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>
136 #include <linux/if_tunnel.h>
137 #include <linux/if_pppox.h>
138 #include <linux/ppp_defs.h>
140 #include "net-sysfs.h"
142 /* Instead of increasing this, you should create a hash table. */
143 #define MAX_GRO_SKBS 8
145 /* This should be increased if a protocol with a bigger head is added. */
146 #define GRO_MAX_HEAD (MAX_HEADER + 128)
149 * The list of packet types we will receive (as opposed to discard)
150 * and the routines to invoke.
152 * Why 16. Because with 16 the only overlap we get on a hash of the
153 * low nibble of the protocol value is RARP/SNAP/X.25.
155 * NOTE: That is no longer true with the addition of VLAN tags. Not
156 * sure which should go first, but I bet it won't make much
157 * difference if we are running VLANs. The good news is that
158 * this protocol won't be in the list unless compiled in, so
159 * the average user (w/out VLANs) will not be adversely affected.
176 #define PTYPE_HASH_SIZE (16)
177 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
179 static DEFINE_SPINLOCK(ptype_lock);
180 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
181 static struct list_head ptype_all __read_mostly; /* Taps */
184 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
187 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
189 * Writers must hold the rtnl semaphore while they loop through the
190 * dev_base_head list, and hold dev_base_lock for writing when they do the
191 * actual updates. This allows pure readers to access the list even
192 * while a writer is preparing to update it.
194 * To put it another way, dev_base_lock is held for writing only to
195 * protect against pure readers; the rtnl semaphore provides the
196 * protection against other writers.
198 * See, for example usages, register_netdevice() and
199 * unregister_netdevice(), which must be called with the rtnl
202 DEFINE_RWLOCK(dev_base_lock);
203 EXPORT_SYMBOL(dev_base_lock);
205 static inline void dev_base_seq_inc(struct net *net)
207 while (++net->dev_base_seq == 0);
210 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
212 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
213 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
216 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
218 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
221 static inline void rps_lock(struct softnet_data *sd)
224 spin_lock(&sd->input_pkt_queue.lock);
228 static inline void rps_unlock(struct softnet_data *sd)
231 spin_unlock(&sd->input_pkt_queue.lock);
235 /* Device list insertion */
236 static int list_netdevice(struct net_device *dev)
238 struct net *net = dev_net(dev);
242 write_lock_bh(&dev_base_lock);
243 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
244 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
245 hlist_add_head_rcu(&dev->index_hlist,
246 dev_index_hash(net, dev->ifindex));
247 write_unlock_bh(&dev_base_lock);
249 dev_base_seq_inc(net);
254 /* Device list removal
255 * caller must respect a RCU grace period before freeing/reusing dev
257 static void unlist_netdevice(struct net_device *dev)
261 /* Unlink dev from the device chain */
262 write_lock_bh(&dev_base_lock);
263 list_del_rcu(&dev->dev_list);
264 hlist_del_rcu(&dev->name_hlist);
265 hlist_del_rcu(&dev->index_hlist);
266 write_unlock_bh(&dev_base_lock);
268 dev_base_seq_inc(dev_net(dev));
275 static RAW_NOTIFIER_HEAD(netdev_chain);
278 * Device drivers call our routines to queue packets here. We empty the
279 * queue in the local softnet handler.
282 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
283 EXPORT_PER_CPU_SYMBOL(softnet_data);
285 #ifdef CONFIG_LOCKDEP
287 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
288 * according to dev->type
290 static const unsigned short netdev_lock_type[] =
291 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
292 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
293 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
294 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
295 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
296 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
297 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
298 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
299 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
300 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
301 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
302 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
303 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
304 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
305 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
306 ARPHRD_VOID, ARPHRD_NONE};
308 static const char *const netdev_lock_name[] =
309 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
310 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
311 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
312 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
313 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
314 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
315 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
316 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
317 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
318 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
319 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
320 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
321 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
322 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
323 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
324 "_xmit_VOID", "_xmit_NONE"};
326 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
327 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
329 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
333 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
334 if (netdev_lock_type[i] == dev_type)
336 /* the last key is used by default */
337 return ARRAY_SIZE(netdev_lock_type) - 1;
340 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
341 unsigned short dev_type)
345 i = netdev_lock_pos(dev_type);
346 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
347 netdev_lock_name[i]);
350 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 i = netdev_lock_pos(dev->type);
355 lockdep_set_class_and_name(&dev->addr_list_lock,
356 &netdev_addr_lock_key[i],
357 netdev_lock_name[i]);
360 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
361 unsigned short dev_type)
364 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
369 /*******************************************************************************
371 Protocol management and registration routines
373 *******************************************************************************/
376 * Add a protocol ID to the list. Now that the input handler is
377 * smarter we can dispense with all the messy stuff that used to be
380 * BEWARE!!! Protocol handlers, mangling input packets,
381 * MUST BE last in hash buckets and checking protocol handlers
382 * MUST start from promiscuous ptype_all chain in net_bh.
383 * It is true now, do not change it.
384 * Explanation follows: if protocol handler, mangling packet, will
385 * be the first on list, it is not able to sense, that packet
386 * is cloned and should be copied-on-write, so that it will
387 * change it and subsequent readers will get broken packet.
391 static inline struct list_head *ptype_head(const struct packet_type *pt)
393 if (pt->type == htons(ETH_P_ALL))
396 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
400 * dev_add_pack - add packet handler
401 * @pt: packet type declaration
403 * Add a protocol handler to the networking stack. The passed &packet_type
404 * is linked into kernel lists and may not be freed until it has been
405 * removed from the kernel lists.
407 * This call does not sleep therefore it can not
408 * guarantee all CPU's that are in middle of receiving packets
409 * will see the new packet type (until the next received packet).
412 void dev_add_pack(struct packet_type *pt)
414 struct list_head *head = ptype_head(pt);
416 spin_lock(&ptype_lock);
417 list_add_rcu(&pt->list, head);
418 spin_unlock(&ptype_lock);
420 EXPORT_SYMBOL(dev_add_pack);
423 * __dev_remove_pack - remove packet handler
424 * @pt: packet type declaration
426 * Remove a protocol handler that was previously added to the kernel
427 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
428 * from the kernel lists and can be freed or reused once this function
431 * The packet type might still be in use by receivers
432 * and must not be freed until after all the CPU's have gone
433 * through a quiescent state.
435 void __dev_remove_pack(struct packet_type *pt)
437 struct list_head *head = ptype_head(pt);
438 struct packet_type *pt1;
440 spin_lock(&ptype_lock);
442 list_for_each_entry(pt1, head, list) {
444 list_del_rcu(&pt->list);
449 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
451 spin_unlock(&ptype_lock);
453 EXPORT_SYMBOL(__dev_remove_pack);
456 * dev_remove_pack - remove packet handler
457 * @pt: packet type declaration
459 * Remove a protocol handler that was previously added to the kernel
460 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
461 * from the kernel lists and can be freed or reused once this function
464 * This call sleeps to guarantee that no CPU is looking at the packet
467 void dev_remove_pack(struct packet_type *pt)
469 __dev_remove_pack(pt);
473 EXPORT_SYMBOL(dev_remove_pack);
475 /******************************************************************************
477 Device Boot-time Settings Routines
479 *******************************************************************************/
481 /* Boot time configuration table */
482 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
485 * netdev_boot_setup_add - add new setup entry
486 * @name: name of the device
487 * @map: configured settings for the device
489 * Adds new setup entry to the dev_boot_setup list. The function
490 * returns 0 on error and 1 on success. This is a generic routine to
493 static int netdev_boot_setup_add(char *name, struct ifmap *map)
495 struct netdev_boot_setup *s;
499 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
500 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
501 memset(s[i].name, 0, sizeof(s[i].name));
502 strlcpy(s[i].name, name, IFNAMSIZ);
503 memcpy(&s[i].map, map, sizeof(s[i].map));
508 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
512 * netdev_boot_setup_check - check boot time settings
513 * @dev: the netdevice
515 * Check boot time settings for the device.
516 * The found settings are set for the device to be used
517 * later in the device probing.
518 * Returns 0 if no settings found, 1 if they are.
520 int netdev_boot_setup_check(struct net_device *dev)
522 struct netdev_boot_setup *s = dev_boot_setup;
525 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
526 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
527 !strcmp(dev->name, s[i].name)) {
528 dev->irq = s[i].map.irq;
529 dev->base_addr = s[i].map.base_addr;
530 dev->mem_start = s[i].map.mem_start;
531 dev->mem_end = s[i].map.mem_end;
537 EXPORT_SYMBOL(netdev_boot_setup_check);
541 * netdev_boot_base - get address from boot time settings
542 * @prefix: prefix for network device
543 * @unit: id for network device
545 * Check boot time settings for the base address of device.
546 * The found settings are set for the device to be used
547 * later in the device probing.
548 * Returns 0 if no settings found.
550 unsigned long netdev_boot_base(const char *prefix, int unit)
552 const struct netdev_boot_setup *s = dev_boot_setup;
556 sprintf(name, "%s%d", prefix, unit);
559 * If device already registered then return base of 1
560 * to indicate not to probe for this interface
562 if (__dev_get_by_name(&init_net, name))
565 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
566 if (!strcmp(name, s[i].name))
567 return s[i].map.base_addr;
572 * Saves at boot time configured settings for any netdevice.
574 int __init netdev_boot_setup(char *str)
579 str = get_options(str, ARRAY_SIZE(ints), ints);
584 memset(&map, 0, sizeof(map));
588 map.base_addr = ints[2];
590 map.mem_start = ints[3];
592 map.mem_end = ints[4];
594 /* Add new entry to the list */
595 return netdev_boot_setup_add(str, &map);
598 __setup("netdev=", netdev_boot_setup);
600 /*******************************************************************************
602 Device Interface Subroutines
604 *******************************************************************************/
607 * __dev_get_by_name - find a device by its name
608 * @net: the applicable net namespace
609 * @name: name to find
611 * Find an interface by name. Must be called under RTNL semaphore
612 * or @dev_base_lock. If the name is found a pointer to the device
613 * is returned. If the name is not found then %NULL is returned. The
614 * reference counters are not incremented so the caller must be
615 * careful with locks.
618 struct net_device *__dev_get_by_name(struct net *net, const char *name)
620 struct hlist_node *p;
621 struct net_device *dev;
622 struct hlist_head *head = dev_name_hash(net, name);
624 hlist_for_each_entry(dev, p, head, name_hlist)
625 if (!strncmp(dev->name, name, IFNAMSIZ))
630 EXPORT_SYMBOL(__dev_get_by_name);
633 * dev_get_by_name_rcu - find a device by its name
634 * @net: the applicable net namespace
635 * @name: name to find
637 * Find an interface by name.
638 * If the name is found a pointer to the device is returned.
639 * If the name is not found then %NULL is returned.
640 * The reference counters are not incremented so the caller must be
641 * careful with locks. The caller must hold RCU lock.
644 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
646 struct hlist_node *p;
647 struct net_device *dev;
648 struct hlist_head *head = dev_name_hash(net, name);
650 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
651 if (!strncmp(dev->name, name, IFNAMSIZ))
656 EXPORT_SYMBOL(dev_get_by_name_rcu);
659 * dev_get_by_name - find a device by its name
660 * @net: the applicable net namespace
661 * @name: name to find
663 * Find an interface by name. This can be called from any
664 * context and does its own locking. The returned handle has
665 * the usage count incremented and the caller must use dev_put() to
666 * release it when it is no longer needed. %NULL is returned if no
667 * matching device is found.
670 struct net_device *dev_get_by_name(struct net *net, const char *name)
672 struct net_device *dev;
675 dev = dev_get_by_name_rcu(net, name);
681 EXPORT_SYMBOL(dev_get_by_name);
684 * __dev_get_by_index - find a device by its ifindex
685 * @net: the applicable net namespace
686 * @ifindex: index of device
688 * Search for an interface by index. Returns %NULL if the device
689 * is not found or a pointer to the device. The device has not
690 * had its reference counter increased so the caller must be careful
691 * about locking. The caller must hold either the RTNL semaphore
695 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
697 struct hlist_node *p;
698 struct net_device *dev;
699 struct hlist_head *head = dev_index_hash(net, ifindex);
701 hlist_for_each_entry(dev, p, head, index_hlist)
702 if (dev->ifindex == ifindex)
707 EXPORT_SYMBOL(__dev_get_by_index);
710 * dev_get_by_index_rcu - find a device by its ifindex
711 * @net: the applicable net namespace
712 * @ifindex: index of device
714 * Search for an interface by index. Returns %NULL if the device
715 * is not found or a pointer to the device. The device has not
716 * had its reference counter increased so the caller must be careful
717 * about locking. The caller must hold RCU lock.
720 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
722 struct hlist_node *p;
723 struct net_device *dev;
724 struct hlist_head *head = dev_index_hash(net, ifindex);
726 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
727 if (dev->ifindex == ifindex)
732 EXPORT_SYMBOL(dev_get_by_index_rcu);
736 * dev_get_by_index - find a device by its ifindex
737 * @net: the applicable net namespace
738 * @ifindex: index of device
740 * Search for an interface by index. Returns NULL if the device
741 * is not found or a pointer to the device. The device returned has
742 * had a reference added and the pointer is safe until the user calls
743 * dev_put to indicate they have finished with it.
746 struct net_device *dev_get_by_index(struct net *net, int ifindex)
748 struct net_device *dev;
751 dev = dev_get_by_index_rcu(net, ifindex);
757 EXPORT_SYMBOL(dev_get_by_index);
760 * dev_getbyhwaddr_rcu - find a device by its hardware address
761 * @net: the applicable net namespace
762 * @type: media type of device
763 * @ha: hardware address
765 * Search for an interface by MAC address. Returns NULL if the device
766 * is not found or a pointer to the device.
767 * The caller must hold RCU or RTNL.
768 * The returned device has not had its ref count increased
769 * and the caller must therefore be careful about locking
773 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
776 struct net_device *dev;
778 for_each_netdev_rcu(net, dev)
779 if (dev->type == type &&
780 !memcmp(dev->dev_addr, ha, dev->addr_len))
785 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
787 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
789 struct net_device *dev;
792 for_each_netdev(net, dev)
793 if (dev->type == type)
798 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
800 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
802 struct net_device *dev, *ret = NULL;
805 for_each_netdev_rcu(net, dev)
806 if (dev->type == type) {
814 EXPORT_SYMBOL(dev_getfirstbyhwtype);
817 * dev_get_by_flags_rcu - find any device with given flags
818 * @net: the applicable net namespace
819 * @if_flags: IFF_* values
820 * @mask: bitmask of bits in if_flags to check
822 * Search for any interface with the given flags. Returns NULL if a device
823 * is not found or a pointer to the device. Must be called inside
824 * rcu_read_lock(), and result refcount is unchanged.
827 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
830 struct net_device *dev, *ret;
833 for_each_netdev_rcu(net, dev) {
834 if (((dev->flags ^ if_flags) & mask) == 0) {
841 EXPORT_SYMBOL(dev_get_by_flags_rcu);
844 * dev_valid_name - check if name is okay for network device
847 * Network device names need to be valid file names to
848 * to allow sysfs to work. We also disallow any kind of
851 int dev_valid_name(const char *name)
855 if (strlen(name) >= IFNAMSIZ)
857 if (!strcmp(name, ".") || !strcmp(name, ".."))
861 if (*name == '/' || isspace(*name))
867 EXPORT_SYMBOL(dev_valid_name);
870 * __dev_alloc_name - allocate a name for a device
871 * @net: network namespace to allocate the device name in
872 * @name: name format string
873 * @buf: scratch buffer and result name string
875 * Passed a format string - eg "lt%d" it will try and find a suitable
876 * id. It scans list of devices to build up a free map, then chooses
877 * the first empty slot. The caller must hold the dev_base or rtnl lock
878 * while allocating the name and adding the device in order to avoid
880 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
881 * Returns the number of the unit assigned or a negative errno code.
884 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
888 const int max_netdevices = 8*PAGE_SIZE;
889 unsigned long *inuse;
890 struct net_device *d;
892 p = strnchr(name, IFNAMSIZ-1, '%');
895 * Verify the string as this thing may have come from
896 * the user. There must be either one "%d" and no other "%"
899 if (p[1] != 'd' || strchr(p + 2, '%'))
902 /* Use one page as a bit array of possible slots */
903 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
907 for_each_netdev(net, d) {
908 if (!sscanf(d->name, name, &i))
910 if (i < 0 || i >= max_netdevices)
913 /* avoid cases where sscanf is not exact inverse of printf */
914 snprintf(buf, IFNAMSIZ, name, i);
915 if (!strncmp(buf, d->name, IFNAMSIZ))
919 i = find_first_zero_bit(inuse, max_netdevices);
920 free_page((unsigned long) inuse);
924 snprintf(buf, IFNAMSIZ, name, i);
925 if (!__dev_get_by_name(net, buf))
928 /* It is possible to run out of possible slots
929 * when the name is long and there isn't enough space left
930 * for the digits, or if all bits are used.
936 * dev_alloc_name - allocate a name for a device
938 * @name: name format string
940 * Passed a format string - eg "lt%d" it will try and find a suitable
941 * id. It scans list of devices to build up a free map, then chooses
942 * the first empty slot. The caller must hold the dev_base or rtnl lock
943 * while allocating the name and adding the device in order to avoid
945 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
946 * Returns the number of the unit assigned or a negative errno code.
949 int dev_alloc_name(struct net_device *dev, const char *name)
955 BUG_ON(!dev_net(dev));
957 ret = __dev_alloc_name(net, name, buf);
959 strlcpy(dev->name, buf, IFNAMSIZ);
962 EXPORT_SYMBOL(dev_alloc_name);
964 static int dev_get_valid_name(struct net_device *dev, const char *name)
968 BUG_ON(!dev_net(dev));
971 if (!dev_valid_name(name))
974 if (strchr(name, '%'))
975 return dev_alloc_name(dev, name);
976 else if (__dev_get_by_name(net, name))
978 else if (dev->name != name)
979 strlcpy(dev->name, name, IFNAMSIZ);
985 * dev_change_name - change name of a device
987 * @newname: name (or format string) must be at least IFNAMSIZ
989 * Change name of a device, can pass format strings "eth%d".
992 int dev_change_name(struct net_device *dev, const char *newname)
994 char oldname[IFNAMSIZ];
1000 BUG_ON(!dev_net(dev));
1003 if (dev->flags & IFF_UP)
1006 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1009 memcpy(oldname, dev->name, IFNAMSIZ);
1011 err = dev_get_valid_name(dev, newname);
1016 ret = device_rename(&dev->dev, dev->name);
1018 memcpy(dev->name, oldname, IFNAMSIZ);
1022 write_lock_bh(&dev_base_lock);
1023 hlist_del_rcu(&dev->name_hlist);
1024 write_unlock_bh(&dev_base_lock);
1028 write_lock_bh(&dev_base_lock);
1029 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1030 write_unlock_bh(&dev_base_lock);
1032 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1033 ret = notifier_to_errno(ret);
1036 /* err >= 0 after dev_alloc_name() or stores the first errno */
1039 memcpy(dev->name, oldname, IFNAMSIZ);
1043 "%s: name change rollback failed: %d.\n",
1052 * dev_set_alias - change ifalias of a device
1054 * @alias: name up to IFALIASZ
1055 * @len: limit of bytes to copy from info
1057 * Set ifalias for a device,
1059 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1063 if (len >= IFALIASZ)
1068 kfree(dev->ifalias);
1069 dev->ifalias = NULL;
1074 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1078 strlcpy(dev->ifalias, alias, len+1);
1084 * netdev_features_change - device changes features
1085 * @dev: device to cause notification
1087 * Called to indicate a device has changed features.
1089 void netdev_features_change(struct net_device *dev)
1091 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1093 EXPORT_SYMBOL(netdev_features_change);
1096 * netdev_state_change - device changes state
1097 * @dev: device to cause notification
1099 * Called to indicate a device has changed state. This function calls
1100 * the notifier chains for netdev_chain and sends a NEWLINK message
1101 * to the routing socket.
1103 void netdev_state_change(struct net_device *dev)
1105 if (dev->flags & IFF_UP) {
1106 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1107 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1110 EXPORT_SYMBOL(netdev_state_change);
1112 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1114 return call_netdevice_notifiers(event, dev);
1116 EXPORT_SYMBOL(netdev_bonding_change);
1119 * dev_load - load a network module
1120 * @net: the applicable net namespace
1121 * @name: name of interface
1123 * If a network interface is not present and the process has suitable
1124 * privileges this function loads the module. If module loading is not
1125 * available in this kernel then it becomes a nop.
1128 void dev_load(struct net *net, const char *name)
1130 struct net_device *dev;
1134 dev = dev_get_by_name_rcu(net, name);
1138 if (no_module && capable(CAP_NET_ADMIN))
1139 no_module = request_module("netdev-%s", name);
1140 if (no_module && capable(CAP_SYS_MODULE)) {
1141 if (!request_module("%s", name))
1142 pr_err("Loading kernel module for a network device "
1143 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1147 EXPORT_SYMBOL(dev_load);
1149 static int __dev_open(struct net_device *dev)
1151 const struct net_device_ops *ops = dev->netdev_ops;
1156 if (!netif_device_present(dev))
1159 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1160 ret = notifier_to_errno(ret);
1164 set_bit(__LINK_STATE_START, &dev->state);
1166 if (ops->ndo_validate_addr)
1167 ret = ops->ndo_validate_addr(dev);
1169 if (!ret && ops->ndo_open)
1170 ret = ops->ndo_open(dev);
1173 clear_bit(__LINK_STATE_START, &dev->state);
1175 dev->flags |= IFF_UP;
1176 net_dmaengine_get();
1177 dev_set_rx_mode(dev);
1185 * dev_open - prepare an interface for use.
1186 * @dev: device to open
1188 * Takes a device from down to up state. The device's private open
1189 * function is invoked and then the multicast lists are loaded. Finally
1190 * the device is moved into the up state and a %NETDEV_UP message is
1191 * sent to the netdev notifier chain.
1193 * Calling this function on an active interface is a nop. On a failure
1194 * a negative errno code is returned.
1196 int dev_open(struct net_device *dev)
1200 if (dev->flags & IFF_UP)
1203 ret = __dev_open(dev);
1207 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1208 call_netdevice_notifiers(NETDEV_UP, dev);
1212 EXPORT_SYMBOL(dev_open);
1214 static int __dev_close_many(struct list_head *head)
1216 struct net_device *dev;
1221 list_for_each_entry(dev, head, unreg_list) {
1222 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1224 clear_bit(__LINK_STATE_START, &dev->state);
1226 /* Synchronize to scheduled poll. We cannot touch poll list, it
1227 * can be even on different cpu. So just clear netif_running().
1229 * dev->stop() will invoke napi_disable() on all of it's
1230 * napi_struct instances on this device.
1232 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1235 dev_deactivate_many(head);
1237 list_for_each_entry(dev, head, unreg_list) {
1238 const struct net_device_ops *ops = dev->netdev_ops;
1241 * Call the device specific close. This cannot fail.
1242 * Only if device is UP
1244 * We allow it to be called even after a DETACH hot-plug
1250 dev->flags &= ~IFF_UP;
1251 net_dmaengine_put();
1257 static int __dev_close(struct net_device *dev)
1262 list_add(&dev->unreg_list, &single);
1263 retval = __dev_close_many(&single);
1268 static int dev_close_many(struct list_head *head)
1270 struct net_device *dev, *tmp;
1271 LIST_HEAD(tmp_list);
1273 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1274 if (!(dev->flags & IFF_UP))
1275 list_move(&dev->unreg_list, &tmp_list);
1277 __dev_close_many(head);
1279 list_for_each_entry(dev, head, unreg_list) {
1280 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1281 call_netdevice_notifiers(NETDEV_DOWN, dev);
1284 /* rollback_registered_many needs the complete original list */
1285 list_splice(&tmp_list, head);
1290 * dev_close - shutdown an interface.
1291 * @dev: device to shutdown
1293 * This function moves an active device into down state. A
1294 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1295 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1298 int dev_close(struct net_device *dev)
1300 if (dev->flags & IFF_UP) {
1303 list_add(&dev->unreg_list, &single);
1304 dev_close_many(&single);
1309 EXPORT_SYMBOL(dev_close);
1313 * dev_disable_lro - disable Large Receive Offload on a device
1316 * Disable Large Receive Offload (LRO) on a net device. Must be
1317 * called under RTNL. This is needed if received packets may be
1318 * forwarded to another interface.
1320 void dev_disable_lro(struct net_device *dev)
1325 * If we're trying to disable lro on a vlan device
1326 * use the underlying physical device instead
1328 if (is_vlan_dev(dev))
1329 dev = vlan_dev_real_dev(dev);
1331 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1332 flags = dev->ethtool_ops->get_flags(dev);
1334 flags = ethtool_op_get_flags(dev);
1336 if (!(flags & ETH_FLAG_LRO))
1339 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1340 if (unlikely(dev->features & NETIF_F_LRO))
1341 netdev_WARN(dev, "failed to disable LRO!\n");
1343 EXPORT_SYMBOL(dev_disable_lro);
1346 static int dev_boot_phase = 1;
1349 * register_netdevice_notifier - register a network notifier block
1352 * Register a notifier to be called when network device events occur.
1353 * The notifier passed is linked into the kernel structures and must
1354 * not be reused until it has been unregistered. A negative errno code
1355 * is returned on a failure.
1357 * When registered all registration and up events are replayed
1358 * to the new notifier to allow device to have a race free
1359 * view of the network device list.
1362 int register_netdevice_notifier(struct notifier_block *nb)
1364 struct net_device *dev;
1365 struct net_device *last;
1370 err = raw_notifier_chain_register(&netdev_chain, nb);
1376 for_each_netdev(net, dev) {
1377 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1378 err = notifier_to_errno(err);
1382 if (!(dev->flags & IFF_UP))
1385 nb->notifier_call(nb, NETDEV_UP, dev);
1396 for_each_netdev(net, dev) {
1400 if (dev->flags & IFF_UP) {
1401 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1402 nb->notifier_call(nb, NETDEV_DOWN, dev);
1404 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1405 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1409 raw_notifier_chain_unregister(&netdev_chain, nb);
1412 EXPORT_SYMBOL(register_netdevice_notifier);
1415 * unregister_netdevice_notifier - unregister a network notifier block
1418 * Unregister a notifier previously registered by
1419 * register_netdevice_notifier(). The notifier is unlinked into the
1420 * kernel structures and may then be reused. A negative errno code
1421 * is returned on a failure.
1424 int unregister_netdevice_notifier(struct notifier_block *nb)
1429 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1433 EXPORT_SYMBOL(unregister_netdevice_notifier);
1436 * call_netdevice_notifiers - call all network notifier blocks
1437 * @val: value passed unmodified to notifier function
1438 * @dev: net_device pointer passed unmodified to notifier function
1440 * Call all network notifier blocks. Parameters and return value
1441 * are as for raw_notifier_call_chain().
1444 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1447 return raw_notifier_call_chain(&netdev_chain, val, dev);
1449 EXPORT_SYMBOL(call_netdevice_notifiers);
1451 /* When > 0 there are consumers of rx skb time stamps */
1452 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1454 void net_enable_timestamp(void)
1456 atomic_inc(&netstamp_needed);
1458 EXPORT_SYMBOL(net_enable_timestamp);
1460 void net_disable_timestamp(void)
1462 atomic_dec(&netstamp_needed);
1464 EXPORT_SYMBOL(net_disable_timestamp);
1466 static inline void net_timestamp_set(struct sk_buff *skb)
1468 if (atomic_read(&netstamp_needed))
1469 __net_timestamp(skb);
1471 skb->tstamp.tv64 = 0;
1474 static inline void net_timestamp_check(struct sk_buff *skb)
1476 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1477 __net_timestamp(skb);
1480 static inline bool is_skb_forwardable(struct net_device *dev,
1481 struct sk_buff *skb)
1485 if (!(dev->flags & IFF_UP))
1488 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1489 if (skb->len <= len)
1492 /* if TSO is enabled, we don't care about the length as the packet
1493 * could be forwarded without being segmented before
1495 if (skb_is_gso(skb))
1502 * dev_forward_skb - loopback an skb to another netif
1504 * @dev: destination network device
1505 * @skb: buffer to forward
1508 * NET_RX_SUCCESS (no congestion)
1509 * NET_RX_DROP (packet was dropped, but freed)
1511 * dev_forward_skb can be used for injecting an skb from the
1512 * start_xmit function of one device into the receive queue
1513 * of another device.
1515 * The receiving device may be in another namespace, so
1516 * we have to clear all information in the skb that could
1517 * impact namespace isolation.
1519 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1521 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1522 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1523 atomic_long_inc(&dev->rx_dropped);
1532 if (unlikely(!is_skb_forwardable(dev, skb))) {
1533 atomic_long_inc(&dev->rx_dropped);
1537 skb_set_dev(skb, dev);
1538 skb->tstamp.tv64 = 0;
1539 skb->pkt_type = PACKET_HOST;
1540 skb->protocol = eth_type_trans(skb, dev);
1541 return netif_rx(skb);
1543 EXPORT_SYMBOL_GPL(dev_forward_skb);
1545 static inline int deliver_skb(struct sk_buff *skb,
1546 struct packet_type *pt_prev,
1547 struct net_device *orig_dev)
1549 atomic_inc(&skb->users);
1550 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1554 * Support routine. Sends outgoing frames to any network
1555 * taps currently in use.
1558 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1560 struct packet_type *ptype;
1561 struct sk_buff *skb2 = NULL;
1562 struct packet_type *pt_prev = NULL;
1565 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1566 /* Never send packets back to the socket
1567 * they originated from - MvS (miquels@drinkel.ow.org)
1569 if ((ptype->dev == dev || !ptype->dev) &&
1570 (ptype->af_packet_priv == NULL ||
1571 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1573 deliver_skb(skb2, pt_prev, skb->dev);
1578 skb2 = skb_clone(skb, GFP_ATOMIC);
1582 net_timestamp_set(skb2);
1584 /* skb->nh should be correctly
1585 set by sender, so that the second statement is
1586 just protection against buggy protocols.
1588 skb_reset_mac_header(skb2);
1590 if (skb_network_header(skb2) < skb2->data ||
1591 skb2->network_header > skb2->tail) {
1592 if (net_ratelimit())
1593 printk(KERN_CRIT "protocol %04x is "
1595 ntohs(skb2->protocol),
1597 skb_reset_network_header(skb2);
1600 skb2->transport_header = skb2->network_header;
1601 skb2->pkt_type = PACKET_OUTGOING;
1606 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1610 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1611 * @dev: Network device
1612 * @txq: number of queues available
1614 * If real_num_tx_queues is changed the tc mappings may no longer be
1615 * valid. To resolve this verify the tc mapping remains valid and if
1616 * not NULL the mapping. With no priorities mapping to this
1617 * offset/count pair it will no longer be used. In the worst case TC0
1618 * is invalid nothing can be done so disable priority mappings. If is
1619 * expected that drivers will fix this mapping if they can before
1620 * calling netif_set_real_num_tx_queues.
1622 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1625 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1627 /* If TC0 is invalidated disable TC mapping */
1628 if (tc->offset + tc->count > txq) {
1629 pr_warning("Number of in use tx queues changed "
1630 "invalidating tc mappings. Priority "
1631 "traffic classification disabled!\n");
1636 /* Invalidated prio to tc mappings set to TC0 */
1637 for (i = 1; i < TC_BITMASK + 1; i++) {
1638 int q = netdev_get_prio_tc_map(dev, i);
1640 tc = &dev->tc_to_txq[q];
1641 if (tc->offset + tc->count > txq) {
1642 pr_warning("Number of in use tx queues "
1643 "changed. Priority %i to tc "
1644 "mapping %i is no longer valid "
1645 "setting map to 0\n",
1647 netdev_set_prio_tc_map(dev, i, 0);
1653 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1654 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1656 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1660 if (txq < 1 || txq > dev->num_tx_queues)
1663 if (dev->reg_state == NETREG_REGISTERED ||
1664 dev->reg_state == NETREG_UNREGISTERING) {
1667 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1673 netif_setup_tc(dev, txq);
1675 if (txq < dev->real_num_tx_queues)
1676 qdisc_reset_all_tx_gt(dev, txq);
1679 dev->real_num_tx_queues = txq;
1682 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1686 * netif_set_real_num_rx_queues - set actual number of RX queues used
1687 * @dev: Network device
1688 * @rxq: Actual number of RX queues
1690 * This must be called either with the rtnl_lock held or before
1691 * registration of the net device. Returns 0 on success, or a
1692 * negative error code. If called before registration, it always
1695 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1699 if (rxq < 1 || rxq > dev->num_rx_queues)
1702 if (dev->reg_state == NETREG_REGISTERED) {
1705 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1711 dev->real_num_rx_queues = rxq;
1714 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1717 static inline void __netif_reschedule(struct Qdisc *q)
1719 struct softnet_data *sd;
1720 unsigned long flags;
1722 local_irq_save(flags);
1723 sd = &__get_cpu_var(softnet_data);
1724 q->next_sched = NULL;
1725 *sd->output_queue_tailp = q;
1726 sd->output_queue_tailp = &q->next_sched;
1727 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1728 local_irq_restore(flags);
1731 void __netif_schedule(struct Qdisc *q)
1733 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1734 __netif_reschedule(q);
1736 EXPORT_SYMBOL(__netif_schedule);
1738 void dev_kfree_skb_irq(struct sk_buff *skb)
1740 if (atomic_dec_and_test(&skb->users)) {
1741 struct softnet_data *sd;
1742 unsigned long flags;
1744 local_irq_save(flags);
1745 sd = &__get_cpu_var(softnet_data);
1746 skb->next = sd->completion_queue;
1747 sd->completion_queue = skb;
1748 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1749 local_irq_restore(flags);
1752 EXPORT_SYMBOL(dev_kfree_skb_irq);
1754 void dev_kfree_skb_any(struct sk_buff *skb)
1756 if (in_irq() || irqs_disabled())
1757 dev_kfree_skb_irq(skb);
1761 EXPORT_SYMBOL(dev_kfree_skb_any);
1765 * netif_device_detach - mark device as removed
1766 * @dev: network device
1768 * Mark device as removed from system and therefore no longer available.
1770 void netif_device_detach(struct net_device *dev)
1772 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1773 netif_running(dev)) {
1774 netif_tx_stop_all_queues(dev);
1777 EXPORT_SYMBOL(netif_device_detach);
1780 * netif_device_attach - mark device as attached
1781 * @dev: network device
1783 * Mark device as attached from system and restart if needed.
1785 void netif_device_attach(struct net_device *dev)
1787 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1788 netif_running(dev)) {
1789 netif_tx_wake_all_queues(dev);
1790 __netdev_watchdog_up(dev);
1793 EXPORT_SYMBOL(netif_device_attach);
1796 * skb_dev_set -- assign a new device to a buffer
1797 * @skb: buffer for the new device
1798 * @dev: network device
1800 * If an skb is owned by a device already, we have to reset
1801 * all data private to the namespace a device belongs to
1802 * before assigning it a new device.
1804 #ifdef CONFIG_NET_NS
1805 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1808 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1811 skb_init_secmark(skb);
1815 skb->ipvs_property = 0;
1816 #ifdef CONFIG_NET_SCHED
1822 EXPORT_SYMBOL(skb_set_dev);
1823 #endif /* CONFIG_NET_NS */
1826 * Invalidate hardware checksum when packet is to be mangled, and
1827 * complete checksum manually on outgoing path.
1829 int skb_checksum_help(struct sk_buff *skb)
1832 int ret = 0, offset;
1834 if (skb->ip_summed == CHECKSUM_COMPLETE)
1835 goto out_set_summed;
1837 if (unlikely(skb_shinfo(skb)->gso_size)) {
1838 /* Let GSO fix up the checksum. */
1839 goto out_set_summed;
1842 offset = skb_checksum_start_offset(skb);
1843 BUG_ON(offset >= skb_headlen(skb));
1844 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1846 offset += skb->csum_offset;
1847 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1849 if (skb_cloned(skb) &&
1850 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1851 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1856 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1858 skb->ip_summed = CHECKSUM_NONE;
1862 EXPORT_SYMBOL(skb_checksum_help);
1865 * skb_gso_segment - Perform segmentation on skb.
1866 * @skb: buffer to segment
1867 * @features: features for the output path (see dev->features)
1869 * This function segments the given skb and returns a list of segments.
1871 * It may return NULL if the skb requires no segmentation. This is
1872 * only possible when GSO is used for verifying header integrity.
1874 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1876 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1877 struct packet_type *ptype;
1878 __be16 type = skb->protocol;
1879 int vlan_depth = ETH_HLEN;
1882 while (type == htons(ETH_P_8021Q)) {
1883 struct vlan_hdr *vh;
1885 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1886 return ERR_PTR(-EINVAL);
1888 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1889 type = vh->h_vlan_encapsulated_proto;
1890 vlan_depth += VLAN_HLEN;
1893 skb_reset_mac_header(skb);
1894 skb->mac_len = skb->network_header - skb->mac_header;
1895 __skb_pull(skb, skb->mac_len);
1897 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1898 struct net_device *dev = skb->dev;
1899 struct ethtool_drvinfo info = {};
1901 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1902 dev->ethtool_ops->get_drvinfo(dev, &info);
1904 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1905 info.driver, dev ? dev->features : 0L,
1906 skb->sk ? skb->sk->sk_route_caps : 0L,
1907 skb->len, skb->data_len, skb->ip_summed);
1909 if (skb_header_cloned(skb) &&
1910 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1911 return ERR_PTR(err);
1915 list_for_each_entry_rcu(ptype,
1916 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1917 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1918 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1919 err = ptype->gso_send_check(skb);
1920 segs = ERR_PTR(err);
1921 if (err || skb_gso_ok(skb, features))
1923 __skb_push(skb, (skb->data -
1924 skb_network_header(skb)));
1926 segs = ptype->gso_segment(skb, features);
1932 __skb_push(skb, skb->data - skb_mac_header(skb));
1936 EXPORT_SYMBOL(skb_gso_segment);
1938 /* Take action when hardware reception checksum errors are detected. */
1940 void netdev_rx_csum_fault(struct net_device *dev)
1942 if (net_ratelimit()) {
1943 printk(KERN_ERR "%s: hw csum failure.\n",
1944 dev ? dev->name : "<unknown>");
1948 EXPORT_SYMBOL(netdev_rx_csum_fault);
1951 /* Actually, we should eliminate this check as soon as we know, that:
1952 * 1. IOMMU is present and allows to map all the memory.
1953 * 2. No high memory really exists on this machine.
1956 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1958 #ifdef CONFIG_HIGHMEM
1960 if (!(dev->features & NETIF_F_HIGHDMA)) {
1961 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1962 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1963 if (PageHighMem(skb_frag_page(frag)))
1968 if (PCI_DMA_BUS_IS_PHYS) {
1969 struct device *pdev = dev->dev.parent;
1973 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1974 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1975 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
1976 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1985 void (*destructor)(struct sk_buff *skb);
1988 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1990 static void dev_gso_skb_destructor(struct sk_buff *skb)
1992 struct dev_gso_cb *cb;
1995 struct sk_buff *nskb = skb->next;
1997 skb->next = nskb->next;
2000 } while (skb->next);
2002 cb = DEV_GSO_CB(skb);
2004 cb->destructor(skb);
2008 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2009 * @skb: buffer to segment
2010 * @features: device features as applicable to this skb
2012 * This function segments the given skb and stores the list of segments
2015 static int dev_gso_segment(struct sk_buff *skb, int features)
2017 struct sk_buff *segs;
2019 segs = skb_gso_segment(skb, features);
2021 /* Verifying header integrity only. */
2026 return PTR_ERR(segs);
2029 DEV_GSO_CB(skb)->destructor = skb->destructor;
2030 skb->destructor = dev_gso_skb_destructor;
2036 * Try to orphan skb early, right before transmission by the device.
2037 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2038 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2040 static inline void skb_orphan_try(struct sk_buff *skb)
2042 struct sock *sk = skb->sk;
2044 if (sk && !skb_shinfo(skb)->tx_flags) {
2045 /* skb_tx_hash() wont be able to get sk.
2046 * We copy sk_hash into skb->rxhash
2049 skb->rxhash = sk->sk_hash;
2054 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2056 return ((features & NETIF_F_GEN_CSUM) ||
2057 ((features & NETIF_F_V4_CSUM) &&
2058 protocol == htons(ETH_P_IP)) ||
2059 ((features & NETIF_F_V6_CSUM) &&
2060 protocol == htons(ETH_P_IPV6)) ||
2061 ((features & NETIF_F_FCOE_CRC) &&
2062 protocol == htons(ETH_P_FCOE)));
2065 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2067 if (!can_checksum_protocol(features, protocol)) {
2068 features &= ~NETIF_F_ALL_CSUM;
2069 features &= ~NETIF_F_SG;
2070 } else if (illegal_highdma(skb->dev, skb)) {
2071 features &= ~NETIF_F_SG;
2077 u32 netif_skb_features(struct sk_buff *skb)
2079 __be16 protocol = skb->protocol;
2080 u32 features = skb->dev->features;
2082 if (protocol == htons(ETH_P_8021Q)) {
2083 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2084 protocol = veh->h_vlan_encapsulated_proto;
2085 } else if (!vlan_tx_tag_present(skb)) {
2086 return harmonize_features(skb, protocol, features);
2089 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2091 if (protocol != htons(ETH_P_8021Q)) {
2092 return harmonize_features(skb, protocol, features);
2094 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2095 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2096 return harmonize_features(skb, protocol, features);
2099 EXPORT_SYMBOL(netif_skb_features);
2102 * Returns true if either:
2103 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2104 * 2. skb is fragmented and the device does not support SG, or if
2105 * at least one of fragments is in highmem and device does not
2106 * support DMA from it.
2108 static inline int skb_needs_linearize(struct sk_buff *skb,
2111 return skb_is_nonlinear(skb) &&
2112 ((skb_has_frag_list(skb) &&
2113 !(features & NETIF_F_FRAGLIST)) ||
2114 (skb_shinfo(skb)->nr_frags &&
2115 !(features & NETIF_F_SG)));
2118 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2119 struct netdev_queue *txq)
2121 const struct net_device_ops *ops = dev->netdev_ops;
2122 int rc = NETDEV_TX_OK;
2123 unsigned int skb_len;
2125 if (likely(!skb->next)) {
2129 * If device doesn't need skb->dst, release it right now while
2130 * its hot in this cpu cache
2132 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2135 if (!list_empty(&ptype_all))
2136 dev_queue_xmit_nit(skb, dev);
2138 skb_orphan_try(skb);
2140 features = netif_skb_features(skb);
2142 if (vlan_tx_tag_present(skb) &&
2143 !(features & NETIF_F_HW_VLAN_TX)) {
2144 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2151 if (netif_needs_gso(skb, features)) {
2152 if (unlikely(dev_gso_segment(skb, features)))
2157 if (skb_needs_linearize(skb, features) &&
2158 __skb_linearize(skb))
2161 /* If packet is not checksummed and device does not
2162 * support checksumming for this protocol, complete
2163 * checksumming here.
2165 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2166 skb_set_transport_header(skb,
2167 skb_checksum_start_offset(skb));
2168 if (!(features & NETIF_F_ALL_CSUM) &&
2169 skb_checksum_help(skb))
2175 rc = ops->ndo_start_xmit(skb, dev);
2176 trace_net_dev_xmit(skb, rc, dev, skb_len);
2177 if (rc == NETDEV_TX_OK)
2178 txq_trans_update(txq);
2184 struct sk_buff *nskb = skb->next;
2186 skb->next = nskb->next;
2190 * If device doesn't need nskb->dst, release it right now while
2191 * its hot in this cpu cache
2193 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2196 skb_len = nskb->len;
2197 rc = ops->ndo_start_xmit(nskb, dev);
2198 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2199 if (unlikely(rc != NETDEV_TX_OK)) {
2200 if (rc & ~NETDEV_TX_MASK)
2201 goto out_kfree_gso_skb;
2202 nskb->next = skb->next;
2206 txq_trans_update(txq);
2207 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2208 return NETDEV_TX_BUSY;
2209 } while (skb->next);
2212 if (likely(skb->next == NULL))
2213 skb->destructor = DEV_GSO_CB(skb)->destructor;
2220 static u32 hashrnd __read_mostly;
2223 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2224 * to be used as a distribution range.
2226 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2227 unsigned int num_tx_queues)
2231 u16 qcount = num_tx_queues;
2233 if (skb_rx_queue_recorded(skb)) {
2234 hash = skb_get_rx_queue(skb);
2235 while (unlikely(hash >= num_tx_queues))
2236 hash -= num_tx_queues;
2241 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2242 qoffset = dev->tc_to_txq[tc].offset;
2243 qcount = dev->tc_to_txq[tc].count;
2246 if (skb->sk && skb->sk->sk_hash)
2247 hash = skb->sk->sk_hash;
2249 hash = (__force u16) skb->protocol ^ skb->rxhash;
2250 hash = jhash_1word(hash, hashrnd);
2252 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2254 EXPORT_SYMBOL(__skb_tx_hash);
2256 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2258 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2259 if (net_ratelimit()) {
2260 pr_warning("%s selects TX queue %d, but "
2261 "real number of TX queues is %d\n",
2262 dev->name, queue_index, dev->real_num_tx_queues);
2269 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2272 struct xps_dev_maps *dev_maps;
2273 struct xps_map *map;
2274 int queue_index = -1;
2277 dev_maps = rcu_dereference(dev->xps_maps);
2279 map = rcu_dereference(
2280 dev_maps->cpu_map[raw_smp_processor_id()]);
2283 queue_index = map->queues[0];
2286 if (skb->sk && skb->sk->sk_hash)
2287 hash = skb->sk->sk_hash;
2289 hash = (__force u16) skb->protocol ^
2291 hash = jhash_1word(hash, hashrnd);
2292 queue_index = map->queues[
2293 ((u64)hash * map->len) >> 32];
2295 if (unlikely(queue_index >= dev->real_num_tx_queues))
2307 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2308 struct sk_buff *skb)
2311 const struct net_device_ops *ops = dev->netdev_ops;
2313 if (dev->real_num_tx_queues == 1)
2315 else if (ops->ndo_select_queue) {
2316 queue_index = ops->ndo_select_queue(dev, skb);
2317 queue_index = dev_cap_txqueue(dev, queue_index);
2319 struct sock *sk = skb->sk;
2320 queue_index = sk_tx_queue_get(sk);
2322 if (queue_index < 0 || skb->ooo_okay ||
2323 queue_index >= dev->real_num_tx_queues) {
2324 int old_index = queue_index;
2326 queue_index = get_xps_queue(dev, skb);
2327 if (queue_index < 0)
2328 queue_index = skb_tx_hash(dev, skb);
2330 if (queue_index != old_index && sk) {
2331 struct dst_entry *dst =
2332 rcu_dereference_check(sk->sk_dst_cache, 1);
2334 if (dst && skb_dst(skb) == dst)
2335 sk_tx_queue_set(sk, queue_index);
2340 skb_set_queue_mapping(skb, queue_index);
2341 return netdev_get_tx_queue(dev, queue_index);
2344 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2345 struct net_device *dev,
2346 struct netdev_queue *txq)
2348 spinlock_t *root_lock = qdisc_lock(q);
2352 qdisc_skb_cb(skb)->pkt_len = skb->len;
2353 qdisc_calculate_pkt_len(skb, q);
2355 * Heuristic to force contended enqueues to serialize on a
2356 * separate lock before trying to get qdisc main lock.
2357 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2358 * and dequeue packets faster.
2360 contended = qdisc_is_running(q);
2361 if (unlikely(contended))
2362 spin_lock(&q->busylock);
2364 spin_lock(root_lock);
2365 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2368 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2369 qdisc_run_begin(q)) {
2371 * This is a work-conserving queue; there are no old skbs
2372 * waiting to be sent out; and the qdisc is not running -
2373 * xmit the skb directly.
2375 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2378 qdisc_bstats_update(q, skb);
2380 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2381 if (unlikely(contended)) {
2382 spin_unlock(&q->busylock);
2389 rc = NET_XMIT_SUCCESS;
2392 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2393 if (qdisc_run_begin(q)) {
2394 if (unlikely(contended)) {
2395 spin_unlock(&q->busylock);
2401 spin_unlock(root_lock);
2402 if (unlikely(contended))
2403 spin_unlock(&q->busylock);
2407 static DEFINE_PER_CPU(int, xmit_recursion);
2408 #define RECURSION_LIMIT 10
2411 * dev_queue_xmit - transmit a buffer
2412 * @skb: buffer to transmit
2414 * Queue a buffer for transmission to a network device. The caller must
2415 * have set the device and priority and built the buffer before calling
2416 * this function. The function can be called from an interrupt.
2418 * A negative errno code is returned on a failure. A success does not
2419 * guarantee the frame will be transmitted as it may be dropped due
2420 * to congestion or traffic shaping.
2422 * -----------------------------------------------------------------------------------
2423 * I notice this method can also return errors from the queue disciplines,
2424 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2427 * Regardless of the return value, the skb is consumed, so it is currently
2428 * difficult to retry a send to this method. (You can bump the ref count
2429 * before sending to hold a reference for retry if you are careful.)
2431 * When calling this method, interrupts MUST be enabled. This is because
2432 * the BH enable code must have IRQs enabled so that it will not deadlock.
2435 int dev_queue_xmit(struct sk_buff *skb)
2437 struct net_device *dev = skb->dev;
2438 struct netdev_queue *txq;
2442 /* Disable soft irqs for various locks below. Also
2443 * stops preemption for RCU.
2447 txq = dev_pick_tx(dev, skb);
2448 q = rcu_dereference_bh(txq->qdisc);
2450 #ifdef CONFIG_NET_CLS_ACT
2451 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2453 trace_net_dev_queue(skb);
2455 rc = __dev_xmit_skb(skb, q, dev, txq);
2459 /* The device has no queue. Common case for software devices:
2460 loopback, all the sorts of tunnels...
2462 Really, it is unlikely that netif_tx_lock protection is necessary
2463 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2465 However, it is possible, that they rely on protection
2468 Check this and shot the lock. It is not prone from deadlocks.
2469 Either shot noqueue qdisc, it is even simpler 8)
2471 if (dev->flags & IFF_UP) {
2472 int cpu = smp_processor_id(); /* ok because BHs are off */
2474 if (txq->xmit_lock_owner != cpu) {
2476 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2477 goto recursion_alert;
2479 HARD_TX_LOCK(dev, txq, cpu);
2481 if (!netif_tx_queue_stopped(txq)) {
2482 __this_cpu_inc(xmit_recursion);
2483 rc = dev_hard_start_xmit(skb, dev, txq);
2484 __this_cpu_dec(xmit_recursion);
2485 if (dev_xmit_complete(rc)) {
2486 HARD_TX_UNLOCK(dev, txq);
2490 HARD_TX_UNLOCK(dev, txq);
2491 if (net_ratelimit())
2492 printk(KERN_CRIT "Virtual device %s asks to "
2493 "queue packet!\n", dev->name);
2495 /* Recursion is detected! It is possible,
2499 if (net_ratelimit())
2500 printk(KERN_CRIT "Dead loop on virtual device "
2501 "%s, fix it urgently!\n", dev->name);
2506 rcu_read_unlock_bh();
2511 rcu_read_unlock_bh();
2514 EXPORT_SYMBOL(dev_queue_xmit);
2517 /*=======================================================================
2519 =======================================================================*/
2521 int netdev_max_backlog __read_mostly = 1000;
2522 int netdev_tstamp_prequeue __read_mostly = 1;
2523 int netdev_budget __read_mostly = 300;
2524 int weight_p __read_mostly = 64; /* old backlog weight */
2526 /* Called with irq disabled */
2527 static inline void ____napi_schedule(struct softnet_data *sd,
2528 struct napi_struct *napi)
2530 list_add_tail(&napi->poll_list, &sd->poll_list);
2531 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2535 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2536 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2537 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2538 * if hash is a canonical 4-tuple hash over transport ports.
2540 void __skb_get_rxhash(struct sk_buff *skb)
2542 int nhoff, hash = 0, poff;
2543 const struct ipv6hdr *ip6;
2544 const struct iphdr *ip;
2545 const struct vlan_hdr *vlan;
2554 nhoff = skb_network_offset(skb);
2555 proto = skb->protocol;
2559 case __constant_htons(ETH_P_IP):
2561 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2564 ip = (const struct iphdr *) (skb->data + nhoff);
2565 if (ip_is_fragment(ip))
2568 ip_proto = ip->protocol;
2569 addr1 = (__force u32) ip->saddr;
2570 addr2 = (__force u32) ip->daddr;
2571 nhoff += ip->ihl * 4;
2573 case __constant_htons(ETH_P_IPV6):
2575 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2578 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2579 ip_proto = ip6->nexthdr;
2580 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2581 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2584 case __constant_htons(ETH_P_8021Q):
2585 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2587 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2588 proto = vlan->h_vlan_encapsulated_proto;
2589 nhoff += sizeof(*vlan);
2591 case __constant_htons(ETH_P_PPP_SES):
2592 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2594 proto = *((__be16 *) (skb->data + nhoff +
2595 sizeof(struct pppoe_hdr)));
2596 nhoff += PPPOE_SES_HLEN;
2598 case __constant_htons(PPP_IP):
2600 case __constant_htons(PPP_IPV6):
2611 if (pskb_may_pull(skb, nhoff + 16)) {
2612 u8 *h = skb->data + nhoff;
2613 __be16 flags = *(__be16 *)h;
2616 * Only look inside GRE if version zero and no
2619 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2620 proto = *(__be16 *)(h + 2);
2622 if (flags & GRE_CSUM)
2624 if (flags & GRE_KEY)
2626 if (flags & GRE_SEQ)
2639 poff = proto_ports_offset(ip_proto);
2642 if (pskb_may_pull(skb, nhoff + 4)) {
2643 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2644 if (ports.v16[1] < ports.v16[0])
2645 swap(ports.v16[0], ports.v16[1]);
2650 /* get a consistent hash (same value on both flow directions) */
2654 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2661 EXPORT_SYMBOL(__skb_get_rxhash);
2665 /* One global table that all flow-based protocols share. */
2666 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2667 EXPORT_SYMBOL(rps_sock_flow_table);
2669 static struct rps_dev_flow *
2670 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2671 struct rps_dev_flow *rflow, u16 next_cpu)
2673 if (next_cpu != RPS_NO_CPU) {
2674 #ifdef CONFIG_RFS_ACCEL
2675 struct netdev_rx_queue *rxqueue;
2676 struct rps_dev_flow_table *flow_table;
2677 struct rps_dev_flow *old_rflow;
2682 /* Should we steer this flow to a different hardware queue? */
2683 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2684 !(dev->features & NETIF_F_NTUPLE))
2686 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2687 if (rxq_index == skb_get_rx_queue(skb))
2690 rxqueue = dev->_rx + rxq_index;
2691 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2694 flow_id = skb->rxhash & flow_table->mask;
2695 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2696 rxq_index, flow_id);
2700 rflow = &flow_table->flows[flow_id];
2702 if (old_rflow->filter == rflow->filter)
2703 old_rflow->filter = RPS_NO_FILTER;
2707 per_cpu(softnet_data, next_cpu).input_queue_head;
2710 rflow->cpu = next_cpu;
2715 * get_rps_cpu is called from netif_receive_skb and returns the target
2716 * CPU from the RPS map of the receiving queue for a given skb.
2717 * rcu_read_lock must be held on entry.
2719 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2720 struct rps_dev_flow **rflowp)
2722 struct netdev_rx_queue *rxqueue;
2723 struct rps_map *map;
2724 struct rps_dev_flow_table *flow_table;
2725 struct rps_sock_flow_table *sock_flow_table;
2729 if (skb_rx_queue_recorded(skb)) {
2730 u16 index = skb_get_rx_queue(skb);
2731 if (unlikely(index >= dev->real_num_rx_queues)) {
2732 WARN_ONCE(dev->real_num_rx_queues > 1,
2733 "%s received packet on queue %u, but number "
2734 "of RX queues is %u\n",
2735 dev->name, index, dev->real_num_rx_queues);
2738 rxqueue = dev->_rx + index;
2742 map = rcu_dereference(rxqueue->rps_map);
2744 if (map->len == 1 &&
2745 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2746 tcpu = map->cpus[0];
2747 if (cpu_online(tcpu))
2751 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2755 skb_reset_network_header(skb);
2756 if (!skb_get_rxhash(skb))
2759 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2760 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2761 if (flow_table && sock_flow_table) {
2763 struct rps_dev_flow *rflow;
2765 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2768 next_cpu = sock_flow_table->ents[skb->rxhash &
2769 sock_flow_table->mask];
2772 * If the desired CPU (where last recvmsg was done) is
2773 * different from current CPU (one in the rx-queue flow
2774 * table entry), switch if one of the following holds:
2775 * - Current CPU is unset (equal to RPS_NO_CPU).
2776 * - Current CPU is offline.
2777 * - The current CPU's queue tail has advanced beyond the
2778 * last packet that was enqueued using this table entry.
2779 * This guarantees that all previous packets for the flow
2780 * have been dequeued, thus preserving in order delivery.
2782 if (unlikely(tcpu != next_cpu) &&
2783 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2784 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2785 rflow->last_qtail)) >= 0))
2786 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2788 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2796 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2798 if (cpu_online(tcpu)) {
2808 #ifdef CONFIG_RFS_ACCEL
2811 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2812 * @dev: Device on which the filter was set
2813 * @rxq_index: RX queue index
2814 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2815 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2817 * Drivers that implement ndo_rx_flow_steer() should periodically call
2818 * this function for each installed filter and remove the filters for
2819 * which it returns %true.
2821 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2822 u32 flow_id, u16 filter_id)
2824 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2825 struct rps_dev_flow_table *flow_table;
2826 struct rps_dev_flow *rflow;
2831 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2832 if (flow_table && flow_id <= flow_table->mask) {
2833 rflow = &flow_table->flows[flow_id];
2834 cpu = ACCESS_ONCE(rflow->cpu);
2835 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2836 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2837 rflow->last_qtail) <
2838 (int)(10 * flow_table->mask)))
2844 EXPORT_SYMBOL(rps_may_expire_flow);
2846 #endif /* CONFIG_RFS_ACCEL */
2848 /* Called from hardirq (IPI) context */
2849 static void rps_trigger_softirq(void *data)
2851 struct softnet_data *sd = data;
2853 ____napi_schedule(sd, &sd->backlog);
2857 #endif /* CONFIG_RPS */
2860 * Check if this softnet_data structure is another cpu one
2861 * If yes, queue it to our IPI list and return 1
2864 static int rps_ipi_queued(struct softnet_data *sd)
2867 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2870 sd->rps_ipi_next = mysd->rps_ipi_list;
2871 mysd->rps_ipi_list = sd;
2873 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2876 #endif /* CONFIG_RPS */
2881 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2882 * queue (may be a remote CPU queue).
2884 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2885 unsigned int *qtail)
2887 struct softnet_data *sd;
2888 unsigned long flags;
2890 sd = &per_cpu(softnet_data, cpu);
2892 local_irq_save(flags);
2895 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2896 if (skb_queue_len(&sd->input_pkt_queue)) {
2898 __skb_queue_tail(&sd->input_pkt_queue, skb);
2899 input_queue_tail_incr_save(sd, qtail);
2901 local_irq_restore(flags);
2902 return NET_RX_SUCCESS;
2905 /* Schedule NAPI for backlog device
2906 * We can use non atomic operation since we own the queue lock
2908 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2909 if (!rps_ipi_queued(sd))
2910 ____napi_schedule(sd, &sd->backlog);
2918 local_irq_restore(flags);
2920 atomic_long_inc(&skb->dev->rx_dropped);
2926 * netif_rx - post buffer to the network code
2927 * @skb: buffer to post
2929 * This function receives a packet from a device driver and queues it for
2930 * the upper (protocol) levels to process. It always succeeds. The buffer
2931 * may be dropped during processing for congestion control or by the
2935 * NET_RX_SUCCESS (no congestion)
2936 * NET_RX_DROP (packet was dropped)
2940 int netif_rx(struct sk_buff *skb)
2944 /* if netpoll wants it, pretend we never saw it */
2945 if (netpoll_rx(skb))
2948 if (netdev_tstamp_prequeue)
2949 net_timestamp_check(skb);
2951 trace_netif_rx(skb);
2954 struct rps_dev_flow voidflow, *rflow = &voidflow;
2960 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2962 cpu = smp_processor_id();
2964 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2972 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2978 EXPORT_SYMBOL(netif_rx);
2980 int netif_rx_ni(struct sk_buff *skb)
2985 err = netif_rx(skb);
2986 if (local_softirq_pending())
2992 EXPORT_SYMBOL(netif_rx_ni);
2994 static void net_tx_action(struct softirq_action *h)
2996 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2998 if (sd->completion_queue) {
2999 struct sk_buff *clist;
3001 local_irq_disable();
3002 clist = sd->completion_queue;
3003 sd->completion_queue = NULL;
3007 struct sk_buff *skb = clist;
3008 clist = clist->next;
3010 WARN_ON(atomic_read(&skb->users));
3011 trace_kfree_skb(skb, net_tx_action);
3016 if (sd->output_queue) {
3019 local_irq_disable();
3020 head = sd->output_queue;
3021 sd->output_queue = NULL;
3022 sd->output_queue_tailp = &sd->output_queue;
3026 struct Qdisc *q = head;
3027 spinlock_t *root_lock;
3029 head = head->next_sched;
3031 root_lock = qdisc_lock(q);
3032 if (spin_trylock(root_lock)) {
3033 smp_mb__before_clear_bit();
3034 clear_bit(__QDISC_STATE_SCHED,
3037 spin_unlock(root_lock);
3039 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3041 __netif_reschedule(q);
3043 smp_mb__before_clear_bit();
3044 clear_bit(__QDISC_STATE_SCHED,
3052 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3053 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3054 /* This hook is defined here for ATM LANE */
3055 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3056 unsigned char *addr) __read_mostly;
3057 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3060 #ifdef CONFIG_NET_CLS_ACT
3061 /* TODO: Maybe we should just force sch_ingress to be compiled in
3062 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3063 * a compare and 2 stores extra right now if we dont have it on
3064 * but have CONFIG_NET_CLS_ACT
3065 * NOTE: This doesn't stop any functionality; if you dont have
3066 * the ingress scheduler, you just can't add policies on ingress.
3069 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3071 struct net_device *dev = skb->dev;
3072 u32 ttl = G_TC_RTTL(skb->tc_verd);
3073 int result = TC_ACT_OK;
3076 if (unlikely(MAX_RED_LOOP < ttl++)) {
3077 if (net_ratelimit())
3078 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3079 skb->skb_iif, dev->ifindex);
3083 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3084 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3087 if (q != &noop_qdisc) {
3088 spin_lock(qdisc_lock(q));
3089 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3090 result = qdisc_enqueue_root(skb, q);
3091 spin_unlock(qdisc_lock(q));
3097 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3098 struct packet_type **pt_prev,
3099 int *ret, struct net_device *orig_dev)
3101 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3103 if (!rxq || rxq->qdisc == &noop_qdisc)
3107 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3111 switch (ing_filter(skb, rxq)) {
3125 * netdev_rx_handler_register - register receive handler
3126 * @dev: device to register a handler for
3127 * @rx_handler: receive handler to register
3128 * @rx_handler_data: data pointer that is used by rx handler
3130 * Register a receive hander for a device. This handler will then be
3131 * called from __netif_receive_skb. A negative errno code is returned
3134 * The caller must hold the rtnl_mutex.
3136 * For a general description of rx_handler, see enum rx_handler_result.
3138 int netdev_rx_handler_register(struct net_device *dev,
3139 rx_handler_func_t *rx_handler,
3140 void *rx_handler_data)
3144 if (dev->rx_handler)
3147 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3148 rcu_assign_pointer(dev->rx_handler, rx_handler);
3152 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3155 * netdev_rx_handler_unregister - unregister receive handler
3156 * @dev: device to unregister a handler from
3158 * Unregister a receive hander from a device.
3160 * The caller must hold the rtnl_mutex.
3162 void netdev_rx_handler_unregister(struct net_device *dev)
3166 RCU_INIT_POINTER(dev->rx_handler, NULL);
3167 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3169 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3171 static int __netif_receive_skb(struct sk_buff *skb)
3173 struct packet_type *ptype, *pt_prev;
3174 rx_handler_func_t *rx_handler;
3175 struct net_device *orig_dev;
3176 struct net_device *null_or_dev;
3177 bool deliver_exact = false;
3178 int ret = NET_RX_DROP;
3181 if (!netdev_tstamp_prequeue)
3182 net_timestamp_check(skb);
3184 trace_netif_receive_skb(skb);
3186 /* if we've gotten here through NAPI, check netpoll */
3187 if (netpoll_receive_skb(skb))
3191 skb->skb_iif = skb->dev->ifindex;
3192 orig_dev = skb->dev;
3194 skb_reset_network_header(skb);
3195 skb_reset_transport_header(skb);
3196 skb_reset_mac_len(skb);
3204 __this_cpu_inc(softnet_data.processed);
3206 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3207 skb = vlan_untag(skb);
3212 #ifdef CONFIG_NET_CLS_ACT
3213 if (skb->tc_verd & TC_NCLS) {
3214 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3219 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3220 if (!ptype->dev || ptype->dev == skb->dev) {
3222 ret = deliver_skb(skb, pt_prev, orig_dev);
3227 #ifdef CONFIG_NET_CLS_ACT
3228 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3234 rx_handler = rcu_dereference(skb->dev->rx_handler);
3237 ret = deliver_skb(skb, pt_prev, orig_dev);
3240 switch (rx_handler(&skb)) {
3241 case RX_HANDLER_CONSUMED:
3243 case RX_HANDLER_ANOTHER:
3245 case RX_HANDLER_EXACT:
3246 deliver_exact = true;
3247 case RX_HANDLER_PASS:
3254 if (vlan_tx_tag_present(skb)) {
3256 ret = deliver_skb(skb, pt_prev, orig_dev);
3259 if (vlan_do_receive(&skb))
3261 else if (unlikely(!skb))
3265 /* deliver only exact match when indicated */
3266 null_or_dev = deliver_exact ? skb->dev : NULL;
3268 type = skb->protocol;
3269 list_for_each_entry_rcu(ptype,
3270 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3271 if (ptype->type == type &&
3272 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3273 ptype->dev == orig_dev)) {
3275 ret = deliver_skb(skb, pt_prev, orig_dev);
3281 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3283 atomic_long_inc(&skb->dev->rx_dropped);
3285 /* Jamal, now you will not able to escape explaining
3286 * me how you were going to use this. :-)
3297 * netif_receive_skb - process receive buffer from network
3298 * @skb: buffer to process
3300 * netif_receive_skb() is the main receive data processing function.
3301 * It always succeeds. The buffer may be dropped during processing
3302 * for congestion control or by the protocol layers.
3304 * This function may only be called from softirq context and interrupts
3305 * should be enabled.
3307 * Return values (usually ignored):
3308 * NET_RX_SUCCESS: no congestion
3309 * NET_RX_DROP: packet was dropped
3311 int netif_receive_skb(struct sk_buff *skb)
3313 if (netdev_tstamp_prequeue)
3314 net_timestamp_check(skb);
3316 if (skb_defer_rx_timestamp(skb))
3317 return NET_RX_SUCCESS;
3321 struct rps_dev_flow voidflow, *rflow = &voidflow;
3326 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3329 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3333 ret = __netif_receive_skb(skb);
3339 return __netif_receive_skb(skb);
3342 EXPORT_SYMBOL(netif_receive_skb);
3344 /* Network device is going away, flush any packets still pending
3345 * Called with irqs disabled.
3347 static void flush_backlog(void *arg)
3349 struct net_device *dev = arg;
3350 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3351 struct sk_buff *skb, *tmp;
3354 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3355 if (skb->dev == dev) {
3356 __skb_unlink(skb, &sd->input_pkt_queue);
3358 input_queue_head_incr(sd);
3363 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3364 if (skb->dev == dev) {
3365 __skb_unlink(skb, &sd->process_queue);
3367 input_queue_head_incr(sd);
3372 static int napi_gro_complete(struct sk_buff *skb)
3374 struct packet_type *ptype;
3375 __be16 type = skb->protocol;
3376 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3379 if (NAPI_GRO_CB(skb)->count == 1) {
3380 skb_shinfo(skb)->gso_size = 0;
3385 list_for_each_entry_rcu(ptype, head, list) {
3386 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3389 err = ptype->gro_complete(skb);
3395 WARN_ON(&ptype->list == head);
3397 return NET_RX_SUCCESS;
3401 return netif_receive_skb(skb);
3404 inline void napi_gro_flush(struct napi_struct *napi)
3406 struct sk_buff *skb, *next;
3408 for (skb = napi->gro_list; skb; skb = next) {
3411 napi_gro_complete(skb);
3414 napi->gro_count = 0;
3415 napi->gro_list = NULL;
3417 EXPORT_SYMBOL(napi_gro_flush);
3419 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3421 struct sk_buff **pp = NULL;
3422 struct packet_type *ptype;
3423 __be16 type = skb->protocol;
3424 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3427 enum gro_result ret;
3429 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3432 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3436 list_for_each_entry_rcu(ptype, head, list) {
3437 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3440 skb_set_network_header(skb, skb_gro_offset(skb));
3441 mac_len = skb->network_header - skb->mac_header;
3442 skb->mac_len = mac_len;
3443 NAPI_GRO_CB(skb)->same_flow = 0;
3444 NAPI_GRO_CB(skb)->flush = 0;
3445 NAPI_GRO_CB(skb)->free = 0;
3447 pp = ptype->gro_receive(&napi->gro_list, skb);
3452 if (&ptype->list == head)
3455 same_flow = NAPI_GRO_CB(skb)->same_flow;
3456 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3459 struct sk_buff *nskb = *pp;
3463 napi_gro_complete(nskb);
3470 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3474 NAPI_GRO_CB(skb)->count = 1;
3475 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3476 skb->next = napi->gro_list;
3477 napi->gro_list = skb;
3481 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3482 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3484 BUG_ON(skb->end - skb->tail < grow);
3486 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3489 skb->data_len -= grow;
3491 skb_shinfo(skb)->frags[0].page_offset += grow;
3492 skb_shinfo(skb)->frags[0].size -= grow;
3494 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3495 skb_frag_unref(skb, 0);
3496 memmove(skb_shinfo(skb)->frags,
3497 skb_shinfo(skb)->frags + 1,
3498 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3509 EXPORT_SYMBOL(dev_gro_receive);
3511 static inline gro_result_t
3512 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3516 for (p = napi->gro_list; p; p = p->next) {
3517 unsigned long diffs;
3519 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3520 diffs |= p->vlan_tci ^ skb->vlan_tci;
3521 diffs |= compare_ether_header(skb_mac_header(p),
3522 skb_gro_mac_header(skb));
3523 NAPI_GRO_CB(p)->same_flow = !diffs;
3524 NAPI_GRO_CB(p)->flush = 0;
3527 return dev_gro_receive(napi, skb);
3530 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3534 if (netif_receive_skb(skb))
3539 case GRO_MERGED_FREE:
3550 EXPORT_SYMBOL(napi_skb_finish);
3552 void skb_gro_reset_offset(struct sk_buff *skb)
3554 NAPI_GRO_CB(skb)->data_offset = 0;
3555 NAPI_GRO_CB(skb)->frag0 = NULL;
3556 NAPI_GRO_CB(skb)->frag0_len = 0;
3558 if (skb->mac_header == skb->tail &&
3559 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3560 NAPI_GRO_CB(skb)->frag0 =
3561 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3562 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3565 EXPORT_SYMBOL(skb_gro_reset_offset);
3567 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3569 skb_gro_reset_offset(skb);
3571 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3573 EXPORT_SYMBOL(napi_gro_receive);
3575 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3577 __skb_pull(skb, skb_headlen(skb));
3578 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3580 skb->dev = napi->dev;
3586 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3588 struct sk_buff *skb = napi->skb;
3591 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3597 EXPORT_SYMBOL(napi_get_frags);
3599 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3605 skb->protocol = eth_type_trans(skb, skb->dev);
3607 if (ret == GRO_HELD)
3608 skb_gro_pull(skb, -ETH_HLEN);
3609 else if (netif_receive_skb(skb))
3614 case GRO_MERGED_FREE:
3615 napi_reuse_skb(napi, skb);
3624 EXPORT_SYMBOL(napi_frags_finish);
3626 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3628 struct sk_buff *skb = napi->skb;
3635 skb_reset_mac_header(skb);
3636 skb_gro_reset_offset(skb);
3638 off = skb_gro_offset(skb);
3639 hlen = off + sizeof(*eth);
3640 eth = skb_gro_header_fast(skb, off);
3641 if (skb_gro_header_hard(skb, hlen)) {
3642 eth = skb_gro_header_slow(skb, hlen, off);
3643 if (unlikely(!eth)) {
3644 napi_reuse_skb(napi, skb);
3650 skb_gro_pull(skb, sizeof(*eth));
3653 * This works because the only protocols we care about don't require
3654 * special handling. We'll fix it up properly at the end.
3656 skb->protocol = eth->h_proto;
3661 EXPORT_SYMBOL(napi_frags_skb);
3663 gro_result_t napi_gro_frags(struct napi_struct *napi)
3665 struct sk_buff *skb = napi_frags_skb(napi);
3670 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3672 EXPORT_SYMBOL(napi_gro_frags);
3675 * net_rps_action sends any pending IPI's for rps.
3676 * Note: called with local irq disabled, but exits with local irq enabled.
3678 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3681 struct softnet_data *remsd = sd->rps_ipi_list;
3684 sd->rps_ipi_list = NULL;
3688 /* Send pending IPI's to kick RPS processing on remote cpus. */
3690 struct softnet_data *next = remsd->rps_ipi_next;
3692 if (cpu_online(remsd->cpu))
3693 __smp_call_function_single(remsd->cpu,
3702 static int process_backlog(struct napi_struct *napi, int quota)
3705 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3708 /* Check if we have pending ipi, its better to send them now,
3709 * not waiting net_rx_action() end.
3711 if (sd->rps_ipi_list) {
3712 local_irq_disable();
3713 net_rps_action_and_irq_enable(sd);
3716 napi->weight = weight_p;
3717 local_irq_disable();
3718 while (work < quota) {
3719 struct sk_buff *skb;
3722 while ((skb = __skb_dequeue(&sd->process_queue))) {
3724 __netif_receive_skb(skb);
3725 local_irq_disable();
3726 input_queue_head_incr(sd);
3727 if (++work >= quota) {
3734 qlen = skb_queue_len(&sd->input_pkt_queue);
3736 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3737 &sd->process_queue);
3739 if (qlen < quota - work) {
3741 * Inline a custom version of __napi_complete().
3742 * only current cpu owns and manipulates this napi,
3743 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3744 * we can use a plain write instead of clear_bit(),
3745 * and we dont need an smp_mb() memory barrier.
3747 list_del(&napi->poll_list);
3750 quota = work + qlen;
3760 * __napi_schedule - schedule for receive
3761 * @n: entry to schedule
3763 * The entry's receive function will be scheduled to run
3765 void __napi_schedule(struct napi_struct *n)
3767 unsigned long flags;
3769 local_irq_save(flags);
3770 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3771 local_irq_restore(flags);
3773 EXPORT_SYMBOL(__napi_schedule);
3775 void __napi_complete(struct napi_struct *n)
3777 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3778 BUG_ON(n->gro_list);
3780 list_del(&n->poll_list);
3781 smp_mb__before_clear_bit();
3782 clear_bit(NAPI_STATE_SCHED, &n->state);
3784 EXPORT_SYMBOL(__napi_complete);
3786 void napi_complete(struct napi_struct *n)
3788 unsigned long flags;
3791 * don't let napi dequeue from the cpu poll list
3792 * just in case its running on a different cpu
3794 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3798 local_irq_save(flags);
3800 local_irq_restore(flags);
3802 EXPORT_SYMBOL(napi_complete);
3804 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3805 int (*poll)(struct napi_struct *, int), int weight)
3807 INIT_LIST_HEAD(&napi->poll_list);
3808 napi->gro_count = 0;
3809 napi->gro_list = NULL;
3812 napi->weight = weight;
3813 list_add(&napi->dev_list, &dev->napi_list);
3815 #ifdef CONFIG_NETPOLL
3816 spin_lock_init(&napi->poll_lock);
3817 napi->poll_owner = -1;
3819 set_bit(NAPI_STATE_SCHED, &napi->state);
3821 EXPORT_SYMBOL(netif_napi_add);
3823 void netif_napi_del(struct napi_struct *napi)
3825 struct sk_buff *skb, *next;
3827 list_del_init(&napi->dev_list);
3828 napi_free_frags(napi);
3830 for (skb = napi->gro_list; skb; skb = next) {
3836 napi->gro_list = NULL;
3837 napi->gro_count = 0;
3839 EXPORT_SYMBOL(netif_napi_del);
3841 static void net_rx_action(struct softirq_action *h)
3843 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3844 unsigned long time_limit = jiffies + 2;
3845 int budget = netdev_budget;
3848 local_irq_disable();
3850 while (!list_empty(&sd->poll_list)) {
3851 struct napi_struct *n;
3854 /* If softirq window is exhuasted then punt.
3855 * Allow this to run for 2 jiffies since which will allow
3856 * an average latency of 1.5/HZ.
3858 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3863 /* Even though interrupts have been re-enabled, this
3864 * access is safe because interrupts can only add new
3865 * entries to the tail of this list, and only ->poll()
3866 * calls can remove this head entry from the list.
3868 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3870 have = netpoll_poll_lock(n);
3874 /* This NAPI_STATE_SCHED test is for avoiding a race
3875 * with netpoll's poll_napi(). Only the entity which
3876 * obtains the lock and sees NAPI_STATE_SCHED set will
3877 * actually make the ->poll() call. Therefore we avoid
3878 * accidentally calling ->poll() when NAPI is not scheduled.
3881 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3882 work = n->poll(n, weight);
3886 WARN_ON_ONCE(work > weight);
3890 local_irq_disable();
3892 /* Drivers must not modify the NAPI state if they
3893 * consume the entire weight. In such cases this code
3894 * still "owns" the NAPI instance and therefore can
3895 * move the instance around on the list at-will.
3897 if (unlikely(work == weight)) {
3898 if (unlikely(napi_disable_pending(n))) {
3901 local_irq_disable();
3903 list_move_tail(&n->poll_list, &sd->poll_list);
3906 netpoll_poll_unlock(have);
3909 net_rps_action_and_irq_enable(sd);
3911 #ifdef CONFIG_NET_DMA
3913 * There may not be any more sk_buffs coming right now, so push
3914 * any pending DMA copies to hardware
3916 dma_issue_pending_all();
3923 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3927 static gifconf_func_t *gifconf_list[NPROTO];
3930 * register_gifconf - register a SIOCGIF handler
3931 * @family: Address family
3932 * @gifconf: Function handler
3934 * Register protocol dependent address dumping routines. The handler
3935 * that is passed must not be freed or reused until it has been replaced
3936 * by another handler.
3938 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3940 if (family >= NPROTO)
3942 gifconf_list[family] = gifconf;
3945 EXPORT_SYMBOL(register_gifconf);
3949 * Map an interface index to its name (SIOCGIFNAME)
3953 * We need this ioctl for efficient implementation of the
3954 * if_indextoname() function required by the IPv6 API. Without
3955 * it, we would have to search all the interfaces to find a
3959 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3961 struct net_device *dev;
3965 * Fetch the caller's info block.
3968 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3972 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3978 strcpy(ifr.ifr_name, dev->name);
3981 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3987 * Perform a SIOCGIFCONF call. This structure will change
3988 * size eventually, and there is nothing I can do about it.
3989 * Thus we will need a 'compatibility mode'.
3992 static int dev_ifconf(struct net *net, char __user *arg)
3995 struct net_device *dev;
4002 * Fetch the caller's info block.
4005 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4012 * Loop over the interfaces, and write an info block for each.
4016 for_each_netdev(net, dev) {
4017 for (i = 0; i < NPROTO; i++) {
4018 if (gifconf_list[i]) {
4021 done = gifconf_list[i](dev, NULL, 0);
4023 done = gifconf_list[i](dev, pos + total,
4033 * All done. Write the updated control block back to the caller.
4035 ifc.ifc_len = total;
4038 * Both BSD and Solaris return 0 here, so we do too.
4040 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4043 #ifdef CONFIG_PROC_FS
4045 * This is invoked by the /proc filesystem handler to display a device
4048 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4051 struct net *net = seq_file_net(seq);
4053 struct net_device *dev;
4057 return SEQ_START_TOKEN;
4060 for_each_netdev_rcu(net, dev)
4067 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4069 struct net_device *dev = v;
4071 if (v == SEQ_START_TOKEN)
4072 dev = first_net_device_rcu(seq_file_net(seq));
4074 dev = next_net_device_rcu(dev);
4080 void dev_seq_stop(struct seq_file *seq, void *v)
4086 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4088 struct rtnl_link_stats64 temp;
4089 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4091 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4092 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4093 dev->name, stats->rx_bytes, stats->rx_packets,
4095 stats->rx_dropped + stats->rx_missed_errors,
4096 stats->rx_fifo_errors,
4097 stats->rx_length_errors + stats->rx_over_errors +
4098 stats->rx_crc_errors + stats->rx_frame_errors,
4099 stats->rx_compressed, stats->multicast,
4100 stats->tx_bytes, stats->tx_packets,
4101 stats->tx_errors, stats->tx_dropped,
4102 stats->tx_fifo_errors, stats->collisions,
4103 stats->tx_carrier_errors +
4104 stats->tx_aborted_errors +
4105 stats->tx_window_errors +
4106 stats->tx_heartbeat_errors,
4107 stats->tx_compressed);
4111 * Called from the PROCfs module. This now uses the new arbitrary sized
4112 * /proc/net interface to create /proc/net/dev
4114 static int dev_seq_show(struct seq_file *seq, void *v)
4116 if (v == SEQ_START_TOKEN)
4117 seq_puts(seq, "Inter-| Receive "
4119 " face |bytes packets errs drop fifo frame "
4120 "compressed multicast|bytes packets errs "
4121 "drop fifo colls carrier compressed\n");
4123 dev_seq_printf_stats(seq, v);
4127 static struct softnet_data *softnet_get_online(loff_t *pos)
4129 struct softnet_data *sd = NULL;
4131 while (*pos < nr_cpu_ids)
4132 if (cpu_online(*pos)) {
4133 sd = &per_cpu(softnet_data, *pos);
4140 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4142 return softnet_get_online(pos);
4145 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4148 return softnet_get_online(pos);
4151 static void softnet_seq_stop(struct seq_file *seq, void *v)
4155 static int softnet_seq_show(struct seq_file *seq, void *v)
4157 struct softnet_data *sd = v;
4159 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4160 sd->processed, sd->dropped, sd->time_squeeze, 0,
4161 0, 0, 0, 0, /* was fastroute */
4162 sd->cpu_collision, sd->received_rps);
4166 static const struct seq_operations dev_seq_ops = {
4167 .start = dev_seq_start,
4168 .next = dev_seq_next,
4169 .stop = dev_seq_stop,
4170 .show = dev_seq_show,
4173 static int dev_seq_open(struct inode *inode, struct file *file)
4175 return seq_open_net(inode, file, &dev_seq_ops,
4176 sizeof(struct seq_net_private));
4179 static const struct file_operations dev_seq_fops = {
4180 .owner = THIS_MODULE,
4181 .open = dev_seq_open,
4183 .llseek = seq_lseek,
4184 .release = seq_release_net,
4187 static const struct seq_operations softnet_seq_ops = {
4188 .start = softnet_seq_start,
4189 .next = softnet_seq_next,
4190 .stop = softnet_seq_stop,
4191 .show = softnet_seq_show,
4194 static int softnet_seq_open(struct inode *inode, struct file *file)
4196 return seq_open(file, &softnet_seq_ops);
4199 static const struct file_operations softnet_seq_fops = {
4200 .owner = THIS_MODULE,
4201 .open = softnet_seq_open,
4203 .llseek = seq_lseek,
4204 .release = seq_release,
4207 static void *ptype_get_idx(loff_t pos)
4209 struct packet_type *pt = NULL;
4213 list_for_each_entry_rcu(pt, &ptype_all, list) {
4219 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4220 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4229 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4233 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4236 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4238 struct packet_type *pt;
4239 struct list_head *nxt;
4243 if (v == SEQ_START_TOKEN)
4244 return ptype_get_idx(0);
4247 nxt = pt->list.next;
4248 if (pt->type == htons(ETH_P_ALL)) {
4249 if (nxt != &ptype_all)
4252 nxt = ptype_base[0].next;
4254 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4256 while (nxt == &ptype_base[hash]) {
4257 if (++hash >= PTYPE_HASH_SIZE)
4259 nxt = ptype_base[hash].next;
4262 return list_entry(nxt, struct packet_type, list);
4265 static void ptype_seq_stop(struct seq_file *seq, void *v)
4271 static int ptype_seq_show(struct seq_file *seq, void *v)
4273 struct packet_type *pt = v;
4275 if (v == SEQ_START_TOKEN)
4276 seq_puts(seq, "Type Device Function\n");
4277 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4278 if (pt->type == htons(ETH_P_ALL))
4279 seq_puts(seq, "ALL ");
4281 seq_printf(seq, "%04x", ntohs(pt->type));
4283 seq_printf(seq, " %-8s %pF\n",
4284 pt->dev ? pt->dev->name : "", pt->func);
4290 static const struct seq_operations ptype_seq_ops = {
4291 .start = ptype_seq_start,
4292 .next = ptype_seq_next,
4293 .stop = ptype_seq_stop,
4294 .show = ptype_seq_show,
4297 static int ptype_seq_open(struct inode *inode, struct file *file)
4299 return seq_open_net(inode, file, &ptype_seq_ops,
4300 sizeof(struct seq_net_private));
4303 static const struct file_operations ptype_seq_fops = {
4304 .owner = THIS_MODULE,
4305 .open = ptype_seq_open,
4307 .llseek = seq_lseek,
4308 .release = seq_release_net,
4312 static int __net_init dev_proc_net_init(struct net *net)
4316 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4318 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4320 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4323 if (wext_proc_init(net))
4329 proc_net_remove(net, "ptype");
4331 proc_net_remove(net, "softnet_stat");
4333 proc_net_remove(net, "dev");
4337 static void __net_exit dev_proc_net_exit(struct net *net)
4339 wext_proc_exit(net);
4341 proc_net_remove(net, "ptype");
4342 proc_net_remove(net, "softnet_stat");
4343 proc_net_remove(net, "dev");
4346 static struct pernet_operations __net_initdata dev_proc_ops = {
4347 .init = dev_proc_net_init,
4348 .exit = dev_proc_net_exit,
4351 static int __init dev_proc_init(void)
4353 return register_pernet_subsys(&dev_proc_ops);
4356 #define dev_proc_init() 0
4357 #endif /* CONFIG_PROC_FS */
4361 * netdev_set_master - set up master pointer
4362 * @slave: slave device
4363 * @master: new master device
4365 * Changes the master device of the slave. Pass %NULL to break the
4366 * bonding. The caller must hold the RTNL semaphore. On a failure
4367 * a negative errno code is returned. On success the reference counts
4368 * are adjusted and the function returns zero.
4370 int netdev_set_master(struct net_device *slave, struct net_device *master)
4372 struct net_device *old = slave->master;
4382 slave->master = master;
4388 EXPORT_SYMBOL(netdev_set_master);
4391 * netdev_set_bond_master - set up bonding master/slave pair
4392 * @slave: slave device
4393 * @master: new master device
4395 * Changes the master device of the slave. Pass %NULL to break the
4396 * bonding. The caller must hold the RTNL semaphore. On a failure
4397 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4398 * to the routing socket and the function returns zero.
4400 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4406 err = netdev_set_master(slave, master);
4410 slave->flags |= IFF_SLAVE;
4412 slave->flags &= ~IFF_SLAVE;
4414 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4417 EXPORT_SYMBOL(netdev_set_bond_master);
4419 static void dev_change_rx_flags(struct net_device *dev, int flags)
4421 const struct net_device_ops *ops = dev->netdev_ops;
4423 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4424 ops->ndo_change_rx_flags(dev, flags);
4427 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4429 unsigned short old_flags = dev->flags;
4435 dev->flags |= IFF_PROMISC;
4436 dev->promiscuity += inc;
4437 if (dev->promiscuity == 0) {
4440 * If inc causes overflow, untouch promisc and return error.
4443 dev->flags &= ~IFF_PROMISC;
4445 dev->promiscuity -= inc;
4446 printk(KERN_WARNING "%s: promiscuity touches roof, "
4447 "set promiscuity failed, promiscuity feature "
4448 "of device might be broken.\n", dev->name);
4452 if (dev->flags != old_flags) {
4453 printk(KERN_INFO "device %s %s promiscuous mode\n",
4454 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4456 if (audit_enabled) {
4457 current_uid_gid(&uid, &gid);
4458 audit_log(current->audit_context, GFP_ATOMIC,
4459 AUDIT_ANOM_PROMISCUOUS,
4460 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4461 dev->name, (dev->flags & IFF_PROMISC),
4462 (old_flags & IFF_PROMISC),
4463 audit_get_loginuid(current),
4465 audit_get_sessionid(current));
4468 dev_change_rx_flags(dev, IFF_PROMISC);
4474 * dev_set_promiscuity - update promiscuity count on a device
4478 * Add or remove promiscuity from a device. While the count in the device
4479 * remains above zero the interface remains promiscuous. Once it hits zero
4480 * the device reverts back to normal filtering operation. A negative inc
4481 * value is used to drop promiscuity on the device.
4482 * Return 0 if successful or a negative errno code on error.
4484 int dev_set_promiscuity(struct net_device *dev, int inc)
4486 unsigned short old_flags = dev->flags;
4489 err = __dev_set_promiscuity(dev, inc);
4492 if (dev->flags != old_flags)
4493 dev_set_rx_mode(dev);
4496 EXPORT_SYMBOL(dev_set_promiscuity);
4499 * dev_set_allmulti - update allmulti count on a device
4503 * Add or remove reception of all multicast frames to a device. While the
4504 * count in the device remains above zero the interface remains listening
4505 * to all interfaces. Once it hits zero the device reverts back to normal
4506 * filtering operation. A negative @inc value is used to drop the counter
4507 * when releasing a resource needing all multicasts.
4508 * Return 0 if successful or a negative errno code on error.
4511 int dev_set_allmulti(struct net_device *dev, int inc)
4513 unsigned short old_flags = dev->flags;
4517 dev->flags |= IFF_ALLMULTI;
4518 dev->allmulti += inc;
4519 if (dev->allmulti == 0) {
4522 * If inc causes overflow, untouch allmulti and return error.
4525 dev->flags &= ~IFF_ALLMULTI;
4527 dev->allmulti -= inc;
4528 printk(KERN_WARNING "%s: allmulti touches roof, "
4529 "set allmulti failed, allmulti feature of "
4530 "device might be broken.\n", dev->name);
4534 if (dev->flags ^ old_flags) {
4535 dev_change_rx_flags(dev, IFF_ALLMULTI);
4536 dev_set_rx_mode(dev);
4540 EXPORT_SYMBOL(dev_set_allmulti);
4543 * Upload unicast and multicast address lists to device and
4544 * configure RX filtering. When the device doesn't support unicast
4545 * filtering it is put in promiscuous mode while unicast addresses
4548 void __dev_set_rx_mode(struct net_device *dev)
4550 const struct net_device_ops *ops = dev->netdev_ops;
4552 /* dev_open will call this function so the list will stay sane. */
4553 if (!(dev->flags&IFF_UP))
4556 if (!netif_device_present(dev))
4559 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4560 /* Unicast addresses changes may only happen under the rtnl,
4561 * therefore calling __dev_set_promiscuity here is safe.
4563 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4564 __dev_set_promiscuity(dev, 1);
4565 dev->uc_promisc = true;
4566 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4567 __dev_set_promiscuity(dev, -1);
4568 dev->uc_promisc = false;
4572 if (ops->ndo_set_rx_mode)
4573 ops->ndo_set_rx_mode(dev);
4576 void dev_set_rx_mode(struct net_device *dev)
4578 netif_addr_lock_bh(dev);
4579 __dev_set_rx_mode(dev);
4580 netif_addr_unlock_bh(dev);
4584 * dev_get_flags - get flags reported to userspace
4587 * Get the combination of flag bits exported through APIs to userspace.
4589 unsigned dev_get_flags(const struct net_device *dev)
4593 flags = (dev->flags & ~(IFF_PROMISC |
4598 (dev->gflags & (IFF_PROMISC |
4601 if (netif_running(dev)) {
4602 if (netif_oper_up(dev))
4603 flags |= IFF_RUNNING;
4604 if (netif_carrier_ok(dev))
4605 flags |= IFF_LOWER_UP;
4606 if (netif_dormant(dev))
4607 flags |= IFF_DORMANT;
4612 EXPORT_SYMBOL(dev_get_flags);
4614 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4616 int old_flags = dev->flags;
4622 * Set the flags on our device.
4625 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4626 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4628 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4632 * Load in the correct multicast list now the flags have changed.
4635 if ((old_flags ^ flags) & IFF_MULTICAST)
4636 dev_change_rx_flags(dev, IFF_MULTICAST);
4638 dev_set_rx_mode(dev);
4641 * Have we downed the interface. We handle IFF_UP ourselves
4642 * according to user attempts to set it, rather than blindly
4647 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4648 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4651 dev_set_rx_mode(dev);
4654 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4655 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4657 dev->gflags ^= IFF_PROMISC;
4658 dev_set_promiscuity(dev, inc);
4661 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4662 is important. Some (broken) drivers set IFF_PROMISC, when
4663 IFF_ALLMULTI is requested not asking us and not reporting.
4665 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4666 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4668 dev->gflags ^= IFF_ALLMULTI;
4669 dev_set_allmulti(dev, inc);
4675 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4677 unsigned int changes = dev->flags ^ old_flags;
4679 if (changes & IFF_UP) {
4680 if (dev->flags & IFF_UP)
4681 call_netdevice_notifiers(NETDEV_UP, dev);
4683 call_netdevice_notifiers(NETDEV_DOWN, dev);
4686 if (dev->flags & IFF_UP &&
4687 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4688 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4692 * dev_change_flags - change device settings
4694 * @flags: device state flags
4696 * Change settings on device based state flags. The flags are
4697 * in the userspace exported format.
4699 int dev_change_flags(struct net_device *dev, unsigned flags)
4702 int old_flags = dev->flags;
4704 ret = __dev_change_flags(dev, flags);
4708 changes = old_flags ^ dev->flags;
4710 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4712 __dev_notify_flags(dev, old_flags);
4715 EXPORT_SYMBOL(dev_change_flags);
4718 * dev_set_mtu - Change maximum transfer unit
4720 * @new_mtu: new transfer unit
4722 * Change the maximum transfer size of the network device.
4724 int dev_set_mtu(struct net_device *dev, int new_mtu)
4726 const struct net_device_ops *ops = dev->netdev_ops;
4729 if (new_mtu == dev->mtu)
4732 /* MTU must be positive. */
4736 if (!netif_device_present(dev))
4740 if (ops->ndo_change_mtu)
4741 err = ops->ndo_change_mtu(dev, new_mtu);
4745 if (!err && dev->flags & IFF_UP)
4746 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4749 EXPORT_SYMBOL(dev_set_mtu);
4752 * dev_set_group - Change group this device belongs to
4754 * @new_group: group this device should belong to
4756 void dev_set_group(struct net_device *dev, int new_group)
4758 dev->group = new_group;
4760 EXPORT_SYMBOL(dev_set_group);
4763 * dev_set_mac_address - Change Media Access Control Address
4767 * Change the hardware (MAC) address of the device
4769 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4771 const struct net_device_ops *ops = dev->netdev_ops;
4774 if (!ops->ndo_set_mac_address)
4776 if (sa->sa_family != dev->type)
4778 if (!netif_device_present(dev))
4780 err = ops->ndo_set_mac_address(dev, sa);
4782 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4785 EXPORT_SYMBOL(dev_set_mac_address);
4788 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4790 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4793 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4799 case SIOCGIFFLAGS: /* Get interface flags */
4800 ifr->ifr_flags = (short) dev_get_flags(dev);
4803 case SIOCGIFMETRIC: /* Get the metric on the interface
4804 (currently unused) */
4805 ifr->ifr_metric = 0;
4808 case SIOCGIFMTU: /* Get the MTU of a device */
4809 ifr->ifr_mtu = dev->mtu;
4814 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4816 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4817 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4818 ifr->ifr_hwaddr.sa_family = dev->type;
4826 ifr->ifr_map.mem_start = dev->mem_start;
4827 ifr->ifr_map.mem_end = dev->mem_end;
4828 ifr->ifr_map.base_addr = dev->base_addr;
4829 ifr->ifr_map.irq = dev->irq;
4830 ifr->ifr_map.dma = dev->dma;
4831 ifr->ifr_map.port = dev->if_port;
4835 ifr->ifr_ifindex = dev->ifindex;
4839 ifr->ifr_qlen = dev->tx_queue_len;
4843 /* dev_ioctl() should ensure this case
4855 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4857 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4860 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4861 const struct net_device_ops *ops;
4866 ops = dev->netdev_ops;
4869 case SIOCSIFFLAGS: /* Set interface flags */
4870 return dev_change_flags(dev, ifr->ifr_flags);
4872 case SIOCSIFMETRIC: /* Set the metric on the interface
4873 (currently unused) */
4876 case SIOCSIFMTU: /* Set the MTU of a device */
4877 return dev_set_mtu(dev, ifr->ifr_mtu);
4880 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4882 case SIOCSIFHWBROADCAST:
4883 if (ifr->ifr_hwaddr.sa_family != dev->type)
4885 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4886 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4887 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4891 if (ops->ndo_set_config) {
4892 if (!netif_device_present(dev))
4894 return ops->ndo_set_config(dev, &ifr->ifr_map);
4899 if (!ops->ndo_set_rx_mode ||
4900 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4902 if (!netif_device_present(dev))
4904 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4907 if (!ops->ndo_set_rx_mode ||
4908 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4910 if (!netif_device_present(dev))
4912 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4915 if (ifr->ifr_qlen < 0)
4917 dev->tx_queue_len = ifr->ifr_qlen;
4921 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4922 return dev_change_name(dev, ifr->ifr_newname);
4925 * Unknown or private ioctl
4928 if ((cmd >= SIOCDEVPRIVATE &&
4929 cmd <= SIOCDEVPRIVATE + 15) ||
4930 cmd == SIOCBONDENSLAVE ||
4931 cmd == SIOCBONDRELEASE ||
4932 cmd == SIOCBONDSETHWADDR ||
4933 cmd == SIOCBONDSLAVEINFOQUERY ||
4934 cmd == SIOCBONDINFOQUERY ||
4935 cmd == SIOCBONDCHANGEACTIVE ||
4936 cmd == SIOCGMIIPHY ||
4937 cmd == SIOCGMIIREG ||
4938 cmd == SIOCSMIIREG ||
4939 cmd == SIOCBRADDIF ||
4940 cmd == SIOCBRDELIF ||
4941 cmd == SIOCSHWTSTAMP ||
4942 cmd == SIOCWANDEV) {
4944 if (ops->ndo_do_ioctl) {
4945 if (netif_device_present(dev))
4946 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4958 * This function handles all "interface"-type I/O control requests. The actual
4959 * 'doing' part of this is dev_ifsioc above.
4963 * dev_ioctl - network device ioctl
4964 * @net: the applicable net namespace
4965 * @cmd: command to issue
4966 * @arg: pointer to a struct ifreq in user space
4968 * Issue ioctl functions to devices. This is normally called by the
4969 * user space syscall interfaces but can sometimes be useful for
4970 * other purposes. The return value is the return from the syscall if
4971 * positive or a negative errno code on error.
4974 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4980 /* One special case: SIOCGIFCONF takes ifconf argument
4981 and requires shared lock, because it sleeps writing
4985 if (cmd == SIOCGIFCONF) {
4987 ret = dev_ifconf(net, (char __user *) arg);
4991 if (cmd == SIOCGIFNAME)
4992 return dev_ifname(net, (struct ifreq __user *)arg);
4994 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4997 ifr.ifr_name[IFNAMSIZ-1] = 0;
4999 colon = strchr(ifr.ifr_name, ':');
5004 * See which interface the caller is talking about.
5009 * These ioctl calls:
5010 * - can be done by all.
5011 * - atomic and do not require locking.
5022 dev_load(net, ifr.ifr_name);
5024 ret = dev_ifsioc_locked(net, &ifr, cmd);
5029 if (copy_to_user(arg, &ifr,
5030 sizeof(struct ifreq)))
5036 dev_load(net, ifr.ifr_name);
5038 ret = dev_ethtool(net, &ifr);
5043 if (copy_to_user(arg, &ifr,
5044 sizeof(struct ifreq)))
5050 * These ioctl calls:
5051 * - require superuser power.
5052 * - require strict serialization.
5058 if (!capable(CAP_NET_ADMIN))
5060 dev_load(net, ifr.ifr_name);
5062 ret = dev_ifsioc(net, &ifr, cmd);
5067 if (copy_to_user(arg, &ifr,
5068 sizeof(struct ifreq)))
5074 * These ioctl calls:
5075 * - require superuser power.
5076 * - require strict serialization.
5077 * - do not return a value
5087 case SIOCSIFHWBROADCAST:
5090 case SIOCBONDENSLAVE:
5091 case SIOCBONDRELEASE:
5092 case SIOCBONDSETHWADDR:
5093 case SIOCBONDCHANGEACTIVE:
5097 if (!capable(CAP_NET_ADMIN))
5100 case SIOCBONDSLAVEINFOQUERY:
5101 case SIOCBONDINFOQUERY:
5102 dev_load(net, ifr.ifr_name);
5104 ret = dev_ifsioc(net, &ifr, cmd);
5109 /* Get the per device memory space. We can add this but
5110 * currently do not support it */
5112 /* Set the per device memory buffer space.
5113 * Not applicable in our case */
5118 * Unknown or private ioctl.
5121 if (cmd == SIOCWANDEV ||
5122 (cmd >= SIOCDEVPRIVATE &&
5123 cmd <= SIOCDEVPRIVATE + 15)) {
5124 dev_load(net, ifr.ifr_name);
5126 ret = dev_ifsioc(net, &ifr, cmd);
5128 if (!ret && copy_to_user(arg, &ifr,
5129 sizeof(struct ifreq)))
5133 /* Take care of Wireless Extensions */
5134 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5135 return wext_handle_ioctl(net, &ifr, cmd, arg);
5142 * dev_new_index - allocate an ifindex
5143 * @net: the applicable net namespace
5145 * Returns a suitable unique value for a new device interface
5146 * number. The caller must hold the rtnl semaphore or the
5147 * dev_base_lock to be sure it remains unique.
5149 static int dev_new_index(struct net *net)
5155 if (!__dev_get_by_index(net, ifindex))
5160 /* Delayed registration/unregisteration */
5161 static LIST_HEAD(net_todo_list);
5163 static void net_set_todo(struct net_device *dev)
5165 list_add_tail(&dev->todo_list, &net_todo_list);
5168 static void rollback_registered_many(struct list_head *head)
5170 struct net_device *dev, *tmp;
5172 BUG_ON(dev_boot_phase);
5175 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5176 /* Some devices call without registering
5177 * for initialization unwind. Remove those
5178 * devices and proceed with the remaining.
5180 if (dev->reg_state == NETREG_UNINITIALIZED) {
5181 pr_debug("unregister_netdevice: device %s/%p never "
5182 "was registered\n", dev->name, dev);
5185 list_del(&dev->unreg_list);
5188 dev->dismantle = true;
5189 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5192 /* If device is running, close it first. */
5193 dev_close_many(head);
5195 list_for_each_entry(dev, head, unreg_list) {
5196 /* And unlink it from device chain. */
5197 unlist_netdevice(dev);
5199 dev->reg_state = NETREG_UNREGISTERING;
5204 list_for_each_entry(dev, head, unreg_list) {
5205 /* Shutdown queueing discipline. */
5209 /* Notify protocols, that we are about to destroy
5210 this device. They should clean all the things.
5212 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5214 if (!dev->rtnl_link_ops ||
5215 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5216 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5219 * Flush the unicast and multicast chains
5224 if (dev->netdev_ops->ndo_uninit)
5225 dev->netdev_ops->ndo_uninit(dev);
5227 /* Notifier chain MUST detach us from master device. */
5228 WARN_ON(dev->master);
5230 /* Remove entries from kobject tree */
5231 netdev_unregister_kobject(dev);
5234 /* Process any work delayed until the end of the batch */
5235 dev = list_first_entry(head, struct net_device, unreg_list);
5236 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5240 list_for_each_entry(dev, head, unreg_list)
5244 static void rollback_registered(struct net_device *dev)
5248 list_add(&dev->unreg_list, &single);
5249 rollback_registered_many(&single);
5253 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5255 /* Fix illegal checksum combinations */
5256 if ((features & NETIF_F_HW_CSUM) &&
5257 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5258 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5259 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5262 if ((features & NETIF_F_NO_CSUM) &&
5263 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5264 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5265 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5268 /* Fix illegal SG+CSUM combinations. */
5269 if ((features & NETIF_F_SG) &&
5270 !(features & NETIF_F_ALL_CSUM)) {
5272 "Dropping NETIF_F_SG since no checksum feature.\n");
5273 features &= ~NETIF_F_SG;
5276 /* TSO requires that SG is present as well. */
5277 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5278 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5279 features &= ~NETIF_F_ALL_TSO;
5282 /* TSO ECN requires that TSO is present as well. */
5283 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5284 features &= ~NETIF_F_TSO_ECN;
5286 /* Software GSO depends on SG. */
5287 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5288 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5289 features &= ~NETIF_F_GSO;
5292 /* UFO needs SG and checksumming */
5293 if (features & NETIF_F_UFO) {
5294 /* maybe split UFO into V4 and V6? */
5295 if (!((features & NETIF_F_GEN_CSUM) ||
5296 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5297 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5299 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5300 features &= ~NETIF_F_UFO;
5303 if (!(features & NETIF_F_SG)) {
5305 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5306 features &= ~NETIF_F_UFO;
5313 int __netdev_update_features(struct net_device *dev)
5320 features = netdev_get_wanted_features(dev);
5322 if (dev->netdev_ops->ndo_fix_features)
5323 features = dev->netdev_ops->ndo_fix_features(dev, features);
5325 /* driver might be less strict about feature dependencies */
5326 features = netdev_fix_features(dev, features);
5328 if (dev->features == features)
5331 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5332 dev->features, features);
5334 if (dev->netdev_ops->ndo_set_features)
5335 err = dev->netdev_ops->ndo_set_features(dev, features);
5337 if (unlikely(err < 0)) {
5339 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5340 err, features, dev->features);
5345 dev->features = features;
5351 * netdev_update_features - recalculate device features
5352 * @dev: the device to check
5354 * Recalculate dev->features set and send notifications if it
5355 * has changed. Should be called after driver or hardware dependent
5356 * conditions might have changed that influence the features.
5358 void netdev_update_features(struct net_device *dev)
5360 if (__netdev_update_features(dev))
5361 netdev_features_change(dev);
5363 EXPORT_SYMBOL(netdev_update_features);
5366 * netdev_change_features - recalculate device features
5367 * @dev: the device to check
5369 * Recalculate dev->features set and send notifications even
5370 * if they have not changed. Should be called instead of
5371 * netdev_update_features() if also dev->vlan_features might
5372 * have changed to allow the changes to be propagated to stacked
5375 void netdev_change_features(struct net_device *dev)
5377 __netdev_update_features(dev);
5378 netdev_features_change(dev);
5380 EXPORT_SYMBOL(netdev_change_features);
5383 * netif_stacked_transfer_operstate - transfer operstate
5384 * @rootdev: the root or lower level device to transfer state from
5385 * @dev: the device to transfer operstate to
5387 * Transfer operational state from root to device. This is normally
5388 * called when a stacking relationship exists between the root
5389 * device and the device(a leaf device).
5391 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5392 struct net_device *dev)
5394 if (rootdev->operstate == IF_OPER_DORMANT)
5395 netif_dormant_on(dev);
5397 netif_dormant_off(dev);
5399 if (netif_carrier_ok(rootdev)) {
5400 if (!netif_carrier_ok(dev))
5401 netif_carrier_on(dev);
5403 if (netif_carrier_ok(dev))
5404 netif_carrier_off(dev);
5407 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5410 static int netif_alloc_rx_queues(struct net_device *dev)
5412 unsigned int i, count = dev->num_rx_queues;
5413 struct netdev_rx_queue *rx;
5417 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5419 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5424 for (i = 0; i < count; i++)
5430 static void netdev_init_one_queue(struct net_device *dev,
5431 struct netdev_queue *queue, void *_unused)
5433 /* Initialize queue lock */
5434 spin_lock_init(&queue->_xmit_lock);
5435 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5436 queue->xmit_lock_owner = -1;
5437 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5441 static int netif_alloc_netdev_queues(struct net_device *dev)
5443 unsigned int count = dev->num_tx_queues;
5444 struct netdev_queue *tx;
5448 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5450 pr_err("netdev: Unable to allocate %u tx queues.\n",
5456 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5457 spin_lock_init(&dev->tx_global_lock);
5463 * register_netdevice - register a network device
5464 * @dev: device to register
5466 * Take a completed network device structure and add it to the kernel
5467 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5468 * chain. 0 is returned on success. A negative errno code is returned
5469 * on a failure to set up the device, or if the name is a duplicate.
5471 * Callers must hold the rtnl semaphore. You may want
5472 * register_netdev() instead of this.
5475 * The locking appears insufficient to guarantee two parallel registers
5476 * will not get the same name.
5479 int register_netdevice(struct net_device *dev)
5482 struct net *net = dev_net(dev);
5484 BUG_ON(dev_boot_phase);
5489 /* When net_device's are persistent, this will be fatal. */
5490 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5493 spin_lock_init(&dev->addr_list_lock);
5494 netdev_set_addr_lockdep_class(dev);
5498 ret = dev_get_valid_name(dev, dev->name);
5502 /* Init, if this function is available */
5503 if (dev->netdev_ops->ndo_init) {
5504 ret = dev->netdev_ops->ndo_init(dev);
5512 dev->ifindex = dev_new_index(net);
5513 if (dev->iflink == -1)
5514 dev->iflink = dev->ifindex;
5516 /* Transfer changeable features to wanted_features and enable
5517 * software offloads (GSO and GRO).
5519 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5520 dev->features |= NETIF_F_SOFT_FEATURES;
5521 dev->wanted_features = dev->features & dev->hw_features;
5523 /* Turn on no cache copy if HW is doing checksum */
5524 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5525 if ((dev->features & NETIF_F_ALL_CSUM) &&
5526 !(dev->features & NETIF_F_NO_CSUM)) {
5527 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5528 dev->features |= NETIF_F_NOCACHE_COPY;
5531 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5533 dev->vlan_features |= NETIF_F_HIGHDMA;
5535 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5536 ret = notifier_to_errno(ret);
5540 ret = netdev_register_kobject(dev);
5543 dev->reg_state = NETREG_REGISTERED;
5545 __netdev_update_features(dev);
5548 * Default initial state at registry is that the
5549 * device is present.
5552 set_bit(__LINK_STATE_PRESENT, &dev->state);
5554 dev_init_scheduler(dev);
5556 list_netdevice(dev);
5558 /* Notify protocols, that a new device appeared. */
5559 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5560 ret = notifier_to_errno(ret);
5562 rollback_registered(dev);
5563 dev->reg_state = NETREG_UNREGISTERED;
5566 * Prevent userspace races by waiting until the network
5567 * device is fully setup before sending notifications.
5569 if (!dev->rtnl_link_ops ||
5570 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5571 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5577 if (dev->netdev_ops->ndo_uninit)
5578 dev->netdev_ops->ndo_uninit(dev);
5581 EXPORT_SYMBOL(register_netdevice);
5584 * init_dummy_netdev - init a dummy network device for NAPI
5585 * @dev: device to init
5587 * This takes a network device structure and initialize the minimum
5588 * amount of fields so it can be used to schedule NAPI polls without
5589 * registering a full blown interface. This is to be used by drivers
5590 * that need to tie several hardware interfaces to a single NAPI
5591 * poll scheduler due to HW limitations.
5593 int init_dummy_netdev(struct net_device *dev)
5595 /* Clear everything. Note we don't initialize spinlocks
5596 * are they aren't supposed to be taken by any of the
5597 * NAPI code and this dummy netdev is supposed to be
5598 * only ever used for NAPI polls
5600 memset(dev, 0, sizeof(struct net_device));
5602 /* make sure we BUG if trying to hit standard
5603 * register/unregister code path
5605 dev->reg_state = NETREG_DUMMY;
5607 /* NAPI wants this */
5608 INIT_LIST_HEAD(&dev->napi_list);
5610 /* a dummy interface is started by default */
5611 set_bit(__LINK_STATE_PRESENT, &dev->state);
5612 set_bit(__LINK_STATE_START, &dev->state);
5614 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5615 * because users of this 'device' dont need to change
5621 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5625 * register_netdev - register a network device
5626 * @dev: device to register
5628 * Take a completed network device structure and add it to the kernel
5629 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5630 * chain. 0 is returned on success. A negative errno code is returned
5631 * on a failure to set up the device, or if the name is a duplicate.
5633 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5634 * and expands the device name if you passed a format string to
5637 int register_netdev(struct net_device *dev)
5642 err = register_netdevice(dev);
5646 EXPORT_SYMBOL(register_netdev);
5648 int netdev_refcnt_read(const struct net_device *dev)
5652 for_each_possible_cpu(i)
5653 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5656 EXPORT_SYMBOL(netdev_refcnt_read);
5659 * netdev_wait_allrefs - wait until all references are gone.
5661 * This is called when unregistering network devices.
5663 * Any protocol or device that holds a reference should register
5664 * for netdevice notification, and cleanup and put back the
5665 * reference if they receive an UNREGISTER event.
5666 * We can get stuck here if buggy protocols don't correctly
5669 static void netdev_wait_allrefs(struct net_device *dev)
5671 unsigned long rebroadcast_time, warning_time;
5674 linkwatch_forget_dev(dev);
5676 rebroadcast_time = warning_time = jiffies;
5677 refcnt = netdev_refcnt_read(dev);
5679 while (refcnt != 0) {
5680 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5683 /* Rebroadcast unregister notification */
5684 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5685 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5686 * should have already handle it the first time */
5688 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5690 /* We must not have linkwatch events
5691 * pending on unregister. If this
5692 * happens, we simply run the queue
5693 * unscheduled, resulting in a noop
5696 linkwatch_run_queue();
5701 rebroadcast_time = jiffies;
5706 refcnt = netdev_refcnt_read(dev);
5708 if (time_after(jiffies, warning_time + 10 * HZ)) {
5709 printk(KERN_EMERG "unregister_netdevice: "
5710 "waiting for %s to become free. Usage "
5713 warning_time = jiffies;
5722 * register_netdevice(x1);
5723 * register_netdevice(x2);
5725 * unregister_netdevice(y1);
5726 * unregister_netdevice(y2);
5732 * We are invoked by rtnl_unlock().
5733 * This allows us to deal with problems:
5734 * 1) We can delete sysfs objects which invoke hotplug
5735 * without deadlocking with linkwatch via keventd.
5736 * 2) Since we run with the RTNL semaphore not held, we can sleep
5737 * safely in order to wait for the netdev refcnt to drop to zero.
5739 * We must not return until all unregister events added during
5740 * the interval the lock was held have been completed.
5742 void netdev_run_todo(void)
5744 struct list_head list;
5746 /* Snapshot list, allow later requests */
5747 list_replace_init(&net_todo_list, &list);
5751 while (!list_empty(&list)) {
5752 struct net_device *dev
5753 = list_first_entry(&list, struct net_device, todo_list);
5754 list_del(&dev->todo_list);
5756 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5757 printk(KERN_ERR "network todo '%s' but state %d\n",
5758 dev->name, dev->reg_state);
5763 dev->reg_state = NETREG_UNREGISTERED;
5765 on_each_cpu(flush_backlog, dev, 1);
5767 netdev_wait_allrefs(dev);
5770 BUG_ON(netdev_refcnt_read(dev));
5771 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5772 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5773 WARN_ON(dev->dn_ptr);
5775 if (dev->destructor)
5776 dev->destructor(dev);
5778 /* Free network device */
5779 kobject_put(&dev->dev.kobj);
5783 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5784 * fields in the same order, with only the type differing.
5786 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5787 const struct net_device_stats *netdev_stats)
5789 #if BITS_PER_LONG == 64
5790 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5791 memcpy(stats64, netdev_stats, sizeof(*stats64));
5793 size_t i, n = sizeof(*stats64) / sizeof(u64);
5794 const unsigned long *src = (const unsigned long *)netdev_stats;
5795 u64 *dst = (u64 *)stats64;
5797 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5798 sizeof(*stats64) / sizeof(u64));
5799 for (i = 0; i < n; i++)
5805 * dev_get_stats - get network device statistics
5806 * @dev: device to get statistics from
5807 * @storage: place to store stats
5809 * Get network statistics from device. Return @storage.
5810 * The device driver may provide its own method by setting
5811 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5812 * otherwise the internal statistics structure is used.
5814 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5815 struct rtnl_link_stats64 *storage)
5817 const struct net_device_ops *ops = dev->netdev_ops;
5819 if (ops->ndo_get_stats64) {
5820 memset(storage, 0, sizeof(*storage));
5821 ops->ndo_get_stats64(dev, storage);
5822 } else if (ops->ndo_get_stats) {
5823 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5825 netdev_stats_to_stats64(storage, &dev->stats);
5827 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5830 EXPORT_SYMBOL(dev_get_stats);
5832 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5834 struct netdev_queue *queue = dev_ingress_queue(dev);
5836 #ifdef CONFIG_NET_CLS_ACT
5839 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5842 netdev_init_one_queue(dev, queue, NULL);
5843 queue->qdisc = &noop_qdisc;
5844 queue->qdisc_sleeping = &noop_qdisc;
5845 rcu_assign_pointer(dev->ingress_queue, queue);
5851 * alloc_netdev_mqs - allocate network device
5852 * @sizeof_priv: size of private data to allocate space for
5853 * @name: device name format string
5854 * @setup: callback to initialize device
5855 * @txqs: the number of TX subqueues to allocate
5856 * @rxqs: the number of RX subqueues to allocate
5858 * Allocates a struct net_device with private data area for driver use
5859 * and performs basic initialization. Also allocates subquue structs
5860 * for each queue on the device.
5862 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5863 void (*setup)(struct net_device *),
5864 unsigned int txqs, unsigned int rxqs)
5866 struct net_device *dev;
5868 struct net_device *p;
5870 BUG_ON(strlen(name) >= sizeof(dev->name));
5873 pr_err("alloc_netdev: Unable to allocate device "
5874 "with zero queues.\n");
5880 pr_err("alloc_netdev: Unable to allocate device "
5881 "with zero RX queues.\n");
5886 alloc_size = sizeof(struct net_device);
5888 /* ensure 32-byte alignment of private area */
5889 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5890 alloc_size += sizeof_priv;
5892 /* ensure 32-byte alignment of whole construct */
5893 alloc_size += NETDEV_ALIGN - 1;
5895 p = kzalloc(alloc_size, GFP_KERNEL);
5897 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5901 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5902 dev->padded = (char *)dev - (char *)p;
5904 dev->pcpu_refcnt = alloc_percpu(int);
5905 if (!dev->pcpu_refcnt)
5908 if (dev_addr_init(dev))
5914 dev_net_set(dev, &init_net);
5916 dev->gso_max_size = GSO_MAX_SIZE;
5918 INIT_LIST_HEAD(&dev->napi_list);
5919 INIT_LIST_HEAD(&dev->unreg_list);
5920 INIT_LIST_HEAD(&dev->link_watch_list);
5921 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5924 dev->num_tx_queues = txqs;
5925 dev->real_num_tx_queues = txqs;
5926 if (netif_alloc_netdev_queues(dev))
5930 dev->num_rx_queues = rxqs;
5931 dev->real_num_rx_queues = rxqs;
5932 if (netif_alloc_rx_queues(dev))
5936 strcpy(dev->name, name);
5937 dev->group = INIT_NETDEV_GROUP;
5945 free_percpu(dev->pcpu_refcnt);
5955 EXPORT_SYMBOL(alloc_netdev_mqs);
5958 * free_netdev - free network device
5961 * This function does the last stage of destroying an allocated device
5962 * interface. The reference to the device object is released.
5963 * If this is the last reference then it will be freed.
5965 void free_netdev(struct net_device *dev)
5967 struct napi_struct *p, *n;
5969 release_net(dev_net(dev));
5976 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5978 /* Flush device addresses */
5979 dev_addr_flush(dev);
5981 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5984 free_percpu(dev->pcpu_refcnt);
5985 dev->pcpu_refcnt = NULL;
5987 /* Compatibility with error handling in drivers */
5988 if (dev->reg_state == NETREG_UNINITIALIZED) {
5989 kfree((char *)dev - dev->padded);
5993 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5994 dev->reg_state = NETREG_RELEASED;
5996 /* will free via device release */
5997 put_device(&dev->dev);
5999 EXPORT_SYMBOL(free_netdev);
6002 * synchronize_net - Synchronize with packet receive processing
6004 * Wait for packets currently being received to be done.
6005 * Does not block later packets from starting.
6007 void synchronize_net(void)
6010 if (rtnl_is_locked())
6011 synchronize_rcu_expedited();
6015 EXPORT_SYMBOL(synchronize_net);
6018 * unregister_netdevice_queue - remove device from the kernel
6022 * This function shuts down a device interface and removes it
6023 * from the kernel tables.
6024 * If head not NULL, device is queued to be unregistered later.
6026 * Callers must hold the rtnl semaphore. You may want
6027 * unregister_netdev() instead of this.
6030 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6035 list_move_tail(&dev->unreg_list, head);
6037 rollback_registered(dev);
6038 /* Finish processing unregister after unlock */
6042 EXPORT_SYMBOL(unregister_netdevice_queue);
6045 * unregister_netdevice_many - unregister many devices
6046 * @head: list of devices
6048 void unregister_netdevice_many(struct list_head *head)
6050 struct net_device *dev;
6052 if (!list_empty(head)) {
6053 rollback_registered_many(head);
6054 list_for_each_entry(dev, head, unreg_list)
6058 EXPORT_SYMBOL(unregister_netdevice_many);
6061 * unregister_netdev - remove device from the kernel
6064 * This function shuts down a device interface and removes it
6065 * from the kernel tables.
6067 * This is just a wrapper for unregister_netdevice that takes
6068 * the rtnl semaphore. In general you want to use this and not
6069 * unregister_netdevice.
6071 void unregister_netdev(struct net_device *dev)
6074 unregister_netdevice(dev);
6077 EXPORT_SYMBOL(unregister_netdev);
6080 * dev_change_net_namespace - move device to different nethost namespace
6082 * @net: network namespace
6083 * @pat: If not NULL name pattern to try if the current device name
6084 * is already taken in the destination network namespace.
6086 * This function shuts down a device interface and moves it
6087 * to a new network namespace. On success 0 is returned, on
6088 * a failure a netagive errno code is returned.
6090 * Callers must hold the rtnl semaphore.
6093 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6099 /* Don't allow namespace local devices to be moved. */
6101 if (dev->features & NETIF_F_NETNS_LOCAL)
6104 /* Ensure the device has been registrered */
6106 if (dev->reg_state != NETREG_REGISTERED)
6109 /* Get out if there is nothing todo */
6111 if (net_eq(dev_net(dev), net))
6114 /* Pick the destination device name, and ensure
6115 * we can use it in the destination network namespace.
6118 if (__dev_get_by_name(net, dev->name)) {
6119 /* We get here if we can't use the current device name */
6122 if (dev_get_valid_name(dev, pat) < 0)
6127 * And now a mini version of register_netdevice unregister_netdevice.
6130 /* If device is running close it first. */
6133 /* And unlink it from device chain */
6135 unlist_netdevice(dev);
6139 /* Shutdown queueing discipline. */
6142 /* Notify protocols, that we are about to destroy
6143 this device. They should clean all the things.
6145 Note that dev->reg_state stays at NETREG_REGISTERED.
6146 This is wanted because this way 8021q and macvlan know
6147 the device is just moving and can keep their slaves up.
6149 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6150 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6153 * Flush the unicast and multicast chains
6158 /* Actually switch the network namespace */
6159 dev_net_set(dev, net);
6161 /* If there is an ifindex conflict assign a new one */
6162 if (__dev_get_by_index(net, dev->ifindex)) {
6163 int iflink = (dev->iflink == dev->ifindex);
6164 dev->ifindex = dev_new_index(net);
6166 dev->iflink = dev->ifindex;
6169 /* Fixup kobjects */
6170 err = device_rename(&dev->dev, dev->name);
6173 /* Add the device back in the hashes */
6174 list_netdevice(dev);
6176 /* Notify protocols, that a new device appeared. */
6177 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6180 * Prevent userspace races by waiting until the network
6181 * device is fully setup before sending notifications.
6183 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6190 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6192 static int dev_cpu_callback(struct notifier_block *nfb,
6193 unsigned long action,
6196 struct sk_buff **list_skb;
6197 struct sk_buff *skb;
6198 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6199 struct softnet_data *sd, *oldsd;
6201 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6204 local_irq_disable();
6205 cpu = smp_processor_id();
6206 sd = &per_cpu(softnet_data, cpu);
6207 oldsd = &per_cpu(softnet_data, oldcpu);
6209 /* Find end of our completion_queue. */
6210 list_skb = &sd->completion_queue;
6212 list_skb = &(*list_skb)->next;
6213 /* Append completion queue from offline CPU. */
6214 *list_skb = oldsd->completion_queue;
6215 oldsd->completion_queue = NULL;
6217 /* Append output queue from offline CPU. */
6218 if (oldsd->output_queue) {
6219 *sd->output_queue_tailp = oldsd->output_queue;
6220 sd->output_queue_tailp = oldsd->output_queue_tailp;
6221 oldsd->output_queue = NULL;
6222 oldsd->output_queue_tailp = &oldsd->output_queue;
6224 /* Append NAPI poll list from offline CPU. */
6225 if (!list_empty(&oldsd->poll_list)) {
6226 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6227 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6230 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6233 /* Process offline CPU's input_pkt_queue */
6234 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6236 input_queue_head_incr(oldsd);
6238 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6240 input_queue_head_incr(oldsd);
6248 * netdev_increment_features - increment feature set by one
6249 * @all: current feature set
6250 * @one: new feature set
6251 * @mask: mask feature set
6253 * Computes a new feature set after adding a device with feature set
6254 * @one to the master device with current feature set @all. Will not
6255 * enable anything that is off in @mask. Returns the new feature set.
6257 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6259 if (mask & NETIF_F_GEN_CSUM)
6260 mask |= NETIF_F_ALL_CSUM;
6261 mask |= NETIF_F_VLAN_CHALLENGED;
6263 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6264 all &= one | ~NETIF_F_ALL_FOR_ALL;
6266 /* If device needs checksumming, downgrade to it. */
6267 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6268 all &= ~NETIF_F_NO_CSUM;
6270 /* If one device supports hw checksumming, set for all. */
6271 if (all & NETIF_F_GEN_CSUM)
6272 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6276 EXPORT_SYMBOL(netdev_increment_features);
6278 static struct hlist_head *netdev_create_hash(void)
6281 struct hlist_head *hash;
6283 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6285 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6286 INIT_HLIST_HEAD(&hash[i]);
6291 /* Initialize per network namespace state */
6292 static int __net_init netdev_init(struct net *net)
6294 INIT_LIST_HEAD(&net->dev_base_head);
6296 net->dev_name_head = netdev_create_hash();
6297 if (net->dev_name_head == NULL)
6300 net->dev_index_head = netdev_create_hash();
6301 if (net->dev_index_head == NULL)
6307 kfree(net->dev_name_head);
6313 * netdev_drivername - network driver for the device
6314 * @dev: network device
6316 * Determine network driver for device.
6318 const char *netdev_drivername(const struct net_device *dev)
6320 const struct device_driver *driver;
6321 const struct device *parent;
6322 const char *empty = "";
6324 parent = dev->dev.parent;
6328 driver = parent->driver;
6329 if (driver && driver->name)
6330 return driver->name;
6334 static int __netdev_printk(const char *level, const struct net_device *dev,
6335 struct va_format *vaf)
6339 if (dev && dev->dev.parent)
6340 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6341 netdev_name(dev), vaf);
6343 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6345 r = printk("%s(NULL net_device): %pV", level, vaf);
6350 int netdev_printk(const char *level, const struct net_device *dev,
6351 const char *format, ...)
6353 struct va_format vaf;
6357 va_start(args, format);
6362 r = __netdev_printk(level, dev, &vaf);
6367 EXPORT_SYMBOL(netdev_printk);
6369 #define define_netdev_printk_level(func, level) \
6370 int func(const struct net_device *dev, const char *fmt, ...) \
6373 struct va_format vaf; \
6376 va_start(args, fmt); \
6381 r = __netdev_printk(level, dev, &vaf); \
6386 EXPORT_SYMBOL(func);
6388 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6389 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6390 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6391 define_netdev_printk_level(netdev_err, KERN_ERR);
6392 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6393 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6394 define_netdev_printk_level(netdev_info, KERN_INFO);
6396 static void __net_exit netdev_exit(struct net *net)
6398 kfree(net->dev_name_head);
6399 kfree(net->dev_index_head);
6402 static struct pernet_operations __net_initdata netdev_net_ops = {
6403 .init = netdev_init,
6404 .exit = netdev_exit,
6407 static void __net_exit default_device_exit(struct net *net)
6409 struct net_device *dev, *aux;
6411 * Push all migratable network devices back to the
6412 * initial network namespace
6415 for_each_netdev_safe(net, dev, aux) {
6417 char fb_name[IFNAMSIZ];
6419 /* Ignore unmoveable devices (i.e. loopback) */
6420 if (dev->features & NETIF_F_NETNS_LOCAL)
6423 /* Leave virtual devices for the generic cleanup */
6424 if (dev->rtnl_link_ops)
6427 /* Push remaining network devices to init_net */
6428 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6429 err = dev_change_net_namespace(dev, &init_net, fb_name);
6431 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6432 __func__, dev->name, err);
6439 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6441 /* At exit all network devices most be removed from a network
6442 * namespace. Do this in the reverse order of registration.
6443 * Do this across as many network namespaces as possible to
6444 * improve batching efficiency.
6446 struct net_device *dev;
6448 LIST_HEAD(dev_kill_list);
6451 list_for_each_entry(net, net_list, exit_list) {
6452 for_each_netdev_reverse(net, dev) {
6453 if (dev->rtnl_link_ops)
6454 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6456 unregister_netdevice_queue(dev, &dev_kill_list);
6459 unregister_netdevice_many(&dev_kill_list);
6460 list_del(&dev_kill_list);
6464 static struct pernet_operations __net_initdata default_device_ops = {
6465 .exit = default_device_exit,
6466 .exit_batch = default_device_exit_batch,
6470 * Initialize the DEV module. At boot time this walks the device list and
6471 * unhooks any devices that fail to initialise (normally hardware not
6472 * present) and leaves us with a valid list of present and active devices.
6477 * This is called single threaded during boot, so no need
6478 * to take the rtnl semaphore.
6480 static int __init net_dev_init(void)
6482 int i, rc = -ENOMEM;
6484 BUG_ON(!dev_boot_phase);
6486 if (dev_proc_init())
6489 if (netdev_kobject_init())
6492 INIT_LIST_HEAD(&ptype_all);
6493 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6494 INIT_LIST_HEAD(&ptype_base[i]);
6496 if (register_pernet_subsys(&netdev_net_ops))
6500 * Initialise the packet receive queues.
6503 for_each_possible_cpu(i) {
6504 struct softnet_data *sd = &per_cpu(softnet_data, i);
6506 memset(sd, 0, sizeof(*sd));
6507 skb_queue_head_init(&sd->input_pkt_queue);
6508 skb_queue_head_init(&sd->process_queue);
6509 sd->completion_queue = NULL;
6510 INIT_LIST_HEAD(&sd->poll_list);
6511 sd->output_queue = NULL;
6512 sd->output_queue_tailp = &sd->output_queue;
6514 sd->csd.func = rps_trigger_softirq;
6520 sd->backlog.poll = process_backlog;
6521 sd->backlog.weight = weight_p;
6522 sd->backlog.gro_list = NULL;
6523 sd->backlog.gro_count = 0;
6528 /* The loopback device is special if any other network devices
6529 * is present in a network namespace the loopback device must
6530 * be present. Since we now dynamically allocate and free the
6531 * loopback device ensure this invariant is maintained by
6532 * keeping the loopback device as the first device on the
6533 * list of network devices. Ensuring the loopback devices
6534 * is the first device that appears and the last network device
6537 if (register_pernet_device(&loopback_net_ops))
6540 if (register_pernet_device(&default_device_ops))
6543 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6544 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6546 hotcpu_notifier(dev_cpu_callback, 0);
6554 subsys_initcall(net_dev_init);
6556 static int __init initialize_hashrnd(void)
6558 get_random_bytes(&hashrnd, sizeof(hashrnd));
6562 late_initcall_sync(initialize_hashrnd);