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/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
103 #include <linux/if_bridge.h>
104 #include <linux/if_macvlan.h>
106 #include <net/pkt_sched.h>
107 #include <net/checksum.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>
132 #include "net-sysfs.h"
134 /* Instead of increasing this, you should create a hash table. */
135 #define MAX_GRO_SKBS 8
137 /* This should be increased if a protocol with a bigger head is added. */
138 #define GRO_MAX_HEAD (MAX_HEADER + 128)
141 * The list of packet types we will receive (as opposed to discard)
142 * and the routines to invoke.
144 * Why 16. Because with 16 the only overlap we get on a hash of the
145 * low nibble of the protocol value is RARP/SNAP/X.25.
147 * NOTE: That is no longer true with the addition of VLAN tags. Not
148 * sure which should go first, but I bet it won't make much
149 * difference if we are running VLANs. The good news is that
150 * this protocol won't be in the list unless compiled in, so
151 * the average user (w/out VLANs) will not be adversely affected.
168 #define PTYPE_HASH_SIZE (16)
169 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
171 static DEFINE_SPINLOCK(ptype_lock);
172 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
173 static struct list_head ptype_all __read_mostly; /* Taps */
176 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
179 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
181 * Writers must hold the rtnl semaphore while they loop through the
182 * dev_base_head list, and hold dev_base_lock for writing when they do the
183 * actual updates. This allows pure readers to access the list even
184 * while a writer is preparing to update it.
186 * To put it another way, dev_base_lock is held for writing only to
187 * protect against pure readers; the rtnl semaphore provides the
188 * protection against other writers.
190 * See, for example usages, register_netdevice() and
191 * unregister_netdevice(), which must be called with the rtnl
194 DEFINE_RWLOCK(dev_base_lock);
195 EXPORT_SYMBOL(dev_base_lock);
197 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
199 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
200 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
203 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
205 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
208 /* Device list insertion */
209 static int list_netdevice(struct net_device *dev)
211 struct net *net = dev_net(dev);
215 write_lock_bh(&dev_base_lock);
216 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
217 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
218 hlist_add_head_rcu(&dev->index_hlist,
219 dev_index_hash(net, dev->ifindex));
220 write_unlock_bh(&dev_base_lock);
224 /* Device list removal
225 * caller must respect a RCU grace period before freeing/reusing dev
227 static void unlist_netdevice(struct net_device *dev)
231 /* Unlink dev from the device chain */
232 write_lock_bh(&dev_base_lock);
233 list_del_rcu(&dev->dev_list);
234 hlist_del_rcu(&dev->name_hlist);
235 hlist_del_rcu(&dev->index_hlist);
236 write_unlock_bh(&dev_base_lock);
243 static RAW_NOTIFIER_HEAD(netdev_chain);
246 * Device drivers call our routines to queue packets here. We empty the
247 * queue in the local softnet handler.
250 DEFINE_PER_CPU(struct softnet_data, softnet_data);
251 EXPORT_PER_CPU_SYMBOL(softnet_data);
253 #ifdef CONFIG_LOCKDEP
255 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
256 * according to dev->type
258 static const unsigned short netdev_lock_type[] =
259 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
260 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
261 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
262 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
263 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
264 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
265 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
266 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
267 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
268 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
269 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
270 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
271 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
272 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
273 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
274 ARPHRD_VOID, ARPHRD_NONE};
276 static const char *const netdev_lock_name[] =
277 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
278 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
279 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
280 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
281 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
282 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
283 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
284 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
285 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
286 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
287 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
288 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
289 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
290 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
291 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
292 "_xmit_VOID", "_xmit_NONE"};
294 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
295 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
297 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
301 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
302 if (netdev_lock_type[i] == dev_type)
304 /* the last key is used by default */
305 return ARRAY_SIZE(netdev_lock_type) - 1;
308 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
309 unsigned short dev_type)
313 i = netdev_lock_pos(dev_type);
314 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
315 netdev_lock_name[i]);
318 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
322 i = netdev_lock_pos(dev->type);
323 lockdep_set_class_and_name(&dev->addr_list_lock,
324 &netdev_addr_lock_key[i],
325 netdev_lock_name[i]);
328 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
329 unsigned short dev_type)
332 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
337 /*******************************************************************************
339 Protocol management and registration routines
341 *******************************************************************************/
344 * Add a protocol ID to the list. Now that the input handler is
345 * smarter we can dispense with all the messy stuff that used to be
348 * BEWARE!!! Protocol handlers, mangling input packets,
349 * MUST BE last in hash buckets and checking protocol handlers
350 * MUST start from promiscuous ptype_all chain in net_bh.
351 * It is true now, do not change it.
352 * Explanation follows: if protocol handler, mangling packet, will
353 * be the first on list, it is not able to sense, that packet
354 * is cloned and should be copied-on-write, so that it will
355 * change it and subsequent readers will get broken packet.
360 * dev_add_pack - add packet handler
361 * @pt: packet type declaration
363 * Add a protocol handler to the networking stack. The passed &packet_type
364 * is linked into kernel lists and may not be freed until it has been
365 * removed from the kernel lists.
367 * This call does not sleep therefore it can not
368 * guarantee all CPU's that are in middle of receiving packets
369 * will see the new packet type (until the next received packet).
372 void dev_add_pack(struct packet_type *pt)
376 spin_lock_bh(&ptype_lock);
377 if (pt->type == htons(ETH_P_ALL))
378 list_add_rcu(&pt->list, &ptype_all);
380 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
381 list_add_rcu(&pt->list, &ptype_base[hash]);
383 spin_unlock_bh(&ptype_lock);
385 EXPORT_SYMBOL(dev_add_pack);
388 * __dev_remove_pack - remove packet handler
389 * @pt: packet type declaration
391 * Remove a protocol handler that was previously added to the kernel
392 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
393 * from the kernel lists and can be freed or reused once this function
396 * The packet type might still be in use by receivers
397 * and must not be freed until after all the CPU's have gone
398 * through a quiescent state.
400 void __dev_remove_pack(struct packet_type *pt)
402 struct list_head *head;
403 struct packet_type *pt1;
405 spin_lock_bh(&ptype_lock);
407 if (pt->type == htons(ETH_P_ALL))
410 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
412 list_for_each_entry(pt1, head, list) {
414 list_del_rcu(&pt->list);
419 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
421 spin_unlock_bh(&ptype_lock);
423 EXPORT_SYMBOL(__dev_remove_pack);
426 * dev_remove_pack - remove packet handler
427 * @pt: packet type declaration
429 * Remove a protocol handler that was previously added to the kernel
430 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
431 * from the kernel lists and can be freed or reused once this function
434 * This call sleeps to guarantee that no CPU is looking at the packet
437 void dev_remove_pack(struct packet_type *pt)
439 __dev_remove_pack(pt);
443 EXPORT_SYMBOL(dev_remove_pack);
445 /******************************************************************************
447 Device Boot-time Settings Routines
449 *******************************************************************************/
451 /* Boot time configuration table */
452 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
455 * netdev_boot_setup_add - add new setup entry
456 * @name: name of the device
457 * @map: configured settings for the device
459 * Adds new setup entry to the dev_boot_setup list. The function
460 * returns 0 on error and 1 on success. This is a generic routine to
463 static int netdev_boot_setup_add(char *name, struct ifmap *map)
465 struct netdev_boot_setup *s;
469 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
470 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
471 memset(s[i].name, 0, sizeof(s[i].name));
472 strlcpy(s[i].name, name, IFNAMSIZ);
473 memcpy(&s[i].map, map, sizeof(s[i].map));
478 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
482 * netdev_boot_setup_check - check boot time settings
483 * @dev: the netdevice
485 * Check boot time settings for the device.
486 * The found settings are set for the device to be used
487 * later in the device probing.
488 * Returns 0 if no settings found, 1 if they are.
490 int netdev_boot_setup_check(struct net_device *dev)
492 struct netdev_boot_setup *s = dev_boot_setup;
495 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
496 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
497 !strcmp(dev->name, s[i].name)) {
498 dev->irq = s[i].map.irq;
499 dev->base_addr = s[i].map.base_addr;
500 dev->mem_start = s[i].map.mem_start;
501 dev->mem_end = s[i].map.mem_end;
507 EXPORT_SYMBOL(netdev_boot_setup_check);
511 * netdev_boot_base - get address from boot time settings
512 * @prefix: prefix for network device
513 * @unit: id for network device
515 * Check boot time settings for the base address of 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.
520 unsigned long netdev_boot_base(const char *prefix, int unit)
522 const struct netdev_boot_setup *s = dev_boot_setup;
526 sprintf(name, "%s%d", prefix, unit);
529 * If device already registered then return base of 1
530 * to indicate not to probe for this interface
532 if (__dev_get_by_name(&init_net, name))
535 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
536 if (!strcmp(name, s[i].name))
537 return s[i].map.base_addr;
542 * Saves at boot time configured settings for any netdevice.
544 int __init netdev_boot_setup(char *str)
549 str = get_options(str, ARRAY_SIZE(ints), ints);
554 memset(&map, 0, sizeof(map));
558 map.base_addr = ints[2];
560 map.mem_start = ints[3];
562 map.mem_end = ints[4];
564 /* Add new entry to the list */
565 return netdev_boot_setup_add(str, &map);
568 __setup("netdev=", netdev_boot_setup);
570 /*******************************************************************************
572 Device Interface Subroutines
574 *******************************************************************************/
577 * __dev_get_by_name - find a device by its name
578 * @net: the applicable net namespace
579 * @name: name to find
581 * Find an interface by name. Must be called under RTNL semaphore
582 * or @dev_base_lock. If the name is found a pointer to the device
583 * is returned. If the name is not found then %NULL is returned. The
584 * reference counters are not incremented so the caller must be
585 * careful with locks.
588 struct net_device *__dev_get_by_name(struct net *net, const char *name)
590 struct hlist_node *p;
591 struct net_device *dev;
592 struct hlist_head *head = dev_name_hash(net, name);
594 hlist_for_each_entry(dev, p, head, name_hlist)
595 if (!strncmp(dev->name, name, IFNAMSIZ))
600 EXPORT_SYMBOL(__dev_get_by_name);
603 * dev_get_by_name_rcu - find a device by its name
604 * @net: the applicable net namespace
605 * @name: name to find
607 * Find an interface by name.
608 * If the name is found a pointer to the device is returned.
609 * If the name is not found then %NULL is returned.
610 * The reference counters are not incremented so the caller must be
611 * careful with locks. The caller must hold RCU lock.
614 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
616 struct hlist_node *p;
617 struct net_device *dev;
618 struct hlist_head *head = dev_name_hash(net, name);
620 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
621 if (!strncmp(dev->name, name, IFNAMSIZ))
626 EXPORT_SYMBOL(dev_get_by_name_rcu);
629 * dev_get_by_name - find a device by its name
630 * @net: the applicable net namespace
631 * @name: name to find
633 * Find an interface by name. This can be called from any
634 * context and does its own locking. The returned handle has
635 * the usage count incremented and the caller must use dev_put() to
636 * release it when it is no longer needed. %NULL is returned if no
637 * matching device is found.
640 struct net_device *dev_get_by_name(struct net *net, const char *name)
642 struct net_device *dev;
645 dev = dev_get_by_name_rcu(net, name);
651 EXPORT_SYMBOL(dev_get_by_name);
654 * __dev_get_by_index - find a device by its ifindex
655 * @net: the applicable net namespace
656 * @ifindex: index of device
658 * Search for an interface by index. Returns %NULL if the device
659 * is not found or a pointer to the device. The device has not
660 * had its reference counter increased so the caller must be careful
661 * about locking. The caller must hold either the RTNL semaphore
665 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
667 struct hlist_node *p;
668 struct net_device *dev;
669 struct hlist_head *head = dev_index_hash(net, ifindex);
671 hlist_for_each_entry(dev, p, head, index_hlist)
672 if (dev->ifindex == ifindex)
677 EXPORT_SYMBOL(__dev_get_by_index);
680 * dev_get_by_index_rcu - find a device by its ifindex
681 * @net: the applicable net namespace
682 * @ifindex: index of device
684 * Search for an interface by index. Returns %NULL if the device
685 * is not found or a pointer to the device. The device has not
686 * had its reference counter increased so the caller must be careful
687 * about locking. The caller must hold RCU lock.
690 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
692 struct hlist_node *p;
693 struct net_device *dev;
694 struct hlist_head *head = dev_index_hash(net, ifindex);
696 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
697 if (dev->ifindex == ifindex)
702 EXPORT_SYMBOL(dev_get_by_index_rcu);
706 * dev_get_by_index - find a device by its ifindex
707 * @net: the applicable net namespace
708 * @ifindex: index of device
710 * Search for an interface by index. Returns NULL if the device
711 * is not found or a pointer to the device. The device returned has
712 * had a reference added and the pointer is safe until the user calls
713 * dev_put to indicate they have finished with it.
716 struct net_device *dev_get_by_index(struct net *net, int ifindex)
718 struct net_device *dev;
721 dev = dev_get_by_index_rcu(net, ifindex);
727 EXPORT_SYMBOL(dev_get_by_index);
730 * dev_getbyhwaddr - find a device by its hardware address
731 * @net: the applicable net namespace
732 * @type: media type of device
733 * @ha: hardware address
735 * Search for an interface by MAC address. Returns NULL if the device
736 * is not found or a pointer to the device. The caller must hold the
737 * rtnl semaphore. The returned device has not had its ref count increased
738 * and the caller must therefore be careful about locking
741 * If the API was consistent this would be __dev_get_by_hwaddr
744 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
746 struct net_device *dev;
750 for_each_netdev(net, dev)
751 if (dev->type == type &&
752 !memcmp(dev->dev_addr, ha, dev->addr_len))
757 EXPORT_SYMBOL(dev_getbyhwaddr);
759 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
761 struct net_device *dev;
764 for_each_netdev(net, dev)
765 if (dev->type == type)
770 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
772 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
774 struct net_device *dev;
777 dev = __dev_getfirstbyhwtype(net, type);
783 EXPORT_SYMBOL(dev_getfirstbyhwtype);
786 * dev_get_by_flags - find any device with given flags
787 * @net: the applicable net namespace
788 * @if_flags: IFF_* values
789 * @mask: bitmask of bits in if_flags to check
791 * Search for any interface with the given flags. Returns NULL if a device
792 * is not found or a pointer to the device. The device returned has
793 * had a reference added and the pointer is safe until the user calls
794 * dev_put to indicate they have finished with it.
797 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
800 struct net_device *dev, *ret;
804 for_each_netdev_rcu(net, dev) {
805 if (((dev->flags ^ if_flags) & mask) == 0) {
814 EXPORT_SYMBOL(dev_get_by_flags);
817 * dev_valid_name - check if name is okay for network device
820 * Network device names need to be valid file names to
821 * to allow sysfs to work. We also disallow any kind of
824 int dev_valid_name(const char *name)
828 if (strlen(name) >= IFNAMSIZ)
830 if (!strcmp(name, ".") || !strcmp(name, ".."))
834 if (*name == '/' || isspace(*name))
840 EXPORT_SYMBOL(dev_valid_name);
843 * __dev_alloc_name - allocate a name for a device
844 * @net: network namespace to allocate the device name in
845 * @name: name format string
846 * @buf: scratch buffer and result name string
848 * Passed a format string - eg "lt%d" it will try and find a suitable
849 * id. It scans list of devices to build up a free map, then chooses
850 * the first empty slot. The caller must hold the dev_base or rtnl lock
851 * while allocating the name and adding the device in order to avoid
853 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
854 * Returns the number of the unit assigned or a negative errno code.
857 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
861 const int max_netdevices = 8*PAGE_SIZE;
862 unsigned long *inuse;
863 struct net_device *d;
865 p = strnchr(name, IFNAMSIZ-1, '%');
868 * Verify the string as this thing may have come from
869 * the user. There must be either one "%d" and no other "%"
872 if (p[1] != 'd' || strchr(p + 2, '%'))
875 /* Use one page as a bit array of possible slots */
876 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
880 for_each_netdev(net, d) {
881 if (!sscanf(d->name, name, &i))
883 if (i < 0 || i >= max_netdevices)
886 /* avoid cases where sscanf is not exact inverse of printf */
887 snprintf(buf, IFNAMSIZ, name, i);
888 if (!strncmp(buf, d->name, IFNAMSIZ))
892 i = find_first_zero_bit(inuse, max_netdevices);
893 free_page((unsigned long) inuse);
896 snprintf(buf, IFNAMSIZ, name, i);
897 if (!__dev_get_by_name(net, buf))
900 /* It is possible to run out of possible slots
901 * when the name is long and there isn't enough space left
902 * for the digits, or if all bits are used.
908 * dev_alloc_name - allocate a name for a device
910 * @name: name format string
912 * Passed a format string - eg "lt%d" it will try and find a suitable
913 * id. It scans list of devices to build up a free map, then chooses
914 * the first empty slot. The caller must hold the dev_base or rtnl lock
915 * while allocating the name and adding the device in order to avoid
917 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
918 * Returns the number of the unit assigned or a negative errno code.
921 int dev_alloc_name(struct net_device *dev, const char *name)
927 BUG_ON(!dev_net(dev));
929 ret = __dev_alloc_name(net, name, buf);
931 strlcpy(dev->name, buf, IFNAMSIZ);
934 EXPORT_SYMBOL(dev_alloc_name);
938 * dev_change_name - change name of a device
940 * @newname: name (or format string) must be at least IFNAMSIZ
942 * Change name of a device, can pass format strings "eth%d".
945 int dev_change_name(struct net_device *dev, const char *newname)
947 char oldname[IFNAMSIZ];
953 BUG_ON(!dev_net(dev));
956 if (dev->flags & IFF_UP)
959 if (!dev_valid_name(newname))
962 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
965 memcpy(oldname, dev->name, IFNAMSIZ);
967 if (strchr(newname, '%')) {
968 err = dev_alloc_name(dev, newname);
971 } else if (__dev_get_by_name(net, newname))
974 strlcpy(dev->name, newname, IFNAMSIZ);
977 /* For now only devices in the initial network namespace
980 if (net == &init_net) {
981 ret = device_rename(&dev->dev, dev->name);
983 memcpy(dev->name, oldname, IFNAMSIZ);
988 write_lock_bh(&dev_base_lock);
989 hlist_del(&dev->name_hlist);
990 write_unlock_bh(&dev_base_lock);
994 write_lock_bh(&dev_base_lock);
995 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
996 write_unlock_bh(&dev_base_lock);
998 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
999 ret = notifier_to_errno(ret);
1004 "%s: name change rollback failed: %d.\n",
1008 memcpy(dev->name, oldname, IFNAMSIZ);
1017 * dev_set_alias - change ifalias of a device
1019 * @alias: name up to IFALIASZ
1020 * @len: limit of bytes to copy from info
1022 * Set ifalias for a device,
1024 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1028 if (len >= IFALIASZ)
1033 kfree(dev->ifalias);
1034 dev->ifalias = NULL;
1039 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1043 strlcpy(dev->ifalias, alias, len+1);
1049 * netdev_features_change - device changes features
1050 * @dev: device to cause notification
1052 * Called to indicate a device has changed features.
1054 void netdev_features_change(struct net_device *dev)
1056 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1058 EXPORT_SYMBOL(netdev_features_change);
1061 * netdev_state_change - device changes state
1062 * @dev: device to cause notification
1064 * Called to indicate a device has changed state. This function calls
1065 * the notifier chains for netdev_chain and sends a NEWLINK message
1066 * to the routing socket.
1068 void netdev_state_change(struct net_device *dev)
1070 if (dev->flags & IFF_UP) {
1071 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1072 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1075 EXPORT_SYMBOL(netdev_state_change);
1077 void netdev_bonding_change(struct net_device *dev, unsigned long event)
1079 call_netdevice_notifiers(event, dev);
1081 EXPORT_SYMBOL(netdev_bonding_change);
1084 * dev_load - load a network module
1085 * @net: the applicable net namespace
1086 * @name: name of interface
1088 * If a network interface is not present and the process has suitable
1089 * privileges this function loads the module. If module loading is not
1090 * available in this kernel then it becomes a nop.
1093 void dev_load(struct net *net, const char *name)
1095 struct net_device *dev;
1098 dev = dev_get_by_name_rcu(net, name);
1101 if (!dev && capable(CAP_NET_ADMIN))
1102 request_module("%s", name);
1104 EXPORT_SYMBOL(dev_load);
1107 * dev_open - prepare an interface for use.
1108 * @dev: device to open
1110 * Takes a device from down to up state. The device's private open
1111 * function is invoked and then the multicast lists are loaded. Finally
1112 * the device is moved into the up state and a %NETDEV_UP message is
1113 * sent to the netdev notifier chain.
1115 * Calling this function on an active interface is a nop. On a failure
1116 * a negative errno code is returned.
1118 int dev_open(struct net_device *dev)
1120 const struct net_device_ops *ops = dev->netdev_ops;
1129 if (dev->flags & IFF_UP)
1133 * Is it even present?
1135 if (!netif_device_present(dev))
1138 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1139 ret = notifier_to_errno(ret);
1144 * Call device private open method
1146 set_bit(__LINK_STATE_START, &dev->state);
1148 if (ops->ndo_validate_addr)
1149 ret = ops->ndo_validate_addr(dev);
1151 if (!ret && ops->ndo_open)
1152 ret = ops->ndo_open(dev);
1155 * If it went open OK then:
1159 clear_bit(__LINK_STATE_START, &dev->state);
1164 dev->flags |= IFF_UP;
1169 net_dmaengine_get();
1172 * Initialize multicasting status
1174 dev_set_rx_mode(dev);
1177 * Wakeup transmit queue engine
1182 * ... and announce new interface.
1184 call_netdevice_notifiers(NETDEV_UP, dev);
1189 EXPORT_SYMBOL(dev_open);
1192 * dev_close - shutdown an interface.
1193 * @dev: device to shutdown
1195 * This function moves an active device into down state. A
1196 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1197 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1200 int dev_close(struct net_device *dev)
1202 const struct net_device_ops *ops = dev->netdev_ops;
1207 if (!(dev->flags & IFF_UP))
1211 * Tell people we are going down, so that they can
1212 * prepare to death, when device is still operating.
1214 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1216 clear_bit(__LINK_STATE_START, &dev->state);
1218 /* Synchronize to scheduled poll. We cannot touch poll list,
1219 * it can be even on different cpu. So just clear netif_running().
1221 * dev->stop() will invoke napi_disable() on all of it's
1222 * napi_struct instances on this device.
1224 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1226 dev_deactivate(dev);
1229 * Call the device specific close. This cannot fail.
1230 * Only if device is UP
1232 * We allow it to be called even after a DETACH hot-plug
1239 * Device is now down.
1242 dev->flags &= ~IFF_UP;
1245 * Tell people we are down
1247 call_netdevice_notifiers(NETDEV_DOWN, dev);
1252 net_dmaengine_put();
1256 EXPORT_SYMBOL(dev_close);
1260 * dev_disable_lro - disable Large Receive Offload on a device
1263 * Disable Large Receive Offload (LRO) on a net device. Must be
1264 * called under RTNL. This is needed if received packets may be
1265 * forwarded to another interface.
1267 void dev_disable_lro(struct net_device *dev)
1269 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1270 dev->ethtool_ops->set_flags) {
1271 u32 flags = dev->ethtool_ops->get_flags(dev);
1272 if (flags & ETH_FLAG_LRO) {
1273 flags &= ~ETH_FLAG_LRO;
1274 dev->ethtool_ops->set_flags(dev, flags);
1277 WARN_ON(dev->features & NETIF_F_LRO);
1279 EXPORT_SYMBOL(dev_disable_lro);
1282 static int dev_boot_phase = 1;
1285 * Device change register/unregister. These are not inline or static
1286 * as we export them to the world.
1290 * register_netdevice_notifier - register a network notifier block
1293 * Register a notifier to be called when network device events occur.
1294 * The notifier passed is linked into the kernel structures and must
1295 * not be reused until it has been unregistered. A negative errno code
1296 * is returned on a failure.
1298 * When registered all registration and up events are replayed
1299 * to the new notifier to allow device to have a race free
1300 * view of the network device list.
1303 int register_netdevice_notifier(struct notifier_block *nb)
1305 struct net_device *dev;
1306 struct net_device *last;
1311 err = raw_notifier_chain_register(&netdev_chain, nb);
1317 for_each_netdev(net, dev) {
1318 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1319 err = notifier_to_errno(err);
1323 if (!(dev->flags & IFF_UP))
1326 nb->notifier_call(nb, NETDEV_UP, dev);
1337 for_each_netdev(net, dev) {
1341 if (dev->flags & IFF_UP) {
1342 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1343 nb->notifier_call(nb, NETDEV_DOWN, dev);
1345 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1349 raw_notifier_chain_unregister(&netdev_chain, nb);
1352 EXPORT_SYMBOL(register_netdevice_notifier);
1355 * unregister_netdevice_notifier - unregister a network notifier block
1358 * Unregister a notifier previously registered by
1359 * register_netdevice_notifier(). The notifier is unlinked into the
1360 * kernel structures and may then be reused. A negative errno code
1361 * is returned on a failure.
1364 int unregister_netdevice_notifier(struct notifier_block *nb)
1369 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1373 EXPORT_SYMBOL(unregister_netdevice_notifier);
1376 * call_netdevice_notifiers - call all network notifier blocks
1377 * @val: value passed unmodified to notifier function
1378 * @dev: net_device pointer passed unmodified to notifier function
1380 * Call all network notifier blocks. Parameters and return value
1381 * are as for raw_notifier_call_chain().
1384 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1386 return raw_notifier_call_chain(&netdev_chain, val, dev);
1389 /* When > 0 there are consumers of rx skb time stamps */
1390 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1392 void net_enable_timestamp(void)
1394 atomic_inc(&netstamp_needed);
1396 EXPORT_SYMBOL(net_enable_timestamp);
1398 void net_disable_timestamp(void)
1400 atomic_dec(&netstamp_needed);
1402 EXPORT_SYMBOL(net_disable_timestamp);
1404 static inline void net_timestamp(struct sk_buff *skb)
1406 if (atomic_read(&netstamp_needed))
1407 __net_timestamp(skb);
1409 skb->tstamp.tv64 = 0;
1413 * Support routine. Sends outgoing frames to any network
1414 * taps currently in use.
1417 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1419 struct packet_type *ptype;
1421 #ifdef CONFIG_NET_CLS_ACT
1422 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1429 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1430 /* Never send packets back to the socket
1431 * they originated from - MvS (miquels@drinkel.ow.org)
1433 if ((ptype->dev == dev || !ptype->dev) &&
1434 (ptype->af_packet_priv == NULL ||
1435 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1436 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1440 /* skb->nh should be correctly
1441 set by sender, so that the second statement is
1442 just protection against buggy protocols.
1444 skb_reset_mac_header(skb2);
1446 if (skb_network_header(skb2) < skb2->data ||
1447 skb2->network_header > skb2->tail) {
1448 if (net_ratelimit())
1449 printk(KERN_CRIT "protocol %04x is "
1451 skb2->protocol, dev->name);
1452 skb_reset_network_header(skb2);
1455 skb2->transport_header = skb2->network_header;
1456 skb2->pkt_type = PACKET_OUTGOING;
1457 ptype->func(skb2, skb->dev, ptype, skb->dev);
1464 static inline void __netif_reschedule(struct Qdisc *q)
1466 struct softnet_data *sd;
1467 unsigned long flags;
1469 local_irq_save(flags);
1470 sd = &__get_cpu_var(softnet_data);
1471 q->next_sched = sd->output_queue;
1472 sd->output_queue = q;
1473 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1474 local_irq_restore(flags);
1477 void __netif_schedule(struct Qdisc *q)
1479 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1480 __netif_reschedule(q);
1482 EXPORT_SYMBOL(__netif_schedule);
1484 void dev_kfree_skb_irq(struct sk_buff *skb)
1486 if (atomic_dec_and_test(&skb->users)) {
1487 struct softnet_data *sd;
1488 unsigned long flags;
1490 local_irq_save(flags);
1491 sd = &__get_cpu_var(softnet_data);
1492 skb->next = sd->completion_queue;
1493 sd->completion_queue = skb;
1494 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1495 local_irq_restore(flags);
1498 EXPORT_SYMBOL(dev_kfree_skb_irq);
1500 void dev_kfree_skb_any(struct sk_buff *skb)
1502 if (in_irq() || irqs_disabled())
1503 dev_kfree_skb_irq(skb);
1507 EXPORT_SYMBOL(dev_kfree_skb_any);
1511 * netif_device_detach - mark device as removed
1512 * @dev: network device
1514 * Mark device as removed from system and therefore no longer available.
1516 void netif_device_detach(struct net_device *dev)
1518 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1519 netif_running(dev)) {
1520 netif_tx_stop_all_queues(dev);
1523 EXPORT_SYMBOL(netif_device_detach);
1526 * netif_device_attach - mark device as attached
1527 * @dev: network device
1529 * Mark device as attached from system and restart if needed.
1531 void netif_device_attach(struct net_device *dev)
1533 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1534 netif_running(dev)) {
1535 netif_tx_wake_all_queues(dev);
1536 __netdev_watchdog_up(dev);
1539 EXPORT_SYMBOL(netif_device_attach);
1541 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1543 return ((features & NETIF_F_GEN_CSUM) ||
1544 ((features & NETIF_F_IP_CSUM) &&
1545 protocol == htons(ETH_P_IP)) ||
1546 ((features & NETIF_F_IPV6_CSUM) &&
1547 protocol == htons(ETH_P_IPV6)) ||
1548 ((features & NETIF_F_FCOE_CRC) &&
1549 protocol == htons(ETH_P_FCOE)));
1552 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1554 if (can_checksum_protocol(dev->features, skb->protocol))
1557 if (skb->protocol == htons(ETH_P_8021Q)) {
1558 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1559 if (can_checksum_protocol(dev->features & dev->vlan_features,
1560 veh->h_vlan_encapsulated_proto))
1568 * Invalidate hardware checksum when packet is to be mangled, and
1569 * complete checksum manually on outgoing path.
1571 int skb_checksum_help(struct sk_buff *skb)
1574 int ret = 0, offset;
1576 if (skb->ip_summed == CHECKSUM_COMPLETE)
1577 goto out_set_summed;
1579 if (unlikely(skb_shinfo(skb)->gso_size)) {
1580 /* Let GSO fix up the checksum. */
1581 goto out_set_summed;
1584 offset = skb->csum_start - skb_headroom(skb);
1585 BUG_ON(offset >= skb_headlen(skb));
1586 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1588 offset += skb->csum_offset;
1589 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1591 if (skb_cloned(skb) &&
1592 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1593 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1598 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1600 skb->ip_summed = CHECKSUM_NONE;
1604 EXPORT_SYMBOL(skb_checksum_help);
1607 * skb_gso_segment - Perform segmentation on skb.
1608 * @skb: buffer to segment
1609 * @features: features for the output path (see dev->features)
1611 * This function segments the given skb and returns a list of segments.
1613 * It may return NULL if the skb requires no segmentation. This is
1614 * only possible when GSO is used for verifying header integrity.
1616 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1618 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1619 struct packet_type *ptype;
1620 __be16 type = skb->protocol;
1623 skb_reset_mac_header(skb);
1624 skb->mac_len = skb->network_header - skb->mac_header;
1625 __skb_pull(skb, skb->mac_len);
1627 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1628 struct net_device *dev = skb->dev;
1629 struct ethtool_drvinfo info = {};
1631 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1632 dev->ethtool_ops->get_drvinfo(dev, &info);
1634 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1636 info.driver, dev ? dev->features : 0L,
1637 skb->sk ? skb->sk->sk_route_caps : 0L,
1638 skb->len, skb->data_len, skb->ip_summed);
1640 if (skb_header_cloned(skb) &&
1641 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1642 return ERR_PTR(err);
1646 list_for_each_entry_rcu(ptype,
1647 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1648 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1649 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1650 err = ptype->gso_send_check(skb);
1651 segs = ERR_PTR(err);
1652 if (err || skb_gso_ok(skb, features))
1654 __skb_push(skb, (skb->data -
1655 skb_network_header(skb)));
1657 segs = ptype->gso_segment(skb, features);
1663 __skb_push(skb, skb->data - skb_mac_header(skb));
1667 EXPORT_SYMBOL(skb_gso_segment);
1669 /* Take action when hardware reception checksum errors are detected. */
1671 void netdev_rx_csum_fault(struct net_device *dev)
1673 if (net_ratelimit()) {
1674 printk(KERN_ERR "%s: hw csum failure.\n",
1675 dev ? dev->name : "<unknown>");
1679 EXPORT_SYMBOL(netdev_rx_csum_fault);
1682 /* Actually, we should eliminate this check as soon as we know, that:
1683 * 1. IOMMU is present and allows to map all the memory.
1684 * 2. No high memory really exists on this machine.
1687 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1689 #ifdef CONFIG_HIGHMEM
1692 if (dev->features & NETIF_F_HIGHDMA)
1695 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1696 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1704 void (*destructor)(struct sk_buff *skb);
1707 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1709 static void dev_gso_skb_destructor(struct sk_buff *skb)
1711 struct dev_gso_cb *cb;
1714 struct sk_buff *nskb = skb->next;
1716 skb->next = nskb->next;
1719 } while (skb->next);
1721 cb = DEV_GSO_CB(skb);
1723 cb->destructor(skb);
1727 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1728 * @skb: buffer to segment
1730 * This function segments the given skb and stores the list of segments
1733 static int dev_gso_segment(struct sk_buff *skb)
1735 struct net_device *dev = skb->dev;
1736 struct sk_buff *segs;
1737 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1740 segs = skb_gso_segment(skb, features);
1742 /* Verifying header integrity only. */
1747 return PTR_ERR(segs);
1750 DEV_GSO_CB(skb)->destructor = skb->destructor;
1751 skb->destructor = dev_gso_skb_destructor;
1756 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1757 struct netdev_queue *txq)
1759 const struct net_device_ops *ops = dev->netdev_ops;
1760 int rc = NETDEV_TX_OK;
1762 if (likely(!skb->next)) {
1763 if (!list_empty(&ptype_all))
1764 dev_queue_xmit_nit(skb, dev);
1766 if (netif_needs_gso(dev, skb)) {
1767 if (unlikely(dev_gso_segment(skb)))
1774 * If device doesnt need skb->dst, release it right now while
1775 * its hot in this cpu cache
1777 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1780 rc = ops->ndo_start_xmit(skb, dev);
1781 if (rc == NETDEV_TX_OK)
1782 txq_trans_update(txq);
1784 * TODO: if skb_orphan() was called by
1785 * dev->hard_start_xmit() (for example, the unmodified
1786 * igb driver does that; bnx2 doesn't), then
1787 * skb_tx_software_timestamp() will be unable to send
1788 * back the time stamp.
1790 * How can this be prevented? Always create another
1791 * reference to the socket before calling
1792 * dev->hard_start_xmit()? Prevent that skb_orphan()
1793 * does anything in dev->hard_start_xmit() by clearing
1794 * the skb destructor before the call and restoring it
1795 * afterwards, then doing the skb_orphan() ourselves?
1802 struct sk_buff *nskb = skb->next;
1804 skb->next = nskb->next;
1806 rc = ops->ndo_start_xmit(nskb, dev);
1807 if (unlikely(rc != NETDEV_TX_OK)) {
1808 if (rc & ~NETDEV_TX_MASK)
1809 goto out_kfree_gso_skb;
1810 nskb->next = skb->next;
1814 txq_trans_update(txq);
1815 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1816 return NETDEV_TX_BUSY;
1817 } while (skb->next);
1820 if (likely(skb->next == NULL))
1821 skb->destructor = DEV_GSO_CB(skb)->destructor;
1827 static u32 skb_tx_hashrnd;
1829 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1833 if (skb_rx_queue_recorded(skb)) {
1834 hash = skb_get_rx_queue(skb);
1835 while (unlikely(hash >= dev->real_num_tx_queues))
1836 hash -= dev->real_num_tx_queues;
1840 if (skb->sk && skb->sk->sk_hash)
1841 hash = skb->sk->sk_hash;
1843 hash = skb->protocol;
1845 hash = jhash_1word(hash, skb_tx_hashrnd);
1847 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1849 EXPORT_SYMBOL(skb_tx_hash);
1851 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1853 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1854 if (net_ratelimit()) {
1855 WARN(1, "%s selects TX queue %d, but "
1856 "real number of TX queues is %d\n",
1857 dev->name, queue_index,
1858 dev->real_num_tx_queues);
1865 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1866 struct sk_buff *skb)
1869 struct sock *sk = skb->sk;
1871 if (sk_tx_queue_recorded(sk)) {
1872 queue_index = sk_tx_queue_get(sk);
1874 const struct net_device_ops *ops = dev->netdev_ops;
1876 if (ops->ndo_select_queue) {
1877 queue_index = ops->ndo_select_queue(dev, skb);
1878 queue_index = dev_cap_txqueue(dev, queue_index);
1881 if (dev->real_num_tx_queues > 1)
1882 queue_index = skb_tx_hash(dev, skb);
1884 if (sk && sk->sk_dst_cache)
1885 sk_tx_queue_set(sk, queue_index);
1889 skb_set_queue_mapping(skb, queue_index);
1890 return netdev_get_tx_queue(dev, queue_index);
1893 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
1894 struct net_device *dev,
1895 struct netdev_queue *txq)
1897 spinlock_t *root_lock = qdisc_lock(q);
1900 spin_lock(root_lock);
1901 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1904 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
1905 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
1907 * This is a work-conserving queue; there are no old skbs
1908 * waiting to be sent out; and the qdisc is not running -
1909 * xmit the skb directly.
1911 __qdisc_update_bstats(q, skb->len);
1912 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
1915 clear_bit(__QDISC_STATE_RUNNING, &q->state);
1917 rc = NET_XMIT_SUCCESS;
1919 rc = qdisc_enqueue_root(skb, q);
1922 spin_unlock(root_lock);
1927 static inline bool dev_xmit_complete(int rc)
1929 /* successful transmission */
1930 if (rc == NETDEV_TX_OK)
1933 /* error while transmitting, driver consumed skb */
1937 /* error while queueing to a different device, driver consumed skb */
1938 if (rc & NET_XMIT_MASK)
1945 * dev_queue_xmit - transmit a buffer
1946 * @skb: buffer to transmit
1948 * Queue a buffer for transmission to a network device. The caller must
1949 * have set the device and priority and built the buffer before calling
1950 * this function. The function can be called from an interrupt.
1952 * A negative errno code is returned on a failure. A success does not
1953 * guarantee the frame will be transmitted as it may be dropped due
1954 * to congestion or traffic shaping.
1956 * -----------------------------------------------------------------------------------
1957 * I notice this method can also return errors from the queue disciplines,
1958 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1961 * Regardless of the return value, the skb is consumed, so it is currently
1962 * difficult to retry a send to this method. (You can bump the ref count
1963 * before sending to hold a reference for retry if you are careful.)
1965 * When calling this method, interrupts MUST be enabled. This is because
1966 * the BH enable code must have IRQs enabled so that it will not deadlock.
1969 int dev_queue_xmit(struct sk_buff *skb)
1971 struct net_device *dev = skb->dev;
1972 struct netdev_queue *txq;
1976 /* GSO will handle the following emulations directly. */
1977 if (netif_needs_gso(dev, skb))
1980 if (skb_has_frags(skb) &&
1981 !(dev->features & NETIF_F_FRAGLIST) &&
1982 __skb_linearize(skb))
1985 /* Fragmented skb is linearized if device does not support SG,
1986 * or if at least one of fragments is in highmem and device
1987 * does not support DMA from it.
1989 if (skb_shinfo(skb)->nr_frags &&
1990 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1991 __skb_linearize(skb))
1994 /* If packet is not checksummed and device does not support
1995 * checksumming for this protocol, complete checksumming here.
1997 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1998 skb_set_transport_header(skb, skb->csum_start -
2000 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2005 /* Disable soft irqs for various locks below. Also
2006 * stops preemption for RCU.
2010 txq = dev_pick_tx(dev, skb);
2011 q = rcu_dereference(txq->qdisc);
2013 #ifdef CONFIG_NET_CLS_ACT
2014 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2017 rc = __dev_xmit_skb(skb, q, dev, txq);
2021 /* The device has no queue. Common case for software devices:
2022 loopback, all the sorts of tunnels...
2024 Really, it is unlikely that netif_tx_lock protection is necessary
2025 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2027 However, it is possible, that they rely on protection
2030 Check this and shot the lock. It is not prone from deadlocks.
2031 Either shot noqueue qdisc, it is even simpler 8)
2033 if (dev->flags & IFF_UP) {
2034 int cpu = smp_processor_id(); /* ok because BHs are off */
2036 if (txq->xmit_lock_owner != cpu) {
2038 HARD_TX_LOCK(dev, txq, cpu);
2040 if (!netif_tx_queue_stopped(txq)) {
2041 rc = dev_hard_start_xmit(skb, dev, txq);
2042 if (dev_xmit_complete(rc)) {
2043 HARD_TX_UNLOCK(dev, txq);
2047 HARD_TX_UNLOCK(dev, txq);
2048 if (net_ratelimit())
2049 printk(KERN_CRIT "Virtual device %s asks to "
2050 "queue packet!\n", dev->name);
2052 /* Recursion is detected! It is possible,
2054 if (net_ratelimit())
2055 printk(KERN_CRIT "Dead loop on virtual device "
2056 "%s, fix it urgently!\n", dev->name);
2061 rcu_read_unlock_bh();
2067 rcu_read_unlock_bh();
2070 EXPORT_SYMBOL(dev_queue_xmit);
2073 /*=======================================================================
2075 =======================================================================*/
2077 int netdev_max_backlog __read_mostly = 1000;
2078 int netdev_budget __read_mostly = 300;
2079 int weight_p __read_mostly = 64; /* old backlog weight */
2081 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2085 * netif_rx - post buffer to the network code
2086 * @skb: buffer to post
2088 * This function receives a packet from a device driver and queues it for
2089 * the upper (protocol) levels to process. It always succeeds. The buffer
2090 * may be dropped during processing for congestion control or by the
2094 * NET_RX_SUCCESS (no congestion)
2095 * NET_RX_DROP (packet was dropped)
2099 int netif_rx(struct sk_buff *skb)
2101 struct softnet_data *queue;
2102 unsigned long flags;
2104 /* if netpoll wants it, pretend we never saw it */
2105 if (netpoll_rx(skb))
2108 if (!skb->tstamp.tv64)
2112 * The code is rearranged so that the path is the most
2113 * short when CPU is congested, but is still operating.
2115 local_irq_save(flags);
2116 queue = &__get_cpu_var(softnet_data);
2118 __get_cpu_var(netdev_rx_stat).total++;
2119 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2120 if (queue->input_pkt_queue.qlen) {
2122 __skb_queue_tail(&queue->input_pkt_queue, skb);
2123 local_irq_restore(flags);
2124 return NET_RX_SUCCESS;
2127 napi_schedule(&queue->backlog);
2131 __get_cpu_var(netdev_rx_stat).dropped++;
2132 local_irq_restore(flags);
2137 EXPORT_SYMBOL(netif_rx);
2139 int netif_rx_ni(struct sk_buff *skb)
2144 err = netif_rx(skb);
2145 if (local_softirq_pending())
2151 EXPORT_SYMBOL(netif_rx_ni);
2153 static void net_tx_action(struct softirq_action *h)
2155 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2157 if (sd->completion_queue) {
2158 struct sk_buff *clist;
2160 local_irq_disable();
2161 clist = sd->completion_queue;
2162 sd->completion_queue = NULL;
2166 struct sk_buff *skb = clist;
2167 clist = clist->next;
2169 WARN_ON(atomic_read(&skb->users));
2174 if (sd->output_queue) {
2177 local_irq_disable();
2178 head = sd->output_queue;
2179 sd->output_queue = NULL;
2183 struct Qdisc *q = head;
2184 spinlock_t *root_lock;
2186 head = head->next_sched;
2188 root_lock = qdisc_lock(q);
2189 if (spin_trylock(root_lock)) {
2190 smp_mb__before_clear_bit();
2191 clear_bit(__QDISC_STATE_SCHED,
2194 spin_unlock(root_lock);
2196 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2198 __netif_reschedule(q);
2200 smp_mb__before_clear_bit();
2201 clear_bit(__QDISC_STATE_SCHED,
2209 static inline int deliver_skb(struct sk_buff *skb,
2210 struct packet_type *pt_prev,
2211 struct net_device *orig_dev)
2213 atomic_inc(&skb->users);
2214 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2217 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2219 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2220 /* This hook is defined here for ATM LANE */
2221 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2222 unsigned char *addr) __read_mostly;
2223 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2227 * If bridge module is loaded call bridging hook.
2228 * returns NULL if packet was consumed.
2230 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2231 struct sk_buff *skb) __read_mostly;
2232 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2234 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2235 struct packet_type **pt_prev, int *ret,
2236 struct net_device *orig_dev)
2238 struct net_bridge_port *port;
2240 if (skb->pkt_type == PACKET_LOOPBACK ||
2241 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2245 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2249 return br_handle_frame_hook(port, skb);
2252 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2255 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2256 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2257 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2259 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2260 struct packet_type **pt_prev,
2262 struct net_device *orig_dev)
2264 if (skb->dev->macvlan_port == NULL)
2268 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2271 return macvlan_handle_frame_hook(skb);
2274 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2277 #ifdef CONFIG_NET_CLS_ACT
2278 /* TODO: Maybe we should just force sch_ingress to be compiled in
2279 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2280 * a compare and 2 stores extra right now if we dont have it on
2281 * but have CONFIG_NET_CLS_ACT
2282 * NOTE: This doesnt stop any functionality; if you dont have
2283 * the ingress scheduler, you just cant add policies on ingress.
2286 static int ing_filter(struct sk_buff *skb)
2288 struct net_device *dev = skb->dev;
2289 u32 ttl = G_TC_RTTL(skb->tc_verd);
2290 struct netdev_queue *rxq;
2291 int result = TC_ACT_OK;
2294 if (MAX_RED_LOOP < ttl++) {
2296 "Redir loop detected Dropping packet (%d->%d)\n",
2297 skb->iif, dev->ifindex);
2301 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2302 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2304 rxq = &dev->rx_queue;
2307 if (q != &noop_qdisc) {
2308 spin_lock(qdisc_lock(q));
2309 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2310 result = qdisc_enqueue_root(skb, q);
2311 spin_unlock(qdisc_lock(q));
2317 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2318 struct packet_type **pt_prev,
2319 int *ret, struct net_device *orig_dev)
2321 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2325 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2328 /* Huh? Why does turning on AF_PACKET affect this? */
2329 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2332 switch (ing_filter(skb)) {
2346 * netif_nit_deliver - deliver received packets to network taps
2349 * This function is used to deliver incoming packets to network
2350 * taps. It should be used when the normal netif_receive_skb path
2351 * is bypassed, for example because of VLAN acceleration.
2353 void netif_nit_deliver(struct sk_buff *skb)
2355 struct packet_type *ptype;
2357 if (list_empty(&ptype_all))
2360 skb_reset_network_header(skb);
2361 skb_reset_transport_header(skb);
2362 skb->mac_len = skb->network_header - skb->mac_header;
2365 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2366 if (!ptype->dev || ptype->dev == skb->dev)
2367 deliver_skb(skb, ptype, skb->dev);
2373 * netif_receive_skb - process receive buffer from network
2374 * @skb: buffer to process
2376 * netif_receive_skb() is the main receive data processing function.
2377 * It always succeeds. The buffer may be dropped during processing
2378 * for congestion control or by the protocol layers.
2380 * This function may only be called from softirq context and interrupts
2381 * should be enabled.
2383 * Return values (usually ignored):
2384 * NET_RX_SUCCESS: no congestion
2385 * NET_RX_DROP: packet was dropped
2387 int netif_receive_skb(struct sk_buff *skb)
2389 struct packet_type *ptype, *pt_prev;
2390 struct net_device *orig_dev;
2391 struct net_device *null_or_orig;
2392 int ret = NET_RX_DROP;
2395 if (!skb->tstamp.tv64)
2398 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2399 return NET_RX_SUCCESS;
2401 /* if we've gotten here through NAPI, check netpoll */
2402 if (netpoll_receive_skb(skb))
2406 skb->iif = skb->dev->ifindex;
2408 null_or_orig = NULL;
2409 orig_dev = skb->dev;
2410 if (orig_dev->master) {
2411 if (skb_bond_should_drop(skb))
2412 null_or_orig = orig_dev; /* deliver only exact match */
2414 skb->dev = orig_dev->master;
2417 __get_cpu_var(netdev_rx_stat).total++;
2419 skb_reset_network_header(skb);
2420 skb_reset_transport_header(skb);
2421 skb->mac_len = skb->network_header - skb->mac_header;
2427 #ifdef CONFIG_NET_CLS_ACT
2428 if (skb->tc_verd & TC_NCLS) {
2429 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2434 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2435 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2436 ptype->dev == orig_dev) {
2438 ret = deliver_skb(skb, pt_prev, orig_dev);
2443 #ifdef CONFIG_NET_CLS_ACT
2444 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2450 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2453 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2457 type = skb->protocol;
2458 list_for_each_entry_rcu(ptype,
2459 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2460 if (ptype->type == type &&
2461 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2462 ptype->dev == orig_dev)) {
2464 ret = deliver_skb(skb, pt_prev, orig_dev);
2470 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2473 /* Jamal, now you will not able to escape explaining
2474 * me how you were going to use this. :-)
2483 EXPORT_SYMBOL(netif_receive_skb);
2485 /* Network device is going away, flush any packets still pending */
2486 static void flush_backlog(void *arg)
2488 struct net_device *dev = arg;
2489 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2490 struct sk_buff *skb, *tmp;
2492 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2493 if (skb->dev == dev) {
2494 __skb_unlink(skb, &queue->input_pkt_queue);
2499 static int napi_gro_complete(struct sk_buff *skb)
2501 struct packet_type *ptype;
2502 __be16 type = skb->protocol;
2503 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2506 if (NAPI_GRO_CB(skb)->count == 1) {
2507 skb_shinfo(skb)->gso_size = 0;
2512 list_for_each_entry_rcu(ptype, head, list) {
2513 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2516 err = ptype->gro_complete(skb);
2522 WARN_ON(&ptype->list == head);
2524 return NET_RX_SUCCESS;
2528 return netif_receive_skb(skb);
2531 void napi_gro_flush(struct napi_struct *napi)
2533 struct sk_buff *skb, *next;
2535 for (skb = napi->gro_list; skb; skb = next) {
2538 napi_gro_complete(skb);
2541 napi->gro_count = 0;
2542 napi->gro_list = NULL;
2544 EXPORT_SYMBOL(napi_gro_flush);
2546 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2548 struct sk_buff **pp = NULL;
2549 struct packet_type *ptype;
2550 __be16 type = skb->protocol;
2551 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2554 enum gro_result ret;
2556 if (!(skb->dev->features & NETIF_F_GRO))
2559 if (skb_is_gso(skb) || skb_has_frags(skb))
2563 list_for_each_entry_rcu(ptype, head, list) {
2564 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2567 skb_set_network_header(skb, skb_gro_offset(skb));
2568 mac_len = skb->network_header - skb->mac_header;
2569 skb->mac_len = mac_len;
2570 NAPI_GRO_CB(skb)->same_flow = 0;
2571 NAPI_GRO_CB(skb)->flush = 0;
2572 NAPI_GRO_CB(skb)->free = 0;
2574 pp = ptype->gro_receive(&napi->gro_list, skb);
2579 if (&ptype->list == head)
2582 same_flow = NAPI_GRO_CB(skb)->same_flow;
2583 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2586 struct sk_buff *nskb = *pp;
2590 napi_gro_complete(nskb);
2597 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2601 NAPI_GRO_CB(skb)->count = 1;
2602 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2603 skb->next = napi->gro_list;
2604 napi->gro_list = skb;
2608 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2609 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2611 BUG_ON(skb->end - skb->tail < grow);
2613 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2616 skb->data_len -= grow;
2618 skb_shinfo(skb)->frags[0].page_offset += grow;
2619 skb_shinfo(skb)->frags[0].size -= grow;
2621 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2622 put_page(skb_shinfo(skb)->frags[0].page);
2623 memmove(skb_shinfo(skb)->frags,
2624 skb_shinfo(skb)->frags + 1,
2625 --skb_shinfo(skb)->nr_frags);
2636 EXPORT_SYMBOL(dev_gro_receive);
2639 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2643 if (netpoll_rx_on(skb))
2646 for (p = napi->gro_list; p; p = p->next) {
2647 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2648 && !compare_ether_header(skb_mac_header(p),
2649 skb_gro_mac_header(skb));
2650 NAPI_GRO_CB(p)->flush = 0;
2653 return dev_gro_receive(napi, skb);
2656 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
2660 if (netif_receive_skb(skb))
2665 case GRO_MERGED_FREE:
2676 EXPORT_SYMBOL(napi_skb_finish);
2678 void skb_gro_reset_offset(struct sk_buff *skb)
2680 NAPI_GRO_CB(skb)->data_offset = 0;
2681 NAPI_GRO_CB(skb)->frag0 = NULL;
2682 NAPI_GRO_CB(skb)->frag0_len = 0;
2684 if (skb->mac_header == skb->tail &&
2685 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2686 NAPI_GRO_CB(skb)->frag0 =
2687 page_address(skb_shinfo(skb)->frags[0].page) +
2688 skb_shinfo(skb)->frags[0].page_offset;
2689 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2692 EXPORT_SYMBOL(skb_gro_reset_offset);
2694 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2696 skb_gro_reset_offset(skb);
2698 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2700 EXPORT_SYMBOL(napi_gro_receive);
2702 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2704 __skb_pull(skb, skb_headlen(skb));
2705 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2709 EXPORT_SYMBOL(napi_reuse_skb);
2711 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2713 struct sk_buff *skb = napi->skb;
2716 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
2722 EXPORT_SYMBOL(napi_get_frags);
2724 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
2730 skb->protocol = eth_type_trans(skb, napi->dev);
2732 if (ret == GRO_HELD)
2733 skb_gro_pull(skb, -ETH_HLEN);
2734 else if (netif_receive_skb(skb))
2739 case GRO_MERGED_FREE:
2740 napi_reuse_skb(napi, skb);
2749 EXPORT_SYMBOL(napi_frags_finish);
2751 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2753 struct sk_buff *skb = napi->skb;
2760 skb_reset_mac_header(skb);
2761 skb_gro_reset_offset(skb);
2763 off = skb_gro_offset(skb);
2764 hlen = off + sizeof(*eth);
2765 eth = skb_gro_header_fast(skb, off);
2766 if (skb_gro_header_hard(skb, hlen)) {
2767 eth = skb_gro_header_slow(skb, hlen, off);
2768 if (unlikely(!eth)) {
2769 napi_reuse_skb(napi, skb);
2775 skb_gro_pull(skb, sizeof(*eth));
2778 * This works because the only protocols we care about don't require
2779 * special handling. We'll fix it up properly at the end.
2781 skb->protocol = eth->h_proto;
2786 EXPORT_SYMBOL(napi_frags_skb);
2788 gro_result_t napi_gro_frags(struct napi_struct *napi)
2790 struct sk_buff *skb = napi_frags_skb(napi);
2795 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2797 EXPORT_SYMBOL(napi_gro_frags);
2799 static int process_backlog(struct napi_struct *napi, int quota)
2802 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2803 unsigned long start_time = jiffies;
2805 napi->weight = weight_p;
2807 struct sk_buff *skb;
2809 local_irq_disable();
2810 skb = __skb_dequeue(&queue->input_pkt_queue);
2812 __napi_complete(napi);
2818 netif_receive_skb(skb);
2819 } while (++work < quota && jiffies == start_time);
2825 * __napi_schedule - schedule for receive
2826 * @n: entry to schedule
2828 * The entry's receive function will be scheduled to run
2830 void __napi_schedule(struct napi_struct *n)
2832 unsigned long flags;
2834 local_irq_save(flags);
2835 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2836 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2837 local_irq_restore(flags);
2839 EXPORT_SYMBOL(__napi_schedule);
2841 void __napi_complete(struct napi_struct *n)
2843 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2844 BUG_ON(n->gro_list);
2846 list_del(&n->poll_list);
2847 smp_mb__before_clear_bit();
2848 clear_bit(NAPI_STATE_SCHED, &n->state);
2850 EXPORT_SYMBOL(__napi_complete);
2852 void napi_complete(struct napi_struct *n)
2854 unsigned long flags;
2857 * don't let napi dequeue from the cpu poll list
2858 * just in case its running on a different cpu
2860 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2864 local_irq_save(flags);
2866 local_irq_restore(flags);
2868 EXPORT_SYMBOL(napi_complete);
2870 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2871 int (*poll)(struct napi_struct *, int), int weight)
2873 INIT_LIST_HEAD(&napi->poll_list);
2874 napi->gro_count = 0;
2875 napi->gro_list = NULL;
2878 napi->weight = weight;
2879 list_add(&napi->dev_list, &dev->napi_list);
2881 #ifdef CONFIG_NETPOLL
2882 spin_lock_init(&napi->poll_lock);
2883 napi->poll_owner = -1;
2885 set_bit(NAPI_STATE_SCHED, &napi->state);
2887 EXPORT_SYMBOL(netif_napi_add);
2889 void netif_napi_del(struct napi_struct *napi)
2891 struct sk_buff *skb, *next;
2893 list_del_init(&napi->dev_list);
2894 napi_free_frags(napi);
2896 for (skb = napi->gro_list; skb; skb = next) {
2902 napi->gro_list = NULL;
2903 napi->gro_count = 0;
2905 EXPORT_SYMBOL(netif_napi_del);
2908 static void net_rx_action(struct softirq_action *h)
2910 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2911 unsigned long time_limit = jiffies + 2;
2912 int budget = netdev_budget;
2915 local_irq_disable();
2917 while (!list_empty(list)) {
2918 struct napi_struct *n;
2921 /* If softirq window is exhuasted then punt.
2922 * Allow this to run for 2 jiffies since which will allow
2923 * an average latency of 1.5/HZ.
2925 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2930 /* Even though interrupts have been re-enabled, this
2931 * access is safe because interrupts can only add new
2932 * entries to the tail of this list, and only ->poll()
2933 * calls can remove this head entry from the list.
2935 n = list_entry(list->next, struct napi_struct, poll_list);
2937 have = netpoll_poll_lock(n);
2941 /* This NAPI_STATE_SCHED test is for avoiding a race
2942 * with netpoll's poll_napi(). Only the entity which
2943 * obtains the lock and sees NAPI_STATE_SCHED set will
2944 * actually make the ->poll() call. Therefore we avoid
2945 * accidently calling ->poll() when NAPI is not scheduled.
2948 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
2949 work = n->poll(n, weight);
2953 WARN_ON_ONCE(work > weight);
2957 local_irq_disable();
2959 /* Drivers must not modify the NAPI state if they
2960 * consume the entire weight. In such cases this code
2961 * still "owns" the NAPI instance and therefore can
2962 * move the instance around on the list at-will.
2964 if (unlikely(work == weight)) {
2965 if (unlikely(napi_disable_pending(n))) {
2968 local_irq_disable();
2970 list_move_tail(&n->poll_list, list);
2973 netpoll_poll_unlock(have);
2978 #ifdef CONFIG_NET_DMA
2980 * There may not be any more sk_buffs coming right now, so push
2981 * any pending DMA copies to hardware
2983 dma_issue_pending_all();
2989 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2990 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2994 static gifconf_func_t *gifconf_list[NPROTO];
2997 * register_gifconf - register a SIOCGIF handler
2998 * @family: Address family
2999 * @gifconf: Function handler
3001 * Register protocol dependent address dumping routines. The handler
3002 * that is passed must not be freed or reused until it has been replaced
3003 * by another handler.
3005 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3007 if (family >= NPROTO)
3009 gifconf_list[family] = gifconf;
3012 EXPORT_SYMBOL(register_gifconf);
3016 * Map an interface index to its name (SIOCGIFNAME)
3020 * We need this ioctl for efficient implementation of the
3021 * if_indextoname() function required by the IPv6 API. Without
3022 * it, we would have to search all the interfaces to find a
3026 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3028 struct net_device *dev;
3032 * Fetch the caller's info block.
3035 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3039 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3045 strcpy(ifr.ifr_name, dev->name);
3048 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3054 * Perform a SIOCGIFCONF call. This structure will change
3055 * size eventually, and there is nothing I can do about it.
3056 * Thus we will need a 'compatibility mode'.
3059 static int dev_ifconf(struct net *net, char __user *arg)
3062 struct net_device *dev;
3069 * Fetch the caller's info block.
3072 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3079 * Loop over the interfaces, and write an info block for each.
3083 for_each_netdev(net, dev) {
3084 for (i = 0; i < NPROTO; i++) {
3085 if (gifconf_list[i]) {
3088 done = gifconf_list[i](dev, NULL, 0);
3090 done = gifconf_list[i](dev, pos + total,
3100 * All done. Write the updated control block back to the caller.
3102 ifc.ifc_len = total;
3105 * Both BSD and Solaris return 0 here, so we do too.
3107 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3110 #ifdef CONFIG_PROC_FS
3112 * This is invoked by the /proc filesystem handler to display a device
3115 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3118 struct net *net = seq_file_net(seq);
3120 struct net_device *dev;
3124 return SEQ_START_TOKEN;
3127 for_each_netdev_rcu(net, dev)
3134 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3136 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3137 first_net_device(seq_file_net(seq)) :
3138 next_net_device((struct net_device *)v);
3141 return rcu_dereference(dev);
3144 void dev_seq_stop(struct seq_file *seq, void *v)
3150 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3152 const struct net_device_stats *stats = dev_get_stats(dev);
3154 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3155 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3156 dev->name, stats->rx_bytes, stats->rx_packets,
3158 stats->rx_dropped + stats->rx_missed_errors,
3159 stats->rx_fifo_errors,
3160 stats->rx_length_errors + stats->rx_over_errors +
3161 stats->rx_crc_errors + stats->rx_frame_errors,
3162 stats->rx_compressed, stats->multicast,
3163 stats->tx_bytes, stats->tx_packets,
3164 stats->tx_errors, stats->tx_dropped,
3165 stats->tx_fifo_errors, stats->collisions,
3166 stats->tx_carrier_errors +
3167 stats->tx_aborted_errors +
3168 stats->tx_window_errors +
3169 stats->tx_heartbeat_errors,
3170 stats->tx_compressed);
3174 * Called from the PROCfs module. This now uses the new arbitrary sized
3175 * /proc/net interface to create /proc/net/dev
3177 static int dev_seq_show(struct seq_file *seq, void *v)
3179 if (v == SEQ_START_TOKEN)
3180 seq_puts(seq, "Inter-| Receive "
3182 " face |bytes packets errs drop fifo frame "
3183 "compressed multicast|bytes packets errs "
3184 "drop fifo colls carrier compressed\n");
3186 dev_seq_printf_stats(seq, v);
3190 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3192 struct netif_rx_stats *rc = NULL;
3194 while (*pos < nr_cpu_ids)
3195 if (cpu_online(*pos)) {
3196 rc = &per_cpu(netdev_rx_stat, *pos);
3203 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3205 return softnet_get_online(pos);
3208 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3211 return softnet_get_online(pos);
3214 static void softnet_seq_stop(struct seq_file *seq, void *v)
3218 static int softnet_seq_show(struct seq_file *seq, void *v)
3220 struct netif_rx_stats *s = v;
3222 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3223 s->total, s->dropped, s->time_squeeze, 0,
3224 0, 0, 0, 0, /* was fastroute */
3229 static const struct seq_operations dev_seq_ops = {
3230 .start = dev_seq_start,
3231 .next = dev_seq_next,
3232 .stop = dev_seq_stop,
3233 .show = dev_seq_show,
3236 static int dev_seq_open(struct inode *inode, struct file *file)
3238 return seq_open_net(inode, file, &dev_seq_ops,
3239 sizeof(struct seq_net_private));
3242 static const struct file_operations dev_seq_fops = {
3243 .owner = THIS_MODULE,
3244 .open = dev_seq_open,
3246 .llseek = seq_lseek,
3247 .release = seq_release_net,
3250 static const struct seq_operations softnet_seq_ops = {
3251 .start = softnet_seq_start,
3252 .next = softnet_seq_next,
3253 .stop = softnet_seq_stop,
3254 .show = softnet_seq_show,
3257 static int softnet_seq_open(struct inode *inode, struct file *file)
3259 return seq_open(file, &softnet_seq_ops);
3262 static const struct file_operations softnet_seq_fops = {
3263 .owner = THIS_MODULE,
3264 .open = softnet_seq_open,
3266 .llseek = seq_lseek,
3267 .release = seq_release,
3270 static void *ptype_get_idx(loff_t pos)
3272 struct packet_type *pt = NULL;
3276 list_for_each_entry_rcu(pt, &ptype_all, list) {
3282 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3283 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3292 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3296 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3299 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3301 struct packet_type *pt;
3302 struct list_head *nxt;
3306 if (v == SEQ_START_TOKEN)
3307 return ptype_get_idx(0);
3310 nxt = pt->list.next;
3311 if (pt->type == htons(ETH_P_ALL)) {
3312 if (nxt != &ptype_all)
3315 nxt = ptype_base[0].next;
3317 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3319 while (nxt == &ptype_base[hash]) {
3320 if (++hash >= PTYPE_HASH_SIZE)
3322 nxt = ptype_base[hash].next;
3325 return list_entry(nxt, struct packet_type, list);
3328 static void ptype_seq_stop(struct seq_file *seq, void *v)
3334 static int ptype_seq_show(struct seq_file *seq, void *v)
3336 struct packet_type *pt = v;
3338 if (v == SEQ_START_TOKEN)
3339 seq_puts(seq, "Type Device Function\n");
3340 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3341 if (pt->type == htons(ETH_P_ALL))
3342 seq_puts(seq, "ALL ");
3344 seq_printf(seq, "%04x", ntohs(pt->type));
3346 seq_printf(seq, " %-8s %pF\n",
3347 pt->dev ? pt->dev->name : "", pt->func);
3353 static const struct seq_operations ptype_seq_ops = {
3354 .start = ptype_seq_start,
3355 .next = ptype_seq_next,
3356 .stop = ptype_seq_stop,
3357 .show = ptype_seq_show,
3360 static int ptype_seq_open(struct inode *inode, struct file *file)
3362 return seq_open_net(inode, file, &ptype_seq_ops,
3363 sizeof(struct seq_net_private));
3366 static const struct file_operations ptype_seq_fops = {
3367 .owner = THIS_MODULE,
3368 .open = ptype_seq_open,
3370 .llseek = seq_lseek,
3371 .release = seq_release_net,
3375 static int __net_init dev_proc_net_init(struct net *net)
3379 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3381 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3383 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3386 if (wext_proc_init(net))
3392 proc_net_remove(net, "ptype");
3394 proc_net_remove(net, "softnet_stat");
3396 proc_net_remove(net, "dev");
3400 static void __net_exit dev_proc_net_exit(struct net *net)
3402 wext_proc_exit(net);
3404 proc_net_remove(net, "ptype");
3405 proc_net_remove(net, "softnet_stat");
3406 proc_net_remove(net, "dev");
3409 static struct pernet_operations __net_initdata dev_proc_ops = {
3410 .init = dev_proc_net_init,
3411 .exit = dev_proc_net_exit,
3414 static int __init dev_proc_init(void)
3416 return register_pernet_subsys(&dev_proc_ops);
3419 #define dev_proc_init() 0
3420 #endif /* CONFIG_PROC_FS */
3424 * netdev_set_master - set up master/slave pair
3425 * @slave: slave device
3426 * @master: new master device
3428 * Changes the master device of the slave. Pass %NULL to break the
3429 * bonding. The caller must hold the RTNL semaphore. On a failure
3430 * a negative errno code is returned. On success the reference counts
3431 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3432 * function returns zero.
3434 int netdev_set_master(struct net_device *slave, struct net_device *master)
3436 struct net_device *old = slave->master;
3446 slave->master = master;
3454 slave->flags |= IFF_SLAVE;
3456 slave->flags &= ~IFF_SLAVE;
3458 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3461 EXPORT_SYMBOL(netdev_set_master);
3463 static void dev_change_rx_flags(struct net_device *dev, int flags)
3465 const struct net_device_ops *ops = dev->netdev_ops;
3467 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3468 ops->ndo_change_rx_flags(dev, flags);
3471 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3473 unsigned short old_flags = dev->flags;
3479 dev->flags |= IFF_PROMISC;
3480 dev->promiscuity += inc;
3481 if (dev->promiscuity == 0) {
3484 * If inc causes overflow, untouch promisc and return error.
3487 dev->flags &= ~IFF_PROMISC;
3489 dev->promiscuity -= inc;
3490 printk(KERN_WARNING "%s: promiscuity touches roof, "
3491 "set promiscuity failed, promiscuity feature "
3492 "of device might be broken.\n", dev->name);
3496 if (dev->flags != old_flags) {
3497 printk(KERN_INFO "device %s %s promiscuous mode\n",
3498 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3500 if (audit_enabled) {
3501 current_uid_gid(&uid, &gid);
3502 audit_log(current->audit_context, GFP_ATOMIC,
3503 AUDIT_ANOM_PROMISCUOUS,
3504 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3505 dev->name, (dev->flags & IFF_PROMISC),
3506 (old_flags & IFF_PROMISC),
3507 audit_get_loginuid(current),
3509 audit_get_sessionid(current));
3512 dev_change_rx_flags(dev, IFF_PROMISC);
3518 * dev_set_promiscuity - update promiscuity count on a device
3522 * Add or remove promiscuity from a device. While the count in the device
3523 * remains above zero the interface remains promiscuous. Once it hits zero
3524 * the device reverts back to normal filtering operation. A negative inc
3525 * value is used to drop promiscuity on the device.
3526 * Return 0 if successful or a negative errno code on error.
3528 int dev_set_promiscuity(struct net_device *dev, int inc)
3530 unsigned short old_flags = dev->flags;
3533 err = __dev_set_promiscuity(dev, inc);
3536 if (dev->flags != old_flags)
3537 dev_set_rx_mode(dev);
3540 EXPORT_SYMBOL(dev_set_promiscuity);
3543 * dev_set_allmulti - update allmulti count on a device
3547 * Add or remove reception of all multicast frames to a device. While the
3548 * count in the device remains above zero the interface remains listening
3549 * to all interfaces. Once it hits zero the device reverts back to normal
3550 * filtering operation. A negative @inc value is used to drop the counter
3551 * when releasing a resource needing all multicasts.
3552 * Return 0 if successful or a negative errno code on error.
3555 int dev_set_allmulti(struct net_device *dev, int inc)
3557 unsigned short old_flags = dev->flags;
3561 dev->flags |= IFF_ALLMULTI;
3562 dev->allmulti += inc;
3563 if (dev->allmulti == 0) {
3566 * If inc causes overflow, untouch allmulti and return error.
3569 dev->flags &= ~IFF_ALLMULTI;
3571 dev->allmulti -= inc;
3572 printk(KERN_WARNING "%s: allmulti touches roof, "
3573 "set allmulti failed, allmulti feature of "
3574 "device might be broken.\n", dev->name);
3578 if (dev->flags ^ old_flags) {
3579 dev_change_rx_flags(dev, IFF_ALLMULTI);
3580 dev_set_rx_mode(dev);
3584 EXPORT_SYMBOL(dev_set_allmulti);
3587 * Upload unicast and multicast address lists to device and
3588 * configure RX filtering. When the device doesn't support unicast
3589 * filtering it is put in promiscuous mode while unicast addresses
3592 void __dev_set_rx_mode(struct net_device *dev)
3594 const struct net_device_ops *ops = dev->netdev_ops;
3596 /* dev_open will call this function so the list will stay sane. */
3597 if (!(dev->flags&IFF_UP))
3600 if (!netif_device_present(dev))
3603 if (ops->ndo_set_rx_mode)
3604 ops->ndo_set_rx_mode(dev);
3606 /* Unicast addresses changes may only happen under the rtnl,
3607 * therefore calling __dev_set_promiscuity here is safe.
3609 if (dev->uc.count > 0 && !dev->uc_promisc) {
3610 __dev_set_promiscuity(dev, 1);
3611 dev->uc_promisc = 1;
3612 } else if (dev->uc.count == 0 && dev->uc_promisc) {
3613 __dev_set_promiscuity(dev, -1);
3614 dev->uc_promisc = 0;
3617 if (ops->ndo_set_multicast_list)
3618 ops->ndo_set_multicast_list(dev);
3622 void dev_set_rx_mode(struct net_device *dev)
3624 netif_addr_lock_bh(dev);
3625 __dev_set_rx_mode(dev);
3626 netif_addr_unlock_bh(dev);
3629 /* hw addresses list handling functions */
3631 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3632 int addr_len, unsigned char addr_type)
3634 struct netdev_hw_addr *ha;
3637 if (addr_len > MAX_ADDR_LEN)
3640 list_for_each_entry(ha, &list->list, list) {
3641 if (!memcmp(ha->addr, addr, addr_len) &&
3642 ha->type == addr_type) {
3649 alloc_size = sizeof(*ha);
3650 if (alloc_size < L1_CACHE_BYTES)
3651 alloc_size = L1_CACHE_BYTES;
3652 ha = kmalloc(alloc_size, GFP_ATOMIC);
3655 memcpy(ha->addr, addr, addr_len);
3656 ha->type = addr_type;
3659 list_add_tail_rcu(&ha->list, &list->list);
3664 static void ha_rcu_free(struct rcu_head *head)
3666 struct netdev_hw_addr *ha;
3668 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3672 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3673 int addr_len, unsigned char addr_type)
3675 struct netdev_hw_addr *ha;
3677 list_for_each_entry(ha, &list->list, list) {
3678 if (!memcmp(ha->addr, addr, addr_len) &&
3679 (ha->type == addr_type || !addr_type)) {
3682 list_del_rcu(&ha->list);
3683 call_rcu(&ha->rcu_head, ha_rcu_free);
3691 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3692 struct netdev_hw_addr_list *from_list,
3694 unsigned char addr_type)
3697 struct netdev_hw_addr *ha, *ha2;
3700 list_for_each_entry(ha, &from_list->list, list) {
3701 type = addr_type ? addr_type : ha->type;
3702 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3709 list_for_each_entry(ha2, &from_list->list, list) {
3712 type = addr_type ? addr_type : ha2->type;
3713 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3718 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3719 struct netdev_hw_addr_list *from_list,
3721 unsigned char addr_type)
3723 struct netdev_hw_addr *ha;
3726 list_for_each_entry(ha, &from_list->list, list) {
3727 type = addr_type ? addr_type : ha->type;
3728 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3732 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3733 struct netdev_hw_addr_list *from_list,
3737 struct netdev_hw_addr *ha, *tmp;
3739 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3741 err = __hw_addr_add(to_list, ha->addr,
3742 addr_len, ha->type);
3747 } else if (ha->refcount == 1) {
3748 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3749 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3755 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3756 struct netdev_hw_addr_list *from_list,
3759 struct netdev_hw_addr *ha, *tmp;
3761 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3763 __hw_addr_del(to_list, ha->addr,
3764 addr_len, ha->type);
3766 __hw_addr_del(from_list, ha->addr,
3767 addr_len, ha->type);
3772 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3774 struct netdev_hw_addr *ha, *tmp;
3776 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3777 list_del_rcu(&ha->list);
3778 call_rcu(&ha->rcu_head, ha_rcu_free);
3783 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3785 INIT_LIST_HEAD(&list->list);
3789 /* Device addresses handling functions */
3791 static void dev_addr_flush(struct net_device *dev)
3793 /* rtnl_mutex must be held here */
3795 __hw_addr_flush(&dev->dev_addrs);
3796 dev->dev_addr = NULL;
3799 static int dev_addr_init(struct net_device *dev)
3801 unsigned char addr[MAX_ADDR_LEN];
3802 struct netdev_hw_addr *ha;
3805 /* rtnl_mutex must be held here */
3807 __hw_addr_init(&dev->dev_addrs);
3808 memset(addr, 0, sizeof(addr));
3809 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3810 NETDEV_HW_ADDR_T_LAN);
3813 * Get the first (previously created) address from the list
3814 * and set dev_addr pointer to this location.
3816 ha = list_first_entry(&dev->dev_addrs.list,
3817 struct netdev_hw_addr, list);
3818 dev->dev_addr = ha->addr;
3824 * dev_addr_add - Add a device address
3826 * @addr: address to add
3827 * @addr_type: address type
3829 * Add a device address to the device or increase the reference count if
3830 * it already exists.
3832 * The caller must hold the rtnl_mutex.
3834 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3835 unsigned char addr_type)
3841 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3843 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3846 EXPORT_SYMBOL(dev_addr_add);
3849 * dev_addr_del - Release a device address.
3851 * @addr: address to delete
3852 * @addr_type: address type
3854 * Release reference to a device address and remove it from the device
3855 * if the reference count drops to zero.
3857 * The caller must hold the rtnl_mutex.
3859 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3860 unsigned char addr_type)
3863 struct netdev_hw_addr *ha;
3868 * We can not remove the first address from the list because
3869 * dev->dev_addr points to that.
3871 ha = list_first_entry(&dev->dev_addrs.list,
3872 struct netdev_hw_addr, list);
3873 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3876 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3879 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3882 EXPORT_SYMBOL(dev_addr_del);
3885 * dev_addr_add_multiple - Add device addresses from another device
3886 * @to_dev: device to which addresses will be added
3887 * @from_dev: device from which addresses will be added
3888 * @addr_type: address type - 0 means type will be used from from_dev
3890 * Add device addresses of the one device to another.
3892 * The caller must hold the rtnl_mutex.
3894 int dev_addr_add_multiple(struct net_device *to_dev,
3895 struct net_device *from_dev,
3896 unsigned char addr_type)
3902 if (from_dev->addr_len != to_dev->addr_len)
3904 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3905 to_dev->addr_len, addr_type);
3907 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3910 EXPORT_SYMBOL(dev_addr_add_multiple);
3913 * dev_addr_del_multiple - Delete device addresses by another device
3914 * @to_dev: device where the addresses will be deleted
3915 * @from_dev: device by which addresses the addresses will be deleted
3916 * @addr_type: address type - 0 means type will used from from_dev
3918 * Deletes addresses in to device by the list of addresses in from device.
3920 * The caller must hold the rtnl_mutex.
3922 int dev_addr_del_multiple(struct net_device *to_dev,
3923 struct net_device *from_dev,
3924 unsigned char addr_type)
3928 if (from_dev->addr_len != to_dev->addr_len)
3930 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3931 to_dev->addr_len, addr_type);
3932 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3935 EXPORT_SYMBOL(dev_addr_del_multiple);
3937 /* multicast addresses handling functions */
3939 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3940 void *addr, int alen, int glbl)
3942 struct dev_addr_list *da;
3944 for (; (da = *list) != NULL; list = &da->next) {
3945 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3946 alen == da->da_addrlen) {
3948 int old_glbl = da->da_gusers;
3965 int __dev_addr_add(struct dev_addr_list **list, int *count,
3966 void *addr, int alen, int glbl)
3968 struct dev_addr_list *da;
3970 for (da = *list; da != NULL; da = da->next) {
3971 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3972 da->da_addrlen == alen) {
3974 int old_glbl = da->da_gusers;
3984 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3987 memcpy(da->da_addr, addr, alen);
3988 da->da_addrlen = alen;
3990 da->da_gusers = glbl ? 1 : 0;
3998 * dev_unicast_delete - Release secondary unicast address.
4000 * @addr: address to delete
4002 * Release reference to a secondary unicast address and remove it
4003 * from the device if the reference count drops to zero.
4005 * The caller must hold the rtnl_mutex.
4007 int dev_unicast_delete(struct net_device *dev, void *addr)
4013 netif_addr_lock_bh(dev);
4014 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
4015 NETDEV_HW_ADDR_T_UNICAST);
4017 __dev_set_rx_mode(dev);
4018 netif_addr_unlock_bh(dev);
4021 EXPORT_SYMBOL(dev_unicast_delete);
4024 * dev_unicast_add - add a secondary unicast address
4026 * @addr: address to add
4028 * Add a secondary unicast address to the device or increase
4029 * the reference count if it already exists.
4031 * The caller must hold the rtnl_mutex.
4033 int dev_unicast_add(struct net_device *dev, void *addr)
4039 netif_addr_lock_bh(dev);
4040 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
4041 NETDEV_HW_ADDR_T_UNICAST);
4043 __dev_set_rx_mode(dev);
4044 netif_addr_unlock_bh(dev);
4047 EXPORT_SYMBOL(dev_unicast_add);
4049 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
4050 struct dev_addr_list **from, int *from_count)
4052 struct dev_addr_list *da, *next;
4056 while (da != NULL) {
4058 if (!da->da_synced) {
4059 err = __dev_addr_add(to, to_count,
4060 da->da_addr, da->da_addrlen, 0);
4065 } else if (da->da_users == 1) {
4066 __dev_addr_delete(to, to_count,
4067 da->da_addr, da->da_addrlen, 0);
4068 __dev_addr_delete(from, from_count,
4069 da->da_addr, da->da_addrlen, 0);
4075 EXPORT_SYMBOL_GPL(__dev_addr_sync);
4077 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4078 struct dev_addr_list **from, int *from_count)
4080 struct dev_addr_list *da, *next;
4083 while (da != NULL) {
4085 if (da->da_synced) {
4086 __dev_addr_delete(to, to_count,
4087 da->da_addr, da->da_addrlen, 0);
4089 __dev_addr_delete(from, from_count,
4090 da->da_addr, da->da_addrlen, 0);
4095 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4098 * dev_unicast_sync - Synchronize device's unicast list to another device
4099 * @to: destination device
4100 * @from: source device
4102 * Add newly added addresses to the destination device and release
4103 * addresses that have no users left. The source device must be
4104 * locked by netif_tx_lock_bh.
4106 * This function is intended to be called from the dev->set_rx_mode
4107 * function of layered software devices.
4109 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4113 if (to->addr_len != from->addr_len)
4116 netif_addr_lock_bh(to);
4117 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4119 __dev_set_rx_mode(to);
4120 netif_addr_unlock_bh(to);
4123 EXPORT_SYMBOL(dev_unicast_sync);
4126 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4127 * @to: destination device
4128 * @from: source device
4130 * Remove all addresses that were added to the destination device by
4131 * dev_unicast_sync(). This function is intended to be called from the
4132 * dev->stop function of layered software devices.
4134 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4136 if (to->addr_len != from->addr_len)
4139 netif_addr_lock_bh(from);
4140 netif_addr_lock(to);
4141 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4142 __dev_set_rx_mode(to);
4143 netif_addr_unlock(to);
4144 netif_addr_unlock_bh(from);
4146 EXPORT_SYMBOL(dev_unicast_unsync);
4148 static void dev_unicast_flush(struct net_device *dev)
4150 netif_addr_lock_bh(dev);
4151 __hw_addr_flush(&dev->uc);
4152 netif_addr_unlock_bh(dev);
4155 static void dev_unicast_init(struct net_device *dev)
4157 __hw_addr_init(&dev->uc);
4161 static void __dev_addr_discard(struct dev_addr_list **list)
4163 struct dev_addr_list *tmp;
4165 while (*list != NULL) {
4168 if (tmp->da_users > tmp->da_gusers)
4169 printk("__dev_addr_discard: address leakage! "
4170 "da_users=%d\n", tmp->da_users);
4175 static void dev_addr_discard(struct net_device *dev)
4177 netif_addr_lock_bh(dev);
4179 __dev_addr_discard(&dev->mc_list);
4182 netif_addr_unlock_bh(dev);
4186 * dev_get_flags - get flags reported to userspace
4189 * Get the combination of flag bits exported through APIs to userspace.
4191 unsigned dev_get_flags(const struct net_device *dev)
4195 flags = (dev->flags & ~(IFF_PROMISC |
4200 (dev->gflags & (IFF_PROMISC |
4203 if (netif_running(dev)) {
4204 if (netif_oper_up(dev))
4205 flags |= IFF_RUNNING;
4206 if (netif_carrier_ok(dev))
4207 flags |= IFF_LOWER_UP;
4208 if (netif_dormant(dev))
4209 flags |= IFF_DORMANT;
4214 EXPORT_SYMBOL(dev_get_flags);
4217 * dev_change_flags - change device settings
4219 * @flags: device state flags
4221 * Change settings on device based state flags. The flags are
4222 * in the userspace exported format.
4224 int dev_change_flags(struct net_device *dev, unsigned flags)
4227 int old_flags = dev->flags;
4232 * Set the flags on our device.
4235 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4236 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4238 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4242 * Load in the correct multicast list now the flags have changed.
4245 if ((old_flags ^ flags) & IFF_MULTICAST)
4246 dev_change_rx_flags(dev, IFF_MULTICAST);
4248 dev_set_rx_mode(dev);
4251 * Have we downed the interface. We handle IFF_UP ourselves
4252 * according to user attempts to set it, rather than blindly
4257 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4258 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4261 dev_set_rx_mode(dev);
4264 if (dev->flags & IFF_UP &&
4265 ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
4267 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4269 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4270 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4272 dev->gflags ^= IFF_PROMISC;
4273 dev_set_promiscuity(dev, inc);
4276 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4277 is important. Some (broken) drivers set IFF_PROMISC, when
4278 IFF_ALLMULTI is requested not asking us and not reporting.
4280 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4281 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4283 dev->gflags ^= IFF_ALLMULTI;
4284 dev_set_allmulti(dev, inc);
4287 /* Exclude state transition flags, already notified */
4288 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4290 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4294 EXPORT_SYMBOL(dev_change_flags);
4297 * dev_set_mtu - Change maximum transfer unit
4299 * @new_mtu: new transfer unit
4301 * Change the maximum transfer size of the network device.
4303 int dev_set_mtu(struct net_device *dev, int new_mtu)
4305 const struct net_device_ops *ops = dev->netdev_ops;
4308 if (new_mtu == dev->mtu)
4311 /* MTU must be positive. */
4315 if (!netif_device_present(dev))
4319 if (ops->ndo_change_mtu)
4320 err = ops->ndo_change_mtu(dev, new_mtu);
4324 if (!err && dev->flags & IFF_UP)
4325 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4328 EXPORT_SYMBOL(dev_set_mtu);
4331 * dev_set_mac_address - Change Media Access Control Address
4335 * Change the hardware (MAC) address of the device
4337 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4339 const struct net_device_ops *ops = dev->netdev_ops;
4342 if (!ops->ndo_set_mac_address)
4344 if (sa->sa_family != dev->type)
4346 if (!netif_device_present(dev))
4348 err = ops->ndo_set_mac_address(dev, sa);
4350 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4353 EXPORT_SYMBOL(dev_set_mac_address);
4356 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4358 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4361 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4367 case SIOCGIFFLAGS: /* Get interface flags */
4368 ifr->ifr_flags = (short) dev_get_flags(dev);
4371 case SIOCGIFMETRIC: /* Get the metric on the interface
4372 (currently unused) */
4373 ifr->ifr_metric = 0;
4376 case SIOCGIFMTU: /* Get the MTU of a device */
4377 ifr->ifr_mtu = dev->mtu;
4382 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4384 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4385 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4386 ifr->ifr_hwaddr.sa_family = dev->type;
4394 ifr->ifr_map.mem_start = dev->mem_start;
4395 ifr->ifr_map.mem_end = dev->mem_end;
4396 ifr->ifr_map.base_addr = dev->base_addr;
4397 ifr->ifr_map.irq = dev->irq;
4398 ifr->ifr_map.dma = dev->dma;
4399 ifr->ifr_map.port = dev->if_port;
4403 ifr->ifr_ifindex = dev->ifindex;
4407 ifr->ifr_qlen = dev->tx_queue_len;
4411 /* dev_ioctl() should ensure this case
4423 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4425 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4428 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4429 const struct net_device_ops *ops;
4434 ops = dev->netdev_ops;
4437 case SIOCSIFFLAGS: /* Set interface flags */
4438 return dev_change_flags(dev, ifr->ifr_flags);
4440 case SIOCSIFMETRIC: /* Set the metric on the interface
4441 (currently unused) */
4444 case SIOCSIFMTU: /* Set the MTU of a device */
4445 return dev_set_mtu(dev, ifr->ifr_mtu);
4448 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4450 case SIOCSIFHWBROADCAST:
4451 if (ifr->ifr_hwaddr.sa_family != dev->type)
4453 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4454 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4455 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4459 if (ops->ndo_set_config) {
4460 if (!netif_device_present(dev))
4462 return ops->ndo_set_config(dev, &ifr->ifr_map);
4467 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4468 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4470 if (!netif_device_present(dev))
4472 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4476 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4477 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4479 if (!netif_device_present(dev))
4481 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4485 if (ifr->ifr_qlen < 0)
4487 dev->tx_queue_len = ifr->ifr_qlen;
4491 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4492 return dev_change_name(dev, ifr->ifr_newname);
4495 * Unknown or private ioctl
4498 if ((cmd >= SIOCDEVPRIVATE &&
4499 cmd <= SIOCDEVPRIVATE + 15) ||
4500 cmd == SIOCBONDENSLAVE ||
4501 cmd == SIOCBONDRELEASE ||
4502 cmd == SIOCBONDSETHWADDR ||
4503 cmd == SIOCBONDSLAVEINFOQUERY ||
4504 cmd == SIOCBONDINFOQUERY ||
4505 cmd == SIOCBONDCHANGEACTIVE ||
4506 cmd == SIOCGMIIPHY ||
4507 cmd == SIOCGMIIREG ||
4508 cmd == SIOCSMIIREG ||
4509 cmd == SIOCBRADDIF ||
4510 cmd == SIOCBRDELIF ||
4511 cmd == SIOCSHWTSTAMP ||
4512 cmd == SIOCWANDEV) {
4514 if (ops->ndo_do_ioctl) {
4515 if (netif_device_present(dev))
4516 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4528 * This function handles all "interface"-type I/O control requests. The actual
4529 * 'doing' part of this is dev_ifsioc above.
4533 * dev_ioctl - network device ioctl
4534 * @net: the applicable net namespace
4535 * @cmd: command to issue
4536 * @arg: pointer to a struct ifreq in user space
4538 * Issue ioctl functions to devices. This is normally called by the
4539 * user space syscall interfaces but can sometimes be useful for
4540 * other purposes. The return value is the return from the syscall if
4541 * positive or a negative errno code on error.
4544 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4550 /* One special case: SIOCGIFCONF takes ifconf argument
4551 and requires shared lock, because it sleeps writing
4555 if (cmd == SIOCGIFCONF) {
4557 ret = dev_ifconf(net, (char __user *) arg);
4561 if (cmd == SIOCGIFNAME)
4562 return dev_ifname(net, (struct ifreq __user *)arg);
4564 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4567 ifr.ifr_name[IFNAMSIZ-1] = 0;
4569 colon = strchr(ifr.ifr_name, ':');
4574 * See which interface the caller is talking about.
4579 * These ioctl calls:
4580 * - can be done by all.
4581 * - atomic and do not require locking.
4592 dev_load(net, ifr.ifr_name);
4594 ret = dev_ifsioc_locked(net, &ifr, cmd);
4599 if (copy_to_user(arg, &ifr,
4600 sizeof(struct ifreq)))
4606 dev_load(net, ifr.ifr_name);
4608 ret = dev_ethtool(net, &ifr);
4613 if (copy_to_user(arg, &ifr,
4614 sizeof(struct ifreq)))
4620 * These ioctl calls:
4621 * - require superuser power.
4622 * - require strict serialization.
4628 if (!capable(CAP_NET_ADMIN))
4630 dev_load(net, ifr.ifr_name);
4632 ret = dev_ifsioc(net, &ifr, cmd);
4637 if (copy_to_user(arg, &ifr,
4638 sizeof(struct ifreq)))
4644 * These ioctl calls:
4645 * - require superuser power.
4646 * - require strict serialization.
4647 * - do not return a value
4657 case SIOCSIFHWBROADCAST:
4660 case SIOCBONDENSLAVE:
4661 case SIOCBONDRELEASE:
4662 case SIOCBONDSETHWADDR:
4663 case SIOCBONDCHANGEACTIVE:
4667 if (!capable(CAP_NET_ADMIN))
4670 case SIOCBONDSLAVEINFOQUERY:
4671 case SIOCBONDINFOQUERY:
4672 dev_load(net, ifr.ifr_name);
4674 ret = dev_ifsioc(net, &ifr, cmd);
4679 /* Get the per device memory space. We can add this but
4680 * currently do not support it */
4682 /* Set the per device memory buffer space.
4683 * Not applicable in our case */
4688 * Unknown or private ioctl.
4691 if (cmd == SIOCWANDEV ||
4692 (cmd >= SIOCDEVPRIVATE &&
4693 cmd <= SIOCDEVPRIVATE + 15)) {
4694 dev_load(net, ifr.ifr_name);
4696 ret = dev_ifsioc(net, &ifr, cmd);
4698 if (!ret && copy_to_user(arg, &ifr,
4699 sizeof(struct ifreq)))
4703 /* Take care of Wireless Extensions */
4704 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4705 return wext_handle_ioctl(net, &ifr, cmd, arg);
4712 * dev_new_index - allocate an ifindex
4713 * @net: the applicable net namespace
4715 * Returns a suitable unique value for a new device interface
4716 * number. The caller must hold the rtnl semaphore or the
4717 * dev_base_lock to be sure it remains unique.
4719 static int dev_new_index(struct net *net)
4725 if (!__dev_get_by_index(net, ifindex))
4730 /* Delayed registration/unregisteration */
4731 static LIST_HEAD(net_todo_list);
4733 static void net_set_todo(struct net_device *dev)
4735 list_add_tail(&dev->todo_list, &net_todo_list);
4738 static void rollback_registered_many(struct list_head *head)
4740 struct net_device *dev;
4742 BUG_ON(dev_boot_phase);
4745 list_for_each_entry(dev, head, unreg_list) {
4746 /* Some devices call without registering
4747 * for initialization unwind.
4749 if (dev->reg_state == NETREG_UNINITIALIZED) {
4750 pr_debug("unregister_netdevice: device %s/%p never "
4751 "was registered\n", dev->name, dev);
4757 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4759 /* If device is running, close it first. */
4762 /* And unlink it from device chain. */
4763 unlist_netdevice(dev);
4765 dev->reg_state = NETREG_UNREGISTERING;
4770 list_for_each_entry(dev, head, unreg_list) {
4771 /* Shutdown queueing discipline. */
4775 /* Notify protocols, that we are about to destroy
4776 this device. They should clean all the things.
4778 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4781 * Flush the unicast and multicast chains
4783 dev_unicast_flush(dev);
4784 dev_addr_discard(dev);
4786 if (dev->netdev_ops->ndo_uninit)
4787 dev->netdev_ops->ndo_uninit(dev);
4789 /* Notifier chain MUST detach us from master device. */
4790 WARN_ON(dev->master);
4792 /* Remove entries from kobject tree */
4793 netdev_unregister_kobject(dev);
4798 list_for_each_entry(dev, head, unreg_list)
4802 static void rollback_registered(struct net_device *dev)
4806 list_add(&dev->unreg_list, &single);
4807 rollback_registered_many(&single);
4810 static void __netdev_init_queue_locks_one(struct net_device *dev,
4811 struct netdev_queue *dev_queue,
4814 spin_lock_init(&dev_queue->_xmit_lock);
4815 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4816 dev_queue->xmit_lock_owner = -1;
4819 static void netdev_init_queue_locks(struct net_device *dev)
4821 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4822 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4825 unsigned long netdev_fix_features(unsigned long features, const char *name)
4827 /* Fix illegal SG+CSUM combinations. */
4828 if ((features & NETIF_F_SG) &&
4829 !(features & NETIF_F_ALL_CSUM)) {
4831 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4832 "checksum feature.\n", name);
4833 features &= ~NETIF_F_SG;
4836 /* TSO requires that SG is present as well. */
4837 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4839 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4840 "SG feature.\n", name);
4841 features &= ~NETIF_F_TSO;
4844 if (features & NETIF_F_UFO) {
4845 if (!(features & NETIF_F_GEN_CSUM)) {
4847 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4848 "since no NETIF_F_HW_CSUM feature.\n",
4850 features &= ~NETIF_F_UFO;
4853 if (!(features & NETIF_F_SG)) {
4855 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4856 "since no NETIF_F_SG feature.\n", name);
4857 features &= ~NETIF_F_UFO;
4863 EXPORT_SYMBOL(netdev_fix_features);
4866 * register_netdevice - register a network device
4867 * @dev: device to register
4869 * Take a completed network device structure and add it to the kernel
4870 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4871 * chain. 0 is returned on success. A negative errno code is returned
4872 * on a failure to set up the device, or if the name is a duplicate.
4874 * Callers must hold the rtnl semaphore. You may want
4875 * register_netdev() instead of this.
4878 * The locking appears insufficient to guarantee two parallel registers
4879 * will not get the same name.
4882 int register_netdevice(struct net_device *dev)
4884 struct hlist_head *head;
4885 struct hlist_node *p;
4887 struct net *net = dev_net(dev);
4889 BUG_ON(dev_boot_phase);
4894 /* When net_device's are persistent, this will be fatal. */
4895 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4898 spin_lock_init(&dev->addr_list_lock);
4899 netdev_set_addr_lockdep_class(dev);
4900 netdev_init_queue_locks(dev);
4904 /* Init, if this function is available */
4905 if (dev->netdev_ops->ndo_init) {
4906 ret = dev->netdev_ops->ndo_init(dev);
4914 if (!dev_valid_name(dev->name)) {
4919 dev->ifindex = dev_new_index(net);
4920 if (dev->iflink == -1)
4921 dev->iflink = dev->ifindex;
4923 /* Check for existence of name */
4924 head = dev_name_hash(net, dev->name);
4925 hlist_for_each(p, head) {
4926 struct net_device *d
4927 = hlist_entry(p, struct net_device, name_hlist);
4928 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4934 /* Fix illegal checksum combinations */
4935 if ((dev->features & NETIF_F_HW_CSUM) &&
4936 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4937 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4939 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4942 if ((dev->features & NETIF_F_NO_CSUM) &&
4943 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4944 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4946 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4949 dev->features = netdev_fix_features(dev->features, dev->name);
4951 /* Enable software GSO if SG is supported. */
4952 if (dev->features & NETIF_F_SG)
4953 dev->features |= NETIF_F_GSO;
4955 netdev_initialize_kobject(dev);
4957 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4958 ret = notifier_to_errno(ret);
4962 ret = netdev_register_kobject(dev);
4965 dev->reg_state = NETREG_REGISTERED;
4968 * Default initial state at registry is that the
4969 * device is present.
4972 set_bit(__LINK_STATE_PRESENT, &dev->state);
4974 dev_init_scheduler(dev);
4976 list_netdevice(dev);
4978 /* Notify protocols, that a new device appeared. */
4979 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4980 ret = notifier_to_errno(ret);
4982 rollback_registered(dev);
4983 dev->reg_state = NETREG_UNREGISTERED;
4990 if (dev->netdev_ops->ndo_uninit)
4991 dev->netdev_ops->ndo_uninit(dev);
4994 EXPORT_SYMBOL(register_netdevice);
4997 * init_dummy_netdev - init a dummy network device for NAPI
4998 * @dev: device to init
5000 * This takes a network device structure and initialize the minimum
5001 * amount of fields so it can be used to schedule NAPI polls without
5002 * registering a full blown interface. This is to be used by drivers
5003 * that need to tie several hardware interfaces to a single NAPI
5004 * poll scheduler due to HW limitations.
5006 int init_dummy_netdev(struct net_device *dev)
5008 /* Clear everything. Note we don't initialize spinlocks
5009 * are they aren't supposed to be taken by any of the
5010 * NAPI code and this dummy netdev is supposed to be
5011 * only ever used for NAPI polls
5013 memset(dev, 0, sizeof(struct net_device));
5015 /* make sure we BUG if trying to hit standard
5016 * register/unregister code path
5018 dev->reg_state = NETREG_DUMMY;
5020 /* initialize the ref count */
5021 atomic_set(&dev->refcnt, 1);
5023 /* NAPI wants this */
5024 INIT_LIST_HEAD(&dev->napi_list);
5026 /* a dummy interface is started by default */
5027 set_bit(__LINK_STATE_PRESENT, &dev->state);
5028 set_bit(__LINK_STATE_START, &dev->state);
5032 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5036 * register_netdev - register a network device
5037 * @dev: device to register
5039 * Take a completed network device structure and add it to the kernel
5040 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5041 * chain. 0 is returned on success. A negative errno code is returned
5042 * on a failure to set up the device, or if the name is a duplicate.
5044 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5045 * and expands the device name if you passed a format string to
5048 int register_netdev(struct net_device *dev)
5055 * If the name is a format string the caller wants us to do a
5058 if (strchr(dev->name, '%')) {
5059 err = dev_alloc_name(dev, dev->name);
5064 err = register_netdevice(dev);
5069 EXPORT_SYMBOL(register_netdev);
5072 * netdev_wait_allrefs - wait until all references are gone.
5074 * This is called when unregistering network devices.
5076 * Any protocol or device that holds a reference should register
5077 * for netdevice notification, and cleanup and put back the
5078 * reference if they receive an UNREGISTER event.
5079 * We can get stuck here if buggy protocols don't correctly
5082 static void netdev_wait_allrefs(struct net_device *dev)
5084 unsigned long rebroadcast_time, warning_time;
5086 rebroadcast_time = warning_time = jiffies;
5087 while (atomic_read(&dev->refcnt) != 0) {
5088 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5091 /* Rebroadcast unregister notification */
5092 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5094 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5096 /* We must not have linkwatch events
5097 * pending on unregister. If this
5098 * happens, we simply run the queue
5099 * unscheduled, resulting in a noop
5102 linkwatch_run_queue();
5107 rebroadcast_time = jiffies;
5112 if (time_after(jiffies, warning_time + 10 * HZ)) {
5113 printk(KERN_EMERG "unregister_netdevice: "
5114 "waiting for %s to become free. Usage "
5116 dev->name, atomic_read(&dev->refcnt));
5117 warning_time = jiffies;
5126 * register_netdevice(x1);
5127 * register_netdevice(x2);
5129 * unregister_netdevice(y1);
5130 * unregister_netdevice(y2);
5136 * We are invoked by rtnl_unlock().
5137 * This allows us to deal with problems:
5138 * 1) We can delete sysfs objects which invoke hotplug
5139 * without deadlocking with linkwatch via keventd.
5140 * 2) Since we run with the RTNL semaphore not held, we can sleep
5141 * safely in order to wait for the netdev refcnt to drop to zero.
5143 * We must not return until all unregister events added during
5144 * the interval the lock was held have been completed.
5146 void netdev_run_todo(void)
5148 struct list_head list;
5150 /* Snapshot list, allow later requests */
5151 list_replace_init(&net_todo_list, &list);
5155 while (!list_empty(&list)) {
5156 struct net_device *dev
5157 = list_entry(list.next, struct net_device, todo_list);
5158 list_del(&dev->todo_list);
5160 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5161 printk(KERN_ERR "network todo '%s' but state %d\n",
5162 dev->name, dev->reg_state);
5167 dev->reg_state = NETREG_UNREGISTERED;
5169 on_each_cpu(flush_backlog, dev, 1);
5171 netdev_wait_allrefs(dev);
5174 BUG_ON(atomic_read(&dev->refcnt));
5175 WARN_ON(dev->ip_ptr);
5176 WARN_ON(dev->ip6_ptr);
5177 WARN_ON(dev->dn_ptr);
5179 if (dev->destructor)
5180 dev->destructor(dev);
5182 /* Free network device */
5183 kobject_put(&dev->dev.kobj);
5188 * dev_get_stats - get network device statistics
5189 * @dev: device to get statistics from
5191 * Get network statistics from device. The device driver may provide
5192 * its own method by setting dev->netdev_ops->get_stats; otherwise
5193 * the internal statistics structure is used.
5195 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5197 const struct net_device_ops *ops = dev->netdev_ops;
5199 if (ops->ndo_get_stats)
5200 return ops->ndo_get_stats(dev);
5202 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5203 struct net_device_stats *stats = &dev->stats;
5205 struct netdev_queue *txq;
5207 for (i = 0; i < dev->num_tx_queues; i++) {
5208 txq = netdev_get_tx_queue(dev, i);
5209 tx_bytes += txq->tx_bytes;
5210 tx_packets += txq->tx_packets;
5211 tx_dropped += txq->tx_dropped;
5213 if (tx_bytes || tx_packets || tx_dropped) {
5214 stats->tx_bytes = tx_bytes;
5215 stats->tx_packets = tx_packets;
5216 stats->tx_dropped = tx_dropped;
5221 EXPORT_SYMBOL(dev_get_stats);
5223 static void netdev_init_one_queue(struct net_device *dev,
5224 struct netdev_queue *queue,
5230 static void netdev_init_queues(struct net_device *dev)
5232 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5233 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5234 spin_lock_init(&dev->tx_global_lock);
5238 * alloc_netdev_mq - allocate network device
5239 * @sizeof_priv: size of private data to allocate space for
5240 * @name: device name format string
5241 * @setup: callback to initialize device
5242 * @queue_count: the number of subqueues to allocate
5244 * Allocates a struct net_device with private data area for driver use
5245 * and performs basic initialization. Also allocates subquue structs
5246 * for each queue on the device at the end of the netdevice.
5248 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5249 void (*setup)(struct net_device *), unsigned int queue_count)
5251 struct netdev_queue *tx;
5252 struct net_device *dev;
5254 struct net_device *p;
5256 BUG_ON(strlen(name) >= sizeof(dev->name));
5258 alloc_size = sizeof(struct net_device);
5260 /* ensure 32-byte alignment of private area */
5261 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5262 alloc_size += sizeof_priv;
5264 /* ensure 32-byte alignment of whole construct */
5265 alloc_size += NETDEV_ALIGN - 1;
5267 p = kzalloc(alloc_size, GFP_KERNEL);
5269 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5273 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5275 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5280 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5281 dev->padded = (char *)dev - (char *)p;
5283 if (dev_addr_init(dev))
5286 dev_unicast_init(dev);
5288 dev_net_set(dev, &init_net);
5291 dev->num_tx_queues = queue_count;
5292 dev->real_num_tx_queues = queue_count;
5294 dev->gso_max_size = GSO_MAX_SIZE;
5296 netdev_init_queues(dev);
5298 INIT_LIST_HEAD(&dev->napi_list);
5299 INIT_LIST_HEAD(&dev->unreg_list);
5300 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5302 strcpy(dev->name, name);
5312 EXPORT_SYMBOL(alloc_netdev_mq);
5315 * free_netdev - free network device
5318 * This function does the last stage of destroying an allocated device
5319 * interface. The reference to the device object is released.
5320 * If this is the last reference then it will be freed.
5322 void free_netdev(struct net_device *dev)
5324 struct napi_struct *p, *n;
5326 release_net(dev_net(dev));
5330 /* Flush device addresses */
5331 dev_addr_flush(dev);
5333 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5336 /* Compatibility with error handling in drivers */
5337 if (dev->reg_state == NETREG_UNINITIALIZED) {
5338 kfree((char *)dev - dev->padded);
5342 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5343 dev->reg_state = NETREG_RELEASED;
5345 /* will free via device release */
5346 put_device(&dev->dev);
5348 EXPORT_SYMBOL(free_netdev);
5351 * synchronize_net - Synchronize with packet receive processing
5353 * Wait for packets currently being received to be done.
5354 * Does not block later packets from starting.
5356 void synchronize_net(void)
5361 EXPORT_SYMBOL(synchronize_net);
5364 * unregister_netdevice_queue - remove device from the kernel
5368 * This function shuts down a device interface and removes it
5369 * from the kernel tables.
5370 * If head not NULL, device is queued to be unregistered later.
5372 * Callers must hold the rtnl semaphore. You may want
5373 * unregister_netdev() instead of this.
5376 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5381 list_move_tail(&dev->unreg_list, head);
5383 rollback_registered(dev);
5384 /* Finish processing unregister after unlock */
5388 EXPORT_SYMBOL(unregister_netdevice_queue);
5391 * unregister_netdevice_many - unregister many devices
5392 * @head: list of devices
5395 void unregister_netdevice_many(struct list_head *head)
5397 struct net_device *dev;
5399 if (!list_empty(head)) {
5400 rollback_registered_many(head);
5401 list_for_each_entry(dev, head, unreg_list)
5405 EXPORT_SYMBOL(unregister_netdevice_many);
5408 * unregister_netdev - remove device from the kernel
5411 * This function shuts down a device interface and removes it
5412 * from the kernel tables.
5414 * This is just a wrapper for unregister_netdevice that takes
5415 * the rtnl semaphore. In general you want to use this and not
5416 * unregister_netdevice.
5418 void unregister_netdev(struct net_device *dev)
5421 unregister_netdevice(dev);
5424 EXPORT_SYMBOL(unregister_netdev);
5427 * dev_change_net_namespace - move device to different nethost namespace
5429 * @net: network namespace
5430 * @pat: If not NULL name pattern to try if the current device name
5431 * is already taken in the destination network namespace.
5433 * This function shuts down a device interface and moves it
5434 * to a new network namespace. On success 0 is returned, on
5435 * a failure a netagive errno code is returned.
5437 * Callers must hold the rtnl semaphore.
5440 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5443 const char *destname;
5448 /* Don't allow namespace local devices to be moved. */
5450 if (dev->features & NETIF_F_NETNS_LOCAL)
5454 /* Don't allow real devices to be moved when sysfs
5458 if (dev->dev.parent)
5462 /* Ensure the device has been registrered */
5464 if (dev->reg_state != NETREG_REGISTERED)
5467 /* Get out if there is nothing todo */
5469 if (net_eq(dev_net(dev), net))
5472 /* Pick the destination device name, and ensure
5473 * we can use it in the destination network namespace.
5476 destname = dev->name;
5477 if (__dev_get_by_name(net, destname)) {
5478 /* We get here if we can't use the current device name */
5481 if (!dev_valid_name(pat))
5483 if (strchr(pat, '%')) {
5484 if (__dev_alloc_name(net, pat, buf) < 0)
5489 if (__dev_get_by_name(net, destname))
5494 * And now a mini version of register_netdevice unregister_netdevice.
5497 /* If device is running close it first. */
5500 /* And unlink it from device chain */
5502 unlist_netdevice(dev);
5506 /* Shutdown queueing discipline. */
5509 /* Notify protocols, that we are about to destroy
5510 this device. They should clean all the things.
5512 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5515 * Flush the unicast and multicast chains
5517 dev_unicast_flush(dev);
5518 dev_addr_discard(dev);
5520 netdev_unregister_kobject(dev);
5522 /* Actually switch the network namespace */
5523 dev_net_set(dev, net);
5525 /* Assign the new device name */
5526 if (destname != dev->name)
5527 strcpy(dev->name, destname);
5529 /* If there is an ifindex conflict assign a new one */
5530 if (__dev_get_by_index(net, dev->ifindex)) {
5531 int iflink = (dev->iflink == dev->ifindex);
5532 dev->ifindex = dev_new_index(net);
5534 dev->iflink = dev->ifindex;
5537 /* Fixup kobjects */
5538 err = netdev_register_kobject(dev);
5541 /* Add the device back in the hashes */
5542 list_netdevice(dev);
5544 /* Notify protocols, that a new device appeared. */
5545 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5552 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5554 static int dev_cpu_callback(struct notifier_block *nfb,
5555 unsigned long action,
5558 struct sk_buff **list_skb;
5559 struct Qdisc **list_net;
5560 struct sk_buff *skb;
5561 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5562 struct softnet_data *sd, *oldsd;
5564 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5567 local_irq_disable();
5568 cpu = smp_processor_id();
5569 sd = &per_cpu(softnet_data, cpu);
5570 oldsd = &per_cpu(softnet_data, oldcpu);
5572 /* Find end of our completion_queue. */
5573 list_skb = &sd->completion_queue;
5575 list_skb = &(*list_skb)->next;
5576 /* Append completion queue from offline CPU. */
5577 *list_skb = oldsd->completion_queue;
5578 oldsd->completion_queue = NULL;
5580 /* Find end of our output_queue. */
5581 list_net = &sd->output_queue;
5583 list_net = &(*list_net)->next_sched;
5584 /* Append output queue from offline CPU. */
5585 *list_net = oldsd->output_queue;
5586 oldsd->output_queue = NULL;
5588 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5591 /* Process offline CPU's input_pkt_queue */
5592 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5600 * netdev_increment_features - increment feature set by one
5601 * @all: current feature set
5602 * @one: new feature set
5603 * @mask: mask feature set
5605 * Computes a new feature set after adding a device with feature set
5606 * @one to the master device with current feature set @all. Will not
5607 * enable anything that is off in @mask. Returns the new feature set.
5609 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5612 /* If device needs checksumming, downgrade to it. */
5613 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5614 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5615 else if (mask & NETIF_F_ALL_CSUM) {
5616 /* If one device supports v4/v6 checksumming, set for all. */
5617 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5618 !(all & NETIF_F_GEN_CSUM)) {
5619 all &= ~NETIF_F_ALL_CSUM;
5620 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5623 /* If one device supports hw checksumming, set for all. */
5624 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5625 all &= ~NETIF_F_ALL_CSUM;
5626 all |= NETIF_F_HW_CSUM;
5630 one |= NETIF_F_ALL_CSUM;
5632 one |= all & NETIF_F_ONE_FOR_ALL;
5633 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5634 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5638 EXPORT_SYMBOL(netdev_increment_features);
5640 static struct hlist_head *netdev_create_hash(void)
5643 struct hlist_head *hash;
5645 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5647 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5648 INIT_HLIST_HEAD(&hash[i]);
5653 /* Initialize per network namespace state */
5654 static int __net_init netdev_init(struct net *net)
5656 INIT_LIST_HEAD(&net->dev_base_head);
5658 net->dev_name_head = netdev_create_hash();
5659 if (net->dev_name_head == NULL)
5662 net->dev_index_head = netdev_create_hash();
5663 if (net->dev_index_head == NULL)
5669 kfree(net->dev_name_head);
5675 * netdev_drivername - network driver for the device
5676 * @dev: network device
5677 * @buffer: buffer for resulting name
5678 * @len: size of buffer
5680 * Determine network driver for device.
5682 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5684 const struct device_driver *driver;
5685 const struct device *parent;
5687 if (len <= 0 || !buffer)
5691 parent = dev->dev.parent;
5696 driver = parent->driver;
5697 if (driver && driver->name)
5698 strlcpy(buffer, driver->name, len);
5702 static void __net_exit netdev_exit(struct net *net)
5704 kfree(net->dev_name_head);
5705 kfree(net->dev_index_head);
5708 static struct pernet_operations __net_initdata netdev_net_ops = {
5709 .init = netdev_init,
5710 .exit = netdev_exit,
5713 static void __net_exit default_device_exit(struct net *net)
5715 struct net_device *dev;
5717 * Push all migratable of the network devices back to the
5718 * initial network namespace
5722 for_each_netdev(net, dev) {
5724 char fb_name[IFNAMSIZ];
5726 /* Ignore unmoveable devices (i.e. loopback) */
5727 if (dev->features & NETIF_F_NETNS_LOCAL)
5730 /* Delete virtual devices */
5731 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5732 dev->rtnl_link_ops->dellink(dev, NULL);
5736 /* Push remaing network devices to init_net */
5737 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5738 err = dev_change_net_namespace(dev, &init_net, fb_name);
5740 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5741 __func__, dev->name, err);
5749 static struct pernet_operations __net_initdata default_device_ops = {
5750 .exit = default_device_exit,
5754 * Initialize the DEV module. At boot time this walks the device list and
5755 * unhooks any devices that fail to initialise (normally hardware not
5756 * present) and leaves us with a valid list of present and active devices.
5761 * This is called single threaded during boot, so no need
5762 * to take the rtnl semaphore.
5764 static int __init net_dev_init(void)
5766 int i, rc = -ENOMEM;
5768 BUG_ON(!dev_boot_phase);
5770 if (dev_proc_init())
5773 if (netdev_kobject_init())
5776 INIT_LIST_HEAD(&ptype_all);
5777 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5778 INIT_LIST_HEAD(&ptype_base[i]);
5780 if (register_pernet_subsys(&netdev_net_ops))
5784 * Initialise the packet receive queues.
5787 for_each_possible_cpu(i) {
5788 struct softnet_data *queue;
5790 queue = &per_cpu(softnet_data, i);
5791 skb_queue_head_init(&queue->input_pkt_queue);
5792 queue->completion_queue = NULL;
5793 INIT_LIST_HEAD(&queue->poll_list);
5795 queue->backlog.poll = process_backlog;
5796 queue->backlog.weight = weight_p;
5797 queue->backlog.gro_list = NULL;
5798 queue->backlog.gro_count = 0;
5803 /* The loopback device is special if any other network devices
5804 * is present in a network namespace the loopback device must
5805 * be present. Since we now dynamically allocate and free the
5806 * loopback device ensure this invariant is maintained by
5807 * keeping the loopback device as the first device on the
5808 * list of network devices. Ensuring the loopback devices
5809 * is the first device that appears and the last network device
5812 if (register_pernet_device(&loopback_net_ops))
5815 if (register_pernet_device(&default_device_ops))
5818 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5819 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5821 hotcpu_notifier(dev_cpu_callback, 0);
5829 subsys_initcall(net_dev_init);
5831 static int __init initialize_hashrnd(void)
5833 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5837 late_initcall_sync(initialize_hashrnd);