_get_by_flags_rcu);
831 * dev_valid_name - check if name is okay for network device
834 * Network device names need to be valid file names to
835 * to allow sysfs to work. We also disallow any kind of
838 int dev_valid_name(const char *name)
842 if (strlen(name) >= IFNAMSIZ)
844 if (!strcmp(name, ".") || !strcmp(name, ".."))
848 if (*name == '/' || isspace(*name))
854 EXPORT_SYMBOL(dev_valid_name);
857 * __dev_alloc_name - allocate a name for a device
858 * @net: network namespace to allocate the device name in
859 * @name: name format string
860 * @buf: scratch buffer and result name string
862 * Passed a format string - eg "lt%d" it will try and find a suitable
863 * id. It scans list of devices to build up a free map, then chooses
864 * the first empty slot. The caller must hold the dev_base or rtnl lock
865 * while allocating the name and adding the device in order to avoid
867 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
868 * Returns the number of the unit assigned or a negative errno code.
871 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
875 const int max_netdevices = 8*PAGE_SIZE;
876 unsigned long *inuse;
877 struct net_device *d;
879 p = strnchr(name, IFNAMSIZ-1, '%');
882 * Verify the string as this thing may have come from
883 * the user. There must be either one "%d" and no other "%"
886 if (p[1] != 'd' || strchr(p + 2, '%'))
889 /* Use one page as a bit array of possible slots */
890 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
894 for_each_netdev(net, d) {
895 if (!sscanf(d->name, name, &i))
897 if (i < 0 || i >= max_netdevices)
900 /* avoid cases where sscanf is not exact inverse of printf */
901 snprintf(buf, IFNAMSIZ, name, i);
902 if (!strncmp(buf, d->name, IFNAMSIZ))
906 i = find_first_zero_bit(inuse, max_netdevices);
907 free_page((unsigned long) inuse);
911 snprintf(buf, IFNAMSIZ, name, i);
912 if (!__dev_get_by_name(net, buf))
915 /* It is possible to run out of possible slots
916 * when the name is long and there isn't enough space left
917 * for the digits, or if all bits are used.
923 * dev_alloc_name - allocate a name for a device
925 * @name: name format string
927 * Passed a format string - eg "lt%d" it will try and find a suitable
928 * id. It scans list of devices to build up a free map, then chooses
929 * the first empty slot. The caller must hold the dev_base or rtnl lock
930 * while allocating the name and adding the device in order to avoid
932 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
933 * Returns the number of the unit assigned or a negative errno code.
936 int dev_alloc_name(struct net_device *dev, const char *name)
942 BUG_ON(!dev_net(dev));
944 ret = __dev_alloc_name(net, name, buf);
946 strlcpy(dev->name, buf, IFNAMSIZ);
949 EXPORT_SYMBOL(dev_alloc_name);
951 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
955 BUG_ON(!dev_net(dev));
958 if (!dev_valid_name(name))
961 if (fmt && strchr(name, '%'))
962 return dev_alloc_name(dev, name);
963 else if (__dev_get_by_name(net, name))
965 else if (dev->name != name)
966 strlcpy(dev->name, name, IFNAMSIZ);
972 * dev_change_name - change name of a device
974 * @newname: name (or format string) must be at least IFNAMSIZ
976 * Change name of a device, can pass format strings "eth%d".
979 int dev_change_name(struct net_device *dev, const char *newname)
981 char oldname[IFNAMSIZ];
987 BUG_ON(!dev_net(dev));
990 if (dev->flags & IFF_UP)
993 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
996 memcpy(oldname, dev->name, IFNAMSIZ);
998 err = dev_get_valid_name(dev, newname, 1);
1003 ret = device_rename(&dev->dev, dev->name);
1005 memcpy(dev->name, oldname, IFNAMSIZ);
1009 write_lock_bh(&dev_base_lock);
1010 hlist_del(&dev->name_hlist);
1011 write_unlock_bh(&dev_base_lock);
1015 write_lock_bh(&dev_base_lock);
1016 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1017 write_unlock_bh(&dev_base_lock);
1019 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1020 ret = notifier_to_errno(ret);
1023 /* err >= 0 after dev_alloc_name() or stores the first errno */
1026 memcpy(dev->name, oldname, IFNAMSIZ);
1030 "%s: name change rollback failed: %d.\n",
1039 * dev_set_alias - change ifalias of a device
1041 * @alias: name up to IFALIASZ
1042 * @len: limit of bytes to copy from info
1044 * Set ifalias for a device,
1046 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1050 if (len >= IFALIASZ)
1055 kfree(dev->ifalias);
1056 dev->ifalias = NULL;
1061 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1065 strlcpy(dev->ifalias, alias, len+1);
1071 * netdev_features_change - device changes features
1072 * @dev: device to cause notification
1074 * Called to indicate a device has changed features.
1076 void netdev_features_change(struct net_device *dev)
1078 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1080 EXPORT_SYMBOL(netdev_features_change);
1083 * netdev_state_change - device changes state
1084 * @dev: device to cause notification
1086 * Called to indicate a device has changed state. This function calls
1087 * the notifier chains for netdev_chain and sends a NEWLINK message
1088 * to the routing socket.
1090 void netdev_state_change(struct net_device *dev)
1092 if (dev->flags & IFF_UP) {
1093 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1094 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1097 EXPORT_SYMBOL(netdev_state_change);
1099 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1101 return call_netdevice_notifiers(event, dev);
1103 EXPORT_SYMBOL(netdev_bonding_change);
1106 * dev_load - load a network module
1107 * @net: the applicable net namespace
1108 * @name: name of interface
1110 * If a network interface is not present and the process has suitable
1111 * privileges this function loads the module. If module loading is not
1112 * available in this kernel then it becomes a nop.
1115 void dev_load(struct net *net, const char *name)
1117 struct net_device *dev;
1120 dev = dev_get_by_name_rcu(net, name);
1123 if (!dev && capable(CAP_NET_ADMIN))
1124 request_module("%s", name);
1126 EXPORT_SYMBOL(dev_load);
1128 static int __dev_open(struct net_device *dev)
1130 const struct net_device_ops *ops = dev->netdev_ops;
1136 * Is it even present?
1138 if (!netif_device_present(dev))
1141 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1142 ret = notifier_to_errno(ret);
1147 * Call device private open method
1149 set_bit(__LINK_STATE_START, &dev->state);
1151 if (ops->ndo_validate_addr)
1152 ret = ops->ndo_validate_addr(dev);
1154 if (!ret && ops->ndo_open)
1155 ret = ops->ndo_open(dev);
1158 * If it went open OK then:
1162 clear_bit(__LINK_STATE_START, &dev->state);
1167 dev->flags |= IFF_UP;
1172 net_dmaengine_get();
1175 * Initialize multicasting status
1177 dev_set_rx_mode(dev);
1180 * Wakeup transmit queue engine
1189 * dev_open - prepare an interface for use.
1190 * @dev: device to open
1192 * Takes a device from down to up state. The device's private open
1193 * function is invoked and then the multicast lists are loaded. Finally
1194 * the device is moved into the up state and a %NETDEV_UP message is
1195 * sent to the netdev notifier chain.
1197 * Calling this function on an active interface is a nop. On a failure
1198 * a negative errno code is returned.
1200 int dev_open(struct net_device *dev)
1207 if (dev->flags & IFF_UP)
1213 ret = __dev_open(dev);
1218 * ... and announce new interface.
1220 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1221 call_netdevice_notifiers(NETDEV_UP, dev);
1225 EXPORT_SYMBOL(dev_open);
1227 static int __dev_close_many(struct list_head *head)
1229 struct net_device *dev;
1234 list_for_each_entry(dev, head, unreg_list) {
1236 * Tell people we are going down, so that they can
1237 * prepare to death, when device is still operating.
1239 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1241 clear_bit(__LINK_STATE_START, &dev->state);
1243 /* Synchronize to scheduled poll. We cannot touch poll list, it
1244 * can be even on different cpu. So just clear netif_running().
1246 * dev->stop() will invoke napi_disable() on all of it's
1247 * napi_struct instances on this device.
1249 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1252 dev_deactivate_many(head);
1254 list_for_each_entry(dev, head, unreg_list) {
1255 const struct net_device_ops *ops = dev->netdev_ops;
1258 * Call the device specific close. This cannot fail.
1259 * Only if device is UP
1261 * We allow it to be called even after a DETACH hot-plug
1268 * Device is now down.
1271 dev->flags &= ~IFF_UP;
1276 net_dmaengine_put();
1282 static int __dev_close(struct net_device *dev)
1286 list_add(&dev->unreg_list, &single);
1287 return __dev_close_many(&single);
1290 static int dev_close_many(struct list_head *head)
1292 struct net_device *dev, *tmp;
1293 LIST_HEAD(tmp_list);
1295 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1296 if (!(dev->flags & IFF_UP))
1297 list_move(&dev->unreg_list, &tmp_list);
1299 __dev_close_many(head);
1302 * Tell people we are down
1304 list_for_each_entry(dev, head, unreg_list) {
1305 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1306 call_netdevice_notifiers(NETDEV_DOWN, dev);
1309 /* rollback_registered_many needs the complete original list */
1310 list_splice(&tmp_list, head);
1315 * dev_close - shutdown an interface.
1316 * @dev: device to shutdown
1318 * This function moves an active device into down state. A
1319 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1320 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1323 int dev_close(struct net_device *dev)
1327 list_add(&dev->unreg_list, &single);
1328 dev_close_many(&single);
1332 EXPORT_SYMBOL(dev_close);
1336 * dev_disable_lro - disable Large Receive Offload on a device
1339 * Disable Large Receive Offload (LRO) on a net device. Must be
1340 * called under RTNL. This is needed if received packets may be
1341 * forwarded to another interface.
1343 void dev_disable_lro(struct net_device *dev)
1345 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1346 dev->ethtool_ops->set_flags) {
1347 u32 flags = dev->ethtool_ops->get_flags(dev);
1348 if (flags & ETH_FLAG_LRO) {
1349 flags &= ~ETH_FLAG_LRO;
1350 dev->ethtool_ops->set_flags(dev, flags);
1353 WARN_ON(dev->features & NETIF_F_LRO);
1355 EXPORT_SYMBOL(dev_disable_lro);
1358 static int dev_boot_phase = 1;
1361 * Device change register/unregister. These are not inline or static
1362 * as we export them to the world.
1366 * register_netdevice_notifier - register a network notifier block
1369 * Register a notifier to be called when network device events occur.
1370 * The notifier passed is linked into the kernel structures and must
1371 * not be reused until it has been unregistered. A negative errno code
1372 * is returned on a failure.
1374 * When registered all registration and up events are replayed
1375 * to the new notifier to allow device to have a race free
1376 * view of the network device list.
1379 int register_netdevice_notifier(struct notifier_block *nb)
1381 struct net_device *dev;
1382 struct net_device *last;
1387 err = raw_notifier_chain_register(&netdev_chain, nb);
1393 for_each_netdev(net, dev) {
1394 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1395 err = notifier_to_errno(err);
1399 if (!(dev->flags & IFF_UP))
1402 nb->notifier_call(nb, NETDEV_UP, dev);
1413 for_each_netdev(net, dev) {
1417 if (dev->flags & IFF_UP) {
1418 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1419 nb->notifier_call(nb, NETDEV_DOWN, dev);
1421 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1422 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1426 raw_notifier_chain_unregister(&netdev_chain, nb);
1429 EXPORT_SYMBOL(register_netdevice_notifier);
1432 * unregister_netdevice_notifier - unregister a network notifier block
1435 * Unregister a notifier previously registered by
1436 * register_netdevice_notifier(). The notifier is unlinked into the
1437 * kernel structures and may then be reused. A negative errno code
1438 * is returned on a failure.
1441 int unregister_netdevice_notifier(struct notifier_block *nb)
1446 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1450 EXPORT_SYMBOL(unregister_netdevice_notifier);
1453 * call_netdevice_notifiers - call all network notifier blocks
1454 * @val: value passed unmodified to notifier function
1455 * @dev: net_device pointer passed unmodified to notifier function
1457 * Call all network notifier blocks. Parameters and return value
1458 * are as for raw_notifier_call_chain().
1461 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1464 return raw_notifier_call_chain(&netdev_chain, val, dev);
1467 /* When > 0 there are consumers of rx skb time stamps */
1468 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1470 void net_enable_timestamp(void)
1472 atomic_inc(&netstamp_needed);
1474 EXPORT_SYMBOL(net_enable_timestamp);
1476 void net_disable_timestamp(void)
1478 atomic_dec(&netstamp_needed);
1480 EXPORT_SYMBOL(net_disable_timestamp);
1482 static inline void net_timestamp_set(struct sk_buff *skb)
1484 if (atomic_read(&netstamp_needed))
1485 __net_timestamp(skb);
1487 skb->tstamp.tv64 = 0;
1490 static inline void net_timestamp_check(struct sk_buff *skb)
1492 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1493 __net_timestamp(skb);
1497 * dev_forward_skb - loopback an skb to another netif
1499 * @dev: destination network device
1500 * @skb: buffer to forward
1503 * NET_RX_SUCCESS (no congestion)
1504 * NET_RX_DROP (packet was dropped, but freed)
1506 * dev_forward_skb can be used for injecting an skb from the
1507 * start_xmit function of one device into the receive queue
1508 * of another device.
1510 * The receiving device may be in another namespace, so
1511 * we have to clear all information in the skb that could
1512 * impact namespace isolation.
1514 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1519 if (unlikely(!(dev->flags & IFF_UP) ||
1520 (skb->len > (dev->mtu + dev->hard_header_len + VLAN_HLEN)))) {
1521 atomic_long_inc(&dev->rx_dropped);
1525 skb_set_dev(skb, dev);
1526 skb->tstamp.tv64 = 0;
1527 skb->pkt_type = PACKET_HOST;
1528 skb->protocol = eth_type_trans(skb, dev);
1529 return netif_rx(skb);
1531 EXPORT_SYMBOL_GPL(dev_forward_skb);
1533 static inline int deliver_skb(struct sk_buff *skb,
1534 struct packet_type *pt_prev,
1535 struct net_device *orig_dev)
1537 atomic_inc(&skb->users);
1538 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1542 * Support routine. Sends outgoing frames to any network
1543 * taps currently in use.
1546 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1548 struct packet_type *ptype;
1549 struct sk_buff *skb2 = NULL;
1550 struct packet_type *pt_prev = NULL;
1553 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1554 /* Never send packets back to the socket
1555 * they originated from - MvS (miquels@drinkel.ow.org)
1557 if ((ptype->dev == dev || !ptype->dev) &&
1558 (ptype->af_packet_priv == NULL ||
1559 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1561 deliver_skb(skb2, pt_prev, skb->dev);
1566 skb2 = skb_clone(skb, GFP_ATOMIC);
1570 net_timestamp_set(skb2);
1572 /* skb->nh should be correctly
1573 set by sender, so that the second statement is
1574 just protection against buggy protocols.
1576 skb_reset_mac_header(skb2);
1578 if (skb_network_header(skb2) < skb2->data ||
1579 skb2->network_header > skb2->tail) {
1580 if (net_ratelimit())
1581 printk(KERN_CRIT "protocol %04x is "
1583 ntohs(skb2->protocol),
1585 skb_reset_network_header(skb2);
1588 skb2->transport_header = skb2->network_header;
1589 skb2->pkt_type = PACKET_OUTGOING;
1594 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1598 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1599 * @dev: Network device
1600 * @txq: number of queues available
1602 * If real_num_tx_queues is changed the tc mappings may no longer be
1603 * valid. To resolve this verify the tc mapping remains valid and if
1604 * not NULL the mapping. With no priorities mapping to this
1605 * offset/count pair it will no longer be used. In the worst case TC0
1606 * is invalid nothing can be done so disable priority mappings. If is
1607 * expected that drivers will fix this mapping if they can before
1608 * calling netif_set_real_num_tx_queues.
1610 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1613 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1615 /* If TC0 is invalidated disable TC mapping */
1616 if (tc->offset + tc->count > txq) {
1617 pr_warning("Number of in use tx queues changed "
1618 "invalidating tc mappings. Priority "
1619 "traffic classification disabled!\n");
1624 /* Invalidated prio to tc mappings set to TC0 */
1625 for (i = 1; i < TC_BITMASK + 1; i++) {
1626 int q = netdev_get_prio_tc_map(dev, i);
1628 tc = &dev->tc_to_txq[q];
1629 if (tc->offset + tc->count > txq) {
1630 pr_warning("Number of in use tx queues "
1631 "changed. Priority %i to tc "
1632 "mapping %i is no longer valid "
1633 "setting map to 0\n",
1635 netdev_set_prio_tc_map(dev, i, 0);
1641 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1642 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1644 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1648 if (txq < 1 || txq > dev->num_tx_queues)
1651 if (dev->reg_state == NETREG_REGISTERED ||
1652 dev->reg_state == NETREG_UNREGISTERING) {
1655 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1661 netif_setup_tc(dev, txq);
1663 if (txq < dev->real_num_tx_queues)
1664 qdisc_reset_all_tx_gt(dev, txq);
1667 dev->real_num_tx_queues = txq;
1670 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1674 * netif_set_real_num_rx_queues - set actual number of RX queues used
1675 * @dev: Network device
1676 * @rxq: Actual number of RX queues
1678 * This must be called either with the rtnl_lock held or before
1679 * registration of the net device. Returns 0 on success, or a
1680 * negative error code. If called before registration, it always
1683 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1687 if (rxq < 1 || rxq > dev->num_rx_queues)
1690 if (dev->reg_state == NETREG_REGISTERED) {
1693 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1699 dev->real_num_rx_queues = rxq;
1702 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1705 static inline void __netif_reschedule(struct Qdisc *q)
1707 struct softnet_data *sd;
1708 unsigned long flags;
1710 local_irq_save(flags);
1711 sd = &__get_cpu_var(softnet_data);
1712 q->next_sched = NULL;
1713 *sd->output_queue_tailp = q;
1714 sd->output_queue_tailp = &q->next_sched;
1715 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1716 local_irq_restore(flags);
1719 void __netif_schedule(struct Qdisc *q)
1721 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1722 __netif_reschedule(q);
1724 EXPORT_SYMBOL(__netif_schedule);
1726 void dev_kfree_skb_irq(struct sk_buff *skb)
1728 if (atomic_dec_and_test(&skb->users)) {
1729 struct softnet_data *sd;
1730 unsigned long flags;
1732 local_irq_save(flags);
1733 sd = &__get_cpu_var(softnet_data);
1734 skb->next = sd->completion_queue;
1735 sd->completion_queue = skb;
1736 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1737 local_irq_restore(flags);
1740 EXPORT_SYMBOL(dev_kfree_skb_irq);
1742 void dev_kfree_skb_any(struct sk_buff *skb)
1744 if (in_irq() || irqs_disabled())
1745 dev_kfree_skb_irq(skb);
1749 EXPORT_SYMBOL(dev_kfree_skb_any);
1753 * netif_device_detach - mark device as removed
1754 * @dev: network device
1756 * Mark device as removed from system and therefore no longer available.
1758 void netif_device_detach(struct net_device *dev)
1760 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1761 netif_running(dev)) {
1762 netif_tx_stop_all_queues(dev);
1765 EXPORT_SYMBOL(netif_device_detach);
1768 * netif_device_attach - mark device as attached
1769 * @dev: network device
1771 * Mark device as attached from system and restart if needed.
1773 void netif_device_attach(struct net_device *dev)
1775 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1776 netif_running(dev)) {
1777 netif_tx_wake_all_queues(dev);
1778 __netdev_watchdog_up(dev);
1781 EXPORT_SYMBOL(netif_device_attach);
1784 * skb_dev_set -- assign a new device to a buffer
1785 * @skb: buffer for the new device
1786 * @dev: network device
1788 * If an skb is owned by a device already, we have to reset
1789 * all data private to the namespace a device belongs to
1790 * before assigning it a new device.
1792 #ifdef CONFIG_NET_NS
1793 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1796 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1799 skb_init_secmark(skb);
1803 skb->ipvs_property = 0;
1804 #ifdef CONFIG_NET_SCHED
1810 EXPORT_SYMBOL(skb_set_dev);
1811 #endif /* CONFIG_NET_NS */
1814 * Invalidate hardware checksum when packet is to be mangled, and
1815 * complete checksum manually on outgoing path.
1817 int skb_checksum_help(struct sk_buff *skb)
1820 int ret = 0, offset;
1822 if (skb->ip_summed == CHECKSUM_COMPLETE)
1823 goto out_set_summed;
1825 if (unlikely(skb_shinfo(skb)->gso_size)) {
1826 /* Let GSO fix up the checksum. */
1827 goto out_set_summed;
1830 offset = skb_checksum_start_offset(skb);
1831 BUG_ON(offset >= skb_headlen(skb));
1832 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1834 offset += skb->csum_offset;
1835 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1837 if (skb_cloned(skb) &&
1838 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1839 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1844 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1846 skb->ip_summed = CHECKSUM_NONE;
1850 EXPORT_SYMBOL(skb_checksum_help);
1853 * skb_gso_segment - Perform segmentation on skb.
1854 * @skb: buffer to segment
1855 * @features: features for the output path (see dev->features)
1857 * This function segments the given skb and returns a list of segments.
1859 * It may return NULL if the skb requires no segmentation. This is
1860 * only possible when GSO is used for verifying header integrity.
1862 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1864 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1865 struct packet_type *ptype;
1866 __be16 type = skb->protocol;
1867 int vlan_depth = ETH_HLEN;
1870 while (type == htons(ETH_P_8021Q)) {
1871 struct vlan_hdr *vh;
1873 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1874 return ERR_PTR(-EINVAL);
1876 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1877 type = vh->h_vlan_encapsulated_proto;
1878 vlan_depth += VLAN_HLEN;
1881 skb_reset_mac_header(skb);
1882 skb->mac_len = skb->network_header - skb->mac_header;
1883 __skb_pull(skb, skb->mac_len);
1885 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1886 struct net_device *dev = skb->dev;
1887 struct ethtool_drvinfo info = {};
1889 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1890 dev->ethtool_ops->get_drvinfo(dev, &info);
1892 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1893 info.driver, dev ? dev->features : 0L,
1894 skb->sk ? skb->sk->sk_route_caps : 0L,
1895 skb->len, skb->data_len, skb->ip_summed);
1897 if (skb_header_cloned(skb) &&
1898 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1899 return ERR_PTR(err);
1903 list_for_each_entry_rcu(ptype,
1904 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1905 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1906 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1907 err = ptype->gso_send_check(skb);
1908 segs = ERR_PTR(err);
1909 if (err || skb_gso_ok(skb, features))
1911 __skb_push(skb, (skb->data -
1912 skb_network_header(skb)));
1914 segs = ptype->gso_segment(skb, features);
1920 __skb_push(skb, skb->data - skb_mac_header(skb));
1924 EXPORT_SYMBOL(skb_gso_segment);
1926 /* Take action when hardware reception checksum errors are detected. */
1928 void netdev_rx_csum_fault(struct net_device *dev)
1930 if (net_ratelimit()) {
1931 printk(KERN_ERR "%s: hw csum failure.\n",
1932 dev ? dev->name : "<unknown>");
1936 EXPORT_SYMBOL(netdev_rx_csum_fault);
1939 /* Actually, we should eliminate this check as soon as we know, that:
1940 * 1. IOMMU is present and allows to map all the memory.
1941 * 2. No high memory really exists on this machine.
1944 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1946 #ifdef CONFIG_HIGHMEM
1948 if (!(dev->features & NETIF_F_HIGHDMA)) {
1949 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1950 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1954 if (PCI_DMA_BUS_IS_PHYS) {
1955 struct device *pdev = dev->dev.parent;
1959 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1960 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1961 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1970 void (*destructor)(struct sk_buff *skb);
1973 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1975 static void dev_gso_skb_destructor(struct sk_buff *skb)
1977 struct dev_gso_cb *cb;
1980 struct sk_buff *nskb = skb->next;
1982 skb->next = nskb->next;
1985 } while (skb->next);
1987 cb = DEV_GSO_CB(skb);
1989 cb->destructor(skb);
1993 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1994 * @skb: buffer to segment
1995 * @features: device features as applicable to this skb
1997 * This function segments the given skb and stores the list of segments
2000 static int dev_gso_segment(struct sk_buff *skb, int features)
2002 struct sk_buff *segs;
2004 segs = skb_gso_segment(skb, features);
2006 /* Verifying header integrity only. */
2011 return PTR_ERR(segs);
2014 DEV_GSO_CB(skb)->destructor = skb->destructor;
2015 skb->destructor = dev_gso_skb_destructor;
2021 * Try to orphan skb early, right before transmission by the device.
2022 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2023 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2025 static inline void skb_orphan_try(struct sk_buff *skb)
2027 struct sock *sk = skb->sk;
2029 if (sk && !skb_shinfo(skb)->tx_flags) {
2030 /* skb_tx_hash() wont be able to get sk.
2031 * We copy sk_hash into skb->rxhash
2034 skb->rxhash = sk->sk_hash;
2039 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2041 return ((features & NETIF_F_GEN_CSUM) ||
2042 ((features & NETIF_F_V4_CSUM) &&
2043 protocol == htons(ETH_P_IP)) ||
2044 ((features & NETIF_F_V6_CSUM) &&
2045 protocol == htons(ETH_P_IPV6)) ||
2046 ((features & NETIF_F_FCOE_CRC) &&
2047 protocol == htons(ETH_P_FCOE)));
2050 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2052 if (!can_checksum_protocol(features, protocol)) {
2053 features &= ~NETIF_F_ALL_CSUM;
2054 features &= ~NETIF_F_SG;
2055 } else if (illegal_highdma(skb->dev, skb)) {
2056 features &= ~NETIF_F_SG;
2062 u32 netif_skb_features(struct sk_buff *skb)
2064 __be16 protocol = skb->protocol;
2065 u32 features = skb->dev->features;
2067 if (protocol == htons(ETH_P_8021Q)) {
2068 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2069 protocol = veh->h_vlan_encapsulated_proto;
2070 } else if (!vlan_tx_tag_present(skb)) {
2071 return harmonize_features(skb, protocol, features);
2074 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2076 if (protocol != htons(ETH_P_8021Q)) {
2077 return harmonize_features(skb, protocol, features);
2079 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2080 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2081 return harmonize_features(skb, protocol, features);
2084 EXPORT_SYMBOL(netif_skb_features);
2087 * Returns true if either:
2088 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2089 * 2. skb is fragmented and the device does not support SG, or if
2090 * at least one of fragments is in highmem and device does not
2091 * support DMA from it.
2093 static inline int skb_needs_linearize(struct sk_buff *skb,
2096 return skb_is_nonlinear(skb) &&
2097 ((skb_has_frag_list(skb) &&
2098 !(features & NETIF_F_FRAGLIST)) ||
2099 (skb_shinfo(skb)->nr_frags &&
2100 !(features & NETIF_F_SG)));
2103 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2104 struct netdev_queue *txq)
2106 const struct net_device_ops *ops = dev->netdev_ops;
2107 int rc = NETDEV_TX_OK;
2109 if (likely(!skb->next)) {
2113 * If device doesnt need skb->dst, release it right now while
2114 * its hot in this cpu cache
2116 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2119 if (!list_empty(&ptype_all))
2120 dev_queue_xmit_nit(skb, dev);
2122 skb_orphan_try(skb);
2124 features = netif_skb_features(skb);
2126 if (vlan_tx_tag_present(skb) &&
2127 !(features & NETIF_F_HW_VLAN_TX)) {
2128 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2135 if (netif_needs_gso(skb, features)) {
2136 if (unlikely(dev_gso_segment(skb, features)))
2141 if (skb_needs_linearize(skb, features) &&
2142 __skb_linearize(skb))
2145 /* If packet is not checksummed and device does not
2146 * support checksumming for this protocol, complete
2147 * checksumming here.
2149 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2150 skb_set_transport_header(skb,
2151 skb_checksum_start_offset(skb));
2152 if (!(features & NETIF_F_ALL_CSUM) &&
2153 skb_checksum_help(skb))
2158 rc = ops->ndo_start_xmit(skb, dev);
2159 trace_net_dev_xmit(skb, rc);
2160 if (rc == NETDEV_TX_OK)
2161 txq_trans_update(txq);
2167 struct sk_buff *nskb = skb->next;
2169 skb->next = nskb->next;
2173 * If device doesnt need nskb->dst, release it right now while
2174 * its hot in this cpu cache
2176 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2179 rc = ops->ndo_start_xmit(nskb, dev);
2180 trace_net_dev_xmit(nskb, rc);
2181 if (unlikely(rc != NETDEV_TX_OK)) {
2182 if (rc & ~NETDEV_TX_MASK)
2183 goto out_kfree_gso_skb;
2184 nskb->next = skb->next;
2188 txq_trans_update(txq);
2189 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2190 return NETDEV_TX_BUSY;
2191 } while (skb->next);
2194 if (likely(skb->next == NULL))
2195 skb->destructor = DEV_GSO_CB(skb)->destructor;
2202 static u32 hashrnd __read_mostly;
2205 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2206 * to be used as a distribution range.
2208 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2209 unsigned int num_tx_queues)
2213 u16 qcount = num_tx_queues;
2215 if (skb_rx_queue_recorded(skb)) {
2216 hash = skb_get_rx_queue(skb);
2217 while (unlikely(hash >= num_tx_queues))
2218 hash -= num_tx_queues;
2223 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2224 qoffset = dev->tc_to_txq[tc].offset;
2225 qcount = dev->tc_to_txq[tc].count;
2228 if (skb->sk && skb->sk->sk_hash)
2229 hash = skb->sk->sk_hash;
2231 hash = (__force u16) skb->protocol ^ skb->rxhash;
2232 hash = jhash_1word(hash, hashrnd);
2234 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2236 EXPORT_SYMBOL(__skb_tx_hash);
2238 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2240 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2241 if (net_ratelimit()) {
2242 pr_warning("%s selects TX queue %d, but "
2243 "real number of TX queues is %d\n",
2244 dev->name, queue_index, dev->real_num_tx_queues);
2251 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2254 struct xps_dev_maps *dev_maps;
2255 struct xps_map *map;
2256 int queue_index = -1;
2259 dev_maps = rcu_dereference(dev->xps_maps);
2261 map = rcu_dereference(
2262 dev_maps->cpu_map[raw_smp_processor_id()]);
2265 queue_index = map->queues[0];
2268 if (skb->sk && skb->sk->sk_hash)
2269 hash = skb->sk->sk_hash;
2271 hash = (__force u16) skb->protocol ^
2273 hash = jhash_1word(hash, hashrnd);
2274 queue_index = map->queues[
2275 ((u64)hash * map->len) >> 32];
2277 if (unlikely(queue_index >= dev->real_num_tx_queues))
2289 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2290 struct sk_buff *skb)
2293 const struct net_device_ops *ops = dev->netdev_ops;
2295 if (dev->real_num_tx_queues == 1)
2297 else if (ops->ndo_select_queue) {
2298 queue_index = ops->ndo_select_queue(dev, skb);
2299 queue_index = dev_cap_txqueue(dev, queue_index);
2301 struct sock *sk = skb->sk;
2302 queue_index = sk_tx_queue_get(sk);
2304 if (queue_index < 0 || skb->ooo_okay ||
2305 queue_index >= dev->real_num_tx_queues) {
2306 int old_index = queue_index;
2308 queue_index = get_xps_queue(dev, skb);
2309 if (queue_index < 0)
2310 queue_index = skb_tx_hash(dev, skb);
2312 if (queue_index != old_index && sk) {
2313 struct dst_entry *dst =
2314 rcu_dereference_check(sk->sk_dst_cache, 1);
2316 if (dst && skb_dst(skb) == dst)
2317 sk_tx_queue_set(sk, queue_index);
2322 skb_set_queue_mapping(skb, queue_index);
2323 return netdev_get_tx_queue(dev, queue_index);
2326 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2327 struct net_device *dev,
2328 struct netdev_queue *txq)
2330 spinlock_t *root_lock = qdisc_lock(q);
2334 qdisc_skb_cb(skb)->pkt_len = skb->len;
2335 qdisc_calculate_pkt_len(skb, q);
2337 * Heuristic to force contended enqueues to serialize on a
2338 * separate lock before trying to get qdisc main lock.
2339 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2340 * and dequeue packets faster.
2342 contended = qdisc_is_running(q);
2343 if (unlikely(contended))
2344 spin_lock(&q->busylock);
2346 spin_lock(root_lock);
2347 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2350 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2351 qdisc_run_begin(q)) {
2353 * This is a work-conserving queue; there are no old skbs
2354 * waiting to be sent out; and the qdisc is not running -
2355 * xmit the skb directly.
2357 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2360 qdisc_bstats_update(q, skb);
2362 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2363 if (unlikely(contended)) {
2364 spin_unlock(&q->busylock);
2371 rc = NET_XMIT_SUCCESS;
2374 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2375 if (qdisc_run_begin(q)) {
2376 if (unlikely(contended)) {
2377 spin_unlock(&q->busylock);
2383 spin_unlock(root_lock);
2384 if (unlikely(contended))
2385 spin_unlock(&q->busylock);
2389 static DEFINE_PER_CPU(int, xmit_recursion);
2390 #define RECURSION_LIMIT 10
2393 * dev_queue_xmit - transmit a buffer
2394 * @skb: buffer to transmit
2396 * Queue a buffer for transmission to a network device. The caller must
2397 * have set the device and priority and built the buffer before calling
2398 * this function. The function can be called from an interrupt.
2400 * A negative errno code is returned on a failure. A success does not
2401 * guarantee the frame will be transmitted as it may be dropped due
2402 * to congestion or traffic shaping.
2404 * -----------------------------------------------------------------------------------
2405 * I notice this method can also return errors from the queue disciplines,
2406 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2409 * Regardless of the return value, the skb is consumed, so it is currently
2410 * difficult to retry a send to this method. (You can bump the ref count
2411 * before sending to hold a reference for retry if you are careful.)
2413 * When calling this method, interrupts MUST be enabled. This is because
2414 * the BH enable code must have IRQs enabled so that it will not deadlock.
2417 int dev_queue_xmit(struct sk_buff *skb)
2419 struct net_device *dev = skb->dev;
2420 struct netdev_queue *txq;
2424 /* Disable soft irqs for various locks below. Also
2425 * stops preemption for RCU.
2429 txq = dev_pick_tx(dev, skb);
2430 q = rcu_dereference_bh(txq->qdisc);
2432 #ifdef CONFIG_NET_CLS_ACT
2433 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2435 trace_net_dev_queue(skb);
2437 rc = __dev_xmit_skb(skb, q, dev, txq);
2441 /* The device has no queue. Common case for software devices:
2442 loopback, all the sorts of tunnels...
2444 Really, it is unlikely that netif_tx_lock protection is necessary
2445 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2447 However, it is possible, that they rely on protection
2450 Check this and shot the lock. It is not prone from deadlocks.
2451 Either shot noqueue qdisc, it is even simpler 8)
2453 if (dev->flags & IFF_UP) {
2454 int cpu = smp_processor_id(); /* ok because BHs are off */
2456 if (txq->xmit_lock_owner != cpu) {
2458 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2459 goto recursion_alert;
2461 HARD_TX_LOCK(dev, txq, cpu);
2463 if (!netif_tx_queue_stopped(txq)) {
2464 __this_cpu_inc(xmit_recursion);
2465 rc = dev_hard_start_xmit(skb, dev, txq);
2466 __this_cpu_dec(xmit_recursion);
2467 if (dev_xmit_complete(rc)) {
2468 HARD_TX_UNLOCK(dev, txq);
2472 HARD_TX_UNLOCK(dev, txq);
2473 if (net_ratelimit())
2474 printk(KERN_CRIT "Virtual device %s asks to "
2475 "queue packet!\n", dev->name);
2477 /* Recursion is detected! It is possible,
2481 if (net_ratelimit())
2482 printk(KERN_CRIT "Dead loop on virtual device "
2483 "%s, fix it urgently!\n", dev->name);
2488 rcu_read_unlock_bh();
2493 rcu_read_unlock_bh();
2496 EXPORT_SYMBOL(dev_queue_xmit);
2499 /*=======================================================================
2501 =======================================================================*/
2503 int netdev_max_backlog __read_mostly = 1000;
2504 int netdev_tstamp_prequeue __read_mostly = 1;
2505 int netdev_budget __read_mostly = 300;
2506 int weight_p __read_mostly = 64; /* old backlog weight */
2508 /* Called with irq disabled */
2509 static inline void ____napi_schedule(struct softnet_data *sd,
2510 struct napi_struct *napi)
2512 list_add_tail(&napi->poll_list, &sd->poll_list);
2513 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2517 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2518 * and src/dst port numbers. Returns a non-zero hash number on success
2521 __u32 __skb_get_rxhash(struct sk_buff *skb)
2523 int nhoff, hash = 0, poff;
2524 struct ipv6hdr *ip6;
2527 u32 addr1, addr2, ihl;
2533 nhoff = skb_network_offset(skb);
2535 switch (skb->protocol) {
2536 case __constant_htons(ETH_P_IP):
2537 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2540 ip = (struct iphdr *) (skb->data + nhoff);
2541 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2544 ip_proto = ip->protocol;
2545 addr1 = (__force u32) ip->saddr;
2546 addr2 = (__force u32) ip->daddr;
2549 case __constant_htons(ETH_P_IPV6):
2550 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2553 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2554 ip_proto = ip6->nexthdr;
2555 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2556 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2564 poff = proto_ports_offset(ip_proto);
2566 nhoff += ihl * 4 + poff;
2567 if (pskb_may_pull(skb, nhoff + 4)) {
2568 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2569 if (ports.v16[1] < ports.v16[0])
2570 swap(ports.v16[0], ports.v16[1]);
2574 /* get a consistent hash (same value on both flow directions) */
2578 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2585 EXPORT_SYMBOL(__skb_get_rxhash);
2589 /* One global table that all flow-based protocols share. */
2590 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2591 EXPORT_SYMBOL(rps_sock_flow_table);
2593 static struct rps_dev_flow *
2594 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2595 struct rps_dev_flow *rflow, u16 next_cpu)
2599 tcpu = rflow->cpu = next_cpu;
2600 if (tcpu != RPS_NO_CPU) {
2601 #ifdef CONFIG_RFS_ACCEL
2602 struct netdev_rx_queue *rxqueue;
2603 struct rps_dev_flow_table *flow_table;
2604 struct rps_dev_flow *old_rflow;
2609 /* Should we steer this flow to a different hardware queue? */
2610 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap)
2612 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2613 if (rxq_index == skb_get_rx_queue(skb))
2616 rxqueue = dev->_rx + rxq_index;
2617 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2620 flow_id = skb->rxhash & flow_table->mask;
2621 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2622 rxq_index, flow_id);
2626 rflow = &flow_table->flows[flow_id];
2627 rflow->cpu = next_cpu;
2629 if (old_rflow->filter == rflow->filter)
2630 old_rflow->filter = RPS_NO_FILTER;
2634 per_cpu(softnet_data, tcpu).input_queue_head;
2641 * get_rps_cpu is called from netif_receive_skb and returns the target
2642 * CPU from the RPS map of the receiving queue for a given skb.
2643 * rcu_read_lock must be held on entry.
2645 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2646 struct rps_dev_flow **rflowp)
2648 struct netdev_rx_queue *rxqueue;
2649 struct rps_map *map;
2650 struct rps_dev_flow_table *flow_table;
2651 struct rps_sock_flow_table *sock_flow_table;
2655 if (skb_rx_queue_recorded(skb)) {
2656 u16 index = skb_get_rx_queue(skb);
2657 if (unlikely(index >= dev->real_num_rx_queues)) {
2658 WARN_ONCE(dev->real_num_rx_queues > 1,
2659 "%s received packet on queue %u, but number "
2660 "of RX queues is %u\n",
2661 dev->name, index, dev->real_num_rx_queues);
2664 rxqueue = dev->_rx + index;
2668 map = rcu_dereference(rxqueue->rps_map);
2670 if (map->len == 1 &&
2671 !rcu_dereference_raw(rxqueue->rps_flow_table)) {
2672 tcpu = map->cpus[0];
2673 if (cpu_online(tcpu))
2677 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2681 skb_reset_network_header(skb);
2682 if (!skb_get_rxhash(skb))
2685 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2686 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2687 if (flow_table && sock_flow_table) {
2689 struct rps_dev_flow *rflow;
2691 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2694 next_cpu = sock_flow_table->ents[skb->rxhash &
2695 sock_flow_table->mask];
2698 * If the desired CPU (where last recvmsg was done) is
2699 * different from current CPU (one in the rx-queue flow
2700 * table entry), switch if one of the following holds:
2701 * - Current CPU is unset (equal to RPS_NO_CPU).
2702 * - Current CPU is offline.
2703 * - The current CPU's queue tail has advanced beyond the
2704 * last packet that was enqueued using this table entry.
2705 * This guarantees that all previous packets for the flow
2706 * have been dequeued, thus preserving in order delivery.
2708 if (unlikely(tcpu != next_cpu) &&
2709 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2710 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2711 rflow->last_qtail)) >= 0))
2712 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2714 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2722 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2724 if (cpu_online(tcpu)) {
2734 #ifdef CONFIG_RFS_ACCEL
2737 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2738 * @dev: Device on which the filter was set
2739 * @rxq_index: RX queue index
2740 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2741 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2743 * Drivers that implement ndo_rx_flow_steer() should periodically call
2744 * this function for each installed filter and remove the filters for
2745 * which it returns %true.
2747 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2748 u32 flow_id, u16 filter_id)
2750 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2751 struct rps_dev_flow_table *flow_table;
2752 struct rps_dev_flow *rflow;
2757 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2758 if (flow_table && flow_id <= flow_table->mask) {
2759 rflow = &flow_table->flows[flow_id];
2760 cpu = ACCESS_ONCE(rflow->cpu);
2761 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2762 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2763 rflow->last_qtail) <
2764 (int)(10 * flow_table->mask)))
2770 EXPORT_SYMBOL(rps_may_expire_flow);
2772 #endif /* CONFIG_RFS_ACCEL */
2774 /* Called from hardirq (IPI) context */
2775 static void rps_trigger_softirq(void *data)
2777 struct softnet_data *sd = data;
2779 ____napi_schedule(sd, &sd->backlog);
2783 #endif /* CONFIG_RPS */
2786 * Check if this softnet_data structure is another cpu one
2787 * If yes, queue it to our IPI list and return 1
2790 static int rps_ipi_queued(struct softnet_data *sd)
2793 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2796 sd->rps_ipi_next = mysd->rps_ipi_list;
2797 mysd->rps_ipi_list = sd;
2799 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2802 #endif /* CONFIG_RPS */
2807 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2808 * queue (may be a remote CPU queue).
2810 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2811 unsigned int *qtail)
2813 struct softnet_data *sd;
2814 unsigned long flags;
2816 sd = &per_cpu(softnet_data, cpu);
2818 local_irq_save(flags);
2821 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2822 if (skb_queue_len(&sd->input_pkt_queue)) {
2824 __skb_queue_tail(&sd->input_pkt_queue, skb);
2825 input_queue_tail_incr_save(sd, qtail);
2827 local_irq_restore(flags);
2828 return NET_RX_SUCCESS;
2831 /* Schedule NAPI for backlog device
2832 * We can use non atomic operation since we own the queue lock
2834 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2835 if (!rps_ipi_queued(sd))
2836 ____napi_schedule(sd, &sd->backlog);
2844 local_irq_restore(flags);
2846 atomic_long_inc(&skb->dev->rx_dropped);
2852 * netif_rx - post buffer to the network code
2853 * @skb: buffer to post
2855 * This function receives a packet from a device driver and queues it for
2856 * the upper (protocol) levels to process. It always succeeds. The buffer
2857 * may be dropped during processing for congestion control or by the
2861 * NET_RX_SUCCESS (no congestion)
2862 * NET_RX_DROP (packet was dropped)
2866 int netif_rx(struct sk_buff *skb)
2870 /* if netpoll wants it, pretend we never saw it */
2871 if (netpoll_rx(skb))
2874 if (netdev_tstamp_prequeue)
2875 net_timestamp_check(skb);
2877 trace_netif_rx(skb);
2880 struct rps_dev_flow voidflow, *rflow = &voidflow;
2886 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2888 cpu = smp_processor_id();
2890 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2898 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2904 EXPORT_SYMBOL(netif_rx);
2906 int netif_rx_ni(struct sk_buff *skb)
2911 err = netif_rx(skb);
2912 if (local_softirq_pending())
2918 EXPORT_SYMBOL(netif_rx_ni);
2920 static void net_tx_action(struct softirq_action *h)
2922 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2924 if (sd->completion_queue) {
2925 struct sk_buff *clist;
2927 local_irq_disable();
2928 clist = sd->completion_queue;
2929 sd->completion_queue = NULL;
2933 struct sk_buff *skb = clist;
2934 clist = clist->next;
2936 WARN_ON(atomic_read(&skb->users));
2937 trace_kfree_skb(skb, net_tx_action);
2942 if (sd->output_queue) {
2945 local_irq_disable();
2946 head = sd->output_queue;
2947 sd->output_queue = NULL;
2948 sd->output_queue_tailp = &sd->output_queue;
2952 struct Qdisc *q = head;
2953 spinlock_t *root_lock;
2955 head = head->next_sched;
2957 root_lock = qdisc_lock(q);
2958 if (spin_trylock(root_lock)) {
2959 smp_mb__before_clear_bit();
2960 clear_bit(__QDISC_STATE_SCHED,
2963 spin_unlock(root_lock);
2965 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2967 __netif_reschedule(q);
2969 smp_mb__before_clear_bit();
2970 clear_bit(__QDISC_STATE_SCHED,
2978 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2979 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2980 /* This hook is defined here for ATM LANE */
2981 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2982 unsigned char *addr) __read_mostly;
2983 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2986 #ifdef CONFIG_NET_CLS_ACT
2987 /* TODO: Maybe we should just force sch_ingress to be compiled in
2988 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2989 * a compare and 2 stores extra right now if we dont have it on
2990 * but have CONFIG_NET_CLS_ACT
2991 * NOTE: This doesnt stop any functionality; if you dont have
2992 * the ingress scheduler, you just cant add policies on ingress.
2995 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2997 struct net_device *dev = skb->dev;
2998 u32 ttl = G_TC_RTTL(skb->tc_verd);
2999 int result = TC_ACT_OK;
3002 if (unlikely(MAX_RED_LOOP < ttl++)) {
3003 if (net_ratelimit())
3004 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3005 skb->skb_iif, dev->ifindex);
3009 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3010 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3013 if (q != &noop_qdisc) {
3014 spin_lock(qdisc_lock(q));
3015 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3016 result = qdisc_enqueue_root(skb, q);
3017 spin_unlock(qdisc_lock(q));
3023 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3024 struct packet_type **pt_prev,
3025 int *ret, struct net_device *orig_dev)
3027 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3029 if (!rxq || rxq->qdisc == &noop_qdisc)
3033 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3037 switch (ing_filter(skb, rxq)) {
3051 * netdev_rx_handler_register - register receive handler
3052 * @dev: device to register a handler for
3053 * @rx_handler: receive handler to register
3054 * @rx_handler_data: data pointer that is used by rx handler
3056 * Register a receive hander for a device. This handler will then be
3057 * called from __netif_receive_skb. A negative errno code is returned
3060 * The caller must hold the rtnl_mutex.
3062 int netdev_rx_handler_register(struct net_device *dev,
3063 rx_handler_func_t *rx_handler,
3064 void *rx_handler_data)
3068 if (dev->rx_handler)
3071 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3072 rcu_assign_pointer(dev->rx_handler, rx_handler);
3076 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3079 * netdev_rx_handler_unregister - unregister receive handler
3080 * @dev: device to unregister a handler from
3082 * Unregister a receive hander from a device.
3084 * The caller must hold the rtnl_mutex.
3086 void netdev_rx_handler_unregister(struct net_device *dev)
3090 rcu_assign_pointer(dev->rx_handler, NULL);
3091 rcu_assign_pointer(dev->rx_handler_data, NULL);
3093 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3095 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
3096 struct net_device *master)
3098 if (skb->pkt_type == PACKET_HOST) {
3099 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
3101 memcpy(dest, master->dev_addr, ETH_ALEN);
3105 /* On bonding slaves other than the currently active slave, suppress
3106 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
3107 * ARP on active-backup slaves with arp_validate enabled.
3109 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
3111 struct net_device *dev = skb->dev;
3113 if (master->priv_flags & IFF_MASTER_ARPMON)
3114 dev->last_rx = jiffies;
3116 if ((master->priv_flags & IFF_MASTER_ALB) &&
3117 (master->priv_flags & IFF_BRIDGE_PORT)) {
3118 /* Do address unmangle. The local destination address
3119 * will be always the one master has. Provides the right
3120 * functionality in a bridge.
3122 skb_bond_set_mac_by_master(skb, master);
3125 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
3126 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
3127 skb->protocol == __cpu_to_be16(ETH_P_ARP))
3130 if (master->priv_flags & IFF_MASTER_ALB) {
3131 if (skb->pkt_type != PACKET_BROADCAST &&
3132 skb->pkt_type != PACKET_MULTICAST)
3135 if (master->priv_flags & IFF_MASTER_8023AD &&
3136 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
3143 EXPORT_SYMBOL(__skb_bond_should_drop);
3145 static int __netif_receive_skb(struct sk_buff *skb)
3147 struct packet_type *ptype, *pt_prev;
3148 rx_handler_func_t *rx_handler;
3149 struct net_device *orig_dev;
3150 struct net_device *master;
3151 struct net_device *null_or_orig;
3152 struct net_device *orig_or_bond;
3153 int ret = NET_RX_DROP;
3156 if (!netdev_tstamp_prequeue)
3157 net_timestamp_check(skb);
3159 trace_netif_receive_skb(skb);
3161 /* if we've gotten here through NAPI, check netpoll */
3162 if (netpoll_receive_skb(skb))
3166 skb->skb_iif = skb->dev->ifindex;
3169 * bonding note: skbs received on inactive slaves should only
3170 * be delivered to pkt handlers that are exact matches. Also
3171 * the deliver_no_wcard flag will be set. If packet handlers
3172 * are sensitive to duplicate packets these skbs will need to
3173 * be dropped at the handler.
3175 null_or_orig = NULL;
3176 orig_dev = skb->dev;
3177 master = ACCESS_ONCE(orig_dev->master);
3178 if (skb->deliver_no_wcard)
3179 null_or_orig = orig_dev;
3181 if (skb_bond_should_drop(skb, master)) {
3182 skb->deliver_no_wcard = 1;
3183 null_or_orig = orig_dev; /* deliver only exact match */
3188 __this_cpu_inc(softnet_data.processed);
3189 skb_reset_network_header(skb);
3190 skb_reset_transport_header(skb);
3191 skb->mac_len = skb->network_header - skb->mac_header;
3197 #ifdef CONFIG_NET_CLS_ACT
3198 if (skb->tc_verd & TC_NCLS) {
3199 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3204 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3205 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
3206 ptype->dev == orig_dev) {
3208 ret = deliver_skb(skb, pt_prev, orig_dev);
3213 #ifdef CONFIG_NET_CLS_ACT
3214 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3220 /* Handle special case of bridge or macvlan */
3221 rx_handler = rcu_dereference(skb->dev->rx_handler);
3224 ret = deliver_skb(skb, pt_prev, orig_dev);
3227 skb = rx_handler(skb);
3232 if (vlan_tx_tag_present(skb)) {
3234 ret = deliver_skb(skb, pt_prev, orig_dev);
3237 if (vlan_hwaccel_do_receive(&skb)) {
3238 ret = __netif_receive_skb(skb);
3240 } else if (unlikely(!skb))
3245 * Make sure frames received on VLAN interfaces stacked on
3246 * bonding interfaces still make their way to any base bonding
3247 * device that may have registered for a specific ptype. The
3248 * handler may have to adjust skb->dev and orig_dev.
3250 orig_or_bond = orig_dev;
3251 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
3252 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
3253 orig_or_bond = vlan_dev_real_dev(skb->dev);
3256 type = skb->protocol;
3257 list_for_each_entry_rcu(ptype,
3258 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3259 if (ptype->type == type && (ptype->dev == null_or_orig ||
3260 ptype->dev == skb->dev || ptype->dev == orig_dev ||
3261 ptype->dev == orig_or_bond)) {
3263 ret = deliver_skb(skb, pt_prev, orig_dev);
3269 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3271 atomic_long_inc(&skb->dev->rx_dropped);
3273 /* Jamal, now you will not able to escape explaining
3274 * me how you were going to use this. :-)
3285 * netif_receive_skb - process receive buffer from network
3286 * @skb: buffer to process
3288 * netif_receive_skb() is the main receive data processing function.
3289 * It always succeeds. The buffer may be dropped during processing
3290 * for congestion control or by the protocol layers.
3292 * This function may only be called from softirq context and interrupts
3293 * should be enabled.
3295 * Return values (usually ignored):
3296 * NET_RX_SUCCESS: no congestion
3297 * NET_RX_DROP: packet was dropped
3299 int netif_receive_skb(struct sk_buff *skb)
3301 if (netdev_tstamp_prequeue)
3302 net_timestamp_check(skb);
3304 if (skb_defer_rx_timestamp(skb))
3305 return NET_RX_SUCCESS;
3309 struct rps_dev_flow voidflow, *rflow = &voidflow;
3314 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3317 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3321 ret = __netif_receive_skb(skb);
3327 return __netif_receive_skb(skb);
3330 EXPORT_SYMBOL(netif_receive_skb);
3332 /* Network device is going away, flush any packets still pending
3333 * Called with irqs disabled.
3335 static void flush_backlog(void *arg)
3337 struct net_device *dev = arg;
3338 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3339 struct sk_buff *skb, *tmp;
3342 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3343 if (skb->dev == dev) {
3344 __skb_unlink(skb, &sd->input_pkt_queue);
3346 input_queue_head_incr(sd);
3351 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3352 if (skb->dev == dev) {
3353 __skb_unlink(skb, &sd->process_queue);
3355 input_queue_head_incr(sd);
3360 static int napi_gro_complete(struct sk_buff *skb)
3362 struct packet_type *ptype;
3363 __be16 type = skb->protocol;
3364 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3367 if (NAPI_GRO_CB(skb)->count == 1) {
3368 skb_shinfo(skb)->gso_size = 0;
3373 list_for_each_entry_rcu(ptype, head, list) {
3374 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3377 err = ptype->gro_complete(skb);
3383 WARN_ON(&ptype->list == head);
3385 return NET_RX_SUCCESS;
3389 return netif_receive_skb(skb);
3392 inline void napi_gro_flush(struct napi_struct *napi)
3394 struct sk_buff *skb, *next;
3396 for (skb = napi->gro_list; skb; skb = next) {
3399 napi_gro_complete(skb);
3402 napi->gro_count = 0;
3403 napi->gro_list = NULL;
3405 EXPORT_SYMBOL(napi_gro_flush);
3407 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3409 struct sk_buff **pp = NULL;
3410 struct packet_type *ptype;
3411 __be16 type = skb->protocol;
3412 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3415 enum gro_result ret;
3417 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3420 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3424 list_for_each_entry_rcu(ptype, head, list) {
3425 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3428 skb_set_network_header(skb, skb_gro_offset(skb));
3429 mac_len = skb->network_header - skb->mac_header;
3430 skb->mac_len = mac_len;
3431 NAPI_GRO_CB(skb)->same_flow = 0;
3432 NAPI_GRO_CB(skb)->flush = 0;
3433 NAPI_GRO_CB(skb)->free = 0;
3435 pp = ptype->gro_receive(&napi->gro_list, skb);
3440 if (&ptype->list == head)
3443 same_flow = NAPI_GRO_CB(skb)->same_flow;
3444 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3447 struct sk_buff *nskb = *pp;
3451 napi_gro_complete(nskb);
3458 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3462 NAPI_GRO_CB(skb)->count = 1;
3463 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3464 skb->next = napi->gro_list;
3465 napi->gro_list = skb;
3469 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3470 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3472 BUG_ON(skb->end - skb->tail < grow);
3474 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3477 skb->data_len -= grow;
3479 skb_shinfo(skb)->frags[0].page_offset += grow;
3480 skb_shinfo(skb)->frags[0].size -= grow;
3482 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3483 put_page(skb_shinfo(skb)->frags[0].page);
3484 memmove(skb_shinfo(skb)->frags,
3485 skb_shinfo(skb)->frags + 1,
3486 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3497 EXPORT_SYMBOL(dev_gro_receive);
3499 static inline gro_result_t
3500 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3504 for (p = napi->gro_list; p; p = p->next) {
3505 unsigned long diffs;
3507 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3508 diffs |= p->vlan_tci ^ skb->vlan_tci;
3509 diffs |= compare_ether_header(skb_mac_header(p),
3510 skb_gro_mac_header(skb));
3511 NAPI_GRO_CB(p)->same_flow = !diffs;
3512 NAPI_GRO_CB(p)->flush = 0;
3515 return dev_gro_receive(napi, skb);
3518 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3522 if (netif_receive_skb(skb))
3527 case GRO_MERGED_FREE:
3538 EXPORT_SYMBOL(napi_skb_finish);
3540 void skb_gro_reset_offset(struct sk_buff *skb)
3542 NAPI_GRO_CB(skb)->data_offset = 0;
3543 NAPI_GRO_CB(skb)->frag0 = NULL;
3544 NAPI_GRO_CB(skb)->frag0_len = 0;
3546 if (skb->mac_header == skb->tail &&
3547 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3548 NAPI_GRO_CB(skb)->frag0 =
3549 page_address(skb_shinfo(skb)->frags[0].page) +
3550 skb_shinfo(skb)->frags[0].page_offset;
3551 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3554 EXPORT_SYMBOL(skb_gro_reset_offset);
3556 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3558 skb_gro_reset_offset(skb);
3560 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3562 EXPORT_SYMBOL(napi_gro_receive);
3564 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3566 __skb_pull(skb, skb_headlen(skb));
3567 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3569 skb->dev = napi->dev;
3575 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3577 struct sk_buff *skb = napi->skb;
3580 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3586 EXPORT_SYMBOL(napi_get_frags);
3588 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3594 skb->protocol = eth_type_trans(skb, skb->dev);
3596 if (ret == GRO_HELD)
3597 skb_gro_pull(skb, -ETH_HLEN);
3598 else if (netif_receive_skb(skb))
3603 case GRO_MERGED_FREE:
3604 napi_reuse_skb(napi, skb);
3613 EXPORT_SYMBOL(napi_frags_finish);
3615 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3617 struct sk_buff *skb = napi->skb;
3624 skb_reset_mac_header(skb);
3625 skb_gro_reset_offset(skb);
3627 off = skb_gro_offset(skb);
3628 hlen = off + sizeof(*eth);
3629 eth = skb_gro_header_fast(skb, off);
3630 if (skb_gro_header_hard(skb, hlen)) {
3631 eth = skb_gro_header_slow(skb, hlen, off);
3632 if (unlikely(!eth)) {
3633 napi_reuse_skb(napi, skb);
3639 skb_gro_pull(skb, sizeof(*eth));
3642 * This works because the only protocols we care about don't require
3643 * special handling. We'll fix it up properly at the end.
3645 skb->protocol = eth->h_proto;
3650 EXPORT_SYMBOL(napi_frags_skb);
3652 gro_result_t napi_gro_frags(struct napi_struct *napi)
3654 struct sk_buff *skb = napi_frags_skb(napi);
3659 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3661 EXPORT_SYMBOL(napi_gro_frags);
3664 * net_rps_action sends any pending IPI's for rps.
3665 * Note: called with local irq disabled, but exits with local irq enabled.
3667 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3670 struct softnet_data *remsd = sd->rps_ipi_list;
3673 sd->rps_ipi_list = NULL;
3677 /* Send pending IPI's to kick RPS processing on remote cpus. */
3679 struct softnet_data *next = remsd->rps_ipi_next;
3681 if (cpu_online(remsd->cpu))
3682 __smp_call_function_single(remsd->cpu,
3691 static int process_backlog(struct napi_struct *napi, int quota)
3694 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3697 /* Check if we have pending ipi, its better to send them now,
3698 * not waiting net_rx_action() end.
3700 if (sd->rps_ipi_list) {
3701 local_irq_disable();
3702 net_rps_action_and_irq_enable(sd);
3705 napi->weight = weight_p;
3706 local_irq_disable();
3707 while (work < quota) {
3708 struct sk_buff *skb;
3711 while ((skb = __skb_dequeue(&sd->process_queue))) {
3713 __netif_receive_skb(skb);
3714 local_irq_disable();
3715 input_queue_head_incr(sd);
3716 if (++work >= quota) {
3723 qlen = skb_queue_len(&sd->input_pkt_queue);
3725 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3726 &sd->process_queue);
3728 if (qlen < quota - work) {
3730 * Inline a custom version of __napi_complete().
3731 * only current cpu owns and manipulates this napi,
3732 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3733 * we can use a plain write instead of clear_bit(),
3734 * and we dont need an smp_mb() memory barrier.
3736 list_del(&napi->poll_list);
3739 quota = work + qlen;
3749 * __napi_schedule - schedule for receive
3750 * @n: entry to schedule
3752 * The entry's receive function will be scheduled to run
3754 void __napi_schedule(struct napi_struct *n)
3756 unsigned long flags;
3758 local_irq_save(flags);
3759 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3760 local_irq_restore(flags);
3762 EXPORT_SYMBOL(__napi_schedule);
3764 void __napi_complete(struct napi_struct *n)
3766 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3767 BUG_ON(n->gro_list);
3769 list_del(&n->poll_list);
3770 smp_mb__before_clear_bit();
3771 clear_bit(NAPI_STATE_SCHED, &n->state);
3773 EXPORT_SYMBOL(__napi_complete);
3775 void napi_complete(struct napi_struct *n)
3777 unsigned long flags;
3780 * don't let napi dequeue from the cpu poll list
3781 * just in case its running on a different cpu
3783 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3787 local_irq_save(flags);
3789 local_irq_restore(flags);
3791 EXPORT_SYMBOL(napi_complete);
3793 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3794 int (*poll)(struct napi_struct *, int), int weight)
3796 INIT_LIST_HEAD(&napi->poll_list);
3797 napi->gro_count = 0;
3798 napi->gro_list = NULL;
3801 napi->weight = weight;
3802 list_add(&napi->dev_list, &dev->napi_list);
3804 #ifdef CONFIG_NETPOLL
3805 spin_lock_init(&napi->poll_lock);
3806 napi->poll_owner = -1;
3808 set_bit(NAPI_STATE_SCHED, &napi->state);
3810 EXPORT_SYMBOL(netif_napi_add);
3812 void netif_napi_del(struct napi_struct *napi)
3814 struct sk_buff *skb, *next;
3816 list_del_init(&napi->dev_list);
3817 napi_free_frags(napi);
3819 for (skb = napi->gro_list; skb; skb = next) {
3825 napi->gro_list = NULL;
3826 napi->gro_count = 0;
3828 EXPORT_SYMBOL(netif_napi_del);
3830 static void net_rx_action(struct softirq_action *h)
3832 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3833 unsigned long time_limit = jiffies + 2;
3834 int budget = netdev_budget;
3837 local_irq_disable();
3839 while (!list_empty(&sd->poll_list)) {
3840 struct napi_struct *n;
3843 /* If softirq window is exhuasted then punt.
3844 * Allow this to run for 2 jiffies since which will allow
3845 * an average latency of 1.5/HZ.
3847 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3852 /* Even though interrupts have been re-enabled, this
3853 * access is safe because interrupts can only add new
3854 * entries to the tail of this list, and only ->poll()
3855 * calls can remove this head entry from the list.
3857 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3859 have = netpoll_poll_lock(n);
3863 /* This NAPI_STATE_SCHED test is for avoiding a race
3864 * with netpoll's poll_napi(). Only the entity which
3865 * obtains the lock and sees NAPI_STATE_SCHED set will
3866 * actually make the ->poll() call. Therefore we avoid
3867 * accidently calling ->poll() when NAPI is not scheduled.
3870 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3871 work = n->poll(n, weight);
3875 WARN_ON_ONCE(work > weight);
3879 local_irq_disable();
3881 /* Drivers must not modify the NAPI state if they
3882 * consume the entire weight. In such cases this code
3883 * still "owns" the NAPI instance and therefore can
3884 * move the instance around on the list at-will.
3886 if (unlikely(work == weight)) {
3887 if (unlikely(napi_disable_pending(n))) {
3890 local_irq_disable();
3892 list_move_tail(&n->poll_list, &sd->poll_list);
3895 netpoll_poll_unlock(have);
3898 net_rps_action_and_irq_enable(sd);
3900 #ifdef CONFIG_NET_DMA
3902 * There may not be any more sk_buffs coming right now, so push
3903 * any pending DMA copies to hardware
3905 dma_issue_pending_all();
3912 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3916 static gifconf_func_t *gifconf_list[NPROTO];
3919 * register_gifconf - register a SIOCGIF handler
3920 * @family: Address family
3921 * @gifconf: Function handler
3923 * Register protocol dependent address dumping routines. The handler
3924 * that is passed must not be freed or reused until it has been replaced
3925 * by another handler.
3927 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3929 if (family >= NPROTO)
3931 gifconf_list[family] = gifconf;
3934 EXPORT_SYMBOL(register_gifconf);
3938 * Map an interface index to its name (SIOCGIFNAME)
3942 * We need this ioctl for efficient implementation of the
3943 * if_indextoname() function required by the IPv6 API. Without
3944 * it, we would have to search all the interfaces to find a
3948 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3950 struct net_device *dev;
3954 * Fetch the caller's info block.
3957 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3961 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3967 strcpy(ifr.ifr_name, dev->name);
3970 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3976 * Perform a SIOCGIFCONF call. This structure will change
3977 * size eventually, and there is nothing I can do about it.
3978 * Thus we will need a 'compatibility mode'.
3981 static int dev_ifconf(struct net *net, char __user *arg)
3984 struct net_device *dev;
3991 * Fetch the caller's info block.
3994 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4001 * Loop over the interfaces, and write an info block for each.
4005 for_each_netdev(net, dev) {
4006 for (i = 0; i < NPROTO; i++) {
4007 if (gifconf_list[i]) {
4010 done = gifconf_list[i](dev, NULL, 0);
4012 done = gifconf_list[i](dev, pos + total,
4022 * All done. Write the updated control block back to the caller.
4024 ifc.ifc_len = total;
4027 * Both BSD and Solaris return 0 here, so we do too.
4029 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4032 #ifdef CONFIG_PROC_FS
4034 * This is invoked by the /proc filesystem handler to display a device
4037 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4040 struct net *net = seq_file_net(seq);
4042 struct net_device *dev;
4046 return SEQ_START_TOKEN;
4049 for_each_netdev_rcu(net, dev)
4056 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4058 struct net_device *dev = v;
4060 if (v == SEQ_START_TOKEN)
4061 dev = first_net_device_rcu(seq_file_net(seq));
4063 dev = next_net_device_rcu(dev);
4069 void dev_seq_stop(struct seq_file *seq, void *v)
4075 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4077 struct rtnl_link_stats64 temp;
4078 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4080 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4081 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4082 dev->name, stats->rx_bytes, stats->rx_packets,
4084 stats->rx_dropped + stats->rx_missed_errors,
4085 stats->rx_fifo_errors,
4086 stats->rx_length_errors + stats->rx_over_errors +
4087 stats->rx_crc_errors + stats->rx_frame_errors,
4088 stats->rx_compressed, stats->multicast,
4089 stats->tx_bytes, stats->tx_packets,
4090 stats->tx_errors, stats->tx_dropped,
4091 stats->tx_fifo_errors, stats->collisions,
4092 stats->tx_carrier_errors +
4093 stats->tx_aborted_errors +
4094 stats->tx_window_errors +
4095 stats->tx_heartbeat_errors,
4096 stats->tx_compressed);
4100 * Called from the PROCfs module. This now uses the new arbitrary sized
4101 * /proc/net interface to create /proc/net/dev
4103 static int dev_seq_show(struct seq_file *seq, void *v)
4105 if (v == SEQ_START_TOKEN)
4106 seq_puts(seq, "Inter-| Receive "
4108 " face |bytes packets errs drop fifo frame "
4109 "compressed multicast|bytes packets errs "
4110 "drop fifo colls carrier compressed\n");
4112 dev_seq_printf_stats(seq, v);
4116 static struct softnet_data *softnet_get_online(loff_t *pos)
4118 struct softnet_data *sd = NULL;
4120 while (*pos < nr_cpu_ids)
4121 if (cpu_online(*pos)) {
4122 sd = &per_cpu(softnet_data, *pos);
4129 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4131 return softnet_get_online(pos);
4134 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4137 return softnet_get_online(pos);
4140 static void softnet_seq_stop(struct seq_file *seq, void *v)
4144 static int softnet_seq_show(struct seq_file *seq, void *v)
4146 struct softnet_data *sd = v;
4148 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4149 sd->processed, sd->dropped, sd->time_squeeze, 0,
4150 0, 0, 0, 0, /* was fastroute */
4151 sd->cpu_collision, sd->received_rps);
4155 static const struct seq_operations dev_seq_ops = {
4156 .start = dev_seq_start,
4157 .next = dev_seq_next,
4158 .stop = dev_seq_stop,
4159 .show = dev_seq_show,
4162 static int dev_seq_open(struct inode *inode, struct file *file)
4164 return seq_open_net(inode, file, &dev_seq_ops,
4165 sizeof(struct seq_net_private));
4168 static const struct file_operations dev_seq_fops = {
4169 .owner = THIS_MODULE,
4170 .open = dev_seq_open,
4172 .llseek = seq_lseek,
4173 .release = seq_release_net,
4176 static const struct seq_operations softnet_seq_ops = {
4177 .start = softnet_seq_start,
4178 .next = softnet_seq_next,
4179 .stop = softnet_seq_stop,
4180 .show = softnet_seq_show,
4183 static int softnet_seq_open(struct inode *inode, struct file *file)
4185 return seq_open(file, &softnet_seq_ops);
4188 static const struct file_operations softnet_seq_fops = {
4189 .owner = THIS_MODULE,
4190 .open = softnet_seq_open,
4192 .llseek = seq_lseek,
4193 .release = seq_release,
4196 static void *ptype_get_idx(loff_t pos)
4198 struct packet_type *pt = NULL;
4202 list_for_each_entry_rcu(pt, &ptype_all, list) {
4208 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4209 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4218 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4222 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4225 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4227 struct packet_type *pt;
4228 struct list_head *nxt;
4232 if (v == SEQ_START_TOKEN)
4233 return ptype_get_idx(0);
4236 nxt = pt->list.next;
4237 if (pt->type == htons(ETH_P_ALL)) {
4238 if (nxt != &ptype_all)
4241 nxt = ptype_base[0].next;
4243 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4245 while (nxt == &ptype_base[hash]) {
4246 if (++hash >= PTYPE_HASH_SIZE)
4248 nxt = ptype_base[hash].next;
4251 return list_entry(nxt, struct packet_type, list);
4254 static void ptype_seq_stop(struct seq_file *seq, void *v)
4260 static int ptype_seq_show(struct seq_file *seq, void *v)
4262 struct packet_type *pt = v;
4264 if (v == SEQ_START_TOKEN)
4265 seq_puts(seq, "Type Device Function\n");
4266 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4267 if (pt->type == htons(ETH_P_ALL))
4268 seq_puts(seq, "ALL ");
4270 seq_printf(seq, "%04x", ntohs(pt->type));
4272 seq_printf(seq, " %-8s %pF\n",
4273 pt->dev ? pt->dev->name : "", pt->func);
4279 static const struct seq_operations ptype_seq_ops = {
4280 .start = ptype_seq_start,
4281 .next = ptype_seq_next,
4282 .stop = ptype_seq_stop,
4283 .show = ptype_seq_show,
4286 static int ptype_seq_open(struct inode *inode, struct file *file)
4288 return seq_open_net(inode, file, &ptype_seq_ops,
4289 sizeof(struct seq_net_private));
4292 static const struct file_operations ptype_seq_fops = {
4293 .owner = THIS_MODULE,
4294 .open = ptype_seq_open,
4296 .llseek = seq_lseek,
4297 .release = seq_release_net,
4301 static int __net_init dev_proc_net_init(struct net *net)
4305 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4307 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4309 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4312 if (wext_proc_init(net))
4318 proc_net_remove(net, "ptype");
4320 proc_net_remove(net, "softnet_stat");
4322 proc_net_remove(net, "dev");
4326 static void __net_exit dev_proc_net_exit(struct net *net)
4328 wext_proc_exit(net);
4330 proc_net_remove(net, "ptype");
4331 proc_net_remove(net, "softnet_stat");
4332 proc_net_remove(net, "dev");
4335 static struct pernet_operations __net_initdata dev_proc_ops = {
4336 .init = dev_proc_net_init,
4337 .exit = dev_proc_net_exit,
4340 static int __init dev_proc_init(void)
4342 return register_pernet_subsys(&dev_proc_ops);
4345 #define dev_proc_init() 0
4346 #endif /* CONFIG_PROC_FS */
4350 * netdev_set_master - set up master/slave pair
4351 * @slave: slave device
4352 * @master: new master device
4354 * Changes the master device of the slave. Pass %NULL to break the
4355 * bonding. The caller must hold the RTNL semaphore. On a failure
4356 * a negative errno code is returned. On success the reference counts
4357 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4358 * function returns zero.
4360 int netdev_set_master(struct net_device *slave, struct net_device *master)
4362 struct net_device *old = slave->master;
4372 slave->master = master;
4379 slave->flags |= IFF_SLAVE;
4381 slave->flags &= ~IFF_SLAVE;
4383 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4386 EXPORT_SYMBOL(netdev_set_master);
4388 static void dev_change_rx_flags(struct net_device *dev, int flags)
4390 const struct net_device_ops *ops = dev->netdev_ops;
4392 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4393 ops->ndo_change_rx_flags(dev, flags);
4396 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4398 unsigned short old_flags = dev->flags;
4404 dev->flags |= IFF_PROMISC;
4405 dev->promiscuity += inc;
4406 if (dev->promiscuity == 0) {
4409 * If inc causes overflow, untouch promisc and return error.
4412 dev->flags &= ~IFF_PROMISC;
4414 dev->promiscuity -= inc;
4415 printk(KERN_WARNING "%s: promiscuity touches roof, "
4416 "set promiscuity failed, promiscuity feature "
4417 "of device might be broken.\n", dev->name);
4421 if (dev->flags != old_flags) {
4422 printk(KERN_INFO "device %s %s promiscuous mode\n",
4423 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4425 if (audit_enabled) {
4426 current_uid_gid(&uid, &gid);
4427 audit_log(current->audit_context, GFP_ATOMIC,
4428 AUDIT_ANOM_PROMISCUOUS,
4429 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4430 dev->name, (dev->flags & IFF_PROMISC),
4431 (old_flags & IFF_PROMISC),
4432 audit_get_loginuid(current),
4434 audit_get_sessionid(current));
4437 dev_change_rx_flags(dev, IFF_PROMISC);
4443 * dev_set_promiscuity - update promiscuity count on a device
4447 * Add or remove promiscuity from a device. While the count in the device
4448 * remains above zero the interface remains promiscuous. Once it hits zero
4449 * the device reverts back to normal filtering operation. A negative inc
4450 * value is used to drop promiscuity on the device.
4451 * Return 0 if successful or a negative errno code on error.
4453 int dev_set_promiscuity(struct net_device *dev, int inc)
4455 unsigned short old_flags = dev->flags;
4458 err = __dev_set_promiscuity(dev, inc);
4461 if (dev->flags != old_flags)
4462 dev_set_rx_mode(dev);
4465 EXPORT_SYMBOL(dev_set_promiscuity);
4468 * dev_set_allmulti - update allmulti count on a device
4472 * Add or remove reception of all multicast frames to a device. While the
4473 * count in the device remains above zero the interface remains listening
4474 * to all interfaces. Once it hits zero the device reverts back to normal
4475 * filtering operation. A negative @inc value is used to drop the counter
4476 * when releasing a resource needing all multicasts.
4477 * Return 0 if successful or a negative errno code on error.
4480 int dev_set_allmulti(struct net_device *dev, int inc)
4482 unsigned short old_flags = dev->flags;
4486 dev->flags |= IFF_ALLMULTI;
4487 dev->allmulti += inc;
4488 if (dev->allmulti == 0) {
4491 * If inc causes overflow, untouch allmulti and return error.
4494 dev->flags &= ~IFF_ALLMULTI;
4496 dev->allmulti -= inc;
4497 printk(KERN_WARNING "%s: allmulti touches roof, "
4498 "set allmulti failed, allmulti feature of "
4499 "device might be broken.\n", dev->name);
4503 if (dev->flags ^ old_flags) {
4504 dev_change_rx_flags(dev, IFF_ALLMULTI);
4505 dev_set_rx_mode(dev);
4509 EXPORT_SYMBOL(dev_set_allmulti);
4512 * Upload unicast and multicast address lists to device and
4513 * configure RX filtering. When the device doesn't support unicast
4514 * filtering it is put in promiscuous mode while unicast addresses
4517 void __dev_set_rx_mode(struct net_device *dev)
4519 const struct net_device_ops *ops = dev->netdev_ops;
4521 /* dev_open will call this function so the list will stay sane. */
4522 if (!(dev->flags&IFF_UP))
4525 if (!netif_device_present(dev))
4528 if (ops->ndo_set_rx_mode)
4529 ops->ndo_set_rx_mode(dev);
4531 /* Unicast addresses changes may only happen under the rtnl,
4532 * therefore calling __dev_set_promiscuity here is safe.
4534 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4535 __dev_set_promiscuity(dev, 1);
4536 dev->uc_promisc = 1;
4537 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4538 __dev_set_promiscuity(dev, -1);
4539 dev->uc_promisc = 0;
4542 if (ops->ndo_set_multicast_list)
4543 ops->ndo_set_multicast_list(dev);
4547 void dev_set_rx_mode(struct net_device *dev)
4549 netif_addr_lock_bh(dev);
4550 __dev_set_rx_mode(dev);
4551 netif_addr_unlock_bh(dev);
4555 * dev_get_flags - get flags reported to userspace
4558 * Get the combination of flag bits exported through APIs to userspace.
4560 unsigned dev_get_flags(const struct net_device *dev)
4564 flags = (dev->flags & ~(IFF_PROMISC |
4569 (dev->gflags & (IFF_PROMISC |
4572 if (netif_running(dev)) {
4573 if (netif_oper_up(dev))
4574 flags |= IFF_RUNNING;
4575 if (netif_carrier_ok(dev))
4576 flags |= IFF_LOWER_UP;
4577 if (netif_dormant(dev))
4578 flags |= IFF_DORMANT;
4583 EXPORT_SYMBOL(dev_get_flags);
4585 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4587 int old_flags = dev->flags;
4593 * Set the flags on our device.
4596 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4597 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4599 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4603 * Load in the correct multicast list now the flags have changed.
4606 if ((old_flags ^ flags) & IFF_MULTICAST)
4607 dev_change_rx_flags(dev, IFF_MULTICAST);
4609 dev_set_rx_mode(dev);
4612 * Have we downed the interface. We handle IFF_UP ourselves
4613 * according to user attempts to set it, rather than blindly
4618 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4619 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4622 dev_set_rx_mode(dev);
4625 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4626 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4628 dev->gflags ^= IFF_PROMISC;
4629 dev_set_promiscuity(dev, inc);
4632 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4633 is important. Some (broken) drivers set IFF_PROMISC, when
4634 IFF_ALLMULTI is requested not asking us and not reporting.
4636 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4637 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4639 dev->gflags ^= IFF_ALLMULTI;
4640 dev_set_allmulti(dev, inc);
4646 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4648 unsigned int changes = dev->flags ^ old_flags;
4650 if (changes & IFF_UP) {
4651 if (dev->flags & IFF_UP)
4652 call_netdevice_notifiers(NETDEV_UP, dev);
4654 call_netdevice_notifiers(NETDEV_DOWN, dev);
4657 if (dev->flags & IFF_UP &&
4658 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4659 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4663 * dev_change_flags - change device settings
4665 * @flags: device state flags
4667 * Change settings on device based state flags. The flags are
4668 * in the userspace exported format.
4670 int dev_change_flags(struct net_device *dev, unsigned flags)
4673 int old_flags = dev->flags;
4675 ret = __dev_change_flags(dev, flags);
4679 changes = old_flags ^ dev->flags;
4681 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4683 __dev_notify_flags(dev, old_flags);
4686 EXPORT_SYMBOL(dev_change_flags);
4689 * dev_set_mtu - Change maximum transfer unit
4691 * @new_mtu: new transfer unit
4693 * Change the maximum transfer size of the network device.
4695 int dev_set_mtu(struct net_device *dev, int new_mtu)
4697 const struct net_device_ops *ops = dev->netdev_ops;
4700 if (new_mtu == dev->mtu)
4703 /* MTU must be positive. */
4707 if (!netif_device_present(dev))
4711 if (ops->ndo_change_mtu)
4712 err = ops->ndo_change_mtu(dev, new_mtu);
4716 if (!err && dev->flags & IFF_UP)
4717 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4720 EXPORT_SYMBOL(dev_set_mtu);
4723 * dev_set_group - Change group this device belongs to
4725 * @new_group: group this device should belong to
4727 void dev_set_group(struct net_device *dev, int new_group)
4729 dev->group = new_group;
4731 EXPORT_SYMBOL(dev_set_group);
4734 * dev_set_mac_address - Change Media Access Control Address
4738 * Change the hardware (MAC) address of the device
4740 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4742 const struct net_device_ops *ops = dev->netdev_ops;
4745 if (!ops->ndo_set_mac_address)
4747 if (sa->sa_family != dev->type)
4749 if (!netif_device_present(dev))
4751 err = ops->ndo_set_mac_address(dev, sa);
4753 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4756 EXPORT_SYMBOL(dev_set_mac_address);
4759 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4761 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4764 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4770 case SIOCGIFFLAGS: /* Get interface flags */
4771 ifr->ifr_flags = (short) dev_get_flags(dev);
4774 case SIOCGIFMETRIC: /* Get the metric on the interface
4775 (currently unused) */
4776 ifr->ifr_metric = 0;
4779 case SIOCGIFMTU: /* Get the MTU of a device */
4780 ifr->ifr_mtu = dev->mtu;
4785 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4787 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4788 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4789 ifr->ifr_hwaddr.sa_family = dev->type;
4797 ifr->ifr_map.mem_start = dev->mem_start;
4798 ifr->ifr_map.mem_end = dev->mem_end;
4799 ifr->ifr_map.base_addr = dev->base_addr;
4800 ifr->ifr_map.irq = dev->irq;
4801 ifr->ifr_map.dma = dev->dma;
4802 ifr->ifr_map.port = dev->if_port;
4806 ifr->ifr_ifindex = dev->ifindex;
4810 ifr->ifr_qlen = dev->tx_queue_len;
4814 /* dev_ioctl() should ensure this case
4826 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4828 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4831 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4832 const struct net_device_ops *ops;
4837 ops = dev->netdev_ops;
4840 case SIOCSIFFLAGS: /* Set interface flags */
4841 return dev_change_flags(dev, ifr->ifr_flags);
4843 case SIOCSIFMETRIC: /* Set the metric on the interface
4844 (currently unused) */
4847 case SIOCSIFMTU: /* Set the MTU of a device */
4848 return dev_set_mtu(dev, ifr->ifr_mtu);
4851 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4853 case SIOCSIFHWBROADCAST:
4854 if (ifr->ifr_hwaddr.sa_family != dev->type)
4856 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4857 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4858 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4862 if (ops->ndo_set_config) {
4863 if (!netif_device_present(dev))
4865 return ops->ndo_set_config(dev, &ifr->ifr_map);
4870 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4871 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4873 if (!netif_device_present(dev))
4875 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4878 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4879 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4881 if (!netif_device_present(dev))
4883 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4886 if (ifr->ifr_qlen < 0)
4888 dev->tx_queue_len = ifr->ifr_qlen;
4892 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4893 return dev_change_name(dev, ifr->ifr_newname);
4896 * Unknown or private ioctl
4899 if ((cmd >= SIOCDEVPRIVATE &&
4900 cmd <= SIOCDEVPRIVATE + 15) ||
4901 cmd == SIOCBONDENSLAVE ||
4902 cmd == SIOCBONDRELEASE ||
4903 cmd == SIOCBONDSETHWADDR ||
4904 cmd == SIOCBONDSLAVEINFOQUERY ||
4905 cmd == SIOCBONDINFOQUERY ||
4906 cmd == SIOCBONDCHANGEACTIVE ||
4907 cmd == SIOCGMIIPHY ||
4908 cmd == SIOCGMIIREG ||
4909 cmd == SIOCSMIIREG ||
4910 cmd == SIOCBRADDIF ||
4911 cmd == SIOCBRDELIF ||
4912 cmd == SIOCSHWTSTAMP ||
4913 cmd == SIOCWANDEV) {
4915 if (ops->ndo_do_ioctl) {
4916 if (netif_device_present(dev))
4917 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4929 * This function handles all "interface"-type I/O control requests. The actual
4930 * 'doing' part of this is dev_ifsioc above.
4934 * dev_ioctl - network device ioctl
4935 * @net: the applicable net namespace
4936 * @cmd: command to issue
4937 * @arg: pointer to a struct ifreq in user space
4939 * Issue ioctl functions to devices. This is normally called by the
4940 * user space syscall interfaces but can sometimes be useful for
4941 * other purposes. The return value is the return from the syscall if
4942 * positive or a negative errno code on error.
4945 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4951 /* One special case: SIOCGIFCONF takes ifconf argument
4952 and requires shared lock, because it sleeps writing
4956 if (cmd == SIOCGIFCONF) {
4958 ret = dev_ifconf(net, (char __user *) arg);
4962 if (cmd == SIOCGIFNAME)
4963 return dev_ifname(net, (struct ifreq __user *)arg);
4965 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4968 ifr.ifr_name[IFNAMSIZ-1] = 0;
4970 colon = strchr(ifr.ifr_name, ':');
4975 * See which interface the caller is talking about.
4980 * These ioctl calls:
4981 * - can be done by all.
4982 * - atomic and do not require locking.
4993 dev_load(net, ifr.ifr_name);
4995 ret = dev_ifsioc_locked(net, &ifr, cmd);
5000 if (copy_to_user(arg, &ifr,
5001 sizeof(struct ifreq)))
5007 dev_load(net, ifr.ifr_name);
5009 ret = dev_ethtool(net, &ifr);
5014 if (copy_to_user(arg, &ifr,
5015 sizeof(struct ifreq)))
5021 * These ioctl calls:
5022 * - require superuser power.
5023 * - require strict serialization.
5029 if (!capable(CAP_NET_ADMIN))
5031 dev_load(net, ifr.ifr_name);
5033 ret = dev_ifsioc(net, &ifr, cmd);
5038 if (copy_to_user(arg, &ifr,
5039 sizeof(struct ifreq)))
5045 * These ioctl calls:
5046 * - require superuser power.
5047 * - require strict serialization.
5048 * - do not return a value
5058 case SIOCSIFHWBROADCAST:
5061 case SIOCBONDENSLAVE:
5062 case SIOCBONDRELEASE:
5063 case SIOCBONDSETHWADDR:
5064 case SIOCBONDCHANGEACTIVE:
5068 if (!capable(CAP_NET_ADMIN))
5071 case SIOCBONDSLAVEINFOQUERY:
5072 case SIOCBONDINFOQUERY:
5073 dev_load(net, ifr.ifr_name);
5075 ret = dev_ifsioc(net, &ifr, cmd);
5080 /* Get the per device memory space. We can add this but
5081 * currently do not support it */
5083 /* Set the per device memory buffer space.
5084 * Not applicable in our case */
5089 * Unknown or private ioctl.
5092 if (cmd == SIOCWANDEV ||
5093 (cmd >= SIOCDEVPRIVATE &&
5094 cmd <= SIOCDEVPRIVATE + 15)) {
5095 dev_load(net, ifr.ifr_name);
5097 ret = dev_ifsioc(net, &ifr, cmd);
5099 if (!ret && copy_to_user(arg, &ifr,
5100 sizeof(struct ifreq)))
5104 /* Take care of Wireless Extensions */
5105 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5106 return wext_handle_ioctl(net, &ifr, cmd, arg);
5113 * dev_new_index - allocate an ifindex
5114 * @net: the applicable net namespace
5116 * Returns a suitable unique value for a new device interface
5117 * number. The caller must hold the rtnl semaphore or the
5118 * dev_base_lock to be sure it remains unique.
5120 static int dev_new_index(struct net *net)
5126 if (!__dev_get_by_index(net, ifindex))
5131 /* Delayed registration/unregisteration */
5132 static LIST_HEAD(net_todo_list);
5134 static void net_set_todo(struct net_device *dev)
5136 list_add_tail(&dev->todo_list, &net_todo_list);
5139 static void rollback_registered_many(struct list_head *head)
5141 struct net_device *dev, *tmp;
5143 BUG_ON(dev_boot_phase);
5146 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5147 /* Some devices call without registering
5148 * for initialization unwind. Remove those
5149 * devices and proceed with the remaining.
5151 if (dev->reg_state == NETREG_UNINITIALIZED) {
5152 pr_debug("unregister_netdevice: device %s/%p never "
5153 "was registered\n", dev->name, dev);
5156 list_del(&dev->unreg_list);
5160 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5163 /* If device is running, close it first. */
5164 dev_close_many(head);
5166 list_for_each_entry(dev, head, unreg_list) {
5167 /* And unlink it from device chain. */
5168 unlist_netdevice(dev);
5170 dev->reg_state = NETREG_UNREGISTERING;
5175 list_for_each_entry(dev, head, unreg_list) {
5176 /* Shutdown queueing discipline. */
5180 /* Notify protocols, that we are about to destroy
5181 this device. They should clean all the things.
5183 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5185 if (!dev->rtnl_link_ops ||
5186 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5187 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5190 * Flush the unicast and multicast chains
5195 if (dev->netdev_ops->ndo_uninit)
5196 dev->netdev_ops->ndo_uninit(dev);
5198 /* Notifier chain MUST detach us from master device. */
5199 WARN_ON(dev->master);
5201 /* Remove entries from kobject tree */
5202 netdev_unregister_kobject(dev);
5205 /* Process any work delayed until the end of the batch */
5206 dev = list_first_entry(head, struct net_device, unreg_list);
5207 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5211 list_for_each_entry(dev, head, unreg_list)
5215 static void rollback_registered(struct net_device *dev)
5219 list_add(&dev->unreg_list, &single);
5220 rollback_registered_many(&single);
5223 u32 netdev_fix_features(struct net_device *dev, u32 features)
5225 /* Fix illegal checksum combinations */
5226 if ((features & NETIF_F_HW_CSUM) &&
5227 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5228 netdev_info(dev, "mixed HW and IP checksum settings.\n");
5229 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5232 if ((features & NETIF_F_NO_CSUM) &&
5233 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5234 netdev_info(dev, "mixed no checksumming and other settings.\n");
5235 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5238 /* Fix illegal SG+CSUM combinations. */
5239 if ((features & NETIF_F_SG) &&
5240 !(features & NETIF_F_ALL_CSUM)) {
5242 "Dropping NETIF_F_SG since no checksum feature.\n");
5243 features &= ~NETIF_F_SG;
5246 /* TSO requires that SG is present as well. */
5247 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
5248 netdev_info(dev, "Dropping NETIF_F_TSO since no SG feature.\n");
5249 features &= ~NETIF_F_TSO;
5252 /* UFO needs SG and checksumming */
5253 if (features & NETIF_F_UFO) {
5254 /* maybe split UFO into V4 and V6? */
5255 if (!((features & NETIF_F_GEN_CSUM) ||
5256 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5257 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5259 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5260 features &= ~NETIF_F_UFO;
5263 if (!(features & NETIF_F_SG)) {
5265 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5266 features &= ~NETIF_F_UFO;
5272 EXPORT_SYMBOL(netdev_fix_features);
5275 * netif_stacked_transfer_operstate - transfer operstate
5276 * @rootdev: the root or lower level device to transfer state from
5277 * @dev: the device to transfer operstate to
5279 * Transfer operational state from root to device. This is normally
5280 * called when a stacking relationship exists between the root
5281 * device and the device(a leaf device).
5283 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5284 struct net_device *dev)
5286 if (rootdev->operstate == IF_OPER_DORMANT)
5287 netif_dormant_on(dev);
5289 netif_dormant_off(dev);
5291 if (netif_carrier_ok(rootdev)) {
5292 if (!netif_carrier_ok(dev))
5293 netif_carrier_on(dev);
5295 if (netif_carrier_ok(dev))
5296 netif_carrier_off(dev);
5299 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5302 static int netif_alloc_rx_queues(struct net_device *dev)
5304 unsigned int i, count = dev->num_rx_queues;
5305 struct netdev_rx_queue *rx;
5309 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5311 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5316 for (i = 0; i < count; i++)
5322 static void netdev_init_one_queue(struct net_device *dev,
5323 struct netdev_queue *queue, void *_unused)
5325 /* Initialize queue lock */
5326 spin_lock_init(&queue->_xmit_lock);
5327 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5328 queue->xmit_lock_owner = -1;
5329 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5333 static int netif_alloc_netdev_queues(struct net_device *dev)
5335 unsigned int count = dev->num_tx_queues;
5336 struct netdev_queue *tx;
5340 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5342 pr_err("netdev: Unable to allocate %u tx queues.\n",
5348 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5349 spin_lock_init(&dev->tx_global_lock);
5355 * register_netdevice - register a network device
5356 * @dev: device to register
5358 * Take a completed network device structure and add it to the kernel
5359 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5360 * chain. 0 is returned on success. A negative errno code is returned
5361 * on a failure to set up the device, or if the name is a duplicate.
5363 * Callers must hold the rtnl semaphore. You may want
5364 * register_netdev() instead of this.
5367 * The locking appears insufficient to guarantee two parallel registers
5368 * will not get the same name.
5371 int register_netdevice(struct net_device *dev)
5374 struct net *net = dev_net(dev);
5376 BUG_ON(dev_boot_phase);
5381 /* When net_device's are persistent, this will be fatal. */
5382 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5385 spin_lock_init(&dev->addr_list_lock);
5386 netdev_set_addr_lockdep_class(dev);
5390 /* Init, if this function is available */
5391 if (dev->netdev_ops->ndo_init) {
5392 ret = dev->netdev_ops->ndo_init(dev);
5400 ret = dev_get_valid_name(dev, dev->name, 0);
5404 dev->ifindex = dev_new_index(net);
5405 if (dev->iflink == -1)
5406 dev->iflink = dev->ifindex;
5408 dev->features = netdev_fix_features(dev, dev->features);
5410 /* Enable software GSO if SG is supported. */
5411 if (dev->features & NETIF_F_SG)
5412 dev->features |= NETIF_F_GSO;
5414 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5415 * vlan_dev_init() will do the dev->features check, so these features
5416 * are enabled only if supported by underlying device.
5418 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5420 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5421 ret = notifier_to_errno(ret);
5425 ret = netdev_register_kobject(dev);
5428 dev->reg_state = NETREG_REGISTERED;
5431 * Default initial state at registry is that the
5432 * device is present.
5435 set_bit(__LINK_STATE_PRESENT, &dev->state);
5437 dev_init_scheduler(dev);
5439 list_netdevice(dev);
5441 /* Notify protocols, that a new device appeared. */
5442 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5443 ret = notifier_to_errno(ret);
5445 rollback_registered(dev);
5446 dev->reg_state = NETREG_UNREGISTERED;
5449 * Prevent userspace races by waiting until the network
5450 * device is fully setup before sending notifications.
5452 if (!dev->rtnl_link_ops ||
5453 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5454 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5460 if (dev->netdev_ops->ndo_uninit)
5461 dev->netdev_ops->ndo_uninit(dev);
5464 EXPORT_SYMBOL(register_netdevice);
5467 * init_dummy_netdev - init a dummy network device for NAPI
5468 * @dev: device to init
5470 * This takes a network device structure and initialize the minimum
5471 * amount of fields so it can be used to schedule NAPI polls without
5472 * registering a full blown interface. This is to be used by drivers
5473 * that need to tie several hardware interfaces to a single NAPI
5474 * poll scheduler due to HW limitations.
5476 int init_dummy_netdev(struct net_device *dev)
5478 /* Clear everything. Note we don't initialize spinlocks
5479 * are they aren't supposed to be taken by any of the
5480 * NAPI code and this dummy netdev is supposed to be
5481 * only ever used for NAPI polls
5483 memset(dev, 0, sizeof(struct net_device));
5485 /* make sure we BUG if trying to hit standard
5486 * register/unregister code path
5488 dev->reg_state = NETREG_DUMMY;
5490 /* NAPI wants this */
5491 INIT_LIST_HEAD(&dev->napi_list);
5493 /* a dummy interface is started by default */
5494 set_bit(__LINK_STATE_PRESENT, &dev->state);
5495 set_bit(__LINK_STATE_START, &dev->state);
5497 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5498 * because users of this 'device' dont need to change
5504 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5508 * register_netdev - register a network device
5509 * @dev: device to register
5511 * Take a completed network device structure and add it to the kernel
5512 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5513 * chain. 0 is returned on success. A negative errno code is returned
5514 * on a failure to set up the device, or if the name is a duplicate.
5516 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5517 * and expands the device name if you passed a format string to
5520 int register_netdev(struct net_device *dev)
5527 * If the name is a format string the caller wants us to do a
5530 if (strchr(dev->name, '%')) {
5531 err = dev_alloc_name(dev, dev->name);
5536 err = register_netdevice(dev);
5541 EXPORT_SYMBOL(register_netdev);
5543 int netdev_refcnt_read(const struct net_device *dev)
5547 for_each_possible_cpu(i)
5548 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5551 EXPORT_SYMBOL(netdev_refcnt_read);
5554 * netdev_wait_allrefs - wait until all references are gone.
5556 * This is called when unregistering network devices.
5558 * Any protocol or device that holds a reference should register
5559 * for netdevice notification, and cleanup and put back the
5560 * reference if they receive an UNREGISTER event.
5561 * We can get stuck here if buggy protocols don't correctly
5564 static void netdev_wait_allrefs(struct net_device *dev)
5566 unsigned long rebroadcast_time, warning_time;
5569 linkwatch_forget_dev(dev);
5571 rebroadcast_time = warning_time = jiffies;
5572 refcnt = netdev_refcnt_read(dev);
5574 while (refcnt != 0) {
5575 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5578 /* Rebroadcast unregister notification */
5579 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5580 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5581 * should have already handle it the first time */
5583 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5585 /* We must not have linkwatch events
5586 * pending on unregister. If this
5587 * happens, we simply run the queue
5588 * unscheduled, resulting in a noop
5591 linkwatch_run_queue();
5596 rebroadcast_time = jiffies;
5601 refcnt = netdev_refcnt_read(dev);
5603 if (time_after(jiffies, warning_time + 10 * HZ)) {
5604 printk(KERN_EMERG "unregister_netdevice: "
5605 "waiting for %s to become free. Usage "
5608 warning_time = jiffies;
5617 * register_netdevice(x1);
5618 * register_netdevice(x2);
5620 * unregister_netdevice(y1);
5621 * unregister_netdevice(y2);
5627 * We are invoked by rtnl_unlock().
5628 * This allows us to deal with problems:
5629 * 1) We can delete sysfs objects which invoke hotplug
5630 * without deadlocking with linkwatch via keventd.
5631 * 2) Since we run with the RTNL semaphore not held, we can sleep
5632 * safely in order to wait for the netdev refcnt to drop to zero.
5634 * We must not return until all unregister events added during
5635 * the interval the lock was held have been completed.
5637 void netdev_run_todo(void)
5639 struct list_head list;
5641 /* Snapshot list, allow later requests */
5642 list_replace_init(&net_todo_list, &list);
5646 while (!list_empty(&list)) {
5647 struct net_device *dev
5648 = list_first_entry(&list, struct net_device, todo_list);
5649 list_del(&dev->todo_list);
5651 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5652 printk(KERN_ERR "network todo '%s' but state %d\n",
5653 dev->name, dev->reg_state);
5658 dev->reg_state = NETREG_UNREGISTERED;
5660 on_each_cpu(flush_backlog, dev, 1);
5662 netdev_wait_allrefs(dev);
5665 BUG_ON(netdev_refcnt_read(dev));
5666 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5667 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5668 WARN_ON(dev->dn_ptr);
5670 if (dev->destructor)
5671 dev->destructor(dev);
5673 /* Free network device */
5674 kobject_put(&dev->dev.kobj);
5678 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5679 * fields in the same order, with only the type differing.
5681 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5682 const struct net_device_stats *netdev_stats)
5684 #if BITS_PER_LONG == 64
5685 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5686 memcpy(stats64, netdev_stats, sizeof(*stats64));
5688 size_t i, n = sizeof(*stats64) / sizeof(u64);
5689 const unsigned long *src = (const unsigned long *)netdev_stats;
5690 u64 *dst = (u64 *)stats64;
5692 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5693 sizeof(*stats64) / sizeof(u64));
5694 for (i = 0; i < n; i++)
5700 * dev_get_stats - get network device statistics
5701 * @dev: device to get statistics from
5702 * @storage: place to store stats
5704 * Get network statistics from device. Return @storage.
5705 * The device driver may provide its own method by setting
5706 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5707 * otherwise the internal statistics structure is used.
5709 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5710 struct rtnl_link_stats64 *storage)
5712 const struct net_device_ops *ops = dev->netdev_ops;
5714 if (ops->ndo_get_stats64) {
5715 memset(storage, 0, sizeof(*storage));
5716 ops->ndo_get_stats64(dev, storage);
5717 } else if (ops->ndo_get_stats) {
5718 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5720 netdev_stats_to_stats64(storage, &dev->stats);
5722 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5725 EXPORT_SYMBOL(dev_get_stats);
5727 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5729 struct netdev_queue *queue = dev_ingress_queue(dev);
5731 #ifdef CONFIG_NET_CLS_ACT
5734 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5737 netdev_init_one_queue(dev, queue, NULL);
5738 queue->qdisc = &noop_qdisc;
5739 queue->qdisc_sleeping = &noop_qdisc;
5740 rcu_assign_pointer(dev->ingress_queue, queue);
5746 * alloc_netdev_mqs - allocate network device
5747 * @sizeof_priv: size of private data to allocate space for
5748 * @name: device name format string
5749 * @setup: callback to initialize device
5750 * @txqs: the number of TX subqueues to allocate
5751 * @rxqs: the number of RX subqueues to allocate
5753 * Allocates a struct net_device with private data area for driver use
5754 * and performs basic initialization. Also allocates subquue structs
5755 * for each queue on the device.
5757 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5758 void (*setup)(struct net_device *),
5759 unsigned int txqs, unsigned int rxqs)
5761 struct net_device *dev;
5763 struct net_device *p;
5765 BUG_ON(strlen(name) >= sizeof(dev->name));
5768 pr_err("alloc_netdev: Unable to allocate device "
5769 "with zero queues.\n");
5775 pr_err("alloc_netdev: Unable to allocate device "
5776 "with zero RX queues.\n");
5781 alloc_size = sizeof(struct net_device);
5783 /* ensure 32-byte alignment of private area */
5784 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5785 alloc_size += sizeof_priv;
5787 /* ensure 32-byte alignment of whole construct */
5788 alloc_size += NETDEV_ALIGN - 1;
5790 p = kzalloc(alloc_size, GFP_KERNEL);
5792 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5796 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5797 dev->padded = (char *)dev - (char *)p;
5799 dev->pcpu_refcnt = alloc_percpu(int);
5800 if (!dev->pcpu_refcnt)
5803 if (dev_addr_init(dev))
5809 dev_net_set(dev, &init_net);
5811 dev->gso_max_size = GSO_MAX_SIZE;
5813 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5814 dev->ethtool_ntuple_list.count = 0;
5815 INIT_LIST_HEAD(&dev->napi_list);
5816 INIT_LIST_HEAD(&dev->unreg_list);
5817 INIT_LIST_HEAD(&dev->link_watch_list);
5818 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5821 dev->num_tx_queues = txqs;
5822 dev->real_num_tx_queues = txqs;
5823 if (netif_alloc_netdev_queues(dev))
5827 dev->num_rx_queues = rxqs;
5828 dev->real_num_rx_queues = rxqs;
5829 if (netif_alloc_rx_queues(dev))
5833 strcpy(dev->name, name);
5834 dev->group = INIT_NETDEV_GROUP;
5842 free_percpu(dev->pcpu_refcnt);
5852 EXPORT_SYMBOL(alloc_netdev_mqs);
5855 * free_netdev - free network device
5858 * This function does the last stage of destroying an allocated device
5859 * interface. The reference to the device object is released.
5860 * If this is the last reference then it will be freed.
5862 void free_netdev(struct net_device *dev)
5864 struct napi_struct *p, *n;
5866 release_net(dev_net(dev));
5873 kfree(rcu_dereference_raw(dev->ingress_queue));
5875 /* Flush device addresses */
5876 dev_addr_flush(dev);
5878 /* Clear ethtool n-tuple list */
5879 ethtool_ntuple_flush(dev);
5881 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5884 free_percpu(dev->pcpu_refcnt);
5885 dev->pcpu_refcnt = NULL;
5887 /* Compatibility with error handling in drivers */
5888 if (dev->reg_state == NETREG_UNINITIALIZED) {
5889 kfree((char *)dev - dev->padded);
5893 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5894 dev->reg_state = NETREG_RELEASED;
5896 /* will free via device release */
5897 put_device(&dev->dev);
5899 EXPORT_SYMBOL(free_netdev);
5902 * synchronize_net - Synchronize with packet receive processing
5904 * Wait for packets currently being received to be done.
5905 * Does not block later packets from starting.
5907 void synchronize_net(void)
5912 EXPORT_SYMBOL(synchronize_net);
5915 * unregister_netdevice_queue - remove device from the kernel
5919 * This function shuts down a device interface and removes it
5920 * from the kernel tables.
5921 * If head not NULL, device is queued to be unregistered later.
5923 * Callers must hold the rtnl semaphore. You may want
5924 * unregister_netdev() instead of this.
5927 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5932 list_move_tail(&dev->unreg_list, head);
5934 rollback_registered(dev);
5935 /* Finish processing unregister after unlock */
5939 EXPORT_SYMBOL(unregister_netdevice_queue);
5942 * unregister_netdevice_many - unregister many devices
5943 * @head: list of devices
5945 void unregister_netdevice_many(struct list_head *head)
5947 struct net_device *dev;
5949 if (!list_empty(head)) {
5950 rollback_registered_many(head);
5951 list_for_each_entry(dev, head, unreg_list)
5955 EXPORT_SYMBOL(unregister_netdevice_many);
5958 * unregister_netdev - remove device from the kernel
5961 * This function shuts down a device interface and removes it
5962 * from the kernel tables.
5964 * This is just a wrapper for unregister_netdevice that takes
5965 * the rtnl semaphore. In general you want to use this and not
5966 * unregister_netdevice.
5968 void unregister_netdev(struct net_device *dev)
5971 unregister_netdevice(dev);
5974 EXPORT_SYMBOL(unregister_netdev);
5977 * dev_change_net_namespace - move device to different nethost namespace
5979 * @net: network namespace
5980 * @pat: If not NULL name pattern to try if the current device name
5981 * is already taken in the destination network namespace.
5983 * This function shuts down a device interface and moves it
5984 * to a new network namespace. On success 0 is returned, on
5985 * a failure a netagive errno code is returned.
5987 * Callers must hold the rtnl semaphore.
5990 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5996 /* Don't allow namespace local devices to be moved. */
5998 if (dev->features & NETIF_F_NETNS_LOCAL)
6001 /* Ensure the device has been registrered */
6003 if (dev->reg_state != NETREG_REGISTERED)
6006 /* Get out if there is nothing todo */
6008 if (net_eq(dev_net(dev), net))
6011 /* Pick the destination device name, and ensure
6012 * we can use it in the destination network namespace.
6015 if (__dev_get_by_name(net, dev->name)) {
6016 /* We get here if we can't use the current device name */
6019 if (dev_get_valid_name(dev, pat, 1))
6024 * And now a mini version of register_netdevice unregister_netdevice.
6027 /* If device is running close it first. */
6030 /* And unlink it from device chain */
6032 unlist_netdevice(dev);
6036 /* Shutdown queueing discipline. */
6039 /* Notify protocols, that we are about to destroy
6040 this device. They should clean all the things.
6042 Note that dev->reg_state stays at NETREG_REGISTERED.
6043 This is wanted because this way 8021q and macvlan know
6044 the device is just moving and can keep their slaves up.
6046 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6047 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6050 * Flush the unicast and multicast chains
6055 /* Actually switch the network namespace */
6056 dev_net_set(dev, net);
6058 /* If there is an ifindex conflict assign a new one */
6059 if (__dev_get_by_index(net, dev->ifindex)) {
6060 int iflink = (dev->iflink == dev->ifindex);
6061 dev->ifindex = dev_new_index(net);
6063 dev->iflink = dev->ifindex;
6066 /* Fixup kobjects */
6067 err = device_rename(&dev->dev, dev->name);
6070 /* Add the device back in the hashes */
6071 list_netdevice(dev);
6073 /* Notify protocols, that a new device appeared. */
6074 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6077 * Prevent userspace races by waiting until the network
6078 * device is fully setup before sending notifications.
6080 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6087 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6089 static int dev_cpu_callback(struct notifier_block *nfb,
6090 unsigned long action,
6093 struct sk_buff **list_skb;
6094 struct sk_buff *skb;
6095 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6096 struct softnet_data *sd, *oldsd;
6098 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6101 local_irq_disable();
6102 cpu = smp_processor_id();
6103 sd = &per_cpu(softnet_data, cpu);
6104 oldsd = &per_cpu(softnet_data, oldcpu);
6106 /* Find end of our completion_queue. */
6107 list_skb = &sd->completion_queue;
6109 list_skb = &(*list_skb)->next;
6110 /* Append completion queue from offline CPU. */
6111 *list_skb = oldsd->completion_queue;
6112 oldsd->completion_queue = NULL;
6114 /* Append output queue from offline CPU. */
6115 if (oldsd->output_queue) {
6116 *sd->output_queue_tailp = oldsd->output_queue;
6117 sd->output_queue_tailp = oldsd->output_queue_tailp;
6118 oldsd->output_queue = NULL;
6119 oldsd->output_queue_tailp = &oldsd->output_queue;
6122 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6125 /* Process offline CPU's input_pkt_queue */
6126 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6128 input_queue_head_incr(oldsd);
6130 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6132 input_queue_head_incr(oldsd);
6140 * netdev_increment_features - increment feature set by one
6141 * @all: current feature set
6142 * @one: new feature set
6143 * @mask: mask feature set
6145 * Computes a new feature set after adding a device with feature set
6146 * @one to the master device with current feature set @all. Will not
6147 * enable anything that is off in @mask. Returns the new feature set.
6149 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6151 /* If device needs checksumming, downgrade to it. */
6152 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
6153 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
6154 else if (mask & NETIF_F_ALL_CSUM) {
6155 /* If one device supports v4/v6 checksumming, set for all. */
6156 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
6157 !(all & NETIF_F_GEN_CSUM)) {
6158 all &= ~NETIF_F_ALL_CSUM;
6159 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
6162 /* If one device supports hw checksumming, set for all. */
6163 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
6164 all &= ~NETIF_F_ALL_CSUM;
6165 all |= NETIF_F_HW_CSUM;
6169 one |= NETIF_F_ALL_CSUM;
6171 one |= all & NETIF_F_ONE_FOR_ALL;
6172 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
6173 all |= one & mask & NETIF_F_ONE_FOR_ALL;
6177 EXPORT_SYMBOL(netdev_increment_features);
6179 static struct hlist_head *netdev_create_hash(void)
6182 struct hlist_head *hash;
6184 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6186 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6187 INIT_HLIST_HEAD(&hash[i]);
6192 /* Initialize per network namespace state */
6193 static int __net_init netdev_init(struct net *net)
6195 INIT_LIST_HEAD(&net->dev_base_head);
6197 net->dev_name_head = netdev_create_hash();
6198 if (net->dev_name_head == NULL)
6201 net->dev_index_head = netdev_create_hash();
6202 if (net->dev_index_head == NULL)
6208 kfree(net->dev_name_head);
6214 * netdev_drivername - network driver for the device
6215 * @dev: network device
6216 * @buffer: buffer for resulting name
6217 * @len: size of buffer
6219 * Determine network driver for device.
6221 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6223 const struct device_driver *driver;
6224 const struct device *parent;
6226 if (len <= 0 || !buffer)
6230 parent = dev->dev.parent;
6235 driver = parent->driver;
6236 if (driver && driver->name)
6237 strlcpy(buffer, driver->name, len);
6241 static int __netdev_printk(const char *level, const struct net_device *dev,
6242 struct va_format *vaf)
6246 if (dev && dev->dev.parent)
6247 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6248 netdev_name(dev), vaf);
6250 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6252 r = printk("%s(NULL net_device): %pV", level, vaf);
6257 int netdev_printk(const char *level, const struct net_device *dev,
6258 const char *format, ...)
6260 struct va_format vaf;
6264 va_start(args, format);
6269 r = __netdev_printk(level, dev, &vaf);
6274 EXPORT_SYMBOL(netdev_printk);
6276 #define define_netdev_printk_level(func, level) \
6277 int func(const struct net_device *dev, const char *fmt, ...) \
6280 struct va_format vaf; \
6283 va_start(args, fmt); \
6288 r = __netdev_printk(level, dev, &vaf); \
6293 EXPORT_SYMBOL(func);
6295 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6296 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6297 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6298 define_netdev_printk_level(netdev_err, KERN_ERR);
6299 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6300 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6301 define_netdev_printk_level(netdev_info, KERN_INFO);
6303 static void __net_exit netdev_exit(struct net *net)
6305 kfree(net->dev_name_head);
6306 kfree(net->dev_index_head);
6309 static struct pernet_operations __net_initdata netdev_net_ops = {
6310 .init = netdev_init,
6311 .exit = netdev_exit,
6314 static void __net_exit default_device_exit(struct net *net)
6316 struct net_device *dev, *aux;
6318 * Push all migratable network devices back to the
6319 * initial network namespace
6322 for_each_netdev_safe(net, dev, aux) {
6324 char fb_name[IFNAMSIZ];
6326 /* Ignore unmoveable devices (i.e. loopback) */
6327 if (dev->features & NETIF_F_NETNS_LOCAL)
6330 /* Leave virtual devices for the generic cleanup */
6331 if (dev->rtnl_link_ops)
6334 /* Push remaing network devices to init_net */
6335 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6336 err = dev_change_net_namespace(dev, &init_net, fb_name);
6338 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6339 __func__, dev->name, err);
6346 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6348 /* At exit all network devices most be removed from a network
6349 * namespace. Do this in the reverse order of registration.
6350 * Do this across as many network namespaces as possible to
6351 * improve batching efficiency.
6353 struct net_device *dev;
6355 LIST_HEAD(dev_kill_list);
6358 list_for_each_entry(net, net_list, exit_list) {
6359 for_each_netdev_reverse(net, dev) {
6360 if (dev->rtnl_link_ops)
6361 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6363 unregister_netdevice_queue(dev, &dev_kill_list);
6366 unregister_netdevice_many(&dev_kill_list);
6370 static struct pernet_operations __net_initdata default_device_ops = {
6371 .exit = default_device_exit,
6372 .exit_batch = default_device_exit_batch,
6376 * Initialize the DEV module. At boot time this walks the device list and
6377 * unhooks any devices that fail to initialise (normally hardware not
6378 * present) and leaves us with a valid list of present and active devices.
6383 * This is called single threaded during boot, so no need
6384 * to take the rtnl semaphore.
6386 static int __init net_dev_init(void)
6388 int i, rc = -ENOMEM;
6390 BUG_ON(!dev_boot_phase);
6392 if (dev_proc_init())
6395 if (netdev_kobject_init())
6398 INIT_LIST_HEAD(&ptype_all);
6399 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6400 INIT_LIST_HEAD(&ptype_base[i]);
6402 if (register_pernet_subsys(&netdev_net_ops))
6406 * Initialise the packet receive queues.
6409 for_each_possible_cpu(i) {
6410 struct softnet_data *sd = &per_cpu(softnet_data, i);
6412 memset(sd, 0, sizeof(*sd));
6413 skb_queue_head_init(&sd->input_pkt_queue);
6414 skb_queue_head_init(&sd->process_queue);
6415 sd->completion_queue = NULL;
6416 INIT_LIST_HEAD(&sd->poll_list);
6417 sd->output_queue = NULL;
6418 sd->output_queue_tailp = &sd->output_queue;
6420 sd->csd.func = rps_trigger_softirq;
6426 sd->backlog.poll = process_backlog;
6427 sd->backlog.weight = weight_p;
6428 sd->backlog.gro_list = NULL;
6429 sd->backlog.gro_count = 0;
6434 /* The loopback device is special if any other network devices
6435 * is present in a network namespace the loopback device must
6436 * be present. Since we now dynamically allocate and free the
6437 * loopback device ensure this invariant is maintained by
6438 * keeping the loopback device as the first device on the
6439 * list of network devices. Ensuring the loopback devices
6440 * is the first device that appears and the last network device
6443 if (register_pernet_device(&loopback_net_ops))
6446 if (register_pernet_device(&default_device_ops))
6449 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6450 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6452 hotcpu_notifier(dev_cpu_callback, 0);
6460 subsys_initcall(net_dev_init);
6462 static int __init initialize_hashrnd(void)
6464 get_random_bytes(&hashrnd, sizeof(hashrnd));
6468 late_initcall_sync(initialize_hashrnd);