2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
135 #include <linux/cpu_rmap.h>
137 #include "net-sysfs.h"
139 /* Instead of increasing this, you should create a hash table. */
140 #define MAX_GRO_SKBS 8
142 /* This should be increased if a protocol with a bigger head is added. */
143 #define GRO_MAX_HEAD (MAX_HEADER + 128)
146 * The list of packet types we will receive (as opposed to discard)
147 * and the routines to invoke.
149 * Why 16. Because with 16 the only overlap we get on a hash of the
150 * low nibble of the protocol value is RARP/SNAP/X.25.
152 * NOTE: That is no longer true with the addition of VLAN tags. Not
153 * sure which should go first, but I bet it won't make much
154 * difference if we are running VLANs. The good news is that
155 * this protocol won't be in the list unless compiled in, so
156 * the average user (w/out VLANs) will not be adversely affected.
173 #define PTYPE_HASH_SIZE (16)
174 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
176 static DEFINE_SPINLOCK(ptype_lock);
177 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
178 static struct list_head ptype_all __read_mostly; /* Taps */
181 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
184 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
186 * Writers must hold the rtnl semaphore while they loop through the
187 * dev_base_head list, and hold dev_base_lock for writing when they do the
188 * actual updates. This allows pure readers to access the list even
189 * while a writer is preparing to update it.
191 * To put it another way, dev_base_lock is held for writing only to
192 * protect against pure readers; the rtnl semaphore provides the
193 * protection against other writers.
195 * See, for example usages, register_netdevice() and
196 * unregister_netdevice(), which must be called with the rtnl
199 DEFINE_RWLOCK(dev_base_lock);
200 EXPORT_SYMBOL(dev_base_lock);
202 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
204 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
205 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
208 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
210 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
213 static inline void rps_lock(struct softnet_data *sd)
216 spin_lock(&sd->input_pkt_queue.lock);
220 static inline void rps_unlock(struct softnet_data *sd)
223 spin_unlock(&sd->input_pkt_queue.lock);
227 /* Device list insertion */
228 static int list_netdevice(struct net_device *dev)
230 struct net *net = dev_net(dev);
234 write_lock_bh(&dev_base_lock);
235 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
236 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
237 hlist_add_head_rcu(&dev->index_hlist,
238 dev_index_hash(net, dev->ifindex));
239 write_unlock_bh(&dev_base_lock);
243 /* Device list removal
244 * caller must respect a RCU grace period before freeing/reusing dev
246 static void unlist_netdevice(struct net_device *dev)
250 /* Unlink dev from the device chain */
251 write_lock_bh(&dev_base_lock);
252 list_del_rcu(&dev->dev_list);
253 hlist_del_rcu(&dev->name_hlist);
254 hlist_del_rcu(&dev->index_hlist);
255 write_unlock_bh(&dev_base_lock);
262 static RAW_NOTIFIER_HEAD(netdev_chain);
265 * Device drivers call our routines to queue packets here. We empty the
266 * queue in the local softnet handler.
269 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
270 EXPORT_PER_CPU_SYMBOL(softnet_data);
272 #ifdef CONFIG_LOCKDEP
274 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
275 * according to dev->type
277 static const unsigned short netdev_lock_type[] =
278 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
279 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
280 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
281 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
282 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
283 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
284 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
285 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
286 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
287 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
288 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
289 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
290 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
291 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
292 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
293 ARPHRD_VOID, ARPHRD_NONE};
295 static const char *const netdev_lock_name[] =
296 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
297 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
298 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
299 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
300 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
301 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
302 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
303 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
304 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
305 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
306 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
307 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
308 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
309 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
310 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
311 "_xmit_VOID", "_xmit_NONE"};
313 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
316 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
320 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
321 if (netdev_lock_type[i] == dev_type)
323 /* the last key is used by default */
324 return ARRAY_SIZE(netdev_lock_type) - 1;
327 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
328 unsigned short dev_type)
332 i = netdev_lock_pos(dev_type);
333 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
334 netdev_lock_name[i]);
337 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
341 i = netdev_lock_pos(dev->type);
342 lockdep_set_class_and_name(&dev->addr_list_lock,
343 &netdev_addr_lock_key[i],
344 netdev_lock_name[i]);
347 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
348 unsigned short dev_type)
351 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
356 /*******************************************************************************
358 Protocol management and registration routines
360 *******************************************************************************/
363 * Add a protocol ID to the list. Now that the input handler is
364 * smarter we can dispense with all the messy stuff that used to be
367 * BEWARE!!! Protocol handlers, mangling input packets,
368 * MUST BE last in hash buckets and checking protocol handlers
369 * MUST start from promiscuous ptype_all chain in net_bh.
370 * It is true now, do not change it.
371 * Explanation follows: if protocol handler, mangling packet, will
372 * be the first on list, it is not able to sense, that packet
373 * is cloned and should be copied-on-write, so that it will
374 * change it and subsequent readers will get broken packet.
378 static inline struct list_head *ptype_head(const struct packet_type *pt)
380 if (pt->type == htons(ETH_P_ALL))
383 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
387 * dev_add_pack - add packet handler
388 * @pt: packet type declaration
390 * Add a protocol handler to the networking stack. The passed &packet_type
391 * is linked into kernel lists and may not be freed until it has been
392 * removed from the kernel lists.
394 * This call does not sleep therefore it can not
395 * guarantee all CPU's that are in middle of receiving packets
396 * will see the new packet type (until the next received packet).
399 void dev_add_pack(struct packet_type *pt)
401 struct list_head *head = ptype_head(pt);
403 spin_lock(&ptype_lock);
404 list_add_rcu(&pt->list, head);
405 spin_unlock(&ptype_lock);
407 EXPORT_SYMBOL(dev_add_pack);
410 * __dev_remove_pack - remove packet handler
411 * @pt: packet type declaration
413 * Remove a protocol handler that was previously added to the kernel
414 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
415 * from the kernel lists and can be freed or reused once this function
418 * The packet type might still be in use by receivers
419 * and must not be freed until after all the CPU's have gone
420 * through a quiescent state.
422 void __dev_remove_pack(struct packet_type *pt)
424 struct list_head *head = ptype_head(pt);
425 struct packet_type *pt1;
427 spin_lock(&ptype_lock);
429 list_for_each_entry(pt1, head, list) {
431 list_del_rcu(&pt->list);
436 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
438 spin_unlock(&ptype_lock);
440 EXPORT_SYMBOL(__dev_remove_pack);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type *pt)
456 __dev_remove_pack(pt);
460 EXPORT_SYMBOL(dev_remove_pack);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name, struct ifmap *map)
482 struct netdev_boot_setup *s;
486 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
487 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
488 memset(s[i].name, 0, sizeof(s[i].name));
489 strlcpy(s[i].name, name, IFNAMSIZ);
490 memcpy(&s[i].map, map, sizeof(s[i].map));
495 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device *dev)
509 struct netdev_boot_setup *s = dev_boot_setup;
512 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
513 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
514 !strcmp(dev->name, s[i].name)) {
515 dev->irq = s[i].map.irq;
516 dev->base_addr = s[i].map.base_addr;
517 dev->mem_start = s[i].map.mem_start;
518 dev->mem_end = s[i].map.mem_end;
524 EXPORT_SYMBOL(netdev_boot_setup_check);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix, int unit)
539 const struct netdev_boot_setup *s = dev_boot_setup;
543 sprintf(name, "%s%d", prefix, unit);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net, name))
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
553 if (!strcmp(name, s[i].name))
554 return s[i].map.base_addr;
559 * Saves at boot time configured settings for any netdevice.
561 int __init netdev_boot_setup(char *str)
566 str = get_options(str, ARRAY_SIZE(ints), ints);
571 memset(&map, 0, sizeof(map));
575 map.base_addr = ints[2];
577 map.mem_start = ints[3];
579 map.mem_end = ints[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str, &map);
585 __setup("netdev=", netdev_boot_setup);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device *__dev_get_by_name(struct net *net, const char *name)
607 struct hlist_node *p;
608 struct net_device *dev;
609 struct hlist_head *head = dev_name_hash(net, name);
611 hlist_for_each_entry(dev, p, head, name_hlist)
612 if (!strncmp(dev->name, name, IFNAMSIZ))
617 EXPORT_SYMBOL(__dev_get_by_name);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
633 struct hlist_node *p;
634 struct net_device *dev;
635 struct hlist_head *head = dev_name_hash(net, name);
637 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
638 if (!strncmp(dev->name, name, IFNAMSIZ))
643 EXPORT_SYMBOL(dev_get_by_name_rcu);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device *dev_get_by_name(struct net *net, const char *name)
659 struct net_device *dev;
662 dev = dev_get_by_name_rcu(net, name);
668 EXPORT_SYMBOL(dev_get_by_name);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
684 struct hlist_node *p;
685 struct net_device *dev;
686 struct hlist_head *head = dev_index_hash(net, ifindex);
688 hlist_for_each_entry(dev, p, head, index_hlist)
689 if (dev->ifindex == ifindex)
694 EXPORT_SYMBOL(__dev_get_by_index);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
709 struct hlist_node *p;
710 struct net_device *dev;
711 struct hlist_head *head = dev_index_hash(net, ifindex);
713 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
714 if (dev->ifindex == ifindex)
719 EXPORT_SYMBOL(dev_get_by_index_rcu);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device *dev_get_by_index(struct net *net, int ifindex)
735 struct net_device *dev;
738 dev = dev_get_by_index_rcu(net, ifindex);
744 EXPORT_SYMBOL(dev_get_by_index);
747 * dev_getbyhwaddr_rcu - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device.
754 * The caller must hold RCU or RTNL.
755 * The returned device has not had its ref count increased
756 * and the caller must therefore be careful about locking
760 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
763 struct net_device *dev;
765 for_each_netdev_rcu(net, dev)
766 if (dev->type == type &&
767 !memcmp(dev->dev_addr, ha, dev->addr_len))
772 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
774 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
776 struct net_device *dev;
779 for_each_netdev(net, dev)
780 if (dev->type == type)
785 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
787 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
789 struct net_device *dev, *ret = NULL;
792 for_each_netdev_rcu(net, dev)
793 if (dev->type == type) {
801 EXPORT_SYMBOL(dev_getfirstbyhwtype);
804 * dev_get_by_flags_rcu - find any device with given flags
805 * @net: the applicable net namespace
806 * @if_flags: IFF_* values
807 * @mask: bitmask of bits in if_flags to check
809 * Search for any interface with the given flags. Returns NULL if a device
810 * is not found or a pointer to the device. Must be called inside
811 * rcu_read_lock(), and result refcount is unchanged.
814 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
817 struct net_device *dev, *ret;
820 for_each_netdev_rcu(net, dev) {
821 if (((dev->flags ^ if_flags) & mask) == 0) {
828 EXPORT_SYMBOL(dev_get_by_flags_rcu);
831 * dev_valid_name - check if name is okay for network device
834 * Network device names need to be valid file names to
835 * to allow sysfs to work. We also disallow any kind of
838 int dev_valid_name(const char *name)
842 if (strlen(name) >= IFNAMSIZ)
844 if (!strcmp(name, ".") || !strcmp(name, ".."))
848 if (*name == '/' || isspace(*name))
854 EXPORT_SYMBOL(dev_valid_name);
857 * __dev_alloc_name - allocate a name for a device
858 * @net: network namespace to allocate the device name in
859 * @name: name format string
860 * @buf: scratch buffer and result name string
862 * Passed a format string - eg "lt%d" it will try and find a suitable
863 * id. It scans list of devices to build up a free map, then chooses
864 * the first empty slot. The caller must hold the dev_base or rtnl lock
865 * while allocating the name and adding the device in order to avoid
867 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
868 * Returns the number of the unit assigned or a negative errno code.
871 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
875 const int max_netdevices = 8*PAGE_SIZE;
876 unsigned long *inuse;
877 struct net_device *d;
879 p = strnchr(name, IFNAMSIZ-1, '%');
882 * Verify the string as this thing may have come from
883 * the user. There must be either one "%d" and no other "%"
886 if (p[1] != 'd' || strchr(p + 2, '%'))
889 /* Use one page as a bit array of possible slots */
890 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
894 for_each_netdev(net, d) {
895 if (!sscanf(d->name, name, &i))
897 if (i < 0 || i >= max_netdevices)
900 /* avoid cases where sscanf is not exact inverse of printf */
901 snprintf(buf, IFNAMSIZ, name, i);
902 if (!strncmp(buf, d->name, IFNAMSIZ))
906 i = find_first_zero_bit(inuse, max_netdevices);
907 free_page((unsigned long) inuse);
911 snprintf(buf, IFNAMSIZ, name, i);
912 if (!__dev_get_by_name(net, buf))
915 /* It is possible to run out of possible slots
916 * when the name is long and there isn't enough space left
917 * for the digits, or if all bits are used.
923 * dev_alloc_name - allocate a name for a device
925 * @name: name format string
927 * Passed a format string - eg "lt%d" it will try and find a suitable
928 * id. It scans list of devices to build up a free map, then chooses
929 * the first empty slot. The caller must hold the dev_base or rtnl lock
930 * while allocating the name and adding the device in order to avoid
932 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
933 * Returns the number of the unit assigned or a negative errno code.
936 int dev_alloc_name(struct net_device *dev, const char *name)
942 BUG_ON(!dev_net(dev));
944 ret = __dev_alloc_name(net, name, buf);
946 strlcpy(dev->name, buf, IFNAMSIZ);
949 EXPORT_SYMBOL(dev_alloc_name);
951 static int dev_get_valid_name(struct net_device *dev, const char *name)
955 BUG_ON(!dev_net(dev));
958 if (!dev_valid_name(name))
961 if (strchr(name, '%'))
962 return dev_alloc_name(dev, name);
963 else if (__dev_get_by_name(net, name))
965 else if (dev->name != name)
966 strlcpy(dev->name, name, IFNAMSIZ);
972 * dev_change_name - change name of a device
974 * @newname: name (or format string) must be at least IFNAMSIZ
976 * Change name of a device, can pass format strings "eth%d".
979 int dev_change_name(struct net_device *dev, const char *newname)
981 char oldname[IFNAMSIZ];
987 BUG_ON(!dev_net(dev));
990 if (dev->flags & IFF_UP)
993 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
996 memcpy(oldname, dev->name, IFNAMSIZ);
998 err = dev_get_valid_name(dev, newname);
1003 ret = device_rename(&dev->dev, dev->name);
1005 memcpy(dev->name, oldname, IFNAMSIZ);
1009 write_lock_bh(&dev_base_lock);
1010 hlist_del_rcu(&dev->name_hlist);
1011 write_unlock_bh(&dev_base_lock);
1015 write_lock_bh(&dev_base_lock);
1016 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1017 write_unlock_bh(&dev_base_lock);
1019 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1020 ret = notifier_to_errno(ret);
1023 /* err >= 0 after dev_alloc_name() or stores the first errno */
1026 memcpy(dev->name, oldname, IFNAMSIZ);
1030 "%s: name change rollback failed: %d.\n",
1039 * dev_set_alias - change ifalias of a device
1041 * @alias: name up to IFALIASZ
1042 * @len: limit of bytes to copy from info
1044 * Set ifalias for a device,
1046 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1050 if (len >= IFALIASZ)
1055 kfree(dev->ifalias);
1056 dev->ifalias = NULL;
1061 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1065 strlcpy(dev->ifalias, alias, len+1);
1071 * netdev_features_change - device changes features
1072 * @dev: device to cause notification
1074 * Called to indicate a device has changed features.
1076 void netdev_features_change(struct net_device *dev)
1078 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1080 EXPORT_SYMBOL(netdev_features_change);
1083 * netdev_state_change - device changes state
1084 * @dev: device to cause notification
1086 * Called to indicate a device has changed state. This function calls
1087 * the notifier chains for netdev_chain and sends a NEWLINK message
1088 * to the routing socket.
1090 void netdev_state_change(struct net_device *dev)
1092 if (dev->flags & IFF_UP) {
1093 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1094 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1097 EXPORT_SYMBOL(netdev_state_change);
1099 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1101 return call_netdevice_notifiers(event, dev);
1103 EXPORT_SYMBOL(netdev_bonding_change);
1106 * dev_load - load a network module
1107 * @net: the applicable net namespace
1108 * @name: name of interface
1110 * If a network interface is not present and the process has suitable
1111 * privileges this function loads the module. If module loading is not
1112 * available in this kernel then it becomes a nop.
1115 void dev_load(struct net *net, const char *name)
1117 struct net_device *dev;
1121 dev = dev_get_by_name_rcu(net, name);
1125 if (no_module && capable(CAP_NET_ADMIN))
1126 no_module = request_module("netdev-%s", name);
1127 if (no_module && capable(CAP_SYS_MODULE)) {
1128 if (!request_module("%s", name))
1129 pr_err("Loading kernel module for a network device "
1130 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1134 EXPORT_SYMBOL(dev_load);
1136 static int __dev_open(struct net_device *dev)
1138 const struct net_device_ops *ops = dev->netdev_ops;
1143 if (!netif_device_present(dev))
1146 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1147 ret = notifier_to_errno(ret);
1151 set_bit(__LINK_STATE_START, &dev->state);
1153 if (ops->ndo_validate_addr)
1154 ret = ops->ndo_validate_addr(dev);
1156 if (!ret && ops->ndo_open)
1157 ret = ops->ndo_open(dev);
1160 clear_bit(__LINK_STATE_START, &dev->state);
1162 dev->flags |= IFF_UP;
1163 net_dmaengine_get();
1164 dev_set_rx_mode(dev);
1166 add_device_randomness(dev->dev_addr, dev->addr_len);
1173 * dev_open - prepare an interface for use.
1174 * @dev: device to open
1176 * Takes a device from down to up state. The device's private open
1177 * function is invoked and then the multicast lists are loaded. Finally
1178 * the device is moved into the up state and a %NETDEV_UP message is
1179 * sent to the netdev notifier chain.
1181 * Calling this function on an active interface is a nop. On a failure
1182 * a negative errno code is returned.
1184 int dev_open(struct net_device *dev)
1188 if (dev->flags & IFF_UP)
1191 ret = __dev_open(dev);
1195 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1196 call_netdevice_notifiers(NETDEV_UP, dev);
1200 EXPORT_SYMBOL(dev_open);
1202 static int __dev_close_many(struct list_head *head)
1204 struct net_device *dev;
1209 list_for_each_entry(dev, head, unreg_list) {
1210 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1212 clear_bit(__LINK_STATE_START, &dev->state);
1214 /* Synchronize to scheduled poll. We cannot touch poll list, it
1215 * can be even on different cpu. So just clear netif_running().
1217 * dev->stop() will invoke napi_disable() on all of it's
1218 * napi_struct instances on this device.
1220 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1223 dev_deactivate_many(head);
1225 list_for_each_entry(dev, head, unreg_list) {
1226 const struct net_device_ops *ops = dev->netdev_ops;
1229 * Call the device specific close. This cannot fail.
1230 * Only if device is UP
1232 * We allow it to be called even after a DETACH hot-plug
1238 dev->flags &= ~IFF_UP;
1239 net_dmaengine_put();
1245 static int __dev_close(struct net_device *dev)
1250 list_add(&dev->unreg_list, &single);
1251 retval = __dev_close_many(&single);
1256 static int dev_close_many(struct list_head *head)
1258 struct net_device *dev, *tmp;
1259 LIST_HEAD(tmp_list);
1261 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1262 if (!(dev->flags & IFF_UP))
1263 list_move(&dev->unreg_list, &tmp_list);
1265 __dev_close_many(head);
1267 list_for_each_entry(dev, head, unreg_list) {
1268 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1269 call_netdevice_notifiers(NETDEV_DOWN, dev);
1272 /* rollback_registered_many needs the complete original list */
1273 list_splice(&tmp_list, head);
1278 * dev_close - shutdown an interface.
1279 * @dev: device to shutdown
1281 * This function moves an active device into down state. A
1282 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1283 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1286 int dev_close(struct net_device *dev)
1288 if (dev->flags & IFF_UP) {
1291 list_add(&dev->unreg_list, &single);
1292 dev_close_many(&single);
1297 EXPORT_SYMBOL(dev_close);
1301 * dev_disable_lro - disable Large Receive Offload on a device
1304 * Disable Large Receive Offload (LRO) on a net device. Must be
1305 * called under RTNL. This is needed if received packets may be
1306 * forwarded to another interface.
1308 void dev_disable_lro(struct net_device *dev)
1313 * If we're trying to disable lro on a vlan device
1314 * use the underlying physical device instead
1316 if (is_vlan_dev(dev))
1317 dev = vlan_dev_real_dev(dev);
1319 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1320 flags = dev->ethtool_ops->get_flags(dev);
1322 flags = ethtool_op_get_flags(dev);
1324 if (!(flags & ETH_FLAG_LRO))
1327 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1328 if (unlikely(dev->features & NETIF_F_LRO))
1329 netdev_WARN(dev, "failed to disable LRO!\n");
1331 EXPORT_SYMBOL(dev_disable_lro);
1334 static int dev_boot_phase = 1;
1337 * register_netdevice_notifier - register a network notifier block
1340 * Register a notifier to be called when network device events occur.
1341 * The notifier passed is linked into the kernel structures and must
1342 * not be reused until it has been unregistered. A negative errno code
1343 * is returned on a failure.
1345 * When registered all registration and up events are replayed
1346 * to the new notifier to allow device to have a race free
1347 * view of the network device list.
1350 int register_netdevice_notifier(struct notifier_block *nb)
1352 struct net_device *dev;
1353 struct net_device *last;
1358 err = raw_notifier_chain_register(&netdev_chain, nb);
1364 for_each_netdev(net, dev) {
1365 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1366 err = notifier_to_errno(err);
1370 if (!(dev->flags & IFF_UP))
1373 nb->notifier_call(nb, NETDEV_UP, dev);
1384 for_each_netdev(net, dev) {
1388 if (dev->flags & IFF_UP) {
1389 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1390 nb->notifier_call(nb, NETDEV_DOWN, dev);
1392 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1393 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1397 raw_notifier_chain_unregister(&netdev_chain, nb);
1400 EXPORT_SYMBOL(register_netdevice_notifier);
1403 * unregister_netdevice_notifier - unregister a network notifier block
1406 * Unregister a notifier previously registered by
1407 * register_netdevice_notifier(). The notifier is unlinked into the
1408 * kernel structures and may then be reused. A negative errno code
1409 * is returned on a failure.
1411 * After unregistering unregister and down device events are synthesized
1412 * for all devices on the device list to the removed notifier to remove
1413 * the need for special case cleanup code.
1416 int unregister_netdevice_notifier(struct notifier_block *nb)
1418 struct net_device *dev;
1423 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1428 for_each_netdev(net, dev) {
1429 if (dev->flags & IFF_UP) {
1430 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1431 nb->notifier_call(nb, NETDEV_DOWN, dev);
1433 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1434 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1441 EXPORT_SYMBOL(unregister_netdevice_notifier);
1444 * call_netdevice_notifiers - call all network notifier blocks
1445 * @val: value passed unmodified to notifier function
1446 * @dev: net_device pointer passed unmodified to notifier function
1448 * Call all network notifier blocks. Parameters and return value
1449 * are as for raw_notifier_call_chain().
1452 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1455 return raw_notifier_call_chain(&netdev_chain, val, dev);
1457 EXPORT_SYMBOL(call_netdevice_notifiers);
1459 /* When > 0 there are consumers of rx skb time stamps */
1460 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1462 void net_enable_timestamp(void)
1464 atomic_inc(&netstamp_needed);
1466 EXPORT_SYMBOL(net_enable_timestamp);
1468 void net_disable_timestamp(void)
1470 atomic_dec(&netstamp_needed);
1472 EXPORT_SYMBOL(net_disable_timestamp);
1474 static inline void net_timestamp_set(struct sk_buff *skb)
1476 if (atomic_read(&netstamp_needed))
1477 __net_timestamp(skb);
1479 skb->tstamp.tv64 = 0;
1482 static inline void net_timestamp_check(struct sk_buff *skb)
1484 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1485 __net_timestamp(skb);
1488 static inline bool is_skb_forwardable(struct net_device *dev,
1489 struct sk_buff *skb)
1493 if (!(dev->flags & IFF_UP))
1496 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1497 if (skb->len <= len)
1500 /* if TSO is enabled, we don't care about the length as the packet
1501 * could be forwarded without being segmented before
1503 if (skb_is_gso(skb))
1510 * dev_forward_skb - loopback an skb to another netif
1512 * @dev: destination network device
1513 * @skb: buffer to forward
1516 * NET_RX_SUCCESS (no congestion)
1517 * NET_RX_DROP (packet was dropped, but freed)
1519 * dev_forward_skb can be used for injecting an skb from the
1520 * start_xmit function of one device into the receive queue
1521 * of another device.
1523 * The receiving device may be in another namespace, so
1524 * we have to clear all information in the skb that could
1525 * impact namespace isolation.
1527 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1532 if (unlikely(!is_skb_forwardable(dev, skb))) {
1533 atomic_long_inc(&dev->rx_dropped);
1539 skb->tstamp.tv64 = 0;
1540 skb->pkt_type = PACKET_HOST;
1541 skb->protocol = eth_type_trans(skb, dev);
1545 return netif_rx(skb);
1547 EXPORT_SYMBOL_GPL(dev_forward_skb);
1549 static inline int deliver_skb(struct sk_buff *skb,
1550 struct packet_type *pt_prev,
1551 struct net_device *orig_dev)
1553 atomic_inc(&skb->users);
1554 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1558 * Support routine. Sends outgoing frames to any network
1559 * taps currently in use.
1562 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1564 struct packet_type *ptype;
1565 struct sk_buff *skb2 = NULL;
1566 struct packet_type *pt_prev = NULL;
1569 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1570 /* Never send packets back to the socket
1571 * they originated from - MvS (miquels@drinkel.ow.org)
1573 if ((ptype->dev == dev || !ptype->dev) &&
1574 (ptype->af_packet_priv == NULL ||
1575 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1577 deliver_skb(skb2, pt_prev, skb->dev);
1582 skb2 = skb_clone(skb, GFP_ATOMIC);
1586 net_timestamp_set(skb2);
1588 /* skb->nh should be correctly
1589 set by sender, so that the second statement is
1590 just protection against buggy protocols.
1592 skb_reset_mac_header(skb2);
1594 if (skb_network_header(skb2) < skb2->data ||
1595 skb2->network_header > skb2->tail) {
1596 if (net_ratelimit())
1597 printk(KERN_CRIT "protocol %04x is "
1599 ntohs(skb2->protocol),
1601 skb_reset_network_header(skb2);
1604 skb2->transport_header = skb2->network_header;
1605 skb2->pkt_type = PACKET_OUTGOING;
1610 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1614 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1615 * @dev: Network device
1616 * @txq: number of queues available
1618 * If real_num_tx_queues is changed the tc mappings may no longer be
1619 * valid. To resolve this verify the tc mapping remains valid and if
1620 * not NULL the mapping. With no priorities mapping to this
1621 * offset/count pair it will no longer be used. In the worst case TC0
1622 * is invalid nothing can be done so disable priority mappings. If is
1623 * expected that drivers will fix this mapping if they can before
1624 * calling netif_set_real_num_tx_queues.
1626 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1629 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1631 /* If TC0 is invalidated disable TC mapping */
1632 if (tc->offset + tc->count > txq) {
1633 pr_warning("Number of in use tx queues changed "
1634 "invalidating tc mappings. Priority "
1635 "traffic classification disabled!\n");
1640 /* Invalidated prio to tc mappings set to TC0 */
1641 for (i = 1; i < TC_BITMASK + 1; i++) {
1642 int q = netdev_get_prio_tc_map(dev, i);
1644 tc = &dev->tc_to_txq[q];
1645 if (tc->offset + tc->count > txq) {
1646 pr_warning("Number of in use tx queues "
1647 "changed. Priority %i to tc "
1648 "mapping %i is no longer valid "
1649 "setting map to 0\n",
1651 netdev_set_prio_tc_map(dev, i, 0);
1657 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1658 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1660 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1664 if (txq < 1 || txq > dev->num_tx_queues)
1667 if (dev->reg_state == NETREG_REGISTERED ||
1668 dev->reg_state == NETREG_UNREGISTERING) {
1671 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1677 netif_setup_tc(dev, txq);
1679 if (txq < dev->real_num_tx_queues)
1680 qdisc_reset_all_tx_gt(dev, txq);
1683 dev->real_num_tx_queues = txq;
1686 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1690 * netif_set_real_num_rx_queues - set actual number of RX queues used
1691 * @dev: Network device
1692 * @rxq: Actual number of RX queues
1694 * This must be called either with the rtnl_lock held or before
1695 * registration of the net device. Returns 0 on success, or a
1696 * negative error code. If called before registration, it always
1699 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1703 if (rxq < 1 || rxq > dev->num_rx_queues)
1706 if (dev->reg_state == NETREG_REGISTERED) {
1709 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1715 dev->real_num_rx_queues = rxq;
1718 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1721 static inline void __netif_reschedule(struct Qdisc *q)
1723 struct softnet_data *sd;
1724 unsigned long flags;
1726 local_irq_save(flags);
1727 sd = &__get_cpu_var(softnet_data);
1728 q->next_sched = NULL;
1729 *sd->output_queue_tailp = q;
1730 sd->output_queue_tailp = &q->next_sched;
1731 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1732 local_irq_restore(flags);
1735 void __netif_schedule(struct Qdisc *q)
1737 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1738 __netif_reschedule(q);
1740 EXPORT_SYMBOL(__netif_schedule);
1742 void dev_kfree_skb_irq(struct sk_buff *skb)
1744 if (atomic_dec_and_test(&skb->users)) {
1745 struct softnet_data *sd;
1746 unsigned long flags;
1748 local_irq_save(flags);
1749 sd = &__get_cpu_var(softnet_data);
1750 skb->next = sd->completion_queue;
1751 sd->completion_queue = skb;
1752 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1753 local_irq_restore(flags);
1756 EXPORT_SYMBOL(dev_kfree_skb_irq);
1758 void dev_kfree_skb_any(struct sk_buff *skb)
1760 if (in_irq() || irqs_disabled())
1761 dev_kfree_skb_irq(skb);
1765 EXPORT_SYMBOL(dev_kfree_skb_any);
1769 * netif_device_detach - mark device as removed
1770 * @dev: network device
1772 * Mark device as removed from system and therefore no longer available.
1774 void netif_device_detach(struct net_device *dev)
1776 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1777 netif_running(dev)) {
1778 netif_tx_stop_all_queues(dev);
1781 EXPORT_SYMBOL(netif_device_detach);
1784 * netif_device_attach - mark device as attached
1785 * @dev: network device
1787 * Mark device as attached from system and restart if needed.
1789 void netif_device_attach(struct net_device *dev)
1791 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1792 netif_running(dev)) {
1793 netif_tx_wake_all_queues(dev);
1794 __netdev_watchdog_up(dev);
1797 EXPORT_SYMBOL(netif_device_attach);
1800 * Invalidate hardware checksum when packet is to be mangled, and
1801 * complete checksum manually on outgoing path.
1803 int skb_checksum_help(struct sk_buff *skb)
1806 int ret = 0, offset;
1808 if (skb->ip_summed == CHECKSUM_COMPLETE)
1809 goto out_set_summed;
1811 if (unlikely(skb_shinfo(skb)->gso_size)) {
1812 /* Let GSO fix up the checksum. */
1813 goto out_set_summed;
1816 offset = skb_checksum_start_offset(skb);
1817 BUG_ON(offset >= skb_headlen(skb));
1818 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1820 offset += skb->csum_offset;
1821 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1823 if (skb_cloned(skb) &&
1824 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1825 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1830 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1832 skb->ip_summed = CHECKSUM_NONE;
1836 EXPORT_SYMBOL(skb_checksum_help);
1839 * skb_gso_segment - Perform segmentation on skb.
1840 * @skb: buffer to segment
1841 * @features: features for the output path (see dev->features)
1843 * This function segments the given skb and returns a list of segments.
1845 * It may return NULL if the skb requires no segmentation. This is
1846 * only possible when GSO is used for verifying header integrity.
1848 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1850 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1851 struct packet_type *ptype;
1852 __be16 type = skb->protocol;
1853 int vlan_depth = ETH_HLEN;
1856 while (type == htons(ETH_P_8021Q)) {
1857 struct vlan_hdr *vh;
1859 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1860 return ERR_PTR(-EINVAL);
1862 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1863 type = vh->h_vlan_encapsulated_proto;
1864 vlan_depth += VLAN_HLEN;
1867 skb_reset_mac_header(skb);
1868 skb->mac_len = skb->network_header - skb->mac_header;
1869 __skb_pull(skb, skb->mac_len);
1871 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1872 struct net_device *dev = skb->dev;
1873 struct ethtool_drvinfo info = {};
1875 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1876 dev->ethtool_ops->get_drvinfo(dev, &info);
1878 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1879 info.driver, dev ? dev->features : 0L,
1880 skb->sk ? skb->sk->sk_route_caps : 0L,
1881 skb->len, skb->data_len, skb->ip_summed);
1883 if (skb_header_cloned(skb) &&
1884 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1885 return ERR_PTR(err);
1889 list_for_each_entry_rcu(ptype,
1890 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1891 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1892 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1893 err = ptype->gso_send_check(skb);
1894 segs = ERR_PTR(err);
1895 if (err || skb_gso_ok(skb, features))
1897 __skb_push(skb, (skb->data -
1898 skb_network_header(skb)));
1900 segs = ptype->gso_segment(skb, features);
1906 __skb_push(skb, skb->data - skb_mac_header(skb));
1910 EXPORT_SYMBOL(skb_gso_segment);
1912 /* Take action when hardware reception checksum errors are detected. */
1914 void netdev_rx_csum_fault(struct net_device *dev)
1916 if (net_ratelimit()) {
1917 printk(KERN_ERR "%s: hw csum failure.\n",
1918 dev ? dev->name : "<unknown>");
1922 EXPORT_SYMBOL(netdev_rx_csum_fault);
1925 /* Actually, we should eliminate this check as soon as we know, that:
1926 * 1. IOMMU is present and allows to map all the memory.
1927 * 2. No high memory really exists on this machine.
1930 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1932 #ifdef CONFIG_HIGHMEM
1934 if (!(dev->features & NETIF_F_HIGHDMA)) {
1935 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1936 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1940 if (PCI_DMA_BUS_IS_PHYS) {
1941 struct device *pdev = dev->dev.parent;
1945 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1946 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1947 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1956 void (*destructor)(struct sk_buff *skb);
1959 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1961 static void dev_gso_skb_destructor(struct sk_buff *skb)
1963 struct dev_gso_cb *cb;
1966 struct sk_buff *nskb = skb->next;
1968 skb->next = nskb->next;
1971 } while (skb->next);
1973 cb = DEV_GSO_CB(skb);
1975 cb->destructor(skb);
1979 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1980 * @skb: buffer to segment
1981 * @features: device features as applicable to this skb
1983 * This function segments the given skb and stores the list of segments
1986 static int dev_gso_segment(struct sk_buff *skb, int features)
1988 struct sk_buff *segs;
1990 segs = skb_gso_segment(skb, features);
1992 /* Verifying header integrity only. */
1997 return PTR_ERR(segs);
2000 DEV_GSO_CB(skb)->destructor = skb->destructor;
2001 skb->destructor = dev_gso_skb_destructor;
2007 * Try to orphan skb early, right before transmission by the device.
2008 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2009 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2011 static inline void skb_orphan_try(struct sk_buff *skb)
2013 struct sock *sk = skb->sk;
2015 if (sk && !skb_shinfo(skb)->tx_flags) {
2016 /* skb_tx_hash() wont be able to get sk.
2017 * We copy sk_hash into skb->rxhash
2020 skb->rxhash = sk->sk_hash;
2025 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2027 return ((features & NETIF_F_GEN_CSUM) ||
2028 ((features & NETIF_F_V4_CSUM) &&
2029 protocol == htons(ETH_P_IP)) ||
2030 ((features & NETIF_F_V6_CSUM) &&
2031 protocol == htons(ETH_P_IPV6)) ||
2032 ((features & NETIF_F_FCOE_CRC) &&
2033 protocol == htons(ETH_P_FCOE)));
2036 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2038 if (!can_checksum_protocol(features, protocol)) {
2039 features &= ~NETIF_F_ALL_CSUM;
2040 features &= ~NETIF_F_SG;
2041 } else if (illegal_highdma(skb->dev, skb)) {
2042 features &= ~NETIF_F_SG;
2048 u32 netif_skb_features(struct sk_buff *skb)
2050 __be16 protocol = skb->protocol;
2051 u32 features = skb->dev->features;
2053 if (protocol == htons(ETH_P_8021Q)) {
2054 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2055 protocol = veh->h_vlan_encapsulated_proto;
2056 } else if (!vlan_tx_tag_present(skb)) {
2057 return harmonize_features(skb, protocol, features);
2060 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2062 if (protocol != htons(ETH_P_8021Q)) {
2063 return harmonize_features(skb, protocol, features);
2065 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2066 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2067 return harmonize_features(skb, protocol, features);
2070 EXPORT_SYMBOL(netif_skb_features);
2073 * Returns true if either:
2074 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2075 * 2. skb is fragmented and the device does not support SG, or if
2076 * at least one of fragments is in highmem and device does not
2077 * support DMA from it.
2079 static inline int skb_needs_linearize(struct sk_buff *skb,
2082 return skb_is_nonlinear(skb) &&
2083 ((skb_has_frag_list(skb) &&
2084 !(features & NETIF_F_FRAGLIST)) ||
2085 (skb_shinfo(skb)->nr_frags &&
2086 !(features & NETIF_F_SG)));
2089 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2090 struct netdev_queue *txq)
2092 const struct net_device_ops *ops = dev->netdev_ops;
2093 int rc = NETDEV_TX_OK;
2094 unsigned int skb_len;
2096 if (likely(!skb->next)) {
2100 * If device doesn't need skb->dst, release it right now while
2101 * its hot in this cpu cache
2103 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2106 if (!list_empty(&ptype_all))
2107 dev_queue_xmit_nit(skb, dev);
2109 skb_orphan_try(skb);
2111 features = netif_skb_features(skb);
2113 if (vlan_tx_tag_present(skb) &&
2114 !(features & NETIF_F_HW_VLAN_TX)) {
2115 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2122 if (netif_needs_gso(skb, features)) {
2123 if (unlikely(dev_gso_segment(skb, features)))
2128 if (skb_needs_linearize(skb, features) &&
2129 __skb_linearize(skb))
2132 /* If packet is not checksummed and device does not
2133 * support checksumming for this protocol, complete
2134 * checksumming here.
2136 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2137 skb_set_transport_header(skb,
2138 skb_checksum_start_offset(skb));
2139 if (!(features & NETIF_F_ALL_CSUM) &&
2140 skb_checksum_help(skb))
2146 rc = ops->ndo_start_xmit(skb, dev);
2147 trace_net_dev_xmit(skb, rc, dev, skb_len);
2148 if (rc == NETDEV_TX_OK)
2149 txq_trans_update(txq);
2155 struct sk_buff *nskb = skb->next;
2157 skb->next = nskb->next;
2161 * If device doesn't need nskb->dst, release it right now while
2162 * its hot in this cpu cache
2164 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2167 skb_len = nskb->len;
2168 rc = ops->ndo_start_xmit(nskb, dev);
2169 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2170 if (unlikely(rc != NETDEV_TX_OK)) {
2171 if (rc & ~NETDEV_TX_MASK)
2172 goto out_kfree_gso_skb;
2173 nskb->next = skb->next;
2177 txq_trans_update(txq);
2178 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2179 return NETDEV_TX_BUSY;
2180 } while (skb->next);
2183 if (likely(skb->next == NULL))
2184 skb->destructor = DEV_GSO_CB(skb)->destructor;
2191 static u32 hashrnd __read_mostly;
2194 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2195 * to be used as a distribution range.
2197 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2198 unsigned int num_tx_queues)
2202 u16 qcount = num_tx_queues;
2204 if (skb_rx_queue_recorded(skb)) {
2205 hash = skb_get_rx_queue(skb);
2206 while (unlikely(hash >= num_tx_queues))
2207 hash -= num_tx_queues;
2212 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2213 qoffset = dev->tc_to_txq[tc].offset;
2214 qcount = dev->tc_to_txq[tc].count;
2217 if (skb->sk && skb->sk->sk_hash)
2218 hash = skb->sk->sk_hash;
2220 hash = (__force u16) skb->protocol ^ skb->rxhash;
2221 hash = jhash_1word(hash, hashrnd);
2223 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2225 EXPORT_SYMBOL(__skb_tx_hash);
2227 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2229 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2230 if (net_ratelimit()) {
2231 pr_warning("%s selects TX queue %d, but "
2232 "real number of TX queues is %d\n",
2233 dev->name, queue_index, dev->real_num_tx_queues);
2240 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2243 struct xps_dev_maps *dev_maps;
2244 struct xps_map *map;
2245 int queue_index = -1;
2248 dev_maps = rcu_dereference(dev->xps_maps);
2250 map = rcu_dereference(
2251 dev_maps->cpu_map[raw_smp_processor_id()]);
2254 queue_index = map->queues[0];
2257 if (skb->sk && skb->sk->sk_hash)
2258 hash = skb->sk->sk_hash;
2260 hash = (__force u16) skb->protocol ^
2262 hash = jhash_1word(hash, hashrnd);
2263 queue_index = map->queues[
2264 ((u64)hash * map->len) >> 32];
2266 if (unlikely(queue_index >= dev->real_num_tx_queues))
2278 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2279 struct sk_buff *skb)
2282 const struct net_device_ops *ops = dev->netdev_ops;
2284 if (dev->real_num_tx_queues == 1)
2286 else if (ops->ndo_select_queue) {
2287 queue_index = ops->ndo_select_queue(dev, skb);
2288 queue_index = dev_cap_txqueue(dev, queue_index);
2290 struct sock *sk = skb->sk;
2291 queue_index = sk_tx_queue_get(sk);
2293 if (queue_index < 0 || skb->ooo_okay ||
2294 queue_index >= dev->real_num_tx_queues) {
2295 int old_index = queue_index;
2297 queue_index = get_xps_queue(dev, skb);
2298 if (queue_index < 0)
2299 queue_index = skb_tx_hash(dev, skb);
2301 if (queue_index != old_index && sk) {
2302 struct dst_entry *dst =
2303 rcu_dereference_check(sk->sk_dst_cache, 1);
2305 if (dst && skb_dst(skb) == dst)
2306 sk_tx_queue_set(sk, queue_index);
2311 skb_set_queue_mapping(skb, queue_index);
2312 return netdev_get_tx_queue(dev, queue_index);
2315 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2316 struct net_device *dev,
2317 struct netdev_queue *txq)
2319 spinlock_t *root_lock = qdisc_lock(q);
2323 qdisc_skb_cb(skb)->pkt_len = skb->len;
2324 qdisc_calculate_pkt_len(skb, q);
2326 * Heuristic to force contended enqueues to serialize on a
2327 * separate lock before trying to get qdisc main lock.
2328 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2329 * and dequeue packets faster.
2331 contended = qdisc_is_running(q);
2332 if (unlikely(contended))
2333 spin_lock(&q->busylock);
2335 spin_lock(root_lock);
2336 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2339 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2340 qdisc_run_begin(q)) {
2342 * This is a work-conserving queue; there are no old skbs
2343 * waiting to be sent out; and the qdisc is not running -
2344 * xmit the skb directly.
2346 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2349 qdisc_bstats_update(q, skb);
2351 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2352 if (unlikely(contended)) {
2353 spin_unlock(&q->busylock);
2360 rc = NET_XMIT_SUCCESS;
2363 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2364 if (qdisc_run_begin(q)) {
2365 if (unlikely(contended)) {
2366 spin_unlock(&q->busylock);
2372 spin_unlock(root_lock);
2373 if (unlikely(contended))
2374 spin_unlock(&q->busylock);
2378 static DEFINE_PER_CPU(int, xmit_recursion);
2379 #define RECURSION_LIMIT 10
2382 * dev_queue_xmit - transmit a buffer
2383 * @skb: buffer to transmit
2385 * Queue a buffer for transmission to a network device. The caller must
2386 * have set the device and priority and built the buffer before calling
2387 * this function. The function can be called from an interrupt.
2389 * A negative errno code is returned on a failure. A success does not
2390 * guarantee the frame will be transmitted as it may be dropped due
2391 * to congestion or traffic shaping.
2393 * -----------------------------------------------------------------------------------
2394 * I notice this method can also return errors from the queue disciplines,
2395 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2398 * Regardless of the return value, the skb is consumed, so it is currently
2399 * difficult to retry a send to this method. (You can bump the ref count
2400 * before sending to hold a reference for retry if you are careful.)
2402 * When calling this method, interrupts MUST be enabled. This is because
2403 * the BH enable code must have IRQs enabled so that it will not deadlock.
2406 int dev_queue_xmit(struct sk_buff *skb)
2408 struct net_device *dev = skb->dev;
2409 struct netdev_queue *txq;
2413 /* Disable soft irqs for various locks below. Also
2414 * stops preemption for RCU.
2418 txq = dev_pick_tx(dev, skb);
2419 q = rcu_dereference_bh(txq->qdisc);
2421 #ifdef CONFIG_NET_CLS_ACT
2422 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2424 trace_net_dev_queue(skb);
2426 rc = __dev_xmit_skb(skb, q, dev, txq);
2430 /* The device has no queue. Common case for software devices:
2431 loopback, all the sorts of tunnels...
2433 Really, it is unlikely that netif_tx_lock protection is necessary
2434 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2436 However, it is possible, that they rely on protection
2439 Check this and shot the lock. It is not prone from deadlocks.
2440 Either shot noqueue qdisc, it is even simpler 8)
2442 if (dev->flags & IFF_UP) {
2443 int cpu = smp_processor_id(); /* ok because BHs are off */
2445 if (txq->xmit_lock_owner != cpu) {
2447 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2448 goto recursion_alert;
2450 HARD_TX_LOCK(dev, txq, cpu);
2452 if (!netif_tx_queue_stopped(txq)) {
2453 __this_cpu_inc(xmit_recursion);
2454 rc = dev_hard_start_xmit(skb, dev, txq);
2455 __this_cpu_dec(xmit_recursion);
2456 if (dev_xmit_complete(rc)) {
2457 HARD_TX_UNLOCK(dev, txq);
2461 HARD_TX_UNLOCK(dev, txq);
2462 if (net_ratelimit())
2463 printk(KERN_CRIT "Virtual device %s asks to "
2464 "queue packet!\n", dev->name);
2466 /* Recursion is detected! It is possible,
2470 if (net_ratelimit())
2471 printk(KERN_CRIT "Dead loop on virtual device "
2472 "%s, fix it urgently!\n", dev->name);
2477 rcu_read_unlock_bh();
2482 rcu_read_unlock_bh();
2485 EXPORT_SYMBOL(dev_queue_xmit);
2488 /*=======================================================================
2490 =======================================================================*/
2492 int netdev_max_backlog __read_mostly = 1000;
2493 int netdev_tstamp_prequeue __read_mostly = 1;
2494 int netdev_budget __read_mostly = 300;
2495 int weight_p __read_mostly = 64; /* old backlog weight */
2497 /* Called with irq disabled */
2498 static inline void ____napi_schedule(struct softnet_data *sd,
2499 struct napi_struct *napi)
2501 list_add_tail(&napi->poll_list, &sd->poll_list);
2502 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2506 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2507 * and src/dst port numbers. Returns a non-zero hash number on success
2510 __u32 __skb_get_rxhash(struct sk_buff *skb)
2512 int nhoff, hash = 0, poff;
2513 const struct ipv6hdr *ip6;
2514 const struct iphdr *ip;
2516 u32 addr1, addr2, ihl;
2522 nhoff = skb_network_offset(skb);
2524 switch (skb->protocol) {
2525 case __constant_htons(ETH_P_IP):
2526 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2529 ip = (const struct iphdr *) (skb->data + nhoff);
2530 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2533 ip_proto = ip->protocol;
2534 addr1 = (__force u32) ip->saddr;
2535 addr2 = (__force u32) ip->daddr;
2538 case __constant_htons(ETH_P_IPV6):
2539 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2542 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2543 ip_proto = ip6->nexthdr;
2544 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2545 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2553 poff = proto_ports_offset(ip_proto);
2555 nhoff += ihl * 4 + poff;
2556 if (pskb_may_pull(skb, nhoff + 4)) {
2557 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2558 if (ports.v16[1] < ports.v16[0])
2559 swap(ports.v16[0], ports.v16[1]);
2563 /* get a consistent hash (same value on both flow directions) */
2567 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2574 EXPORT_SYMBOL(__skb_get_rxhash);
2578 /* One global table that all flow-based protocols share. */
2579 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2580 EXPORT_SYMBOL(rps_sock_flow_table);
2582 static struct rps_dev_flow *
2583 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2584 struct rps_dev_flow *rflow, u16 next_cpu)
2588 tcpu = rflow->cpu = next_cpu;
2589 if (tcpu != RPS_NO_CPU) {
2590 #ifdef CONFIG_RFS_ACCEL
2591 struct netdev_rx_queue *rxqueue;
2592 struct rps_dev_flow_table *flow_table;
2593 struct rps_dev_flow *old_rflow;
2598 /* Should we steer this flow to a different hardware queue? */
2599 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2600 !(dev->features & NETIF_F_NTUPLE))
2602 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2603 if (rxq_index == skb_get_rx_queue(skb))
2606 rxqueue = dev->_rx + rxq_index;
2607 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2610 flow_id = skb->rxhash & flow_table->mask;
2611 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2612 rxq_index, flow_id);
2616 rflow = &flow_table->flows[flow_id];
2617 rflow->cpu = next_cpu;
2619 if (old_rflow->filter == rflow->filter)
2620 old_rflow->filter = RPS_NO_FILTER;
2624 per_cpu(softnet_data, tcpu).input_queue_head;
2631 * get_rps_cpu is called from netif_receive_skb and returns the target
2632 * CPU from the RPS map of the receiving queue for a given skb.
2633 * rcu_read_lock must be held on entry.
2635 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2636 struct rps_dev_flow **rflowp)
2638 struct netdev_rx_queue *rxqueue;
2639 struct rps_map *map;
2640 struct rps_dev_flow_table *flow_table;
2641 struct rps_sock_flow_table *sock_flow_table;
2645 if (skb_rx_queue_recorded(skb)) {
2646 u16 index = skb_get_rx_queue(skb);
2647 if (unlikely(index >= dev->real_num_rx_queues)) {
2648 WARN_ONCE(dev->real_num_rx_queues > 1,
2649 "%s received packet on queue %u, but number "
2650 "of RX queues is %u\n",
2651 dev->name, index, dev->real_num_rx_queues);
2654 rxqueue = dev->_rx + index;
2658 map = rcu_dereference(rxqueue->rps_map);
2660 if (map->len == 1 &&
2661 !rcu_dereference_raw(rxqueue->rps_flow_table)) {
2662 tcpu = map->cpus[0];
2663 if (cpu_online(tcpu))
2667 } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2671 skb_reset_network_header(skb);
2672 if (!skb_get_rxhash(skb))
2675 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2676 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2677 if (flow_table && sock_flow_table) {
2679 struct rps_dev_flow *rflow;
2681 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2684 next_cpu = sock_flow_table->ents[skb->rxhash &
2685 sock_flow_table->mask];
2688 * If the desired CPU (where last recvmsg was done) is
2689 * different from current CPU (one in the rx-queue flow
2690 * table entry), switch if one of the following holds:
2691 * - Current CPU is unset (equal to RPS_NO_CPU).
2692 * - Current CPU is offline.
2693 * - The current CPU's queue tail has advanced beyond the
2694 * last packet that was enqueued using this table entry.
2695 * This guarantees that all previous packets for the flow
2696 * have been dequeued, thus preserving in order delivery.
2698 if (unlikely(tcpu != next_cpu) &&
2699 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2700 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2701 rflow->last_qtail)) >= 0))
2702 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2704 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2712 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2714 if (cpu_online(tcpu)) {
2724 #ifdef CONFIG_RFS_ACCEL
2727 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2728 * @dev: Device on which the filter was set
2729 * @rxq_index: RX queue index
2730 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2731 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2733 * Drivers that implement ndo_rx_flow_steer() should periodically call
2734 * this function for each installed filter and remove the filters for
2735 * which it returns %true.
2737 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2738 u32 flow_id, u16 filter_id)
2740 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2741 struct rps_dev_flow_table *flow_table;
2742 struct rps_dev_flow *rflow;
2747 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2748 if (flow_table && flow_id <= flow_table->mask) {
2749 rflow = &flow_table->flows[flow_id];
2750 cpu = ACCESS_ONCE(rflow->cpu);
2751 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2752 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2753 rflow->last_qtail) <
2754 (int)(10 * flow_table->mask)))
2760 EXPORT_SYMBOL(rps_may_expire_flow);
2762 #endif /* CONFIG_RFS_ACCEL */
2764 /* Called from hardirq (IPI) context */
2765 static void rps_trigger_softirq(void *data)
2767 struct softnet_data *sd = data;
2769 ____napi_schedule(sd, &sd->backlog);
2773 #endif /* CONFIG_RPS */
2776 * Check if this softnet_data structure is another cpu one
2777 * If yes, queue it to our IPI list and return 1
2780 static int rps_ipi_queued(struct softnet_data *sd)
2783 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2786 sd->rps_ipi_next = mysd->rps_ipi_list;
2787 mysd->rps_ipi_list = sd;
2789 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2792 #endif /* CONFIG_RPS */
2797 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2798 * queue (may be a remote CPU queue).
2800 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2801 unsigned int *qtail)
2803 struct softnet_data *sd;
2804 unsigned long flags;
2806 sd = &per_cpu(softnet_data, cpu);
2808 local_irq_save(flags);
2811 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2812 if (skb_queue_len(&sd->input_pkt_queue)) {
2814 __skb_queue_tail(&sd->input_pkt_queue, skb);
2815 input_queue_tail_incr_save(sd, qtail);
2817 local_irq_restore(flags);
2818 return NET_RX_SUCCESS;
2821 /* Schedule NAPI for backlog device
2822 * We can use non atomic operation since we own the queue lock
2824 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2825 if (!rps_ipi_queued(sd))
2826 ____napi_schedule(sd, &sd->backlog);
2834 local_irq_restore(flags);
2836 atomic_long_inc(&skb->dev->rx_dropped);
2842 * netif_rx - post buffer to the network code
2843 * @skb: buffer to post
2845 * This function receives a packet from a device driver and queues it for
2846 * the upper (protocol) levels to process. It always succeeds. The buffer
2847 * may be dropped during processing for congestion control or by the
2851 * NET_RX_SUCCESS (no congestion)
2852 * NET_RX_DROP (packet was dropped)
2856 int netif_rx(struct sk_buff *skb)
2860 /* if netpoll wants it, pretend we never saw it */
2861 if (netpoll_rx(skb))
2864 if (netdev_tstamp_prequeue)
2865 net_timestamp_check(skb);
2867 trace_netif_rx(skb);
2870 struct rps_dev_flow voidflow, *rflow = &voidflow;
2876 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2878 cpu = smp_processor_id();
2880 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2888 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2894 EXPORT_SYMBOL(netif_rx);
2896 int netif_rx_ni(struct sk_buff *skb)
2901 err = netif_rx(skb);
2902 if (local_softirq_pending())
2908 EXPORT_SYMBOL(netif_rx_ni);
2910 static void net_tx_action(struct softirq_action *h)
2912 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2914 if (sd->completion_queue) {
2915 struct sk_buff *clist;
2917 local_irq_disable();
2918 clist = sd->completion_queue;
2919 sd->completion_queue = NULL;
2923 struct sk_buff *skb = clist;
2924 clist = clist->next;
2926 WARN_ON(atomic_read(&skb->users));
2927 trace_kfree_skb(skb, net_tx_action);
2932 if (sd->output_queue) {
2935 local_irq_disable();
2936 head = sd->output_queue;
2937 sd->output_queue = NULL;
2938 sd->output_queue_tailp = &sd->output_queue;
2942 struct Qdisc *q = head;
2943 spinlock_t *root_lock;
2945 head = head->next_sched;
2947 root_lock = qdisc_lock(q);
2948 if (spin_trylock(root_lock)) {
2949 smp_mb__before_clear_bit();
2950 clear_bit(__QDISC_STATE_SCHED,
2953 spin_unlock(root_lock);
2955 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2957 __netif_reschedule(q);
2959 smp_mb__before_clear_bit();
2960 clear_bit(__QDISC_STATE_SCHED,
2968 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2969 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2970 /* This hook is defined here for ATM LANE */
2971 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2972 unsigned char *addr) __read_mostly;
2973 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2976 #ifdef CONFIG_NET_CLS_ACT
2977 /* TODO: Maybe we should just force sch_ingress to be compiled in
2978 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2979 * a compare and 2 stores extra right now if we dont have it on
2980 * but have CONFIG_NET_CLS_ACT
2981 * NOTE: This doesn't stop any functionality; if you dont have
2982 * the ingress scheduler, you just can't add policies on ingress.
2985 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2987 struct net_device *dev = skb->dev;
2988 u32 ttl = G_TC_RTTL(skb->tc_verd);
2989 int result = TC_ACT_OK;
2992 if (unlikely(MAX_RED_LOOP < ttl++)) {
2993 if (net_ratelimit())
2994 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2995 skb->skb_iif, dev->ifindex);
2999 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3000 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3003 if (q != &noop_qdisc) {
3004 spin_lock(qdisc_lock(q));
3005 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3006 result = qdisc_enqueue_root(skb, q);
3007 spin_unlock(qdisc_lock(q));
3013 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3014 struct packet_type **pt_prev,
3015 int *ret, struct net_device *orig_dev)
3017 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3019 if (!rxq || rxq->qdisc == &noop_qdisc)
3023 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3027 switch (ing_filter(skb, rxq)) {
3041 * netdev_rx_handler_register - register receive handler
3042 * @dev: device to register a handler for
3043 * @rx_handler: receive handler to register
3044 * @rx_handler_data: data pointer that is used by rx handler
3046 * Register a receive hander for a device. This handler will then be
3047 * called from __netif_receive_skb. A negative errno code is returned
3050 * The caller must hold the rtnl_mutex.
3052 * For a general description of rx_handler, see enum rx_handler_result.
3054 int netdev_rx_handler_register(struct net_device *dev,
3055 rx_handler_func_t *rx_handler,
3056 void *rx_handler_data)
3060 if (dev->rx_handler)
3063 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3064 rcu_assign_pointer(dev->rx_handler, rx_handler);
3068 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3071 * netdev_rx_handler_unregister - unregister receive handler
3072 * @dev: device to unregister a handler from
3074 * Unregister a receive hander from a device.
3076 * The caller must hold the rtnl_mutex.
3078 void netdev_rx_handler_unregister(struct net_device *dev)
3082 rcu_assign_pointer(dev->rx_handler, NULL);
3083 rcu_assign_pointer(dev->rx_handler_data, NULL);
3085 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3087 static int __netif_receive_skb(struct sk_buff *skb)
3089 struct packet_type *ptype, *pt_prev;
3090 rx_handler_func_t *rx_handler;
3091 struct net_device *orig_dev;
3092 struct net_device *null_or_dev;
3093 bool deliver_exact = false;
3094 int ret = NET_RX_DROP;
3097 if (!netdev_tstamp_prequeue)
3098 net_timestamp_check(skb);
3100 trace_netif_receive_skb(skb);
3102 /* if we've gotten here through NAPI, check netpoll */
3103 if (netpoll_receive_skb(skb))
3107 skb->skb_iif = skb->dev->ifindex;
3108 orig_dev = skb->dev;
3110 skb_reset_network_header(skb);
3111 skb_reset_transport_header(skb);
3112 skb_reset_mac_len(skb);
3120 __this_cpu_inc(softnet_data.processed);
3122 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3123 skb = vlan_untag(skb);
3128 #ifdef CONFIG_NET_CLS_ACT
3129 if (skb->tc_verd & TC_NCLS) {
3130 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3135 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3136 if (!ptype->dev || ptype->dev == skb->dev) {
3138 ret = deliver_skb(skb, pt_prev, orig_dev);
3143 #ifdef CONFIG_NET_CLS_ACT
3144 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3150 rx_handler = rcu_dereference(skb->dev->rx_handler);
3153 ret = deliver_skb(skb, pt_prev, orig_dev);
3156 switch (rx_handler(&skb)) {
3157 case RX_HANDLER_CONSUMED:
3159 case RX_HANDLER_ANOTHER:
3161 case RX_HANDLER_EXACT:
3162 deliver_exact = true;
3163 case RX_HANDLER_PASS:
3170 if (vlan_tx_tag_present(skb)) {
3172 ret = deliver_skb(skb, pt_prev, orig_dev);
3175 if (vlan_do_receive(&skb)) {
3176 ret = __netif_receive_skb(skb);
3178 } else if (unlikely(!skb))
3182 /* deliver only exact match when indicated */
3183 null_or_dev = deliver_exact ? skb->dev : NULL;
3185 type = skb->protocol;
3186 list_for_each_entry_rcu(ptype,
3187 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3188 if (ptype->type == type &&
3189 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3190 ptype->dev == orig_dev)) {
3192 ret = deliver_skb(skb, pt_prev, orig_dev);
3198 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3200 atomic_long_inc(&skb->dev->rx_dropped);
3202 /* Jamal, now you will not able to escape explaining
3203 * me how you were going to use this. :-)
3214 * netif_receive_skb - process receive buffer from network
3215 * @skb: buffer to process
3217 * netif_receive_skb() is the main receive data processing function.
3218 * It always succeeds. The buffer may be dropped during processing
3219 * for congestion control or by the protocol layers.
3221 * This function may only be called from softirq context and interrupts
3222 * should be enabled.
3224 * Return values (usually ignored):
3225 * NET_RX_SUCCESS: no congestion
3226 * NET_RX_DROP: packet was dropped
3228 int netif_receive_skb(struct sk_buff *skb)
3230 if (netdev_tstamp_prequeue)
3231 net_timestamp_check(skb);
3233 if (skb_defer_rx_timestamp(skb))
3234 return NET_RX_SUCCESS;
3238 struct rps_dev_flow voidflow, *rflow = &voidflow;
3243 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3246 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3250 ret = __netif_receive_skb(skb);
3256 return __netif_receive_skb(skb);
3259 EXPORT_SYMBOL(netif_receive_skb);
3261 /* Network device is going away, flush any packets still pending
3262 * Called with irqs disabled.
3264 static void flush_backlog(void *arg)
3266 struct net_device *dev = arg;
3267 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3268 struct sk_buff *skb, *tmp;
3271 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3272 if (skb->dev == dev) {
3273 __skb_unlink(skb, &sd->input_pkt_queue);
3275 input_queue_head_incr(sd);
3280 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3281 if (skb->dev == dev) {
3282 __skb_unlink(skb, &sd->process_queue);
3284 input_queue_head_incr(sd);
3289 static int napi_gro_complete(struct sk_buff *skb)
3291 struct packet_type *ptype;
3292 __be16 type = skb->protocol;
3293 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3296 if (NAPI_GRO_CB(skb)->count == 1) {
3297 skb_shinfo(skb)->gso_size = 0;
3302 list_for_each_entry_rcu(ptype, head, list) {
3303 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3306 err = ptype->gro_complete(skb);
3312 WARN_ON(&ptype->list == head);
3314 return NET_RX_SUCCESS;
3318 return netif_receive_skb(skb);
3321 inline void napi_gro_flush(struct napi_struct *napi)
3323 struct sk_buff *skb, *next;
3325 for (skb = napi->gro_list; skb; skb = next) {
3328 napi_gro_complete(skb);
3331 napi->gro_count = 0;
3332 napi->gro_list = NULL;
3334 EXPORT_SYMBOL(napi_gro_flush);
3336 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3338 struct sk_buff **pp = NULL;
3339 struct packet_type *ptype;
3340 __be16 type = skb->protocol;
3341 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3344 enum gro_result ret;
3346 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3349 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3353 list_for_each_entry_rcu(ptype, head, list) {
3354 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3357 skb_set_network_header(skb, skb_gro_offset(skb));
3358 mac_len = skb->network_header - skb->mac_header;
3359 skb->mac_len = mac_len;
3360 NAPI_GRO_CB(skb)->same_flow = 0;
3361 NAPI_GRO_CB(skb)->flush = 0;
3362 NAPI_GRO_CB(skb)->free = 0;
3364 pp = ptype->gro_receive(&napi->gro_list, skb);
3369 if (&ptype->list == head)
3372 same_flow = NAPI_GRO_CB(skb)->same_flow;
3373 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3376 struct sk_buff *nskb = *pp;
3380 napi_gro_complete(nskb);
3387 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3391 NAPI_GRO_CB(skb)->count = 1;
3392 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3393 skb->next = napi->gro_list;
3394 napi->gro_list = skb;
3398 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3399 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3401 BUG_ON(skb->end - skb->tail < grow);
3403 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3406 skb->data_len -= grow;
3408 skb_shinfo(skb)->frags[0].page_offset += grow;
3409 skb_shinfo(skb)->frags[0].size -= grow;
3411 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3412 put_page(skb_shinfo(skb)->frags[0].page);
3413 memmove(skb_shinfo(skb)->frags,
3414 skb_shinfo(skb)->frags + 1,
3415 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3426 EXPORT_SYMBOL(dev_gro_receive);
3428 static inline gro_result_t
3429 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3432 unsigned int maclen = skb->dev->hard_header_len;
3434 for (p = napi->gro_list; p; p = p->next) {
3435 unsigned long diffs;
3437 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3438 diffs |= p->vlan_tci ^ skb->vlan_tci;
3439 if (maclen == ETH_HLEN)
3440 diffs |= compare_ether_header(skb_mac_header(p),
3441 skb_gro_mac_header(skb));
3443 diffs = memcmp(skb_mac_header(p),
3444 skb_gro_mac_header(skb),
3446 NAPI_GRO_CB(p)->same_flow = !diffs;
3447 NAPI_GRO_CB(p)->flush = 0;
3450 return dev_gro_receive(napi, skb);
3453 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3457 if (netif_receive_skb(skb))
3462 case GRO_MERGED_FREE:
3473 EXPORT_SYMBOL(napi_skb_finish);
3475 void skb_gro_reset_offset(struct sk_buff *skb)
3477 NAPI_GRO_CB(skb)->data_offset = 0;
3478 NAPI_GRO_CB(skb)->frag0 = NULL;
3479 NAPI_GRO_CB(skb)->frag0_len = 0;
3481 if (skb->mac_header == skb->tail &&
3482 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3483 NAPI_GRO_CB(skb)->frag0 =
3484 page_address(skb_shinfo(skb)->frags[0].page) +
3485 skb_shinfo(skb)->frags[0].page_offset;
3486 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3489 EXPORT_SYMBOL(skb_gro_reset_offset);
3491 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3493 skb_gro_reset_offset(skb);
3495 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3497 EXPORT_SYMBOL(napi_gro_receive);
3499 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3501 __skb_pull(skb, skb_headlen(skb));
3502 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3503 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3505 skb->dev = napi->dev;
3511 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3513 struct sk_buff *skb = napi->skb;
3516 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3522 EXPORT_SYMBOL(napi_get_frags);
3524 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3530 skb->protocol = eth_type_trans(skb, skb->dev);
3532 if (ret == GRO_HELD)
3533 skb_gro_pull(skb, -ETH_HLEN);
3534 else if (netif_receive_skb(skb))
3539 case GRO_MERGED_FREE:
3540 napi_reuse_skb(napi, skb);
3549 EXPORT_SYMBOL(napi_frags_finish);
3551 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3553 struct sk_buff *skb = napi->skb;
3560 skb_reset_mac_header(skb);
3561 skb_gro_reset_offset(skb);
3563 off = skb_gro_offset(skb);
3564 hlen = off + sizeof(*eth);
3565 eth = skb_gro_header_fast(skb, off);
3566 if (skb_gro_header_hard(skb, hlen)) {
3567 eth = skb_gro_header_slow(skb, hlen, off);
3568 if (unlikely(!eth)) {
3569 napi_reuse_skb(napi, skb);
3575 skb_gro_pull(skb, sizeof(*eth));
3578 * This works because the only protocols we care about don't require
3579 * special handling. We'll fix it up properly at the end.
3581 skb->protocol = eth->h_proto;
3586 EXPORT_SYMBOL(napi_frags_skb);
3588 gro_result_t napi_gro_frags(struct napi_struct *napi)
3590 struct sk_buff *skb = napi_frags_skb(napi);
3595 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3597 EXPORT_SYMBOL(napi_gro_frags);
3600 * net_rps_action sends any pending IPI's for rps.
3601 * Note: called with local irq disabled, but exits with local irq enabled.
3603 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3606 struct softnet_data *remsd = sd->rps_ipi_list;
3609 sd->rps_ipi_list = NULL;
3613 /* Send pending IPI's to kick RPS processing on remote cpus. */
3615 struct softnet_data *next = remsd->rps_ipi_next;
3617 if (cpu_online(remsd->cpu))
3618 __smp_call_function_single(remsd->cpu,
3627 static int process_backlog(struct napi_struct *napi, int quota)
3630 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3633 /* Check if we have pending ipi, its better to send them now,
3634 * not waiting net_rx_action() end.
3636 if (sd->rps_ipi_list) {
3637 local_irq_disable();
3638 net_rps_action_and_irq_enable(sd);
3641 napi->weight = weight_p;
3642 local_irq_disable();
3643 while (work < quota) {
3644 struct sk_buff *skb;
3647 while ((skb = __skb_dequeue(&sd->process_queue))) {
3649 __netif_receive_skb(skb);
3650 local_irq_disable();
3651 input_queue_head_incr(sd);
3652 if (++work >= quota) {
3659 qlen = skb_queue_len(&sd->input_pkt_queue);
3661 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3662 &sd->process_queue);
3664 if (qlen < quota - work) {
3666 * Inline a custom version of __napi_complete().
3667 * only current cpu owns and manipulates this napi,
3668 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3669 * we can use a plain write instead of clear_bit(),
3670 * and we dont need an smp_mb() memory barrier.
3672 list_del(&napi->poll_list);
3675 quota = work + qlen;
3685 * __napi_schedule - schedule for receive
3686 * @n: entry to schedule
3688 * The entry's receive function will be scheduled to run
3690 void __napi_schedule(struct napi_struct *n)
3692 unsigned long flags;
3694 local_irq_save(flags);
3695 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3696 local_irq_restore(flags);
3698 EXPORT_SYMBOL(__napi_schedule);
3700 void __napi_complete(struct napi_struct *n)
3702 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3703 BUG_ON(n->gro_list);
3705 list_del(&n->poll_list);
3706 smp_mb__before_clear_bit();
3707 clear_bit(NAPI_STATE_SCHED, &n->state);
3709 EXPORT_SYMBOL(__napi_complete);
3711 void napi_complete(struct napi_struct *n)
3713 unsigned long flags;
3716 * don't let napi dequeue from the cpu poll list
3717 * just in case its running on a different cpu
3719 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3723 local_irq_save(flags);
3725 local_irq_restore(flags);
3727 EXPORT_SYMBOL(napi_complete);
3729 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3730 int (*poll)(struct napi_struct *, int), int weight)
3732 INIT_LIST_HEAD(&napi->poll_list);
3733 napi->gro_count = 0;
3734 napi->gro_list = NULL;
3737 napi->weight = weight;
3738 list_add(&napi->dev_list, &dev->napi_list);
3740 #ifdef CONFIG_NETPOLL
3741 spin_lock_init(&napi->poll_lock);
3742 napi->poll_owner = -1;
3744 set_bit(NAPI_STATE_SCHED, &napi->state);
3746 EXPORT_SYMBOL(netif_napi_add);
3748 void netif_napi_del(struct napi_struct *napi)
3750 struct sk_buff *skb, *next;
3752 list_del_init(&napi->dev_list);
3753 napi_free_frags(napi);
3755 for (skb = napi->gro_list; skb; skb = next) {
3761 napi->gro_list = NULL;
3762 napi->gro_count = 0;
3764 EXPORT_SYMBOL(netif_napi_del);
3766 static void net_rx_action(struct softirq_action *h)
3768 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3769 unsigned long time_limit = jiffies + 2;
3770 int budget = netdev_budget;
3773 local_irq_disable();
3775 while (!list_empty(&sd->poll_list)) {
3776 struct napi_struct *n;
3779 /* If softirq window is exhuasted then punt.
3780 * Allow this to run for 2 jiffies since which will allow
3781 * an average latency of 1.5/HZ.
3783 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3788 /* Even though interrupts have been re-enabled, this
3789 * access is safe because interrupts can only add new
3790 * entries to the tail of this list, and only ->poll()
3791 * calls can remove this head entry from the list.
3793 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3795 have = netpoll_poll_lock(n);
3799 /* This NAPI_STATE_SCHED test is for avoiding a race
3800 * with netpoll's poll_napi(). Only the entity which
3801 * obtains the lock and sees NAPI_STATE_SCHED set will
3802 * actually make the ->poll() call. Therefore we avoid
3803 * accidentally calling ->poll() when NAPI is not scheduled.
3806 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3807 work = n->poll(n, weight);
3811 WARN_ON_ONCE(work > weight);
3815 local_irq_disable();
3817 /* Drivers must not modify the NAPI state if they
3818 * consume the entire weight. In such cases this code
3819 * still "owns" the NAPI instance and therefore can
3820 * move the instance around on the list at-will.
3822 if (unlikely(work == weight)) {
3823 if (unlikely(napi_disable_pending(n))) {
3826 local_irq_disable();
3828 list_move_tail(&n->poll_list, &sd->poll_list);
3831 netpoll_poll_unlock(have);
3834 net_rps_action_and_irq_enable(sd);
3836 #ifdef CONFIG_NET_DMA
3838 * There may not be any more sk_buffs coming right now, so push
3839 * any pending DMA copies to hardware
3841 dma_issue_pending_all();
3848 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3852 static gifconf_func_t *gifconf_list[NPROTO];
3855 * register_gifconf - register a SIOCGIF handler
3856 * @family: Address family
3857 * @gifconf: Function handler
3859 * Register protocol dependent address dumping routines. The handler
3860 * that is passed must not be freed or reused until it has been replaced
3861 * by another handler.
3863 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3865 if (family >= NPROTO)
3867 gifconf_list[family] = gifconf;
3870 EXPORT_SYMBOL(register_gifconf);
3874 * Map an interface index to its name (SIOCGIFNAME)
3878 * We need this ioctl for efficient implementation of the
3879 * if_indextoname() function required by the IPv6 API. Without
3880 * it, we would have to search all the interfaces to find a
3884 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3886 struct net_device *dev;
3890 * Fetch the caller's info block.
3893 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3897 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3903 strcpy(ifr.ifr_name, dev->name);
3906 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3912 * Perform a SIOCGIFCONF call. This structure will change
3913 * size eventually, and there is nothing I can do about it.
3914 * Thus we will need a 'compatibility mode'.
3917 static int dev_ifconf(struct net *net, char __user *arg)
3920 struct net_device *dev;
3927 * Fetch the caller's info block.
3930 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3937 * Loop over the interfaces, and write an info block for each.
3941 for_each_netdev(net, dev) {
3942 for (i = 0; i < NPROTO; i++) {
3943 if (gifconf_list[i]) {
3946 done = gifconf_list[i](dev, NULL, 0);
3948 done = gifconf_list[i](dev, pos + total,
3958 * All done. Write the updated control block back to the caller.
3960 ifc.ifc_len = total;
3963 * Both BSD and Solaris return 0 here, so we do too.
3965 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3968 #ifdef CONFIG_PROC_FS
3970 * This is invoked by the /proc filesystem handler to display a device
3973 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3976 struct net *net = seq_file_net(seq);
3978 struct net_device *dev;
3982 return SEQ_START_TOKEN;
3985 for_each_netdev_rcu(net, dev)
3992 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3994 struct net_device *dev = v;
3996 if (v == SEQ_START_TOKEN)
3997 dev = first_net_device_rcu(seq_file_net(seq));
3999 dev = next_net_device_rcu(dev);
4005 void dev_seq_stop(struct seq_file *seq, void *v)
4011 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4013 struct rtnl_link_stats64 temp;
4014 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4016 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4017 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4018 dev->name, stats->rx_bytes, stats->rx_packets,
4020 stats->rx_dropped + stats->rx_missed_errors,
4021 stats->rx_fifo_errors,
4022 stats->rx_length_errors + stats->rx_over_errors +
4023 stats->rx_crc_errors + stats->rx_frame_errors,
4024 stats->rx_compressed, stats->multicast,
4025 stats->tx_bytes, stats->tx_packets,
4026 stats->tx_errors, stats->tx_dropped,
4027 stats->tx_fifo_errors, stats->collisions,
4028 stats->tx_carrier_errors +
4029 stats->tx_aborted_errors +
4030 stats->tx_window_errors +
4031 stats->tx_heartbeat_errors,
4032 stats->tx_compressed);
4036 * Called from the PROCfs module. This now uses the new arbitrary sized
4037 * /proc/net interface to create /proc/net/dev
4039 static int dev_seq_show(struct seq_file *seq, void *v)
4041 if (v == SEQ_START_TOKEN)
4042 seq_puts(seq, "Inter-| Receive "
4044 " face |bytes packets errs drop fifo frame "
4045 "compressed multicast|bytes packets errs "
4046 "drop fifo colls carrier compressed\n");
4048 dev_seq_printf_stats(seq, v);
4052 static struct softnet_data *softnet_get_online(loff_t *pos)
4054 struct softnet_data *sd = NULL;
4056 while (*pos < nr_cpu_ids)
4057 if (cpu_online(*pos)) {
4058 sd = &per_cpu(softnet_data, *pos);
4065 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4067 return softnet_get_online(pos);
4070 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4073 return softnet_get_online(pos);
4076 static void softnet_seq_stop(struct seq_file *seq, void *v)
4080 static int softnet_seq_show(struct seq_file *seq, void *v)
4082 struct softnet_data *sd = v;
4084 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4085 sd->processed, sd->dropped, sd->time_squeeze, 0,
4086 0, 0, 0, 0, /* was fastroute */
4087 sd->cpu_collision, sd->received_rps);
4091 static const struct seq_operations dev_seq_ops = {
4092 .start = dev_seq_start,
4093 .next = dev_seq_next,
4094 .stop = dev_seq_stop,
4095 .show = dev_seq_show,
4098 static int dev_seq_open(struct inode *inode, struct file *file)
4100 return seq_open_net(inode, file, &dev_seq_ops,
4101 sizeof(struct seq_net_private));
4104 static const struct file_operations dev_seq_fops = {
4105 .owner = THIS_MODULE,
4106 .open = dev_seq_open,
4108 .llseek = seq_lseek,
4109 .release = seq_release_net,
4112 static const struct seq_operations softnet_seq_ops = {
4113 .start = softnet_seq_start,
4114 .next = softnet_seq_next,
4115 .stop = softnet_seq_stop,
4116 .show = softnet_seq_show,
4119 static int softnet_seq_open(struct inode *inode, struct file *file)
4121 return seq_open(file, &softnet_seq_ops);
4124 static const struct file_operations softnet_seq_fops = {
4125 .owner = THIS_MODULE,
4126 .open = softnet_seq_open,
4128 .llseek = seq_lseek,
4129 .release = seq_release,
4132 static void *ptype_get_idx(loff_t pos)
4134 struct packet_type *pt = NULL;
4138 list_for_each_entry_rcu(pt, &ptype_all, list) {
4144 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4145 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4154 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4158 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4161 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4163 struct packet_type *pt;
4164 struct list_head *nxt;
4168 if (v == SEQ_START_TOKEN)
4169 return ptype_get_idx(0);
4172 nxt = pt->list.next;
4173 if (pt->type == htons(ETH_P_ALL)) {
4174 if (nxt != &ptype_all)
4177 nxt = ptype_base[0].next;
4179 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4181 while (nxt == &ptype_base[hash]) {
4182 if (++hash >= PTYPE_HASH_SIZE)
4184 nxt = ptype_base[hash].next;
4187 return list_entry(nxt, struct packet_type, list);
4190 static void ptype_seq_stop(struct seq_file *seq, void *v)
4196 static int ptype_seq_show(struct seq_file *seq, void *v)
4198 struct packet_type *pt = v;
4200 if (v == SEQ_START_TOKEN)
4201 seq_puts(seq, "Type Device Function\n");
4202 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4203 if (pt->type == htons(ETH_P_ALL))
4204 seq_puts(seq, "ALL ");
4206 seq_printf(seq, "%04x", ntohs(pt->type));
4208 seq_printf(seq, " %-8s %pF\n",
4209 pt->dev ? pt->dev->name : "", pt->func);
4215 static const struct seq_operations ptype_seq_ops = {
4216 .start = ptype_seq_start,
4217 .next = ptype_seq_next,
4218 .stop = ptype_seq_stop,
4219 .show = ptype_seq_show,
4222 static int ptype_seq_open(struct inode *inode, struct file *file)
4224 return seq_open_net(inode, file, &ptype_seq_ops,
4225 sizeof(struct seq_net_private));
4228 static const struct file_operations ptype_seq_fops = {
4229 .owner = THIS_MODULE,
4230 .open = ptype_seq_open,
4232 .llseek = seq_lseek,
4233 .release = seq_release_net,
4237 static int __net_init dev_proc_net_init(struct net *net)
4241 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4243 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4245 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4248 if (wext_proc_init(net))
4254 proc_net_remove(net, "ptype");
4256 proc_net_remove(net, "softnet_stat");
4258 proc_net_remove(net, "dev");
4262 static void __net_exit dev_proc_net_exit(struct net *net)
4264 wext_proc_exit(net);
4266 proc_net_remove(net, "ptype");
4267 proc_net_remove(net, "softnet_stat");
4268 proc_net_remove(net, "dev");
4271 static struct pernet_operations __net_initdata dev_proc_ops = {
4272 .init = dev_proc_net_init,
4273 .exit = dev_proc_net_exit,
4276 static int __init dev_proc_init(void)
4278 return register_pernet_subsys(&dev_proc_ops);
4281 #define dev_proc_init() 0
4282 #endif /* CONFIG_PROC_FS */
4286 * netdev_set_master - set up master pointer
4287 * @slave: slave device
4288 * @master: new master device
4290 * Changes the master device of the slave. Pass %NULL to break the
4291 * bonding. The caller must hold the RTNL semaphore. On a failure
4292 * a negative errno code is returned. On success the reference counts
4293 * are adjusted and the function returns zero.
4295 int netdev_set_master(struct net_device *slave, struct net_device *master)
4297 struct net_device *old = slave->master;
4307 slave->master = master;
4313 EXPORT_SYMBOL(netdev_set_master);
4316 * netdev_set_bond_master - set up bonding master/slave pair
4317 * @slave: slave device
4318 * @master: new master device
4320 * Changes the master device of the slave. Pass %NULL to break the
4321 * bonding. The caller must hold the RTNL semaphore. On a failure
4322 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4323 * to the routing socket and the function returns zero.
4325 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4331 err = netdev_set_master(slave, master);
4335 slave->flags |= IFF_SLAVE;
4337 slave->flags &= ~IFF_SLAVE;
4339 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4342 EXPORT_SYMBOL(netdev_set_bond_master);
4344 static void dev_change_rx_flags(struct net_device *dev, int flags)
4346 const struct net_device_ops *ops = dev->netdev_ops;
4348 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4349 ops->ndo_change_rx_flags(dev, flags);
4352 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4354 unsigned short old_flags = dev->flags;
4360 dev->flags |= IFF_PROMISC;
4361 dev->promiscuity += inc;
4362 if (dev->promiscuity == 0) {
4365 * If inc causes overflow, untouch promisc and return error.
4368 dev->flags &= ~IFF_PROMISC;
4370 dev->promiscuity -= inc;
4371 printk(KERN_WARNING "%s: promiscuity touches roof, "
4372 "set promiscuity failed, promiscuity feature "
4373 "of device might be broken.\n", dev->name);
4377 if (dev->flags != old_flags) {
4378 printk(KERN_INFO "device %s %s promiscuous mode\n",
4379 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4381 if (audit_enabled) {
4382 current_uid_gid(&uid, &gid);
4383 audit_log(current->audit_context, GFP_ATOMIC,
4384 AUDIT_ANOM_PROMISCUOUS,
4385 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4386 dev->name, (dev->flags & IFF_PROMISC),
4387 (old_flags & IFF_PROMISC),
4388 audit_get_loginuid(current),
4390 audit_get_sessionid(current));
4393 dev_change_rx_flags(dev, IFF_PROMISC);
4399 * dev_set_promiscuity - update promiscuity count on a device
4403 * Add or remove promiscuity from a device. While the count in the device
4404 * remains above zero the interface remains promiscuous. Once it hits zero
4405 * the device reverts back to normal filtering operation. A negative inc
4406 * value is used to drop promiscuity on the device.
4407 * Return 0 if successful or a negative errno code on error.
4409 int dev_set_promiscuity(struct net_device *dev, int inc)
4411 unsigned short old_flags = dev->flags;
4414 err = __dev_set_promiscuity(dev, inc);
4417 if (dev->flags != old_flags)
4418 dev_set_rx_mode(dev);
4421 EXPORT_SYMBOL(dev_set_promiscuity);
4424 * dev_set_allmulti - update allmulti count on a device
4428 * Add or remove reception of all multicast frames to a device. While the
4429 * count in the device remains above zero the interface remains listening
4430 * to all interfaces. Once it hits zero the device reverts back to normal
4431 * filtering operation. A negative @inc value is used to drop the counter
4432 * when releasing a resource needing all multicasts.
4433 * Return 0 if successful or a negative errno code on error.
4436 int dev_set_allmulti(struct net_device *dev, int inc)
4438 unsigned short old_flags = dev->flags;
4442 dev->flags |= IFF_ALLMULTI;
4443 dev->allmulti += inc;
4444 if (dev->allmulti == 0) {
4447 * If inc causes overflow, untouch allmulti and return error.
4450 dev->flags &= ~IFF_ALLMULTI;
4452 dev->allmulti -= inc;
4453 printk(KERN_WARNING "%s: allmulti touches roof, "
4454 "set allmulti failed, allmulti feature of "
4455 "device might be broken.\n", dev->name);
4459 if (dev->flags ^ old_flags) {
4460 dev_change_rx_flags(dev, IFF_ALLMULTI);
4461 dev_set_rx_mode(dev);
4465 EXPORT_SYMBOL(dev_set_allmulti);
4468 * Upload unicast and multicast address lists to device and
4469 * configure RX filtering. When the device doesn't support unicast
4470 * filtering it is put in promiscuous mode while unicast addresses
4473 void __dev_set_rx_mode(struct net_device *dev)
4475 const struct net_device_ops *ops = dev->netdev_ops;
4477 /* dev_open will call this function so the list will stay sane. */
4478 if (!(dev->flags&IFF_UP))
4481 if (!netif_device_present(dev))
4484 if (ops->ndo_set_rx_mode)
4485 ops->ndo_set_rx_mode(dev);
4487 /* Unicast addresses changes may only happen under the rtnl,
4488 * therefore calling __dev_set_promiscuity here is safe.
4490 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4491 __dev_set_promiscuity(dev, 1);
4492 dev->uc_promisc = 1;
4493 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4494 __dev_set_promiscuity(dev, -1);
4495 dev->uc_promisc = 0;
4498 if (ops->ndo_set_multicast_list)
4499 ops->ndo_set_multicast_list(dev);
4503 void dev_set_rx_mode(struct net_device *dev)
4505 netif_addr_lock_bh(dev);
4506 __dev_set_rx_mode(dev);
4507 netif_addr_unlock_bh(dev);
4511 * dev_ethtool_get_settings - call device's ethtool_ops::get_settings()
4513 * @cmd: memory area for ethtool_ops::get_settings() result
4515 * The cmd arg is initialized properly (cleared and
4516 * ethtool_cmd::cmd field set to ETHTOOL_GSET).
4518 * Return device's ethtool_ops::get_settings() result value or
4519 * -EOPNOTSUPP when device doesn't expose
4520 * ethtool_ops::get_settings() operation.
4522 int dev_ethtool_get_settings(struct net_device *dev,
4523 struct ethtool_cmd *cmd)
4525 if (!dev->ethtool_ops || !dev->ethtool_ops->get_settings)
4528 memset(cmd, 0, sizeof(struct ethtool_cmd));
4529 cmd->cmd = ETHTOOL_GSET;
4530 return dev->ethtool_ops->get_settings(dev, cmd);
4532 EXPORT_SYMBOL(dev_ethtool_get_settings);
4535 * dev_get_flags - get flags reported to userspace
4538 * Get the combination of flag bits exported through APIs to userspace.
4540 unsigned dev_get_flags(const struct net_device *dev)
4544 flags = (dev->flags & ~(IFF_PROMISC |
4549 (dev->gflags & (IFF_PROMISC |
4552 if (netif_running(dev)) {
4553 if (netif_oper_up(dev))
4554 flags |= IFF_RUNNING;
4555 if (netif_carrier_ok(dev))
4556 flags |= IFF_LOWER_UP;
4557 if (netif_dormant(dev))
4558 flags |= IFF_DORMANT;
4563 EXPORT_SYMBOL(dev_get_flags);
4565 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4567 int old_flags = dev->flags;
4573 * Set the flags on our device.
4576 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4577 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4579 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4583 * Load in the correct multicast list now the flags have changed.
4586 if ((old_flags ^ flags) & IFF_MULTICAST)
4587 dev_change_rx_flags(dev, IFF_MULTICAST);
4589 dev_set_rx_mode(dev);
4592 * Have we downed the interface. We handle IFF_UP ourselves
4593 * according to user attempts to set it, rather than blindly
4598 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4599 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4602 dev_set_rx_mode(dev);
4605 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4606 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4608 dev->gflags ^= IFF_PROMISC;
4609 dev_set_promiscuity(dev, inc);
4612 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4613 is important. Some (broken) drivers set IFF_PROMISC, when
4614 IFF_ALLMULTI is requested not asking us and not reporting.
4616 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4617 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4619 dev->gflags ^= IFF_ALLMULTI;
4620 dev_set_allmulti(dev, inc);
4626 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4628 unsigned int changes = dev->flags ^ old_flags;
4630 if (changes & IFF_UP) {
4631 if (dev->flags & IFF_UP)
4632 call_netdevice_notifiers(NETDEV_UP, dev);
4634 call_netdevice_notifiers(NETDEV_DOWN, dev);
4637 if (dev->flags & IFF_UP &&
4638 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4639 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4643 * dev_change_flags - change device settings
4645 * @flags: device state flags
4647 * Change settings on device based state flags. The flags are
4648 * in the userspace exported format.
4650 int dev_change_flags(struct net_device *dev, unsigned flags)
4653 int old_flags = dev->flags;
4655 ret = __dev_change_flags(dev, flags);
4659 changes = old_flags ^ dev->flags;
4661 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4663 __dev_notify_flags(dev, old_flags);
4666 EXPORT_SYMBOL(dev_change_flags);
4669 * dev_set_mtu - Change maximum transfer unit
4671 * @new_mtu: new transfer unit
4673 * Change the maximum transfer size of the network device.
4675 int dev_set_mtu(struct net_device *dev, int new_mtu)
4677 const struct net_device_ops *ops = dev->netdev_ops;
4680 if (new_mtu == dev->mtu)
4683 /* MTU must be positive. */
4687 if (!netif_device_present(dev))
4691 if (ops->ndo_change_mtu)
4692 err = ops->ndo_change_mtu(dev, new_mtu);
4696 if (!err && dev->flags & IFF_UP)
4697 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4700 EXPORT_SYMBOL(dev_set_mtu);
4703 * dev_set_group - Change group this device belongs to
4705 * @new_group: group this device should belong to
4707 void dev_set_group(struct net_device *dev, int new_group)
4709 dev->group = new_group;
4711 EXPORT_SYMBOL(dev_set_group);
4714 * dev_set_mac_address - Change Media Access Control Address
4718 * Change the hardware (MAC) address of the device
4720 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4722 const struct net_device_ops *ops = dev->netdev_ops;
4725 if (!ops->ndo_set_mac_address)
4727 if (sa->sa_family != dev->type)
4729 if (!netif_device_present(dev))
4731 err = ops->ndo_set_mac_address(dev, sa);
4733 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4734 add_device_randomness(dev->dev_addr, dev->addr_len);
4737 EXPORT_SYMBOL(dev_set_mac_address);
4740 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4742 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4745 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4751 case SIOCGIFFLAGS: /* Get interface flags */
4752 ifr->ifr_flags = (short) dev_get_flags(dev);
4755 case SIOCGIFMETRIC: /* Get the metric on the interface
4756 (currently unused) */
4757 ifr->ifr_metric = 0;
4760 case SIOCGIFMTU: /* Get the MTU of a device */
4761 ifr->ifr_mtu = dev->mtu;
4766 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4768 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4769 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4770 ifr->ifr_hwaddr.sa_family = dev->type;
4778 ifr->ifr_map.mem_start = dev->mem_start;
4779 ifr->ifr_map.mem_end = dev->mem_end;
4780 ifr->ifr_map.base_addr = dev->base_addr;
4781 ifr->ifr_map.irq = dev->irq;
4782 ifr->ifr_map.dma = dev->dma;
4783 ifr->ifr_map.port = dev->if_port;
4787 ifr->ifr_ifindex = dev->ifindex;
4791 ifr->ifr_qlen = dev->tx_queue_len;
4795 /* dev_ioctl() should ensure this case
4807 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4809 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4812 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4813 const struct net_device_ops *ops;
4818 ops = dev->netdev_ops;
4821 case SIOCSIFFLAGS: /* Set interface flags */
4822 return dev_change_flags(dev, ifr->ifr_flags);
4824 case SIOCSIFMETRIC: /* Set the metric on the interface
4825 (currently unused) */
4828 case SIOCSIFMTU: /* Set the MTU of a device */
4829 return dev_set_mtu(dev, ifr->ifr_mtu);
4832 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4834 case SIOCSIFHWBROADCAST:
4835 if (ifr->ifr_hwaddr.sa_family != dev->type)
4837 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4838 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4839 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4843 if (ops->ndo_set_config) {
4844 if (!netif_device_present(dev))
4846 return ops->ndo_set_config(dev, &ifr->ifr_map);
4851 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4852 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4854 if (!netif_device_present(dev))
4856 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4859 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4860 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4862 if (!netif_device_present(dev))
4864 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4867 if (ifr->ifr_qlen < 0)
4869 dev->tx_queue_len = ifr->ifr_qlen;
4873 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4874 return dev_change_name(dev, ifr->ifr_newname);
4877 * Unknown or private ioctl
4880 if ((cmd >= SIOCDEVPRIVATE &&
4881 cmd <= SIOCDEVPRIVATE + 15) ||
4882 cmd == SIOCBONDENSLAVE ||
4883 cmd == SIOCBONDRELEASE ||
4884 cmd == SIOCBONDSETHWADDR ||
4885 cmd == SIOCBONDSLAVEINFOQUERY ||
4886 cmd == SIOCBONDINFOQUERY ||
4887 cmd == SIOCBONDCHANGEACTIVE ||
4888 cmd == SIOCGMIIPHY ||
4889 cmd == SIOCGMIIREG ||
4890 cmd == SIOCSMIIREG ||
4891 cmd == SIOCBRADDIF ||
4892 cmd == SIOCBRDELIF ||
4893 cmd == SIOCSHWTSTAMP ||
4894 cmd == SIOCWANDEV) {
4896 if (ops->ndo_do_ioctl) {
4897 if (netif_device_present(dev))
4898 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4910 * This function handles all "interface"-type I/O control requests. The actual
4911 * 'doing' part of this is dev_ifsioc above.
4915 * dev_ioctl - network device ioctl
4916 * @net: the applicable net namespace
4917 * @cmd: command to issue
4918 * @arg: pointer to a struct ifreq in user space
4920 * Issue ioctl functions to devices. This is normally called by the
4921 * user space syscall interfaces but can sometimes be useful for
4922 * other purposes. The return value is the return from the syscall if
4923 * positive or a negative errno code on error.
4926 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4932 /* One special case: SIOCGIFCONF takes ifconf argument
4933 and requires shared lock, because it sleeps writing
4937 if (cmd == SIOCGIFCONF) {
4939 ret = dev_ifconf(net, (char __user *) arg);
4943 if (cmd == SIOCGIFNAME)
4944 return dev_ifname(net, (struct ifreq __user *)arg);
4946 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4949 ifr.ifr_name[IFNAMSIZ-1] = 0;
4951 colon = strchr(ifr.ifr_name, ':');
4956 * See which interface the caller is talking about.
4961 * These ioctl calls:
4962 * - can be done by all.
4963 * - atomic and do not require locking.
4974 dev_load(net, ifr.ifr_name);
4976 ret = dev_ifsioc_locked(net, &ifr, cmd);
4981 if (copy_to_user(arg, &ifr,
4982 sizeof(struct ifreq)))
4988 dev_load(net, ifr.ifr_name);
4990 ret = dev_ethtool(net, &ifr);
4995 if (copy_to_user(arg, &ifr,
4996 sizeof(struct ifreq)))
5002 * These ioctl calls:
5003 * - require superuser power.
5004 * - require strict serialization.
5010 if (!capable(CAP_NET_ADMIN))
5012 dev_load(net, ifr.ifr_name);
5014 ret = dev_ifsioc(net, &ifr, cmd);
5019 if (copy_to_user(arg, &ifr,
5020 sizeof(struct ifreq)))
5026 * These ioctl calls:
5027 * - require superuser power.
5028 * - require strict serialization.
5029 * - do not return a value
5039 case SIOCSIFHWBROADCAST:
5042 case SIOCBONDENSLAVE:
5043 case SIOCBONDRELEASE:
5044 case SIOCBONDSETHWADDR:
5045 case SIOCBONDCHANGEACTIVE:
5049 if (!capable(CAP_NET_ADMIN))
5052 case SIOCBONDSLAVEINFOQUERY:
5053 case SIOCBONDINFOQUERY:
5054 dev_load(net, ifr.ifr_name);
5056 ret = dev_ifsioc(net, &ifr, cmd);
5061 /* Get the per device memory space. We can add this but
5062 * currently do not support it */
5064 /* Set the per device memory buffer space.
5065 * Not applicable in our case */
5070 * Unknown or private ioctl.
5073 if (cmd == SIOCWANDEV ||
5074 (cmd >= SIOCDEVPRIVATE &&
5075 cmd <= SIOCDEVPRIVATE + 15)) {
5076 dev_load(net, ifr.ifr_name);
5078 ret = dev_ifsioc(net, &ifr, cmd);
5080 if (!ret && copy_to_user(arg, &ifr,
5081 sizeof(struct ifreq)))
5085 /* Take care of Wireless Extensions */
5086 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5087 return wext_handle_ioctl(net, &ifr, cmd, arg);
5094 * dev_new_index - allocate an ifindex
5095 * @net: the applicable net namespace
5097 * Returns a suitable unique value for a new device interface
5098 * number. The caller must hold the rtnl semaphore or the
5099 * dev_base_lock to be sure it remains unique.
5101 static int dev_new_index(struct net *net)
5107 if (!__dev_get_by_index(net, ifindex))
5112 /* Delayed registration/unregisteration */
5113 static LIST_HEAD(net_todo_list);
5115 static void net_set_todo(struct net_device *dev)
5117 list_add_tail(&dev->todo_list, &net_todo_list);
5120 static void rollback_registered_many(struct list_head *head)
5122 struct net_device *dev, *tmp;
5124 BUG_ON(dev_boot_phase);
5127 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5128 /* Some devices call without registering
5129 * for initialization unwind. Remove those
5130 * devices and proceed with the remaining.
5132 if (dev->reg_state == NETREG_UNINITIALIZED) {
5133 pr_debug("unregister_netdevice: device %s/%p never "
5134 "was registered\n", dev->name, dev);
5137 list_del(&dev->unreg_list);
5140 dev->dismantle = true;
5141 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5144 /* If device is running, close it first. */
5145 dev_close_many(head);
5147 list_for_each_entry(dev, head, unreg_list) {
5148 /* And unlink it from device chain. */
5149 unlist_netdevice(dev);
5151 dev->reg_state = NETREG_UNREGISTERING;
5156 list_for_each_entry(dev, head, unreg_list) {
5157 /* Shutdown queueing discipline. */
5161 /* Notify protocols, that we are about to destroy
5162 this device. They should clean all the things.
5164 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5166 if (!dev->rtnl_link_ops ||
5167 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5168 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5171 * Flush the unicast and multicast chains
5176 if (dev->netdev_ops->ndo_uninit)
5177 dev->netdev_ops->ndo_uninit(dev);
5179 /* Notifier chain MUST detach us from master device. */
5180 WARN_ON(dev->master);
5182 /* Remove entries from kobject tree */
5183 netdev_unregister_kobject(dev);
5186 /* Process any work delayed until the end of the batch */
5187 dev = list_first_entry(head, struct net_device, unreg_list);
5188 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5192 list_for_each_entry(dev, head, unreg_list)
5196 static void rollback_registered(struct net_device *dev)
5200 list_add(&dev->unreg_list, &single);
5201 rollback_registered_many(&single);
5205 u32 netdev_fix_features(struct net_device *dev, u32 features)
5207 /* Fix illegal checksum combinations */
5208 if ((features & NETIF_F_HW_CSUM) &&
5209 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5210 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5211 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5214 if ((features & NETIF_F_NO_CSUM) &&
5215 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5216 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5217 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5220 /* Fix illegal SG+CSUM combinations. */
5221 if ((features & NETIF_F_SG) &&
5222 !(features & NETIF_F_ALL_CSUM)) {
5224 "Dropping NETIF_F_SG since no checksum feature.\n");
5225 features &= ~NETIF_F_SG;
5228 /* TSO requires that SG is present as well. */
5229 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5230 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5231 features &= ~NETIF_F_ALL_TSO;
5234 /* TSO ECN requires that TSO is present as well. */
5235 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5236 features &= ~NETIF_F_TSO_ECN;
5238 /* Software GSO depends on SG. */
5239 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5240 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5241 features &= ~NETIF_F_GSO;
5244 /* UFO needs SG and checksumming */
5245 if (features & NETIF_F_UFO) {
5246 /* maybe split UFO into V4 and V6? */
5247 if (!((features & NETIF_F_GEN_CSUM) ||
5248 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5249 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5251 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5252 features &= ~NETIF_F_UFO;
5255 if (!(features & NETIF_F_SG)) {
5257 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5258 features &= ~NETIF_F_UFO;
5264 EXPORT_SYMBOL(netdev_fix_features);
5266 int __netdev_update_features(struct net_device *dev)
5273 features = netdev_get_wanted_features(dev);
5275 if (dev->netdev_ops->ndo_fix_features)
5276 features = dev->netdev_ops->ndo_fix_features(dev, features);
5278 /* driver might be less strict about feature dependencies */
5279 features = netdev_fix_features(dev, features);
5281 if (dev->features == features)
5284 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5285 dev->features, features);
5287 if (dev->netdev_ops->ndo_set_features)
5288 err = dev->netdev_ops->ndo_set_features(dev, features);
5290 if (unlikely(err < 0)) {
5292 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5293 err, features, dev->features);
5298 dev->features = features;
5304 * netdev_update_features - recalculate device features
5305 * @dev: the device to check
5307 * Recalculate dev->features set and send notifications if it
5308 * has changed. Should be called after driver or hardware dependent
5309 * conditions might have changed that influence the features.
5311 void netdev_update_features(struct net_device *dev)
5313 if (__netdev_update_features(dev))
5314 netdev_features_change(dev);
5316 EXPORT_SYMBOL(netdev_update_features);
5319 * netdev_change_features - recalculate device features
5320 * @dev: the device to check
5322 * Recalculate dev->features set and send notifications even
5323 * if they have not changed. Should be called instead of
5324 * netdev_update_features() if also dev->vlan_features might
5325 * have changed to allow the changes to be propagated to stacked
5328 void netdev_change_features(struct net_device *dev)
5330 __netdev_update_features(dev);
5331 netdev_features_change(dev);
5333 EXPORT_SYMBOL(netdev_change_features);
5336 * netif_stacked_transfer_operstate - transfer operstate
5337 * @rootdev: the root or lower level device to transfer state from
5338 * @dev: the device to transfer operstate to
5340 * Transfer operational state from root to device. This is normally
5341 * called when a stacking relationship exists between the root
5342 * device and the device(a leaf device).
5344 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5345 struct net_device *dev)
5347 if (rootdev->operstate == IF_OPER_DORMANT)
5348 netif_dormant_on(dev);
5350 netif_dormant_off(dev);
5352 if (netif_carrier_ok(rootdev)) {
5353 if (!netif_carrier_ok(dev))
5354 netif_carrier_on(dev);
5356 if (netif_carrier_ok(dev))
5357 netif_carrier_off(dev);
5360 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5363 static int netif_alloc_rx_queues(struct net_device *dev)
5365 unsigned int i, count = dev->num_rx_queues;
5366 struct netdev_rx_queue *rx;
5370 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5372 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5377 for (i = 0; i < count; i++)
5383 static void netdev_init_one_queue(struct net_device *dev,
5384 struct netdev_queue *queue, void *_unused)
5386 /* Initialize queue lock */
5387 spin_lock_init(&queue->_xmit_lock);
5388 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5389 queue->xmit_lock_owner = -1;
5390 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5394 static int netif_alloc_netdev_queues(struct net_device *dev)
5396 unsigned int count = dev->num_tx_queues;
5397 struct netdev_queue *tx;
5401 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5403 pr_err("netdev: Unable to allocate %u tx queues.\n",
5409 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5410 spin_lock_init(&dev->tx_global_lock);
5416 * register_netdevice - register a network device
5417 * @dev: device to register
5419 * Take a completed network device structure and add it to the kernel
5420 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5421 * chain. 0 is returned on success. A negative errno code is returned
5422 * on a failure to set up the device, or if the name is a duplicate.
5424 * Callers must hold the rtnl semaphore. You may want
5425 * register_netdev() instead of this.
5428 * The locking appears insufficient to guarantee two parallel registers
5429 * will not get the same name.
5432 int register_netdevice(struct net_device *dev)
5435 struct net *net = dev_net(dev);
5437 BUG_ON(dev_boot_phase);
5442 /* When net_device's are persistent, this will be fatal. */
5443 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5446 spin_lock_init(&dev->addr_list_lock);
5447 netdev_set_addr_lockdep_class(dev);
5451 ret = dev_get_valid_name(dev, dev->name);
5455 /* Init, if this function is available */
5456 if (dev->netdev_ops->ndo_init) {
5457 ret = dev->netdev_ops->ndo_init(dev);
5465 dev->ifindex = dev_new_index(net);
5466 if (dev->iflink == -1)
5467 dev->iflink = dev->ifindex;
5469 /* Transfer changeable features to wanted_features and enable
5470 * software offloads (GSO and GRO).
5472 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5473 dev->features |= NETIF_F_SOFT_FEATURES;
5474 dev->wanted_features = dev->features & dev->hw_features;
5476 /* Turn on no cache copy if HW is doing checksum */
5477 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5478 if ((dev->features & NETIF_F_ALL_CSUM) &&
5479 !(dev->features & NETIF_F_NO_CSUM)) {
5480 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5481 dev->features |= NETIF_F_NOCACHE_COPY;
5484 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5485 * vlan_dev_init() will do the dev->features check, so these features
5486 * are enabled only if supported by underlying device.
5488 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5490 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5491 ret = notifier_to_errno(ret);
5495 ret = netdev_register_kobject(dev);
5498 dev->reg_state = NETREG_REGISTERED;
5500 __netdev_update_features(dev);
5503 * Default initial state at registry is that the
5504 * device is present.
5507 set_bit(__LINK_STATE_PRESENT, &dev->state);
5509 dev_init_scheduler(dev);
5511 list_netdevice(dev);
5512 add_device_randomness(dev->dev_addr, dev->addr_len);
5514 /* Notify protocols, that a new device appeared. */
5515 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5516 ret = notifier_to_errno(ret);
5518 rollback_registered(dev);
5519 dev->reg_state = NETREG_UNREGISTERED;
5522 * Prevent userspace races by waiting until the network
5523 * device is fully setup before sending notifications.
5525 if (!dev->rtnl_link_ops ||
5526 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5527 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5533 if (dev->netdev_ops->ndo_uninit)
5534 dev->netdev_ops->ndo_uninit(dev);
5537 EXPORT_SYMBOL(register_netdevice);
5540 * init_dummy_netdev - init a dummy network device for NAPI
5541 * @dev: device to init
5543 * This takes a network device structure and initialize the minimum
5544 * amount of fields so it can be used to schedule NAPI polls without
5545 * registering a full blown interface. This is to be used by drivers
5546 * that need to tie several hardware interfaces to a single NAPI
5547 * poll scheduler due to HW limitations.
5549 int init_dummy_netdev(struct net_device *dev)
5551 /* Clear everything. Note we don't initialize spinlocks
5552 * are they aren't supposed to be taken by any of the
5553 * NAPI code and this dummy netdev is supposed to be
5554 * only ever used for NAPI polls
5556 memset(dev, 0, sizeof(struct net_device));
5558 /* make sure we BUG if trying to hit standard
5559 * register/unregister code path
5561 dev->reg_state = NETREG_DUMMY;
5563 /* NAPI wants this */
5564 INIT_LIST_HEAD(&dev->napi_list);
5566 /* a dummy interface is started by default */
5567 set_bit(__LINK_STATE_PRESENT, &dev->state);
5568 set_bit(__LINK_STATE_START, &dev->state);
5570 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5571 * because users of this 'device' dont need to change
5577 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5581 * register_netdev - register a network device
5582 * @dev: device to register
5584 * Take a completed network device structure and add it to the kernel
5585 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5586 * chain. 0 is returned on success. A negative errno code is returned
5587 * on a failure to set up the device, or if the name is a duplicate.
5589 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5590 * and expands the device name if you passed a format string to
5593 int register_netdev(struct net_device *dev)
5598 err = register_netdevice(dev);
5602 EXPORT_SYMBOL(register_netdev);
5604 int netdev_refcnt_read(const struct net_device *dev)
5608 for_each_possible_cpu(i)
5609 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5612 EXPORT_SYMBOL(netdev_refcnt_read);
5615 * netdev_wait_allrefs - wait until all references are gone.
5617 * This is called when unregistering network devices.
5619 * Any protocol or device that holds a reference should register
5620 * for netdevice notification, and cleanup and put back the
5621 * reference if they receive an UNREGISTER event.
5622 * We can get stuck here if buggy protocols don't correctly
5625 static void netdev_wait_allrefs(struct net_device *dev)
5627 unsigned long rebroadcast_time, warning_time;
5630 linkwatch_forget_dev(dev);
5632 rebroadcast_time = warning_time = jiffies;
5633 refcnt = netdev_refcnt_read(dev);
5635 while (refcnt != 0) {
5636 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5639 /* Rebroadcast unregister notification */
5640 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5641 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5642 * should have already handle it the first time */
5644 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5646 /* We must not have linkwatch events
5647 * pending on unregister. If this
5648 * happens, we simply run the queue
5649 * unscheduled, resulting in a noop
5652 linkwatch_run_queue();
5657 rebroadcast_time = jiffies;
5662 refcnt = netdev_refcnt_read(dev);
5664 if (time_after(jiffies, warning_time + 10 * HZ)) {
5665 printk(KERN_EMERG "unregister_netdevice: "
5666 "waiting for %s to become free. Usage "
5669 warning_time = jiffies;
5678 * register_netdevice(x1);
5679 * register_netdevice(x2);
5681 * unregister_netdevice(y1);
5682 * unregister_netdevice(y2);
5688 * We are invoked by rtnl_unlock().
5689 * This allows us to deal with problems:
5690 * 1) We can delete sysfs objects which invoke hotplug
5691 * without deadlocking with linkwatch via keventd.
5692 * 2) Since we run with the RTNL semaphore not held, we can sleep
5693 * safely in order to wait for the netdev refcnt to drop to zero.
5695 * We must not return until all unregister events added during
5696 * the interval the lock was held have been completed.
5698 void netdev_run_todo(void)
5700 struct list_head list;
5702 /* Snapshot list, allow later requests */
5703 list_replace_init(&net_todo_list, &list);
5707 while (!list_empty(&list)) {
5708 struct net_device *dev
5709 = list_first_entry(&list, struct net_device, todo_list);
5710 list_del(&dev->todo_list);
5712 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5713 printk(KERN_ERR "network todo '%s' but state %d\n",
5714 dev->name, dev->reg_state);
5719 dev->reg_state = NETREG_UNREGISTERED;
5721 on_each_cpu(flush_backlog, dev, 1);
5723 netdev_wait_allrefs(dev);
5726 BUG_ON(netdev_refcnt_read(dev));
5727 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5728 WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5729 WARN_ON(dev->dn_ptr);
5731 if (dev->destructor)
5732 dev->destructor(dev);
5734 /* Free network device */
5735 kobject_put(&dev->dev.kobj);
5739 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5740 * fields in the same order, with only the type differing.
5742 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5743 const struct net_device_stats *netdev_stats)
5745 #if BITS_PER_LONG == 64
5746 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5747 memcpy(stats64, netdev_stats, sizeof(*stats64));
5749 size_t i, n = sizeof(*stats64) / sizeof(u64);
5750 const unsigned long *src = (const unsigned long *)netdev_stats;
5751 u64 *dst = (u64 *)stats64;
5753 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5754 sizeof(*stats64) / sizeof(u64));
5755 for (i = 0; i < n; i++)
5761 * dev_get_stats - get network device statistics
5762 * @dev: device to get statistics from
5763 * @storage: place to store stats
5765 * Get network statistics from device. Return @storage.
5766 * The device driver may provide its own method by setting
5767 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5768 * otherwise the internal statistics structure is used.
5770 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5771 struct rtnl_link_stats64 *storage)
5773 const struct net_device_ops *ops = dev->netdev_ops;
5775 if (ops->ndo_get_stats64) {
5776 memset(storage, 0, sizeof(*storage));
5777 ops->ndo_get_stats64(dev, storage);
5778 } else if (ops->ndo_get_stats) {
5779 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5781 netdev_stats_to_stats64(storage, &dev->stats);
5783 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5786 EXPORT_SYMBOL(dev_get_stats);
5788 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5790 struct netdev_queue *queue = dev_ingress_queue(dev);
5792 #ifdef CONFIG_NET_CLS_ACT
5795 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5798 netdev_init_one_queue(dev, queue, NULL);
5799 queue->qdisc = &noop_qdisc;
5800 queue->qdisc_sleeping = &noop_qdisc;
5801 rcu_assign_pointer(dev->ingress_queue, queue);
5807 * alloc_netdev_mqs - allocate network device
5808 * @sizeof_priv: size of private data to allocate space for
5809 * @name: device name format string
5810 * @setup: callback to initialize device
5811 * @txqs: the number of TX subqueues to allocate
5812 * @rxqs: the number of RX subqueues to allocate
5814 * Allocates a struct net_device with private data area for driver use
5815 * and performs basic initialization. Also allocates subquue structs
5816 * for each queue on the device.
5818 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5819 void (*setup)(struct net_device *),
5820 unsigned int txqs, unsigned int rxqs)
5822 struct net_device *dev;
5824 struct net_device *p;
5826 BUG_ON(strlen(name) >= sizeof(dev->name));
5829 pr_err("alloc_netdev: Unable to allocate device "
5830 "with zero queues.\n");
5836 pr_err("alloc_netdev: Unable to allocate device "
5837 "with zero RX queues.\n");
5842 alloc_size = sizeof(struct net_device);
5844 /* ensure 32-byte alignment of private area */
5845 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5846 alloc_size += sizeof_priv;
5848 /* ensure 32-byte alignment of whole construct */
5849 alloc_size += NETDEV_ALIGN - 1;
5851 p = kzalloc(alloc_size, GFP_KERNEL);
5853 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5857 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5858 dev->padded = (char *)dev - (char *)p;
5860 dev->pcpu_refcnt = alloc_percpu(int);
5861 if (!dev->pcpu_refcnt)
5864 if (dev_addr_init(dev))
5870 dev_net_set(dev, &init_net);
5872 dev->gso_max_size = GSO_MAX_SIZE;
5874 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5875 dev->ethtool_ntuple_list.count = 0;
5876 INIT_LIST_HEAD(&dev->napi_list);
5877 INIT_LIST_HEAD(&dev->unreg_list);
5878 INIT_LIST_HEAD(&dev->link_watch_list);
5879 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5882 dev->num_tx_queues = txqs;
5883 dev->real_num_tx_queues = txqs;
5884 if (netif_alloc_netdev_queues(dev))
5888 dev->num_rx_queues = rxqs;
5889 dev->real_num_rx_queues = rxqs;
5890 if (netif_alloc_rx_queues(dev))
5894 strcpy(dev->name, name);
5895 dev->group = INIT_NETDEV_GROUP;
5903 free_percpu(dev->pcpu_refcnt);
5913 EXPORT_SYMBOL(alloc_netdev_mqs);
5916 * free_netdev - free network device
5919 * This function does the last stage of destroying an allocated device
5920 * interface. The reference to the device object is released.
5921 * If this is the last reference then it will be freed.
5923 void free_netdev(struct net_device *dev)
5925 struct napi_struct *p, *n;
5927 release_net(dev_net(dev));
5934 kfree(rcu_dereference_raw(dev->ingress_queue));
5936 /* Flush device addresses */
5937 dev_addr_flush(dev);
5939 /* Clear ethtool n-tuple list */
5940 ethtool_ntuple_flush(dev);
5942 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5945 free_percpu(dev->pcpu_refcnt);
5946 dev->pcpu_refcnt = NULL;
5948 /* Compatibility with error handling in drivers */
5949 if (dev->reg_state == NETREG_UNINITIALIZED) {
5950 kfree((char *)dev - dev->padded);
5954 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5955 dev->reg_state = NETREG_RELEASED;
5957 /* will free via device release */
5958 put_device(&dev->dev);
5960 EXPORT_SYMBOL(free_netdev);
5963 * synchronize_net - Synchronize with packet receive processing
5965 * Wait for packets currently being received to be done.
5966 * Does not block later packets from starting.
5968 void synchronize_net(void)
5971 if (rtnl_is_locked())
5972 synchronize_rcu_expedited();
5976 EXPORT_SYMBOL(synchronize_net);
5979 * unregister_netdevice_queue - remove device from the kernel
5983 * This function shuts down a device interface and removes it
5984 * from the kernel tables.
5985 * If head not NULL, device is queued to be unregistered later.
5987 * Callers must hold the rtnl semaphore. You may want
5988 * unregister_netdev() instead of this.
5991 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5996 list_move_tail(&dev->unreg_list, head);
5998 rollback_registered(dev);
5999 /* Finish processing unregister after unlock */
6003 EXPORT_SYMBOL(unregister_netdevice_queue);
6006 * unregister_netdevice_many - unregister many devices
6007 * @head: list of devices
6009 void unregister_netdevice_many(struct list_head *head)
6011 struct net_device *dev;
6013 if (!list_empty(head)) {
6014 rollback_registered_many(head);
6015 list_for_each_entry(dev, head, unreg_list)
6019 EXPORT_SYMBOL(unregister_netdevice_many);
6022 * unregister_netdev - remove device from the kernel
6025 * This function shuts down a device interface and removes it
6026 * from the kernel tables.
6028 * This is just a wrapper for unregister_netdevice that takes
6029 * the rtnl semaphore. In general you want to use this and not
6030 * unregister_netdevice.
6032 void unregister_netdev(struct net_device *dev)
6035 unregister_netdevice(dev);
6038 EXPORT_SYMBOL(unregister_netdev);
6041 * dev_change_net_namespace - move device to different nethost namespace
6043 * @net: network namespace
6044 * @pat: If not NULL name pattern to try if the current device name
6045 * is already taken in the destination network namespace.
6047 * This function shuts down a device interface and moves it
6048 * to a new network namespace. On success 0 is returned, on
6049 * a failure a netagive errno code is returned.
6051 * Callers must hold the rtnl semaphore.
6054 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6060 /* Don't allow namespace local devices to be moved. */
6062 if (dev->features & NETIF_F_NETNS_LOCAL)
6065 /* Ensure the device has been registrered */
6067 if (dev->reg_state != NETREG_REGISTERED)
6070 /* Get out if there is nothing todo */
6072 if (net_eq(dev_net(dev), net))
6075 /* Pick the destination device name, and ensure
6076 * we can use it in the destination network namespace.
6079 if (__dev_get_by_name(net, dev->name)) {
6080 /* We get here if we can't use the current device name */
6083 if (dev_get_valid_name(dev, pat) < 0)
6088 * And now a mini version of register_netdevice unregister_netdevice.
6091 /* If device is running close it first. */
6094 /* And unlink it from device chain */
6096 unlist_netdevice(dev);
6100 /* Shutdown queueing discipline. */
6103 /* Notify protocols, that we are about to destroy
6104 this device. They should clean all the things.
6106 Note that dev->reg_state stays at NETREG_REGISTERED.
6107 This is wanted because this way 8021q and macvlan know
6108 the device is just moving and can keep their slaves up.
6110 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6111 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6112 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6115 * Flush the unicast and multicast chains
6120 /* Actually switch the network namespace */
6121 dev_net_set(dev, net);
6123 /* If there is an ifindex conflict assign a new one */
6124 if (__dev_get_by_index(net, dev->ifindex)) {
6125 int iflink = (dev->iflink == dev->ifindex);
6126 dev->ifindex = dev_new_index(net);
6128 dev->iflink = dev->ifindex;
6131 /* Fixup kobjects */
6132 err = device_rename(&dev->dev, dev->name);
6135 /* Add the device back in the hashes */
6136 list_netdevice(dev);
6138 /* Notify protocols, that a new device appeared. */
6139 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6142 * Prevent userspace races by waiting until the network
6143 * device is fully setup before sending notifications.
6145 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6152 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6154 static int dev_cpu_callback(struct notifier_block *nfb,
6155 unsigned long action,
6158 struct sk_buff **list_skb;
6159 struct sk_buff *skb;
6160 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6161 struct softnet_data *sd, *oldsd;
6163 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6166 local_irq_disable();
6167 cpu = smp_processor_id();
6168 sd = &per_cpu(softnet_data, cpu);
6169 oldsd = &per_cpu(softnet_data, oldcpu);
6171 /* Find end of our completion_queue. */
6172 list_skb = &sd->completion_queue;
6174 list_skb = &(*list_skb)->next;
6175 /* Append completion queue from offline CPU. */
6176 *list_skb = oldsd->completion_queue;
6177 oldsd->completion_queue = NULL;
6179 /* Append output queue from offline CPU. */
6180 if (oldsd->output_queue) {
6181 *sd->output_queue_tailp = oldsd->output_queue;
6182 sd->output_queue_tailp = oldsd->output_queue_tailp;
6183 oldsd->output_queue = NULL;
6184 oldsd->output_queue_tailp = &oldsd->output_queue;
6186 /* Append NAPI poll list from offline CPU. */
6187 if (!list_empty(&oldsd->poll_list)) {
6188 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6189 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6192 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6195 /* Process offline CPU's input_pkt_queue */
6196 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6198 input_queue_head_incr(oldsd);
6200 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6202 input_queue_head_incr(oldsd);
6210 * netdev_increment_features - increment feature set by one
6211 * @all: current feature set
6212 * @one: new feature set
6213 * @mask: mask feature set
6215 * Computes a new feature set after adding a device with feature set
6216 * @one to the master device with current feature set @all. Will not
6217 * enable anything that is off in @mask. Returns the new feature set.
6219 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6221 if (mask & NETIF_F_GEN_CSUM)
6222 mask |= NETIF_F_ALL_CSUM;
6223 mask |= NETIF_F_VLAN_CHALLENGED;
6225 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6226 all &= one | ~NETIF_F_ALL_FOR_ALL;
6228 /* If device needs checksumming, downgrade to it. */
6229 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6230 all &= ~NETIF_F_NO_CSUM;
6232 /* If one device supports hw checksumming, set for all. */
6233 if (all & NETIF_F_GEN_CSUM)
6234 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6238 EXPORT_SYMBOL(netdev_increment_features);
6240 static struct hlist_head *netdev_create_hash(void)
6243 struct hlist_head *hash;
6245 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6247 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6248 INIT_HLIST_HEAD(&hash[i]);
6253 /* Initialize per network namespace state */
6254 static int __net_init netdev_init(struct net *net)
6256 INIT_LIST_HEAD(&net->dev_base_head);
6258 net->dev_name_head = netdev_create_hash();
6259 if (net->dev_name_head == NULL)
6262 net->dev_index_head = netdev_create_hash();
6263 if (net->dev_index_head == NULL)
6269 kfree(net->dev_name_head);
6275 * netdev_drivername - network driver for the device
6276 * @dev: network device
6278 * Determine network driver for device.
6280 const char *netdev_drivername(const struct net_device *dev)
6282 const struct device_driver *driver;
6283 const struct device *parent;
6284 const char *empty = "";
6286 parent = dev->dev.parent;
6290 driver = parent->driver;
6291 if (driver && driver->name)
6292 return driver->name;
6296 static int __netdev_printk(const char *level, const struct net_device *dev,
6297 struct va_format *vaf)
6301 if (dev && dev->dev.parent)
6302 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6303 netdev_name(dev), vaf);
6305 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6307 r = printk("%s(NULL net_device): %pV", level, vaf);
6312 int netdev_printk(const char *level, const struct net_device *dev,
6313 const char *format, ...)
6315 struct va_format vaf;
6319 va_start(args, format);
6324 r = __netdev_printk(level, dev, &vaf);
6329 EXPORT_SYMBOL(netdev_printk);
6331 #define define_netdev_printk_level(func, level) \
6332 int func(const struct net_device *dev, const char *fmt, ...) \
6335 struct va_format vaf; \
6338 va_start(args, fmt); \
6343 r = __netdev_printk(level, dev, &vaf); \
6348 EXPORT_SYMBOL(func);
6350 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6351 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6352 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6353 define_netdev_printk_level(netdev_err, KERN_ERR);
6354 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6355 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6356 define_netdev_printk_level(netdev_info, KERN_INFO);
6358 static void __net_exit netdev_exit(struct net *net)
6360 kfree(net->dev_name_head);
6361 kfree(net->dev_index_head);
6364 static struct pernet_operations __net_initdata netdev_net_ops = {
6365 .init = netdev_init,
6366 .exit = netdev_exit,
6369 static void __net_exit default_device_exit(struct net *net)
6371 struct net_device *dev, *aux;
6373 * Push all migratable network devices back to the
6374 * initial network namespace
6377 for_each_netdev_safe(net, dev, aux) {
6379 char fb_name[IFNAMSIZ];
6381 /* Ignore unmoveable devices (i.e. loopback) */
6382 if (dev->features & NETIF_F_NETNS_LOCAL)
6385 /* Leave virtual devices for the generic cleanup */
6386 if (dev->rtnl_link_ops)
6389 /* Push remaining network devices to init_net */
6390 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6391 err = dev_change_net_namespace(dev, &init_net, fb_name);
6393 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6394 __func__, dev->name, err);
6401 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6403 /* At exit all network devices most be removed from a network
6404 * namespace. Do this in the reverse order of registration.
6405 * Do this across as many network namespaces as possible to
6406 * improve batching efficiency.
6408 struct net_device *dev;
6410 LIST_HEAD(dev_kill_list);
6413 list_for_each_entry(net, net_list, exit_list) {
6414 for_each_netdev_reverse(net, dev) {
6415 if (dev->rtnl_link_ops)
6416 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6418 unregister_netdevice_queue(dev, &dev_kill_list);
6421 unregister_netdevice_many(&dev_kill_list);
6422 list_del(&dev_kill_list);
6426 static struct pernet_operations __net_initdata default_device_ops = {
6427 .exit = default_device_exit,
6428 .exit_batch = default_device_exit_batch,
6432 * Initialize the DEV module. At boot time this walks the device list and
6433 * unhooks any devices that fail to initialise (normally hardware not
6434 * present) and leaves us with a valid list of present and active devices.
6439 * This is called single threaded during boot, so no need
6440 * to take the rtnl semaphore.
6442 static int __init net_dev_init(void)
6444 int i, rc = -ENOMEM;
6446 BUG_ON(!dev_boot_phase);
6448 if (dev_proc_init())
6451 if (netdev_kobject_init())
6454 INIT_LIST_HEAD(&ptype_all);
6455 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6456 INIT_LIST_HEAD(&ptype_base[i]);
6458 if (register_pernet_subsys(&netdev_net_ops))
6462 * Initialise the packet receive queues.
6465 for_each_possible_cpu(i) {
6466 struct softnet_data *sd = &per_cpu(softnet_data, i);
6468 memset(sd, 0, sizeof(*sd));
6469 skb_queue_head_init(&sd->input_pkt_queue);
6470 skb_queue_head_init(&sd->process_queue);
6471 sd->completion_queue = NULL;
6472 INIT_LIST_HEAD(&sd->poll_list);
6473 sd->output_queue = NULL;
6474 sd->output_queue_tailp = &sd->output_queue;
6476 sd->csd.func = rps_trigger_softirq;
6482 sd->backlog.poll = process_backlog;
6483 sd->backlog.weight = weight_p;
6484 sd->backlog.gro_list = NULL;
6485 sd->backlog.gro_count = 0;
6490 /* The loopback device is special if any other network devices
6491 * is present in a network namespace the loopback device must
6492 * be present. Since we now dynamically allocate and free the
6493 * loopback device ensure this invariant is maintained by
6494 * keeping the loopback device as the first device on the
6495 * list of network devices. Ensuring the loopback devices
6496 * is the first device that appears and the last network device
6499 if (register_pernet_device(&loopback_net_ops))
6502 if (register_pernet_device(&default_device_ops))
6505 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6506 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6508 hotcpu_notifier(dev_cpu_callback, 0);
6516 subsys_initcall(net_dev_init);
6518 static int __init initialize_hashrnd(void)
6520 get_random_bytes(&hashrnd, sizeof(hashrnd));
6524 late_initcall_sync(initialize_hashrnd);