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;
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 nskb->next = skb->next;
1812 txq_trans_update(txq);
1813 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1814 return NETDEV_TX_BUSY;
1815 } while (skb->next);
1817 skb->destructor = DEV_GSO_CB(skb)->destructor;
1821 return NETDEV_TX_OK;
1824 static u32 skb_tx_hashrnd;
1826 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1830 if (skb_rx_queue_recorded(skb)) {
1831 hash = skb_get_rx_queue(skb);
1832 while (unlikely(hash >= dev->real_num_tx_queues))
1833 hash -= dev->real_num_tx_queues;
1837 if (skb->sk && skb->sk->sk_hash)
1838 hash = skb->sk->sk_hash;
1840 hash = skb->protocol;
1842 hash = jhash_1word(hash, skb_tx_hashrnd);
1844 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1846 EXPORT_SYMBOL(skb_tx_hash);
1848 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1849 struct sk_buff *skb)
1852 struct sock *sk = skb->sk;
1854 if (sk_tx_queue_recorded(sk)) {
1855 queue_index = sk_tx_queue_get(sk);
1857 const struct net_device_ops *ops = dev->netdev_ops;
1859 if (ops->ndo_select_queue) {
1860 queue_index = ops->ndo_select_queue(dev, skb);
1863 if (dev->real_num_tx_queues > 1)
1864 queue_index = skb_tx_hash(dev, skb);
1866 if (sk && sk->sk_dst_cache)
1867 sk_tx_queue_set(sk, queue_index);
1871 skb_set_queue_mapping(skb, queue_index);
1872 return netdev_get_tx_queue(dev, queue_index);
1875 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
1876 struct net_device *dev,
1877 struct netdev_queue *txq)
1879 spinlock_t *root_lock = qdisc_lock(q);
1882 spin_lock(root_lock);
1883 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1886 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
1887 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
1889 * This is a work-conserving queue; there are no old skbs
1890 * waiting to be sent out; and the qdisc is not running -
1891 * xmit the skb directly.
1893 __qdisc_update_bstats(q, skb->len);
1894 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
1897 clear_bit(__QDISC_STATE_RUNNING, &q->state);
1899 rc = NET_XMIT_SUCCESS;
1901 rc = qdisc_enqueue_root(skb, q);
1904 spin_unlock(root_lock);
1910 * dev_queue_xmit - transmit a buffer
1911 * @skb: buffer to transmit
1913 * Queue a buffer for transmission to a network device. The caller must
1914 * have set the device and priority and built the buffer before calling
1915 * this function. The function can be called from an interrupt.
1917 * A negative errno code is returned on a failure. A success does not
1918 * guarantee the frame will be transmitted as it may be dropped due
1919 * to congestion or traffic shaping.
1921 * -----------------------------------------------------------------------------------
1922 * I notice this method can also return errors from the queue disciplines,
1923 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1926 * Regardless of the return value, the skb is consumed, so it is currently
1927 * difficult to retry a send to this method. (You can bump the ref count
1928 * before sending to hold a reference for retry if you are careful.)
1930 * When calling this method, interrupts MUST be enabled. This is because
1931 * the BH enable code must have IRQs enabled so that it will not deadlock.
1934 int dev_queue_xmit(struct sk_buff *skb)
1936 struct net_device *dev = skb->dev;
1937 struct netdev_queue *txq;
1941 /* GSO will handle the following emulations directly. */
1942 if (netif_needs_gso(dev, skb))
1945 if (skb_has_frags(skb) &&
1946 !(dev->features & NETIF_F_FRAGLIST) &&
1947 __skb_linearize(skb))
1950 /* Fragmented skb is linearized if device does not support SG,
1951 * or if at least one of fragments is in highmem and device
1952 * does not support DMA from it.
1954 if (skb_shinfo(skb)->nr_frags &&
1955 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1956 __skb_linearize(skb))
1959 /* If packet is not checksummed and device does not support
1960 * checksumming for this protocol, complete checksumming here.
1962 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1963 skb_set_transport_header(skb, skb->csum_start -
1965 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1970 /* Disable soft irqs for various locks below. Also
1971 * stops preemption for RCU.
1975 txq = dev_pick_tx(dev, skb);
1976 q = rcu_dereference(txq->qdisc);
1978 #ifdef CONFIG_NET_CLS_ACT
1979 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
1982 rc = __dev_xmit_skb(skb, q, dev, txq);
1986 /* The device has no queue. Common case for software devices:
1987 loopback, all the sorts of tunnels...
1989 Really, it is unlikely that netif_tx_lock protection is necessary
1990 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1992 However, it is possible, that they rely on protection
1995 Check this and shot the lock. It is not prone from deadlocks.
1996 Either shot noqueue qdisc, it is even simpler 8)
1998 if (dev->flags & IFF_UP) {
1999 int cpu = smp_processor_id(); /* ok because BHs are off */
2001 if (txq->xmit_lock_owner != cpu) {
2003 HARD_TX_LOCK(dev, txq, cpu);
2005 if (!netif_tx_queue_stopped(txq)) {
2006 rc = NET_XMIT_SUCCESS;
2007 if (!dev_hard_start_xmit(skb, dev, txq)) {
2008 HARD_TX_UNLOCK(dev, txq);
2012 HARD_TX_UNLOCK(dev, txq);
2013 if (net_ratelimit())
2014 printk(KERN_CRIT "Virtual device %s asks to "
2015 "queue packet!\n", dev->name);
2017 /* Recursion is detected! It is possible,
2019 if (net_ratelimit())
2020 printk(KERN_CRIT "Dead loop on virtual device "
2021 "%s, fix it urgently!\n", dev->name);
2026 rcu_read_unlock_bh();
2032 rcu_read_unlock_bh();
2035 EXPORT_SYMBOL(dev_queue_xmit);
2038 /*=======================================================================
2040 =======================================================================*/
2042 int netdev_max_backlog __read_mostly = 1000;
2043 int netdev_budget __read_mostly = 300;
2044 int weight_p __read_mostly = 64; /* old backlog weight */
2046 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2050 * netif_rx - post buffer to the network code
2051 * @skb: buffer to post
2053 * This function receives a packet from a device driver and queues it for
2054 * the upper (protocol) levels to process. It always succeeds. The buffer
2055 * may be dropped during processing for congestion control or by the
2059 * NET_RX_SUCCESS (no congestion)
2060 * NET_RX_DROP (packet was dropped)
2064 int netif_rx(struct sk_buff *skb)
2066 struct softnet_data *queue;
2067 unsigned long flags;
2069 /* if netpoll wants it, pretend we never saw it */
2070 if (netpoll_rx(skb))
2073 if (!skb->tstamp.tv64)
2077 * The code is rearranged so that the path is the most
2078 * short when CPU is congested, but is still operating.
2080 local_irq_save(flags);
2081 queue = &__get_cpu_var(softnet_data);
2083 __get_cpu_var(netdev_rx_stat).total++;
2084 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2085 if (queue->input_pkt_queue.qlen) {
2087 __skb_queue_tail(&queue->input_pkt_queue, skb);
2088 local_irq_restore(flags);
2089 return NET_RX_SUCCESS;
2092 napi_schedule(&queue->backlog);
2096 __get_cpu_var(netdev_rx_stat).dropped++;
2097 local_irq_restore(flags);
2102 EXPORT_SYMBOL(netif_rx);
2104 int netif_rx_ni(struct sk_buff *skb)
2109 err = netif_rx(skb);
2110 if (local_softirq_pending())
2116 EXPORT_SYMBOL(netif_rx_ni);
2118 static void net_tx_action(struct softirq_action *h)
2120 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2122 if (sd->completion_queue) {
2123 struct sk_buff *clist;
2125 local_irq_disable();
2126 clist = sd->completion_queue;
2127 sd->completion_queue = NULL;
2131 struct sk_buff *skb = clist;
2132 clist = clist->next;
2134 WARN_ON(atomic_read(&skb->users));
2139 if (sd->output_queue) {
2142 local_irq_disable();
2143 head = sd->output_queue;
2144 sd->output_queue = NULL;
2148 struct Qdisc *q = head;
2149 spinlock_t *root_lock;
2151 head = head->next_sched;
2153 root_lock = qdisc_lock(q);
2154 if (spin_trylock(root_lock)) {
2155 smp_mb__before_clear_bit();
2156 clear_bit(__QDISC_STATE_SCHED,
2159 spin_unlock(root_lock);
2161 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2163 __netif_reschedule(q);
2165 smp_mb__before_clear_bit();
2166 clear_bit(__QDISC_STATE_SCHED,
2174 static inline int deliver_skb(struct sk_buff *skb,
2175 struct packet_type *pt_prev,
2176 struct net_device *orig_dev)
2178 atomic_inc(&skb->users);
2179 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2182 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2184 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2185 /* This hook is defined here for ATM LANE */
2186 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2187 unsigned char *addr) __read_mostly;
2188 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2192 * If bridge module is loaded call bridging hook.
2193 * returns NULL if packet was consumed.
2195 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2196 struct sk_buff *skb) __read_mostly;
2197 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2199 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2200 struct packet_type **pt_prev, int *ret,
2201 struct net_device *orig_dev)
2203 struct net_bridge_port *port;
2205 if (skb->pkt_type == PACKET_LOOPBACK ||
2206 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2210 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2214 return br_handle_frame_hook(port, skb);
2217 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2220 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2221 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2222 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2224 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2225 struct packet_type **pt_prev,
2227 struct net_device *orig_dev)
2229 if (skb->dev->macvlan_port == NULL)
2233 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2236 return macvlan_handle_frame_hook(skb);
2239 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2242 #ifdef CONFIG_NET_CLS_ACT
2243 /* TODO: Maybe we should just force sch_ingress to be compiled in
2244 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2245 * a compare and 2 stores extra right now if we dont have it on
2246 * but have CONFIG_NET_CLS_ACT
2247 * NOTE: This doesnt stop any functionality; if you dont have
2248 * the ingress scheduler, you just cant add policies on ingress.
2251 static int ing_filter(struct sk_buff *skb)
2253 struct net_device *dev = skb->dev;
2254 u32 ttl = G_TC_RTTL(skb->tc_verd);
2255 struct netdev_queue *rxq;
2256 int result = TC_ACT_OK;
2259 if (MAX_RED_LOOP < ttl++) {
2261 "Redir loop detected Dropping packet (%d->%d)\n",
2262 skb->iif, dev->ifindex);
2266 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2267 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2269 rxq = &dev->rx_queue;
2272 if (q != &noop_qdisc) {
2273 spin_lock(qdisc_lock(q));
2274 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2275 result = qdisc_enqueue_root(skb, q);
2276 spin_unlock(qdisc_lock(q));
2282 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2283 struct packet_type **pt_prev,
2284 int *ret, struct net_device *orig_dev)
2286 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2290 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2293 /* Huh? Why does turning on AF_PACKET affect this? */
2294 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2297 switch (ing_filter(skb)) {
2311 * netif_nit_deliver - deliver received packets to network taps
2314 * This function is used to deliver incoming packets to network
2315 * taps. It should be used when the normal netif_receive_skb path
2316 * is bypassed, for example because of VLAN acceleration.
2318 void netif_nit_deliver(struct sk_buff *skb)
2320 struct packet_type *ptype;
2322 if (list_empty(&ptype_all))
2325 skb_reset_network_header(skb);
2326 skb_reset_transport_header(skb);
2327 skb->mac_len = skb->network_header - skb->mac_header;
2330 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2331 if (!ptype->dev || ptype->dev == skb->dev)
2332 deliver_skb(skb, ptype, skb->dev);
2338 * netif_receive_skb - process receive buffer from network
2339 * @skb: buffer to process
2341 * netif_receive_skb() is the main receive data processing function.
2342 * It always succeeds. The buffer may be dropped during processing
2343 * for congestion control or by the protocol layers.
2345 * This function may only be called from softirq context and interrupts
2346 * should be enabled.
2348 * Return values (usually ignored):
2349 * NET_RX_SUCCESS: no congestion
2350 * NET_RX_DROP: packet was dropped
2352 int netif_receive_skb(struct sk_buff *skb)
2354 struct packet_type *ptype, *pt_prev;
2355 struct net_device *orig_dev;
2356 struct net_device *null_or_orig;
2357 int ret = NET_RX_DROP;
2360 if (!skb->tstamp.tv64)
2363 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2364 return NET_RX_SUCCESS;
2366 /* if we've gotten here through NAPI, check netpoll */
2367 if (netpoll_receive_skb(skb))
2371 skb->iif = skb->dev->ifindex;
2373 null_or_orig = NULL;
2374 orig_dev = skb->dev;
2375 if (orig_dev->master) {
2376 if (skb_bond_should_drop(skb))
2377 null_or_orig = orig_dev; /* deliver only exact match */
2379 skb->dev = orig_dev->master;
2382 __get_cpu_var(netdev_rx_stat).total++;
2384 skb_reset_network_header(skb);
2385 skb_reset_transport_header(skb);
2386 skb->mac_len = skb->network_header - skb->mac_header;
2392 #ifdef CONFIG_NET_CLS_ACT
2393 if (skb->tc_verd & TC_NCLS) {
2394 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2399 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2400 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2401 ptype->dev == orig_dev) {
2403 ret = deliver_skb(skb, pt_prev, orig_dev);
2408 #ifdef CONFIG_NET_CLS_ACT
2409 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2415 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2418 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2422 type = skb->protocol;
2423 list_for_each_entry_rcu(ptype,
2424 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2425 if (ptype->type == type &&
2426 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2427 ptype->dev == orig_dev)) {
2429 ret = deliver_skb(skb, pt_prev, orig_dev);
2435 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2438 /* Jamal, now you will not able to escape explaining
2439 * me how you were going to use this. :-)
2448 EXPORT_SYMBOL(netif_receive_skb);
2450 /* Network device is going away, flush any packets still pending */
2451 static void flush_backlog(void *arg)
2453 struct net_device *dev = arg;
2454 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2455 struct sk_buff *skb, *tmp;
2457 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2458 if (skb->dev == dev) {
2459 __skb_unlink(skb, &queue->input_pkt_queue);
2464 static int napi_gro_complete(struct sk_buff *skb)
2466 struct packet_type *ptype;
2467 __be16 type = skb->protocol;
2468 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2471 if (NAPI_GRO_CB(skb)->count == 1) {
2472 skb_shinfo(skb)->gso_size = 0;
2477 list_for_each_entry_rcu(ptype, head, list) {
2478 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2481 err = ptype->gro_complete(skb);
2487 WARN_ON(&ptype->list == head);
2489 return NET_RX_SUCCESS;
2493 return netif_receive_skb(skb);
2496 void napi_gro_flush(struct napi_struct *napi)
2498 struct sk_buff *skb, *next;
2500 for (skb = napi->gro_list; skb; skb = next) {
2503 napi_gro_complete(skb);
2506 napi->gro_count = 0;
2507 napi->gro_list = NULL;
2509 EXPORT_SYMBOL(napi_gro_flush);
2511 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2513 struct sk_buff **pp = NULL;
2514 struct packet_type *ptype;
2515 __be16 type = skb->protocol;
2516 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2519 enum gro_result ret;
2521 if (!(skb->dev->features & NETIF_F_GRO))
2524 if (skb_is_gso(skb) || skb_has_frags(skb))
2528 list_for_each_entry_rcu(ptype, head, list) {
2529 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2532 skb_set_network_header(skb, skb_gro_offset(skb));
2533 mac_len = skb->network_header - skb->mac_header;
2534 skb->mac_len = mac_len;
2535 NAPI_GRO_CB(skb)->same_flow = 0;
2536 NAPI_GRO_CB(skb)->flush = 0;
2537 NAPI_GRO_CB(skb)->free = 0;
2539 pp = ptype->gro_receive(&napi->gro_list, skb);
2544 if (&ptype->list == head)
2547 same_flow = NAPI_GRO_CB(skb)->same_flow;
2548 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2551 struct sk_buff *nskb = *pp;
2555 napi_gro_complete(nskb);
2562 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2566 NAPI_GRO_CB(skb)->count = 1;
2567 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2568 skb->next = napi->gro_list;
2569 napi->gro_list = skb;
2573 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2574 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2576 BUG_ON(skb->end - skb->tail < grow);
2578 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2581 skb->data_len -= grow;
2583 skb_shinfo(skb)->frags[0].page_offset += grow;
2584 skb_shinfo(skb)->frags[0].size -= grow;
2586 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2587 put_page(skb_shinfo(skb)->frags[0].page);
2588 memmove(skb_shinfo(skb)->frags,
2589 skb_shinfo(skb)->frags + 1,
2590 --skb_shinfo(skb)->nr_frags);
2601 EXPORT_SYMBOL(dev_gro_receive);
2604 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2608 if (netpoll_rx_on(skb))
2611 for (p = napi->gro_list; p; p = p->next) {
2612 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2613 && !compare_ether_header(skb_mac_header(p),
2614 skb_gro_mac_header(skb));
2615 NAPI_GRO_CB(p)->flush = 0;
2618 return dev_gro_receive(napi, skb);
2621 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
2625 if (netif_receive_skb(skb))
2630 case GRO_MERGED_FREE:
2641 EXPORT_SYMBOL(napi_skb_finish);
2643 void skb_gro_reset_offset(struct sk_buff *skb)
2645 NAPI_GRO_CB(skb)->data_offset = 0;
2646 NAPI_GRO_CB(skb)->frag0 = NULL;
2647 NAPI_GRO_CB(skb)->frag0_len = 0;
2649 if (skb->mac_header == skb->tail &&
2650 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2651 NAPI_GRO_CB(skb)->frag0 =
2652 page_address(skb_shinfo(skb)->frags[0].page) +
2653 skb_shinfo(skb)->frags[0].page_offset;
2654 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2657 EXPORT_SYMBOL(skb_gro_reset_offset);
2659 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2661 skb_gro_reset_offset(skb);
2663 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2665 EXPORT_SYMBOL(napi_gro_receive);
2667 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2669 __skb_pull(skb, skb_headlen(skb));
2670 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2674 EXPORT_SYMBOL(napi_reuse_skb);
2676 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2678 struct sk_buff *skb = napi->skb;
2681 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
2687 EXPORT_SYMBOL(napi_get_frags);
2689 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
2695 skb->protocol = eth_type_trans(skb, napi->dev);
2697 if (ret == GRO_HELD)
2698 skb_gro_pull(skb, -ETH_HLEN);
2699 else if (netif_receive_skb(skb))
2704 case GRO_MERGED_FREE:
2705 napi_reuse_skb(napi, skb);
2714 EXPORT_SYMBOL(napi_frags_finish);
2716 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2718 struct sk_buff *skb = napi->skb;
2725 skb_reset_mac_header(skb);
2726 skb_gro_reset_offset(skb);
2728 off = skb_gro_offset(skb);
2729 hlen = off + sizeof(*eth);
2730 eth = skb_gro_header_fast(skb, off);
2731 if (skb_gro_header_hard(skb, hlen)) {
2732 eth = skb_gro_header_slow(skb, hlen, off);
2733 if (unlikely(!eth)) {
2734 napi_reuse_skb(napi, skb);
2740 skb_gro_pull(skb, sizeof(*eth));
2743 * This works because the only protocols we care about don't require
2744 * special handling. We'll fix it up properly at the end.
2746 skb->protocol = eth->h_proto;
2751 EXPORT_SYMBOL(napi_frags_skb);
2753 gro_result_t napi_gro_frags(struct napi_struct *napi)
2755 struct sk_buff *skb = napi_frags_skb(napi);
2760 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2762 EXPORT_SYMBOL(napi_gro_frags);
2764 static int process_backlog(struct napi_struct *napi, int quota)
2767 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2768 unsigned long start_time = jiffies;
2770 napi->weight = weight_p;
2772 struct sk_buff *skb;
2774 local_irq_disable();
2775 skb = __skb_dequeue(&queue->input_pkt_queue);
2777 __napi_complete(napi);
2783 netif_receive_skb(skb);
2784 } while (++work < quota && jiffies == start_time);
2790 * __napi_schedule - schedule for receive
2791 * @n: entry to schedule
2793 * The entry's receive function will be scheduled to run
2795 void __napi_schedule(struct napi_struct *n)
2797 unsigned long flags;
2799 local_irq_save(flags);
2800 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2801 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2802 local_irq_restore(flags);
2804 EXPORT_SYMBOL(__napi_schedule);
2806 void __napi_complete(struct napi_struct *n)
2808 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2809 BUG_ON(n->gro_list);
2811 list_del(&n->poll_list);
2812 smp_mb__before_clear_bit();
2813 clear_bit(NAPI_STATE_SCHED, &n->state);
2815 EXPORT_SYMBOL(__napi_complete);
2817 void napi_complete(struct napi_struct *n)
2819 unsigned long flags;
2822 * don't let napi dequeue from the cpu poll list
2823 * just in case its running on a different cpu
2825 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2829 local_irq_save(flags);
2831 local_irq_restore(flags);
2833 EXPORT_SYMBOL(napi_complete);
2835 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2836 int (*poll)(struct napi_struct *, int), int weight)
2838 INIT_LIST_HEAD(&napi->poll_list);
2839 napi->gro_count = 0;
2840 napi->gro_list = NULL;
2843 napi->weight = weight;
2844 list_add(&napi->dev_list, &dev->napi_list);
2846 #ifdef CONFIG_NETPOLL
2847 spin_lock_init(&napi->poll_lock);
2848 napi->poll_owner = -1;
2850 set_bit(NAPI_STATE_SCHED, &napi->state);
2852 EXPORT_SYMBOL(netif_napi_add);
2854 void netif_napi_del(struct napi_struct *napi)
2856 struct sk_buff *skb, *next;
2858 list_del_init(&napi->dev_list);
2859 napi_free_frags(napi);
2861 for (skb = napi->gro_list; skb; skb = next) {
2867 napi->gro_list = NULL;
2868 napi->gro_count = 0;
2870 EXPORT_SYMBOL(netif_napi_del);
2873 static void net_rx_action(struct softirq_action *h)
2875 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2876 unsigned long time_limit = jiffies + 2;
2877 int budget = netdev_budget;
2880 local_irq_disable();
2882 while (!list_empty(list)) {
2883 struct napi_struct *n;
2886 /* If softirq window is exhuasted then punt.
2887 * Allow this to run for 2 jiffies since which will allow
2888 * an average latency of 1.5/HZ.
2890 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2895 /* Even though interrupts have been re-enabled, this
2896 * access is safe because interrupts can only add new
2897 * entries to the tail of this list, and only ->poll()
2898 * calls can remove this head entry from the list.
2900 n = list_entry(list->next, struct napi_struct, poll_list);
2902 have = netpoll_poll_lock(n);
2906 /* This NAPI_STATE_SCHED test is for avoiding a race
2907 * with netpoll's poll_napi(). Only the entity which
2908 * obtains the lock and sees NAPI_STATE_SCHED set will
2909 * actually make the ->poll() call. Therefore we avoid
2910 * accidently calling ->poll() when NAPI is not scheduled.
2913 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
2914 work = n->poll(n, weight);
2918 WARN_ON_ONCE(work > weight);
2922 local_irq_disable();
2924 /* Drivers must not modify the NAPI state if they
2925 * consume the entire weight. In such cases this code
2926 * still "owns" the NAPI instance and therefore can
2927 * move the instance around on the list at-will.
2929 if (unlikely(work == weight)) {
2930 if (unlikely(napi_disable_pending(n))) {
2933 local_irq_disable();
2935 list_move_tail(&n->poll_list, list);
2938 netpoll_poll_unlock(have);
2943 #ifdef CONFIG_NET_DMA
2945 * There may not be any more sk_buffs coming right now, so push
2946 * any pending DMA copies to hardware
2948 dma_issue_pending_all();
2954 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2955 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2959 static gifconf_func_t *gifconf_list[NPROTO];
2962 * register_gifconf - register a SIOCGIF handler
2963 * @family: Address family
2964 * @gifconf: Function handler
2966 * Register protocol dependent address dumping routines. The handler
2967 * that is passed must not be freed or reused until it has been replaced
2968 * by another handler.
2970 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
2972 if (family >= NPROTO)
2974 gifconf_list[family] = gifconf;
2977 EXPORT_SYMBOL(register_gifconf);
2981 * Map an interface index to its name (SIOCGIFNAME)
2985 * We need this ioctl for efficient implementation of the
2986 * if_indextoname() function required by the IPv6 API. Without
2987 * it, we would have to search all the interfaces to find a
2991 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2993 struct net_device *dev;
2997 * Fetch the caller's info block.
3000 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3004 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3010 strcpy(ifr.ifr_name, dev->name);
3013 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3019 * Perform a SIOCGIFCONF call. This structure will change
3020 * size eventually, and there is nothing I can do about it.
3021 * Thus we will need a 'compatibility mode'.
3024 static int dev_ifconf(struct net *net, char __user *arg)
3027 struct net_device *dev;
3034 * Fetch the caller's info block.
3037 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3044 * Loop over the interfaces, and write an info block for each.
3048 for_each_netdev(net, dev) {
3049 for (i = 0; i < NPROTO; i++) {
3050 if (gifconf_list[i]) {
3053 done = gifconf_list[i](dev, NULL, 0);
3055 done = gifconf_list[i](dev, pos + total,
3065 * All done. Write the updated control block back to the caller.
3067 ifc.ifc_len = total;
3070 * Both BSD and Solaris return 0 here, so we do too.
3072 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3075 #ifdef CONFIG_PROC_FS
3077 * This is invoked by the /proc filesystem handler to display a device
3080 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3083 struct net *net = seq_file_net(seq);
3085 struct net_device *dev;
3089 return SEQ_START_TOKEN;
3092 for_each_netdev_rcu(net, dev)
3099 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3101 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3102 first_net_device(seq_file_net(seq)) :
3103 next_net_device((struct net_device *)v);
3106 return rcu_dereference(dev);
3109 void dev_seq_stop(struct seq_file *seq, void *v)
3115 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3117 const struct net_device_stats *stats = dev_get_stats(dev);
3119 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3120 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3121 dev->name, stats->rx_bytes, stats->rx_packets,
3123 stats->rx_dropped + stats->rx_missed_errors,
3124 stats->rx_fifo_errors,
3125 stats->rx_length_errors + stats->rx_over_errors +
3126 stats->rx_crc_errors + stats->rx_frame_errors,
3127 stats->rx_compressed, stats->multicast,
3128 stats->tx_bytes, stats->tx_packets,
3129 stats->tx_errors, stats->tx_dropped,
3130 stats->tx_fifo_errors, stats->collisions,
3131 stats->tx_carrier_errors +
3132 stats->tx_aborted_errors +
3133 stats->tx_window_errors +
3134 stats->tx_heartbeat_errors,
3135 stats->tx_compressed);
3139 * Called from the PROCfs module. This now uses the new arbitrary sized
3140 * /proc/net interface to create /proc/net/dev
3142 static int dev_seq_show(struct seq_file *seq, void *v)
3144 if (v == SEQ_START_TOKEN)
3145 seq_puts(seq, "Inter-| Receive "
3147 " face |bytes packets errs drop fifo frame "
3148 "compressed multicast|bytes packets errs "
3149 "drop fifo colls carrier compressed\n");
3151 dev_seq_printf_stats(seq, v);
3155 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3157 struct netif_rx_stats *rc = NULL;
3159 while (*pos < nr_cpu_ids)
3160 if (cpu_online(*pos)) {
3161 rc = &per_cpu(netdev_rx_stat, *pos);
3168 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3170 return softnet_get_online(pos);
3173 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3176 return softnet_get_online(pos);
3179 static void softnet_seq_stop(struct seq_file *seq, void *v)
3183 static int softnet_seq_show(struct seq_file *seq, void *v)
3185 struct netif_rx_stats *s = v;
3187 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3188 s->total, s->dropped, s->time_squeeze, 0,
3189 0, 0, 0, 0, /* was fastroute */
3194 static const struct seq_operations dev_seq_ops = {
3195 .start = dev_seq_start,
3196 .next = dev_seq_next,
3197 .stop = dev_seq_stop,
3198 .show = dev_seq_show,
3201 static int dev_seq_open(struct inode *inode, struct file *file)
3203 return seq_open_net(inode, file, &dev_seq_ops,
3204 sizeof(struct seq_net_private));
3207 static const struct file_operations dev_seq_fops = {
3208 .owner = THIS_MODULE,
3209 .open = dev_seq_open,
3211 .llseek = seq_lseek,
3212 .release = seq_release_net,
3215 static const struct seq_operations softnet_seq_ops = {
3216 .start = softnet_seq_start,
3217 .next = softnet_seq_next,
3218 .stop = softnet_seq_stop,
3219 .show = softnet_seq_show,
3222 static int softnet_seq_open(struct inode *inode, struct file *file)
3224 return seq_open(file, &softnet_seq_ops);
3227 static const struct file_operations softnet_seq_fops = {
3228 .owner = THIS_MODULE,
3229 .open = softnet_seq_open,
3231 .llseek = seq_lseek,
3232 .release = seq_release,
3235 static void *ptype_get_idx(loff_t pos)
3237 struct packet_type *pt = NULL;
3241 list_for_each_entry_rcu(pt, &ptype_all, list) {
3247 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3248 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3257 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3261 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3264 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3266 struct packet_type *pt;
3267 struct list_head *nxt;
3271 if (v == SEQ_START_TOKEN)
3272 return ptype_get_idx(0);
3275 nxt = pt->list.next;
3276 if (pt->type == htons(ETH_P_ALL)) {
3277 if (nxt != &ptype_all)
3280 nxt = ptype_base[0].next;
3282 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3284 while (nxt == &ptype_base[hash]) {
3285 if (++hash >= PTYPE_HASH_SIZE)
3287 nxt = ptype_base[hash].next;
3290 return list_entry(nxt, struct packet_type, list);
3293 static void ptype_seq_stop(struct seq_file *seq, void *v)
3299 static int ptype_seq_show(struct seq_file *seq, void *v)
3301 struct packet_type *pt = v;
3303 if (v == SEQ_START_TOKEN)
3304 seq_puts(seq, "Type Device Function\n");
3305 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3306 if (pt->type == htons(ETH_P_ALL))
3307 seq_puts(seq, "ALL ");
3309 seq_printf(seq, "%04x", ntohs(pt->type));
3311 seq_printf(seq, " %-8s %pF\n",
3312 pt->dev ? pt->dev->name : "", pt->func);
3318 static const struct seq_operations ptype_seq_ops = {
3319 .start = ptype_seq_start,
3320 .next = ptype_seq_next,
3321 .stop = ptype_seq_stop,
3322 .show = ptype_seq_show,
3325 static int ptype_seq_open(struct inode *inode, struct file *file)
3327 return seq_open_net(inode, file, &ptype_seq_ops,
3328 sizeof(struct seq_net_private));
3331 static const struct file_operations ptype_seq_fops = {
3332 .owner = THIS_MODULE,
3333 .open = ptype_seq_open,
3335 .llseek = seq_lseek,
3336 .release = seq_release_net,
3340 static int __net_init dev_proc_net_init(struct net *net)
3344 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3346 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3348 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3351 if (wext_proc_init(net))
3357 proc_net_remove(net, "ptype");
3359 proc_net_remove(net, "softnet_stat");
3361 proc_net_remove(net, "dev");
3365 static void __net_exit dev_proc_net_exit(struct net *net)
3367 wext_proc_exit(net);
3369 proc_net_remove(net, "ptype");
3370 proc_net_remove(net, "softnet_stat");
3371 proc_net_remove(net, "dev");
3374 static struct pernet_operations __net_initdata dev_proc_ops = {
3375 .init = dev_proc_net_init,
3376 .exit = dev_proc_net_exit,
3379 static int __init dev_proc_init(void)
3381 return register_pernet_subsys(&dev_proc_ops);
3384 #define dev_proc_init() 0
3385 #endif /* CONFIG_PROC_FS */
3389 * netdev_set_master - set up master/slave pair
3390 * @slave: slave device
3391 * @master: new master device
3393 * Changes the master device of the slave. Pass %NULL to break the
3394 * bonding. The caller must hold the RTNL semaphore. On a failure
3395 * a negative errno code is returned. On success the reference counts
3396 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3397 * function returns zero.
3399 int netdev_set_master(struct net_device *slave, struct net_device *master)
3401 struct net_device *old = slave->master;
3411 slave->master = master;
3419 slave->flags |= IFF_SLAVE;
3421 slave->flags &= ~IFF_SLAVE;
3423 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3426 EXPORT_SYMBOL(netdev_set_master);
3428 static void dev_change_rx_flags(struct net_device *dev, int flags)
3430 const struct net_device_ops *ops = dev->netdev_ops;
3432 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3433 ops->ndo_change_rx_flags(dev, flags);
3436 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3438 unsigned short old_flags = dev->flags;
3444 dev->flags |= IFF_PROMISC;
3445 dev->promiscuity += inc;
3446 if (dev->promiscuity == 0) {
3449 * If inc causes overflow, untouch promisc and return error.
3452 dev->flags &= ~IFF_PROMISC;
3454 dev->promiscuity -= inc;
3455 printk(KERN_WARNING "%s: promiscuity touches roof, "
3456 "set promiscuity failed, promiscuity feature "
3457 "of device might be broken.\n", dev->name);
3461 if (dev->flags != old_flags) {
3462 printk(KERN_INFO "device %s %s promiscuous mode\n",
3463 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3465 if (audit_enabled) {
3466 current_uid_gid(&uid, &gid);
3467 audit_log(current->audit_context, GFP_ATOMIC,
3468 AUDIT_ANOM_PROMISCUOUS,
3469 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3470 dev->name, (dev->flags & IFF_PROMISC),
3471 (old_flags & IFF_PROMISC),
3472 audit_get_loginuid(current),
3474 audit_get_sessionid(current));
3477 dev_change_rx_flags(dev, IFF_PROMISC);
3483 * dev_set_promiscuity - update promiscuity count on a device
3487 * Add or remove promiscuity from a device. While the count in the device
3488 * remains above zero the interface remains promiscuous. Once it hits zero
3489 * the device reverts back to normal filtering operation. A negative inc
3490 * value is used to drop promiscuity on the device.
3491 * Return 0 if successful or a negative errno code on error.
3493 int dev_set_promiscuity(struct net_device *dev, int inc)
3495 unsigned short old_flags = dev->flags;
3498 err = __dev_set_promiscuity(dev, inc);
3501 if (dev->flags != old_flags)
3502 dev_set_rx_mode(dev);
3505 EXPORT_SYMBOL(dev_set_promiscuity);
3508 * dev_set_allmulti - update allmulti count on a device
3512 * Add or remove reception of all multicast frames to a device. While the
3513 * count in the device remains above zero the interface remains listening
3514 * to all interfaces. Once it hits zero the device reverts back to normal
3515 * filtering operation. A negative @inc value is used to drop the counter
3516 * when releasing a resource needing all multicasts.
3517 * Return 0 if successful or a negative errno code on error.
3520 int dev_set_allmulti(struct net_device *dev, int inc)
3522 unsigned short old_flags = dev->flags;
3526 dev->flags |= IFF_ALLMULTI;
3527 dev->allmulti += inc;
3528 if (dev->allmulti == 0) {
3531 * If inc causes overflow, untouch allmulti and return error.
3534 dev->flags &= ~IFF_ALLMULTI;
3536 dev->allmulti -= inc;
3537 printk(KERN_WARNING "%s: allmulti touches roof, "
3538 "set allmulti failed, allmulti feature of "
3539 "device might be broken.\n", dev->name);
3543 if (dev->flags ^ old_flags) {
3544 dev_change_rx_flags(dev, IFF_ALLMULTI);
3545 dev_set_rx_mode(dev);
3549 EXPORT_SYMBOL(dev_set_allmulti);
3552 * Upload unicast and multicast address lists to device and
3553 * configure RX filtering. When the device doesn't support unicast
3554 * filtering it is put in promiscuous mode while unicast addresses
3557 void __dev_set_rx_mode(struct net_device *dev)
3559 const struct net_device_ops *ops = dev->netdev_ops;
3561 /* dev_open will call this function so the list will stay sane. */
3562 if (!(dev->flags&IFF_UP))
3565 if (!netif_device_present(dev))
3568 if (ops->ndo_set_rx_mode)
3569 ops->ndo_set_rx_mode(dev);
3571 /* Unicast addresses changes may only happen under the rtnl,
3572 * therefore calling __dev_set_promiscuity here is safe.
3574 if (dev->uc.count > 0 && !dev->uc_promisc) {
3575 __dev_set_promiscuity(dev, 1);
3576 dev->uc_promisc = 1;
3577 } else if (dev->uc.count == 0 && dev->uc_promisc) {
3578 __dev_set_promiscuity(dev, -1);
3579 dev->uc_promisc = 0;
3582 if (ops->ndo_set_multicast_list)
3583 ops->ndo_set_multicast_list(dev);
3587 void dev_set_rx_mode(struct net_device *dev)
3589 netif_addr_lock_bh(dev);
3590 __dev_set_rx_mode(dev);
3591 netif_addr_unlock_bh(dev);
3594 /* hw addresses list handling functions */
3596 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3597 int addr_len, unsigned char addr_type)
3599 struct netdev_hw_addr *ha;
3602 if (addr_len > MAX_ADDR_LEN)
3605 list_for_each_entry(ha, &list->list, list) {
3606 if (!memcmp(ha->addr, addr, addr_len) &&
3607 ha->type == addr_type) {
3614 alloc_size = sizeof(*ha);
3615 if (alloc_size < L1_CACHE_BYTES)
3616 alloc_size = L1_CACHE_BYTES;
3617 ha = kmalloc(alloc_size, GFP_ATOMIC);
3620 memcpy(ha->addr, addr, addr_len);
3621 ha->type = addr_type;
3624 list_add_tail_rcu(&ha->list, &list->list);
3629 static void ha_rcu_free(struct rcu_head *head)
3631 struct netdev_hw_addr *ha;
3633 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3637 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3638 int addr_len, unsigned char addr_type)
3640 struct netdev_hw_addr *ha;
3642 list_for_each_entry(ha, &list->list, list) {
3643 if (!memcmp(ha->addr, addr, addr_len) &&
3644 (ha->type == addr_type || !addr_type)) {
3647 list_del_rcu(&ha->list);
3648 call_rcu(&ha->rcu_head, ha_rcu_free);
3656 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3657 struct netdev_hw_addr_list *from_list,
3659 unsigned char addr_type)
3662 struct netdev_hw_addr *ha, *ha2;
3665 list_for_each_entry(ha, &from_list->list, list) {
3666 type = addr_type ? addr_type : ha->type;
3667 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3674 list_for_each_entry(ha2, &from_list->list, list) {
3677 type = addr_type ? addr_type : ha2->type;
3678 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3683 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3684 struct netdev_hw_addr_list *from_list,
3686 unsigned char addr_type)
3688 struct netdev_hw_addr *ha;
3691 list_for_each_entry(ha, &from_list->list, list) {
3692 type = addr_type ? addr_type : ha->type;
3693 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3697 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3698 struct netdev_hw_addr_list *from_list,
3702 struct netdev_hw_addr *ha, *tmp;
3704 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3706 err = __hw_addr_add(to_list, ha->addr,
3707 addr_len, ha->type);
3712 } else if (ha->refcount == 1) {
3713 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3714 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3720 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3721 struct netdev_hw_addr_list *from_list,
3724 struct netdev_hw_addr *ha, *tmp;
3726 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3728 __hw_addr_del(to_list, ha->addr,
3729 addr_len, ha->type);
3731 __hw_addr_del(from_list, ha->addr,
3732 addr_len, ha->type);
3737 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3739 struct netdev_hw_addr *ha, *tmp;
3741 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3742 list_del_rcu(&ha->list);
3743 call_rcu(&ha->rcu_head, ha_rcu_free);
3748 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3750 INIT_LIST_HEAD(&list->list);
3754 /* Device addresses handling functions */
3756 static void dev_addr_flush(struct net_device *dev)
3758 /* rtnl_mutex must be held here */
3760 __hw_addr_flush(&dev->dev_addrs);
3761 dev->dev_addr = NULL;
3764 static int dev_addr_init(struct net_device *dev)
3766 unsigned char addr[MAX_ADDR_LEN];
3767 struct netdev_hw_addr *ha;
3770 /* rtnl_mutex must be held here */
3772 __hw_addr_init(&dev->dev_addrs);
3773 memset(addr, 0, sizeof(addr));
3774 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3775 NETDEV_HW_ADDR_T_LAN);
3778 * Get the first (previously created) address from the list
3779 * and set dev_addr pointer to this location.
3781 ha = list_first_entry(&dev->dev_addrs.list,
3782 struct netdev_hw_addr, list);
3783 dev->dev_addr = ha->addr;
3789 * dev_addr_add - Add a device address
3791 * @addr: address to add
3792 * @addr_type: address type
3794 * Add a device address to the device or increase the reference count if
3795 * it already exists.
3797 * The caller must hold the rtnl_mutex.
3799 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3800 unsigned char addr_type)
3806 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3808 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3811 EXPORT_SYMBOL(dev_addr_add);
3814 * dev_addr_del - Release a device address.
3816 * @addr: address to delete
3817 * @addr_type: address type
3819 * Release reference to a device address and remove it from the device
3820 * if the reference count drops to zero.
3822 * The caller must hold the rtnl_mutex.
3824 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3825 unsigned char addr_type)
3828 struct netdev_hw_addr *ha;
3833 * We can not remove the first address from the list because
3834 * dev->dev_addr points to that.
3836 ha = list_first_entry(&dev->dev_addrs.list,
3837 struct netdev_hw_addr, list);
3838 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3841 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3844 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3847 EXPORT_SYMBOL(dev_addr_del);
3850 * dev_addr_add_multiple - Add device addresses from another device
3851 * @to_dev: device to which addresses will be added
3852 * @from_dev: device from which addresses will be added
3853 * @addr_type: address type - 0 means type will be used from from_dev
3855 * Add device addresses of the one device to another.
3857 * The caller must hold the rtnl_mutex.
3859 int dev_addr_add_multiple(struct net_device *to_dev,
3860 struct net_device *from_dev,
3861 unsigned char addr_type)
3867 if (from_dev->addr_len != to_dev->addr_len)
3869 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3870 to_dev->addr_len, addr_type);
3872 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3875 EXPORT_SYMBOL(dev_addr_add_multiple);
3878 * dev_addr_del_multiple - Delete device addresses by another device
3879 * @to_dev: device where the addresses will be deleted
3880 * @from_dev: device by which addresses the addresses will be deleted
3881 * @addr_type: address type - 0 means type will used from from_dev
3883 * Deletes addresses in to device by the list of addresses in from device.
3885 * The caller must hold the rtnl_mutex.
3887 int dev_addr_del_multiple(struct net_device *to_dev,
3888 struct net_device *from_dev,
3889 unsigned char addr_type)
3893 if (from_dev->addr_len != to_dev->addr_len)
3895 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3896 to_dev->addr_len, addr_type);
3897 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3900 EXPORT_SYMBOL(dev_addr_del_multiple);
3902 /* multicast addresses handling functions */
3904 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3905 void *addr, int alen, int glbl)
3907 struct dev_addr_list *da;
3909 for (; (da = *list) != NULL; list = &da->next) {
3910 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3911 alen == da->da_addrlen) {
3913 int old_glbl = da->da_gusers;
3930 int __dev_addr_add(struct dev_addr_list **list, int *count,
3931 void *addr, int alen, int glbl)
3933 struct dev_addr_list *da;
3935 for (da = *list; da != NULL; da = da->next) {
3936 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3937 da->da_addrlen == alen) {
3939 int old_glbl = da->da_gusers;
3949 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3952 memcpy(da->da_addr, addr, alen);
3953 da->da_addrlen = alen;
3955 da->da_gusers = glbl ? 1 : 0;
3963 * dev_unicast_delete - Release secondary unicast address.
3965 * @addr: address to delete
3967 * Release reference to a secondary unicast address and remove it
3968 * from the device if the reference count drops to zero.
3970 * The caller must hold the rtnl_mutex.
3972 int dev_unicast_delete(struct net_device *dev, void *addr)
3978 netif_addr_lock_bh(dev);
3979 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
3980 NETDEV_HW_ADDR_T_UNICAST);
3982 __dev_set_rx_mode(dev);
3983 netif_addr_unlock_bh(dev);
3986 EXPORT_SYMBOL(dev_unicast_delete);
3989 * dev_unicast_add - add a secondary unicast address
3991 * @addr: address to add
3993 * Add a secondary unicast address to the device or increase
3994 * the reference count if it already exists.
3996 * The caller must hold the rtnl_mutex.
3998 int dev_unicast_add(struct net_device *dev, void *addr)
4004 netif_addr_lock_bh(dev);
4005 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
4006 NETDEV_HW_ADDR_T_UNICAST);
4008 __dev_set_rx_mode(dev);
4009 netif_addr_unlock_bh(dev);
4012 EXPORT_SYMBOL(dev_unicast_add);
4014 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
4015 struct dev_addr_list **from, int *from_count)
4017 struct dev_addr_list *da, *next;
4021 while (da != NULL) {
4023 if (!da->da_synced) {
4024 err = __dev_addr_add(to, to_count,
4025 da->da_addr, da->da_addrlen, 0);
4030 } else if (da->da_users == 1) {
4031 __dev_addr_delete(to, to_count,
4032 da->da_addr, da->da_addrlen, 0);
4033 __dev_addr_delete(from, from_count,
4034 da->da_addr, da->da_addrlen, 0);
4040 EXPORT_SYMBOL_GPL(__dev_addr_sync);
4042 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4043 struct dev_addr_list **from, int *from_count)
4045 struct dev_addr_list *da, *next;
4048 while (da != NULL) {
4050 if (da->da_synced) {
4051 __dev_addr_delete(to, to_count,
4052 da->da_addr, da->da_addrlen, 0);
4054 __dev_addr_delete(from, from_count,
4055 da->da_addr, da->da_addrlen, 0);
4060 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4063 * dev_unicast_sync - Synchronize device's unicast list to another device
4064 * @to: destination device
4065 * @from: source device
4067 * Add newly added addresses to the destination device and release
4068 * addresses that have no users left. The source device must be
4069 * locked by netif_tx_lock_bh.
4071 * This function is intended to be called from the dev->set_rx_mode
4072 * function of layered software devices.
4074 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4078 if (to->addr_len != from->addr_len)
4081 netif_addr_lock_bh(to);
4082 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4084 __dev_set_rx_mode(to);
4085 netif_addr_unlock_bh(to);
4088 EXPORT_SYMBOL(dev_unicast_sync);
4091 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4092 * @to: destination device
4093 * @from: source device
4095 * Remove all addresses that were added to the destination device by
4096 * dev_unicast_sync(). This function is intended to be called from the
4097 * dev->stop function of layered software devices.
4099 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4101 if (to->addr_len != from->addr_len)
4104 netif_addr_lock_bh(from);
4105 netif_addr_lock(to);
4106 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4107 __dev_set_rx_mode(to);
4108 netif_addr_unlock(to);
4109 netif_addr_unlock_bh(from);
4111 EXPORT_SYMBOL(dev_unicast_unsync);
4113 static void dev_unicast_flush(struct net_device *dev)
4115 netif_addr_lock_bh(dev);
4116 __hw_addr_flush(&dev->uc);
4117 netif_addr_unlock_bh(dev);
4120 static void dev_unicast_init(struct net_device *dev)
4122 __hw_addr_init(&dev->uc);
4126 static void __dev_addr_discard(struct dev_addr_list **list)
4128 struct dev_addr_list *tmp;
4130 while (*list != NULL) {
4133 if (tmp->da_users > tmp->da_gusers)
4134 printk("__dev_addr_discard: address leakage! "
4135 "da_users=%d\n", tmp->da_users);
4140 static void dev_addr_discard(struct net_device *dev)
4142 netif_addr_lock_bh(dev);
4144 __dev_addr_discard(&dev->mc_list);
4147 netif_addr_unlock_bh(dev);
4151 * dev_get_flags - get flags reported to userspace
4154 * Get the combination of flag bits exported through APIs to userspace.
4156 unsigned dev_get_flags(const struct net_device *dev)
4160 flags = (dev->flags & ~(IFF_PROMISC |
4165 (dev->gflags & (IFF_PROMISC |
4168 if (netif_running(dev)) {
4169 if (netif_oper_up(dev))
4170 flags |= IFF_RUNNING;
4171 if (netif_carrier_ok(dev))
4172 flags |= IFF_LOWER_UP;
4173 if (netif_dormant(dev))
4174 flags |= IFF_DORMANT;
4179 EXPORT_SYMBOL(dev_get_flags);
4182 * dev_change_flags - change device settings
4184 * @flags: device state flags
4186 * Change settings on device based state flags. The flags are
4187 * in the userspace exported format.
4189 int dev_change_flags(struct net_device *dev, unsigned flags)
4192 int old_flags = dev->flags;
4197 * Set the flags on our device.
4200 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4201 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4203 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4207 * Load in the correct multicast list now the flags have changed.
4210 if ((old_flags ^ flags) & IFF_MULTICAST)
4211 dev_change_rx_flags(dev, IFF_MULTICAST);
4213 dev_set_rx_mode(dev);
4216 * Have we downed the interface. We handle IFF_UP ourselves
4217 * according to user attempts to set it, rather than blindly
4222 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4223 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4226 dev_set_rx_mode(dev);
4229 if (dev->flags & IFF_UP &&
4230 ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
4232 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4234 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4235 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4237 dev->gflags ^= IFF_PROMISC;
4238 dev_set_promiscuity(dev, inc);
4241 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4242 is important. Some (broken) drivers set IFF_PROMISC, when
4243 IFF_ALLMULTI is requested not asking us and not reporting.
4245 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4246 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4248 dev->gflags ^= IFF_ALLMULTI;
4249 dev_set_allmulti(dev, inc);
4252 /* Exclude state transition flags, already notified */
4253 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4255 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4259 EXPORT_SYMBOL(dev_change_flags);
4262 * dev_set_mtu - Change maximum transfer unit
4264 * @new_mtu: new transfer unit
4266 * Change the maximum transfer size of the network device.
4268 int dev_set_mtu(struct net_device *dev, int new_mtu)
4270 const struct net_device_ops *ops = dev->netdev_ops;
4273 if (new_mtu == dev->mtu)
4276 /* MTU must be positive. */
4280 if (!netif_device_present(dev))
4284 if (ops->ndo_change_mtu)
4285 err = ops->ndo_change_mtu(dev, new_mtu);
4289 if (!err && dev->flags & IFF_UP)
4290 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4293 EXPORT_SYMBOL(dev_set_mtu);
4296 * dev_set_mac_address - Change Media Access Control Address
4300 * Change the hardware (MAC) address of the device
4302 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4304 const struct net_device_ops *ops = dev->netdev_ops;
4307 if (!ops->ndo_set_mac_address)
4309 if (sa->sa_family != dev->type)
4311 if (!netif_device_present(dev))
4313 err = ops->ndo_set_mac_address(dev, sa);
4315 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4318 EXPORT_SYMBOL(dev_set_mac_address);
4321 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4323 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4326 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4332 case SIOCGIFFLAGS: /* Get interface flags */
4333 ifr->ifr_flags = (short) dev_get_flags(dev);
4336 case SIOCGIFMETRIC: /* Get the metric on the interface
4337 (currently unused) */
4338 ifr->ifr_metric = 0;
4341 case SIOCGIFMTU: /* Get the MTU of a device */
4342 ifr->ifr_mtu = dev->mtu;
4347 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4349 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4350 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4351 ifr->ifr_hwaddr.sa_family = dev->type;
4359 ifr->ifr_map.mem_start = dev->mem_start;
4360 ifr->ifr_map.mem_end = dev->mem_end;
4361 ifr->ifr_map.base_addr = dev->base_addr;
4362 ifr->ifr_map.irq = dev->irq;
4363 ifr->ifr_map.dma = dev->dma;
4364 ifr->ifr_map.port = dev->if_port;
4368 ifr->ifr_ifindex = dev->ifindex;
4372 ifr->ifr_qlen = dev->tx_queue_len;
4376 /* dev_ioctl() should ensure this case
4388 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4390 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4393 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4394 const struct net_device_ops *ops;
4399 ops = dev->netdev_ops;
4402 case SIOCSIFFLAGS: /* Set interface flags */
4403 return dev_change_flags(dev, ifr->ifr_flags);
4405 case SIOCSIFMETRIC: /* Set the metric on the interface
4406 (currently unused) */
4409 case SIOCSIFMTU: /* Set the MTU of a device */
4410 return dev_set_mtu(dev, ifr->ifr_mtu);
4413 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4415 case SIOCSIFHWBROADCAST:
4416 if (ifr->ifr_hwaddr.sa_family != dev->type)
4418 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4419 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4420 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4424 if (ops->ndo_set_config) {
4425 if (!netif_device_present(dev))
4427 return ops->ndo_set_config(dev, &ifr->ifr_map);
4432 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4433 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4435 if (!netif_device_present(dev))
4437 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4441 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4442 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4444 if (!netif_device_present(dev))
4446 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4450 if (ifr->ifr_qlen < 0)
4452 dev->tx_queue_len = ifr->ifr_qlen;
4456 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4457 return dev_change_name(dev, ifr->ifr_newname);
4460 * Unknown or private ioctl
4463 if ((cmd >= SIOCDEVPRIVATE &&
4464 cmd <= SIOCDEVPRIVATE + 15) ||
4465 cmd == SIOCBONDENSLAVE ||
4466 cmd == SIOCBONDRELEASE ||
4467 cmd == SIOCBONDSETHWADDR ||
4468 cmd == SIOCBONDSLAVEINFOQUERY ||
4469 cmd == SIOCBONDINFOQUERY ||
4470 cmd == SIOCBONDCHANGEACTIVE ||
4471 cmd == SIOCGMIIPHY ||
4472 cmd == SIOCGMIIREG ||
4473 cmd == SIOCSMIIREG ||
4474 cmd == SIOCBRADDIF ||
4475 cmd == SIOCBRDELIF ||
4476 cmd == SIOCSHWTSTAMP ||
4477 cmd == SIOCWANDEV) {
4479 if (ops->ndo_do_ioctl) {
4480 if (netif_device_present(dev))
4481 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4493 * This function handles all "interface"-type I/O control requests. The actual
4494 * 'doing' part of this is dev_ifsioc above.
4498 * dev_ioctl - network device ioctl
4499 * @net: the applicable net namespace
4500 * @cmd: command to issue
4501 * @arg: pointer to a struct ifreq in user space
4503 * Issue ioctl functions to devices. This is normally called by the
4504 * user space syscall interfaces but can sometimes be useful for
4505 * other purposes. The return value is the return from the syscall if
4506 * positive or a negative errno code on error.
4509 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4515 /* One special case: SIOCGIFCONF takes ifconf argument
4516 and requires shared lock, because it sleeps writing
4520 if (cmd == SIOCGIFCONF) {
4522 ret = dev_ifconf(net, (char __user *) arg);
4526 if (cmd == SIOCGIFNAME)
4527 return dev_ifname(net, (struct ifreq __user *)arg);
4529 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4532 ifr.ifr_name[IFNAMSIZ-1] = 0;
4534 colon = strchr(ifr.ifr_name, ':');
4539 * See which interface the caller is talking about.
4544 * These ioctl calls:
4545 * - can be done by all.
4546 * - atomic and do not require locking.
4557 dev_load(net, ifr.ifr_name);
4559 ret = dev_ifsioc_locked(net, &ifr, cmd);
4564 if (copy_to_user(arg, &ifr,
4565 sizeof(struct ifreq)))
4571 dev_load(net, ifr.ifr_name);
4573 ret = dev_ethtool(net, &ifr);
4578 if (copy_to_user(arg, &ifr,
4579 sizeof(struct ifreq)))
4585 * These ioctl calls:
4586 * - require superuser power.
4587 * - require strict serialization.
4593 if (!capable(CAP_NET_ADMIN))
4595 dev_load(net, ifr.ifr_name);
4597 ret = dev_ifsioc(net, &ifr, cmd);
4602 if (copy_to_user(arg, &ifr,
4603 sizeof(struct ifreq)))
4609 * These ioctl calls:
4610 * - require superuser power.
4611 * - require strict serialization.
4612 * - do not return a value
4622 case SIOCSIFHWBROADCAST:
4625 case SIOCBONDENSLAVE:
4626 case SIOCBONDRELEASE:
4627 case SIOCBONDSETHWADDR:
4628 case SIOCBONDCHANGEACTIVE:
4632 if (!capable(CAP_NET_ADMIN))
4635 case SIOCBONDSLAVEINFOQUERY:
4636 case SIOCBONDINFOQUERY:
4637 dev_load(net, ifr.ifr_name);
4639 ret = dev_ifsioc(net, &ifr, cmd);
4644 /* Get the per device memory space. We can add this but
4645 * currently do not support it */
4647 /* Set the per device memory buffer space.
4648 * Not applicable in our case */
4653 * Unknown or private ioctl.
4656 if (cmd == SIOCWANDEV ||
4657 (cmd >= SIOCDEVPRIVATE &&
4658 cmd <= SIOCDEVPRIVATE + 15)) {
4659 dev_load(net, ifr.ifr_name);
4661 ret = dev_ifsioc(net, &ifr, cmd);
4663 if (!ret && copy_to_user(arg, &ifr,
4664 sizeof(struct ifreq)))
4668 /* Take care of Wireless Extensions */
4669 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4670 return wext_handle_ioctl(net, &ifr, cmd, arg);
4677 * dev_new_index - allocate an ifindex
4678 * @net: the applicable net namespace
4680 * Returns a suitable unique value for a new device interface
4681 * number. The caller must hold the rtnl semaphore or the
4682 * dev_base_lock to be sure it remains unique.
4684 static int dev_new_index(struct net *net)
4690 if (!__dev_get_by_index(net, ifindex))
4695 /* Delayed registration/unregisteration */
4696 static LIST_HEAD(net_todo_list);
4698 static void net_set_todo(struct net_device *dev)
4700 list_add_tail(&dev->todo_list, &net_todo_list);
4703 static void rollback_registered_many(struct list_head *head)
4705 struct net_device *dev;
4707 BUG_ON(dev_boot_phase);
4710 list_for_each_entry(dev, head, unreg_list) {
4711 /* Some devices call without registering
4712 * for initialization unwind.
4714 if (dev->reg_state == NETREG_UNINITIALIZED) {
4715 pr_debug("unregister_netdevice: device %s/%p never "
4716 "was registered\n", dev->name, dev);
4722 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4724 /* If device is running, close it first. */
4727 /* And unlink it from device chain. */
4728 unlist_netdevice(dev);
4730 dev->reg_state = NETREG_UNREGISTERING;
4735 list_for_each_entry(dev, head, unreg_list) {
4736 /* Shutdown queueing discipline. */
4740 /* Notify protocols, that we are about to destroy
4741 this device. They should clean all the things.
4743 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4746 * Flush the unicast and multicast chains
4748 dev_unicast_flush(dev);
4749 dev_addr_discard(dev);
4751 if (dev->netdev_ops->ndo_uninit)
4752 dev->netdev_ops->ndo_uninit(dev);
4754 /* Notifier chain MUST detach us from master device. */
4755 WARN_ON(dev->master);
4757 /* Remove entries from kobject tree */
4758 netdev_unregister_kobject(dev);
4763 list_for_each_entry(dev, head, unreg_list)
4767 static void rollback_registered(struct net_device *dev)
4771 list_add(&dev->unreg_list, &single);
4772 rollback_registered_many(&single);
4775 static void __netdev_init_queue_locks_one(struct net_device *dev,
4776 struct netdev_queue *dev_queue,
4779 spin_lock_init(&dev_queue->_xmit_lock);
4780 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4781 dev_queue->xmit_lock_owner = -1;
4784 static void netdev_init_queue_locks(struct net_device *dev)
4786 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4787 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4790 unsigned long netdev_fix_features(unsigned long features, const char *name)
4792 /* Fix illegal SG+CSUM combinations. */
4793 if ((features & NETIF_F_SG) &&
4794 !(features & NETIF_F_ALL_CSUM)) {
4796 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4797 "checksum feature.\n", name);
4798 features &= ~NETIF_F_SG;
4801 /* TSO requires that SG is present as well. */
4802 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4804 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4805 "SG feature.\n", name);
4806 features &= ~NETIF_F_TSO;
4809 if (features & NETIF_F_UFO) {
4810 if (!(features & NETIF_F_GEN_CSUM)) {
4812 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4813 "since no NETIF_F_HW_CSUM feature.\n",
4815 features &= ~NETIF_F_UFO;
4818 if (!(features & NETIF_F_SG)) {
4820 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4821 "since no NETIF_F_SG feature.\n", name);
4822 features &= ~NETIF_F_UFO;
4828 EXPORT_SYMBOL(netdev_fix_features);
4831 * register_netdevice - register a network device
4832 * @dev: device to register
4834 * Take a completed network device structure and add it to the kernel
4835 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4836 * chain. 0 is returned on success. A negative errno code is returned
4837 * on a failure to set up the device, or if the name is a duplicate.
4839 * Callers must hold the rtnl semaphore. You may want
4840 * register_netdev() instead of this.
4843 * The locking appears insufficient to guarantee two parallel registers
4844 * will not get the same name.
4847 int register_netdevice(struct net_device *dev)
4849 struct hlist_head *head;
4850 struct hlist_node *p;
4852 struct net *net = dev_net(dev);
4854 BUG_ON(dev_boot_phase);
4859 /* When net_device's are persistent, this will be fatal. */
4860 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4863 spin_lock_init(&dev->addr_list_lock);
4864 netdev_set_addr_lockdep_class(dev);
4865 netdev_init_queue_locks(dev);
4869 /* Init, if this function is available */
4870 if (dev->netdev_ops->ndo_init) {
4871 ret = dev->netdev_ops->ndo_init(dev);
4879 if (!dev_valid_name(dev->name)) {
4884 dev->ifindex = dev_new_index(net);
4885 if (dev->iflink == -1)
4886 dev->iflink = dev->ifindex;
4888 /* Check for existence of name */
4889 head = dev_name_hash(net, dev->name);
4890 hlist_for_each(p, head) {
4891 struct net_device *d
4892 = hlist_entry(p, struct net_device, name_hlist);
4893 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4899 /* Fix illegal checksum combinations */
4900 if ((dev->features & NETIF_F_HW_CSUM) &&
4901 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4902 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4904 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4907 if ((dev->features & NETIF_F_NO_CSUM) &&
4908 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4909 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4911 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4914 dev->features = netdev_fix_features(dev->features, dev->name);
4916 /* Enable software GSO if SG is supported. */
4917 if (dev->features & NETIF_F_SG)
4918 dev->features |= NETIF_F_GSO;
4920 netdev_initialize_kobject(dev);
4922 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4923 ret = notifier_to_errno(ret);
4927 ret = netdev_register_kobject(dev);
4930 dev->reg_state = NETREG_REGISTERED;
4933 * Default initial state at registry is that the
4934 * device is present.
4937 set_bit(__LINK_STATE_PRESENT, &dev->state);
4939 dev_init_scheduler(dev);
4941 list_netdevice(dev);
4943 /* Notify protocols, that a new device appeared. */
4944 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4945 ret = notifier_to_errno(ret);
4947 rollback_registered(dev);
4948 dev->reg_state = NETREG_UNREGISTERED;
4955 if (dev->netdev_ops->ndo_uninit)
4956 dev->netdev_ops->ndo_uninit(dev);
4959 EXPORT_SYMBOL(register_netdevice);
4962 * init_dummy_netdev - init a dummy network device for NAPI
4963 * @dev: device to init
4965 * This takes a network device structure and initialize the minimum
4966 * amount of fields so it can be used to schedule NAPI polls without
4967 * registering a full blown interface. This is to be used by drivers
4968 * that need to tie several hardware interfaces to a single NAPI
4969 * poll scheduler due to HW limitations.
4971 int init_dummy_netdev(struct net_device *dev)
4973 /* Clear everything. Note we don't initialize spinlocks
4974 * are they aren't supposed to be taken by any of the
4975 * NAPI code and this dummy netdev is supposed to be
4976 * only ever used for NAPI polls
4978 memset(dev, 0, sizeof(struct net_device));
4980 /* make sure we BUG if trying to hit standard
4981 * register/unregister code path
4983 dev->reg_state = NETREG_DUMMY;
4985 /* initialize the ref count */
4986 atomic_set(&dev->refcnt, 1);
4988 /* NAPI wants this */
4989 INIT_LIST_HEAD(&dev->napi_list);
4991 /* a dummy interface is started by default */
4992 set_bit(__LINK_STATE_PRESENT, &dev->state);
4993 set_bit(__LINK_STATE_START, &dev->state);
4997 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5001 * register_netdev - register a network device
5002 * @dev: device to register
5004 * Take a completed network device structure and add it to the kernel
5005 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5006 * chain. 0 is returned on success. A negative errno code is returned
5007 * on a failure to set up the device, or if the name is a duplicate.
5009 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5010 * and expands the device name if you passed a format string to
5013 int register_netdev(struct net_device *dev)
5020 * If the name is a format string the caller wants us to do a
5023 if (strchr(dev->name, '%')) {
5024 err = dev_alloc_name(dev, dev->name);
5029 err = register_netdevice(dev);
5034 EXPORT_SYMBOL(register_netdev);
5037 * netdev_wait_allrefs - wait until all references are gone.
5039 * This is called when unregistering network devices.
5041 * Any protocol or device that holds a reference should register
5042 * for netdevice notification, and cleanup and put back the
5043 * reference if they receive an UNREGISTER event.
5044 * We can get stuck here if buggy protocols don't correctly
5047 static void netdev_wait_allrefs(struct net_device *dev)
5049 unsigned long rebroadcast_time, warning_time;
5051 rebroadcast_time = warning_time = jiffies;
5052 while (atomic_read(&dev->refcnt) != 0) {
5053 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5056 /* Rebroadcast unregister notification */
5057 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5059 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5061 /* We must not have linkwatch events
5062 * pending on unregister. If this
5063 * happens, we simply run the queue
5064 * unscheduled, resulting in a noop
5067 linkwatch_run_queue();
5072 rebroadcast_time = jiffies;
5077 if (time_after(jiffies, warning_time + 10 * HZ)) {
5078 printk(KERN_EMERG "unregister_netdevice: "
5079 "waiting for %s to become free. Usage "
5081 dev->name, atomic_read(&dev->refcnt));
5082 warning_time = jiffies;
5091 * register_netdevice(x1);
5092 * register_netdevice(x2);
5094 * unregister_netdevice(y1);
5095 * unregister_netdevice(y2);
5101 * We are invoked by rtnl_unlock().
5102 * This allows us to deal with problems:
5103 * 1) We can delete sysfs objects which invoke hotplug
5104 * without deadlocking with linkwatch via keventd.
5105 * 2) Since we run with the RTNL semaphore not held, we can sleep
5106 * safely in order to wait for the netdev refcnt to drop to zero.
5108 * We must not return until all unregister events added during
5109 * the interval the lock was held have been completed.
5111 void netdev_run_todo(void)
5113 struct list_head list;
5115 /* Snapshot list, allow later requests */
5116 list_replace_init(&net_todo_list, &list);
5120 while (!list_empty(&list)) {
5121 struct net_device *dev
5122 = list_entry(list.next, struct net_device, todo_list);
5123 list_del(&dev->todo_list);
5125 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5126 printk(KERN_ERR "network todo '%s' but state %d\n",
5127 dev->name, dev->reg_state);
5132 dev->reg_state = NETREG_UNREGISTERED;
5134 on_each_cpu(flush_backlog, dev, 1);
5136 netdev_wait_allrefs(dev);
5139 BUG_ON(atomic_read(&dev->refcnt));
5140 WARN_ON(dev->ip_ptr);
5141 WARN_ON(dev->ip6_ptr);
5142 WARN_ON(dev->dn_ptr);
5144 if (dev->destructor)
5145 dev->destructor(dev);
5147 /* Free network device */
5148 kobject_put(&dev->dev.kobj);
5153 * dev_get_stats - get network device statistics
5154 * @dev: device to get statistics from
5156 * Get network statistics from device. The device driver may provide
5157 * its own method by setting dev->netdev_ops->get_stats; otherwise
5158 * the internal statistics structure is used.
5160 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5162 const struct net_device_ops *ops = dev->netdev_ops;
5164 if (ops->ndo_get_stats)
5165 return ops->ndo_get_stats(dev);
5167 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5168 struct net_device_stats *stats = &dev->stats;
5170 struct netdev_queue *txq;
5172 for (i = 0; i < dev->num_tx_queues; i++) {
5173 txq = netdev_get_tx_queue(dev, i);
5174 tx_bytes += txq->tx_bytes;
5175 tx_packets += txq->tx_packets;
5176 tx_dropped += txq->tx_dropped;
5178 if (tx_bytes || tx_packets || tx_dropped) {
5179 stats->tx_bytes = tx_bytes;
5180 stats->tx_packets = tx_packets;
5181 stats->tx_dropped = tx_dropped;
5186 EXPORT_SYMBOL(dev_get_stats);
5188 static void netdev_init_one_queue(struct net_device *dev,
5189 struct netdev_queue *queue,
5195 static void netdev_init_queues(struct net_device *dev)
5197 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5198 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5199 spin_lock_init(&dev->tx_global_lock);
5203 * alloc_netdev_mq - allocate network device
5204 * @sizeof_priv: size of private data to allocate space for
5205 * @name: device name format string
5206 * @setup: callback to initialize device
5207 * @queue_count: the number of subqueues to allocate
5209 * Allocates a struct net_device with private data area for driver use
5210 * and performs basic initialization. Also allocates subquue structs
5211 * for each queue on the device at the end of the netdevice.
5213 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5214 void (*setup)(struct net_device *), unsigned int queue_count)
5216 struct netdev_queue *tx;
5217 struct net_device *dev;
5219 struct net_device *p;
5221 BUG_ON(strlen(name) >= sizeof(dev->name));
5223 alloc_size = sizeof(struct net_device);
5225 /* ensure 32-byte alignment of private area */
5226 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5227 alloc_size += sizeof_priv;
5229 /* ensure 32-byte alignment of whole construct */
5230 alloc_size += NETDEV_ALIGN - 1;
5232 p = kzalloc(alloc_size, GFP_KERNEL);
5234 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5238 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5240 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5245 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5246 dev->padded = (char *)dev - (char *)p;
5248 if (dev_addr_init(dev))
5251 dev_unicast_init(dev);
5253 dev_net_set(dev, &init_net);
5256 dev->num_tx_queues = queue_count;
5257 dev->real_num_tx_queues = queue_count;
5259 dev->gso_max_size = GSO_MAX_SIZE;
5261 netdev_init_queues(dev);
5263 INIT_LIST_HEAD(&dev->napi_list);
5264 INIT_LIST_HEAD(&dev->unreg_list);
5265 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5267 strcpy(dev->name, name);
5277 EXPORT_SYMBOL(alloc_netdev_mq);
5280 * free_netdev - free network device
5283 * This function does the last stage of destroying an allocated device
5284 * interface. The reference to the device object is released.
5285 * If this is the last reference then it will be freed.
5287 void free_netdev(struct net_device *dev)
5289 struct napi_struct *p, *n;
5291 release_net(dev_net(dev));
5295 /* Flush device addresses */
5296 dev_addr_flush(dev);
5298 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5301 /* Compatibility with error handling in drivers */
5302 if (dev->reg_state == NETREG_UNINITIALIZED) {
5303 kfree((char *)dev - dev->padded);
5307 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5308 dev->reg_state = NETREG_RELEASED;
5310 /* will free via device release */
5311 put_device(&dev->dev);
5313 EXPORT_SYMBOL(free_netdev);
5316 * synchronize_net - Synchronize with packet receive processing
5318 * Wait for packets currently being received to be done.
5319 * Does not block later packets from starting.
5321 void synchronize_net(void)
5326 EXPORT_SYMBOL(synchronize_net);
5329 * unregister_netdevice_queue - remove device from the kernel
5333 * This function shuts down a device interface and removes it
5334 * from the kernel tables.
5335 * If head not NULL, device is queued to be unregistered later.
5337 * Callers must hold the rtnl semaphore. You may want
5338 * unregister_netdev() instead of this.
5341 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5346 list_move_tail(&dev->unreg_list, head);
5348 rollback_registered(dev);
5349 /* Finish processing unregister after unlock */
5353 EXPORT_SYMBOL(unregister_netdevice_queue);
5356 * unregister_netdevice_many - unregister many devices
5357 * @head: list of devices
5360 void unregister_netdevice_many(struct list_head *head)
5362 struct net_device *dev;
5364 if (!list_empty(head)) {
5365 rollback_registered_many(head);
5366 list_for_each_entry(dev, head, unreg_list)
5370 EXPORT_SYMBOL(unregister_netdevice_many);
5373 * unregister_netdev - remove device from the kernel
5376 * This function shuts down a device interface and removes it
5377 * from the kernel tables.
5379 * This is just a wrapper for unregister_netdevice that takes
5380 * the rtnl semaphore. In general you want to use this and not
5381 * unregister_netdevice.
5383 void unregister_netdev(struct net_device *dev)
5386 unregister_netdevice(dev);
5389 EXPORT_SYMBOL(unregister_netdev);
5392 * dev_change_net_namespace - move device to different nethost namespace
5394 * @net: network namespace
5395 * @pat: If not NULL name pattern to try if the current device name
5396 * is already taken in the destination network namespace.
5398 * This function shuts down a device interface and moves it
5399 * to a new network namespace. On success 0 is returned, on
5400 * a failure a netagive errno code is returned.
5402 * Callers must hold the rtnl semaphore.
5405 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5408 const char *destname;
5413 /* Don't allow namespace local devices to be moved. */
5415 if (dev->features & NETIF_F_NETNS_LOCAL)
5419 /* Don't allow real devices to be moved when sysfs
5423 if (dev->dev.parent)
5427 /* Ensure the device has been registrered */
5429 if (dev->reg_state != NETREG_REGISTERED)
5432 /* Get out if there is nothing todo */
5434 if (net_eq(dev_net(dev), net))
5437 /* Pick the destination device name, and ensure
5438 * we can use it in the destination network namespace.
5441 destname = dev->name;
5442 if (__dev_get_by_name(net, destname)) {
5443 /* We get here if we can't use the current device name */
5446 if (!dev_valid_name(pat))
5448 if (strchr(pat, '%')) {
5449 if (__dev_alloc_name(net, pat, buf) < 0)
5454 if (__dev_get_by_name(net, destname))
5459 * And now a mini version of register_netdevice unregister_netdevice.
5462 /* If device is running close it first. */
5465 /* And unlink it from device chain */
5467 unlist_netdevice(dev);
5471 /* Shutdown queueing discipline. */
5474 /* Notify protocols, that we are about to destroy
5475 this device. They should clean all the things.
5477 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5480 * Flush the unicast and multicast chains
5482 dev_unicast_flush(dev);
5483 dev_addr_discard(dev);
5485 netdev_unregister_kobject(dev);
5487 /* Actually switch the network namespace */
5488 dev_net_set(dev, net);
5490 /* Assign the new device name */
5491 if (destname != dev->name)
5492 strcpy(dev->name, destname);
5494 /* If there is an ifindex conflict assign a new one */
5495 if (__dev_get_by_index(net, dev->ifindex)) {
5496 int iflink = (dev->iflink == dev->ifindex);
5497 dev->ifindex = dev_new_index(net);
5499 dev->iflink = dev->ifindex;
5502 /* Fixup kobjects */
5503 err = netdev_register_kobject(dev);
5506 /* Add the device back in the hashes */
5507 list_netdevice(dev);
5509 /* Notify protocols, that a new device appeared. */
5510 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5517 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5519 static int dev_cpu_callback(struct notifier_block *nfb,
5520 unsigned long action,
5523 struct sk_buff **list_skb;
5524 struct Qdisc **list_net;
5525 struct sk_buff *skb;
5526 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5527 struct softnet_data *sd, *oldsd;
5529 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5532 local_irq_disable();
5533 cpu = smp_processor_id();
5534 sd = &per_cpu(softnet_data, cpu);
5535 oldsd = &per_cpu(softnet_data, oldcpu);
5537 /* Find end of our completion_queue. */
5538 list_skb = &sd->completion_queue;
5540 list_skb = &(*list_skb)->next;
5541 /* Append completion queue from offline CPU. */
5542 *list_skb = oldsd->completion_queue;
5543 oldsd->completion_queue = NULL;
5545 /* Find end of our output_queue. */
5546 list_net = &sd->output_queue;
5548 list_net = &(*list_net)->next_sched;
5549 /* Append output queue from offline CPU. */
5550 *list_net = oldsd->output_queue;
5551 oldsd->output_queue = NULL;
5553 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5556 /* Process offline CPU's input_pkt_queue */
5557 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5565 * netdev_increment_features - increment feature set by one
5566 * @all: current feature set
5567 * @one: new feature set
5568 * @mask: mask feature set
5570 * Computes a new feature set after adding a device with feature set
5571 * @one to the master device with current feature set @all. Will not
5572 * enable anything that is off in @mask. Returns the new feature set.
5574 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5577 /* If device needs checksumming, downgrade to it. */
5578 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5579 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5580 else if (mask & NETIF_F_ALL_CSUM) {
5581 /* If one device supports v4/v6 checksumming, set for all. */
5582 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5583 !(all & NETIF_F_GEN_CSUM)) {
5584 all &= ~NETIF_F_ALL_CSUM;
5585 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5588 /* If one device supports hw checksumming, set for all. */
5589 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5590 all &= ~NETIF_F_ALL_CSUM;
5591 all |= NETIF_F_HW_CSUM;
5595 one |= NETIF_F_ALL_CSUM;
5597 one |= all & NETIF_F_ONE_FOR_ALL;
5598 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5599 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5603 EXPORT_SYMBOL(netdev_increment_features);
5605 static struct hlist_head *netdev_create_hash(void)
5608 struct hlist_head *hash;
5610 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5612 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5613 INIT_HLIST_HEAD(&hash[i]);
5618 /* Initialize per network namespace state */
5619 static int __net_init netdev_init(struct net *net)
5621 INIT_LIST_HEAD(&net->dev_base_head);
5623 net->dev_name_head = netdev_create_hash();
5624 if (net->dev_name_head == NULL)
5627 net->dev_index_head = netdev_create_hash();
5628 if (net->dev_index_head == NULL)
5634 kfree(net->dev_name_head);
5640 * netdev_drivername - network driver for the device
5641 * @dev: network device
5642 * @buffer: buffer for resulting name
5643 * @len: size of buffer
5645 * Determine network driver for device.
5647 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5649 const struct device_driver *driver;
5650 const struct device *parent;
5652 if (len <= 0 || !buffer)
5656 parent = dev->dev.parent;
5661 driver = parent->driver;
5662 if (driver && driver->name)
5663 strlcpy(buffer, driver->name, len);
5667 static void __net_exit netdev_exit(struct net *net)
5669 kfree(net->dev_name_head);
5670 kfree(net->dev_index_head);
5673 static struct pernet_operations __net_initdata netdev_net_ops = {
5674 .init = netdev_init,
5675 .exit = netdev_exit,
5678 static void __net_exit default_device_exit(struct net *net)
5680 struct net_device *dev;
5682 * Push all migratable of the network devices back to the
5683 * initial network namespace
5687 for_each_netdev(net, dev) {
5689 char fb_name[IFNAMSIZ];
5691 /* Ignore unmoveable devices (i.e. loopback) */
5692 if (dev->features & NETIF_F_NETNS_LOCAL)
5695 /* Delete virtual devices */
5696 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5697 dev->rtnl_link_ops->dellink(dev, NULL);
5701 /* Push remaing network devices to init_net */
5702 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5703 err = dev_change_net_namespace(dev, &init_net, fb_name);
5705 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5706 __func__, dev->name, err);
5714 static struct pernet_operations __net_initdata default_device_ops = {
5715 .exit = default_device_exit,
5719 * Initialize the DEV module. At boot time this walks the device list and
5720 * unhooks any devices that fail to initialise (normally hardware not
5721 * present) and leaves us with a valid list of present and active devices.
5726 * This is called single threaded during boot, so no need
5727 * to take the rtnl semaphore.
5729 static int __init net_dev_init(void)
5731 int i, rc = -ENOMEM;
5733 BUG_ON(!dev_boot_phase);
5735 if (dev_proc_init())
5738 if (netdev_kobject_init())
5741 INIT_LIST_HEAD(&ptype_all);
5742 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5743 INIT_LIST_HEAD(&ptype_base[i]);
5745 if (register_pernet_subsys(&netdev_net_ops))
5749 * Initialise the packet receive queues.
5752 for_each_possible_cpu(i) {
5753 struct softnet_data *queue;
5755 queue = &per_cpu(softnet_data, i);
5756 skb_queue_head_init(&queue->input_pkt_queue);
5757 queue->completion_queue = NULL;
5758 INIT_LIST_HEAD(&queue->poll_list);
5760 queue->backlog.poll = process_backlog;
5761 queue->backlog.weight = weight_p;
5762 queue->backlog.gro_list = NULL;
5763 queue->backlog.gro_count = 0;
5768 /* The loopback device is special if any other network devices
5769 * is present in a network namespace the loopback device must
5770 * be present. Since we now dynamically allocate and free the
5771 * loopback device ensure this invariant is maintained by
5772 * keeping the loopback device as the first device on the
5773 * list of network devices. Ensuring the loopback devices
5774 * is the first device that appears and the last network device
5777 if (register_pernet_device(&loopback_net_ops))
5780 if (register_pernet_device(&default_device_ops))
5783 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5784 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5786 hotcpu_notifier(dev_cpu_callback, 0);
5794 subsys_initcall(net_dev_init);
5796 static int __init initialize_hashrnd(void)
5798 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5802 late_initcall_sync(initialize_hashrnd);