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);
1002 /* err >= 0 after dev_alloc_name() or stores the first errno */
1005 memcpy(dev->name, oldname, IFNAMSIZ);
1009 "%s: name change rollback failed: %d.\n",
1018 * dev_set_alias - change ifalias of a device
1020 * @alias: name up to IFALIASZ
1021 * @len: limit of bytes to copy from info
1023 * Set ifalias for a device,
1025 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1029 if (len >= IFALIASZ)
1034 kfree(dev->ifalias);
1035 dev->ifalias = NULL;
1040 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1044 strlcpy(dev->ifalias, alias, len+1);
1050 * netdev_features_change - device changes features
1051 * @dev: device to cause notification
1053 * Called to indicate a device has changed features.
1055 void netdev_features_change(struct net_device *dev)
1057 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1059 EXPORT_SYMBOL(netdev_features_change);
1062 * netdev_state_change - device changes state
1063 * @dev: device to cause notification
1065 * Called to indicate a device has changed state. This function calls
1066 * the notifier chains for netdev_chain and sends a NEWLINK message
1067 * to the routing socket.
1069 void netdev_state_change(struct net_device *dev)
1071 if (dev->flags & IFF_UP) {
1072 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1073 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1076 EXPORT_SYMBOL(netdev_state_change);
1078 void netdev_bonding_change(struct net_device *dev, unsigned long event)
1080 call_netdevice_notifiers(event, dev);
1082 EXPORT_SYMBOL(netdev_bonding_change);
1085 * dev_load - load a network module
1086 * @net: the applicable net namespace
1087 * @name: name of interface
1089 * If a network interface is not present and the process has suitable
1090 * privileges this function loads the module. If module loading is not
1091 * available in this kernel then it becomes a nop.
1094 void dev_load(struct net *net, const char *name)
1096 struct net_device *dev;
1099 dev = dev_get_by_name_rcu(net, name);
1102 if (!dev && capable(CAP_NET_ADMIN))
1103 request_module("%s", name);
1105 EXPORT_SYMBOL(dev_load);
1108 * dev_open - prepare an interface for use.
1109 * @dev: device to open
1111 * Takes a device from down to up state. The device's private open
1112 * function is invoked and then the multicast lists are loaded. Finally
1113 * the device is moved into the up state and a %NETDEV_UP message is
1114 * sent to the netdev notifier chain.
1116 * Calling this function on an active interface is a nop. On a failure
1117 * a negative errno code is returned.
1119 int dev_open(struct net_device *dev)
1121 const struct net_device_ops *ops = dev->netdev_ops;
1130 if (dev->flags & IFF_UP)
1134 * Is it even present?
1136 if (!netif_device_present(dev))
1139 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1140 ret = notifier_to_errno(ret);
1145 * Call device private open method
1147 set_bit(__LINK_STATE_START, &dev->state);
1149 if (ops->ndo_validate_addr)
1150 ret = ops->ndo_validate_addr(dev);
1152 if (!ret && ops->ndo_open)
1153 ret = ops->ndo_open(dev);
1156 * If it went open OK then:
1160 clear_bit(__LINK_STATE_START, &dev->state);
1165 dev->flags |= IFF_UP;
1170 net_dmaengine_get();
1173 * Initialize multicasting status
1175 dev_set_rx_mode(dev);
1178 * Wakeup transmit queue engine
1183 * ... and announce new interface.
1185 call_netdevice_notifiers(NETDEV_UP, dev);
1190 EXPORT_SYMBOL(dev_open);
1193 * dev_close - shutdown an interface.
1194 * @dev: device to shutdown
1196 * This function moves an active device into down state. A
1197 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1198 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1201 int dev_close(struct net_device *dev)
1203 const struct net_device_ops *ops = dev->netdev_ops;
1208 if (!(dev->flags & IFF_UP))
1212 * Tell people we are going down, so that they can
1213 * prepare to death, when device is still operating.
1215 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1217 clear_bit(__LINK_STATE_START, &dev->state);
1219 /* Synchronize to scheduled poll. We cannot touch poll list,
1220 * it can be even on different cpu. So just clear netif_running().
1222 * dev->stop() will invoke napi_disable() on all of it's
1223 * napi_struct instances on this device.
1225 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1227 dev_deactivate(dev);
1230 * Call the device specific close. This cannot fail.
1231 * Only if device is UP
1233 * We allow it to be called even after a DETACH hot-plug
1240 * Device is now down.
1243 dev->flags &= ~IFF_UP;
1246 * Tell people we are down
1248 call_netdevice_notifiers(NETDEV_DOWN, dev);
1253 net_dmaengine_put();
1257 EXPORT_SYMBOL(dev_close);
1261 * dev_disable_lro - disable Large Receive Offload on a device
1264 * Disable Large Receive Offload (LRO) on a net device. Must be
1265 * called under RTNL. This is needed if received packets may be
1266 * forwarded to another interface.
1268 void dev_disable_lro(struct net_device *dev)
1270 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1271 dev->ethtool_ops->set_flags) {
1272 u32 flags = dev->ethtool_ops->get_flags(dev);
1273 if (flags & ETH_FLAG_LRO) {
1274 flags &= ~ETH_FLAG_LRO;
1275 dev->ethtool_ops->set_flags(dev, flags);
1278 WARN_ON(dev->features & NETIF_F_LRO);
1280 EXPORT_SYMBOL(dev_disable_lro);
1283 static int dev_boot_phase = 1;
1286 * Device change register/unregister. These are not inline or static
1287 * as we export them to the world.
1291 * register_netdevice_notifier - register a network notifier block
1294 * Register a notifier to be called when network device events occur.
1295 * The notifier passed is linked into the kernel structures and must
1296 * not be reused until it has been unregistered. A negative errno code
1297 * is returned on a failure.
1299 * When registered all registration and up events are replayed
1300 * to the new notifier to allow device to have a race free
1301 * view of the network device list.
1304 int register_netdevice_notifier(struct notifier_block *nb)
1306 struct net_device *dev;
1307 struct net_device *last;
1312 err = raw_notifier_chain_register(&netdev_chain, nb);
1318 for_each_netdev(net, dev) {
1319 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1320 err = notifier_to_errno(err);
1324 if (!(dev->flags & IFF_UP))
1327 nb->notifier_call(nb, NETDEV_UP, dev);
1338 for_each_netdev(net, dev) {
1342 if (dev->flags & IFF_UP) {
1343 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1344 nb->notifier_call(nb, NETDEV_DOWN, dev);
1346 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1347 nb->notifier_call(nb, NETDEV_UNREGISTER_PERNET, dev);
1351 raw_notifier_chain_unregister(&netdev_chain, nb);
1354 EXPORT_SYMBOL(register_netdevice_notifier);
1357 * unregister_netdevice_notifier - unregister a network notifier block
1360 * Unregister a notifier previously registered by
1361 * register_netdevice_notifier(). The notifier is unlinked into the
1362 * kernel structures and may then be reused. A negative errno code
1363 * is returned on a failure.
1366 int unregister_netdevice_notifier(struct notifier_block *nb)
1371 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1375 EXPORT_SYMBOL(unregister_netdevice_notifier);
1378 * call_netdevice_notifiers - call all network notifier blocks
1379 * @val: value passed unmodified to notifier function
1380 * @dev: net_device pointer passed unmodified to notifier function
1382 * Call all network notifier blocks. Parameters and return value
1383 * are as for raw_notifier_call_chain().
1386 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1388 return raw_notifier_call_chain(&netdev_chain, val, dev);
1391 /* When > 0 there are consumers of rx skb time stamps */
1392 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1394 void net_enable_timestamp(void)
1396 atomic_inc(&netstamp_needed);
1398 EXPORT_SYMBOL(net_enable_timestamp);
1400 void net_disable_timestamp(void)
1402 atomic_dec(&netstamp_needed);
1404 EXPORT_SYMBOL(net_disable_timestamp);
1406 static inline void net_timestamp(struct sk_buff *skb)
1408 if (atomic_read(&netstamp_needed))
1409 __net_timestamp(skb);
1411 skb->tstamp.tv64 = 0;
1415 * Support routine. Sends outgoing frames to any network
1416 * taps currently in use.
1419 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1421 struct packet_type *ptype;
1423 #ifdef CONFIG_NET_CLS_ACT
1424 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1431 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1432 /* Never send packets back to the socket
1433 * they originated from - MvS (miquels@drinkel.ow.org)
1435 if ((ptype->dev == dev || !ptype->dev) &&
1436 (ptype->af_packet_priv == NULL ||
1437 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1438 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1442 /* skb->nh should be correctly
1443 set by sender, so that the second statement is
1444 just protection against buggy protocols.
1446 skb_reset_mac_header(skb2);
1448 if (skb_network_header(skb2) < skb2->data ||
1449 skb2->network_header > skb2->tail) {
1450 if (net_ratelimit())
1451 printk(KERN_CRIT "protocol %04x is "
1453 skb2->protocol, dev->name);
1454 skb_reset_network_header(skb2);
1457 skb2->transport_header = skb2->network_header;
1458 skb2->pkt_type = PACKET_OUTGOING;
1459 ptype->func(skb2, skb->dev, ptype, skb->dev);
1466 static inline void __netif_reschedule(struct Qdisc *q)
1468 struct softnet_data *sd;
1469 unsigned long flags;
1471 local_irq_save(flags);
1472 sd = &__get_cpu_var(softnet_data);
1473 q->next_sched = sd->output_queue;
1474 sd->output_queue = q;
1475 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1476 local_irq_restore(flags);
1479 void __netif_schedule(struct Qdisc *q)
1481 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1482 __netif_reschedule(q);
1484 EXPORT_SYMBOL(__netif_schedule);
1486 void dev_kfree_skb_irq(struct sk_buff *skb)
1488 if (atomic_dec_and_test(&skb->users)) {
1489 struct softnet_data *sd;
1490 unsigned long flags;
1492 local_irq_save(flags);
1493 sd = &__get_cpu_var(softnet_data);
1494 skb->next = sd->completion_queue;
1495 sd->completion_queue = skb;
1496 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1497 local_irq_restore(flags);
1500 EXPORT_SYMBOL(dev_kfree_skb_irq);
1502 void dev_kfree_skb_any(struct sk_buff *skb)
1504 if (in_irq() || irqs_disabled())
1505 dev_kfree_skb_irq(skb);
1509 EXPORT_SYMBOL(dev_kfree_skb_any);
1513 * netif_device_detach - mark device as removed
1514 * @dev: network device
1516 * Mark device as removed from system and therefore no longer available.
1518 void netif_device_detach(struct net_device *dev)
1520 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1521 netif_running(dev)) {
1522 netif_tx_stop_all_queues(dev);
1525 EXPORT_SYMBOL(netif_device_detach);
1528 * netif_device_attach - mark device as attached
1529 * @dev: network device
1531 * Mark device as attached from system and restart if needed.
1533 void netif_device_attach(struct net_device *dev)
1535 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1536 netif_running(dev)) {
1537 netif_tx_wake_all_queues(dev);
1538 __netdev_watchdog_up(dev);
1541 EXPORT_SYMBOL(netif_device_attach);
1543 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1545 return ((features & NETIF_F_GEN_CSUM) ||
1546 ((features & NETIF_F_IP_CSUM) &&
1547 protocol == htons(ETH_P_IP)) ||
1548 ((features & NETIF_F_IPV6_CSUM) &&
1549 protocol == htons(ETH_P_IPV6)) ||
1550 ((features & NETIF_F_FCOE_CRC) &&
1551 protocol == htons(ETH_P_FCOE)));
1554 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1556 if (can_checksum_protocol(dev->features, skb->protocol))
1559 if (skb->protocol == htons(ETH_P_8021Q)) {
1560 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1561 if (can_checksum_protocol(dev->features & dev->vlan_features,
1562 veh->h_vlan_encapsulated_proto))
1570 * Invalidate hardware checksum when packet is to be mangled, and
1571 * complete checksum manually on outgoing path.
1573 int skb_checksum_help(struct sk_buff *skb)
1576 int ret = 0, offset;
1578 if (skb->ip_summed == CHECKSUM_COMPLETE)
1579 goto out_set_summed;
1581 if (unlikely(skb_shinfo(skb)->gso_size)) {
1582 /* Let GSO fix up the checksum. */
1583 goto out_set_summed;
1586 offset = skb->csum_start - skb_headroom(skb);
1587 BUG_ON(offset >= skb_headlen(skb));
1588 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1590 offset += skb->csum_offset;
1591 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1593 if (skb_cloned(skb) &&
1594 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1595 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1600 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1602 skb->ip_summed = CHECKSUM_NONE;
1606 EXPORT_SYMBOL(skb_checksum_help);
1609 * skb_gso_segment - Perform segmentation on skb.
1610 * @skb: buffer to segment
1611 * @features: features for the output path (see dev->features)
1613 * This function segments the given skb and returns a list of segments.
1615 * It may return NULL if the skb requires no segmentation. This is
1616 * only possible when GSO is used for verifying header integrity.
1618 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1620 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1621 struct packet_type *ptype;
1622 __be16 type = skb->protocol;
1625 skb_reset_mac_header(skb);
1626 skb->mac_len = skb->network_header - skb->mac_header;
1627 __skb_pull(skb, skb->mac_len);
1629 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1630 struct net_device *dev = skb->dev;
1631 struct ethtool_drvinfo info = {};
1633 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1634 dev->ethtool_ops->get_drvinfo(dev, &info);
1636 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1638 info.driver, dev ? dev->features : 0L,
1639 skb->sk ? skb->sk->sk_route_caps : 0L,
1640 skb->len, skb->data_len, skb->ip_summed);
1642 if (skb_header_cloned(skb) &&
1643 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1644 return ERR_PTR(err);
1648 list_for_each_entry_rcu(ptype,
1649 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1650 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1651 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1652 err = ptype->gso_send_check(skb);
1653 segs = ERR_PTR(err);
1654 if (err || skb_gso_ok(skb, features))
1656 __skb_push(skb, (skb->data -
1657 skb_network_header(skb)));
1659 segs = ptype->gso_segment(skb, features);
1665 __skb_push(skb, skb->data - skb_mac_header(skb));
1669 EXPORT_SYMBOL(skb_gso_segment);
1671 /* Take action when hardware reception checksum errors are detected. */
1673 void netdev_rx_csum_fault(struct net_device *dev)
1675 if (net_ratelimit()) {
1676 printk(KERN_ERR "%s: hw csum failure.\n",
1677 dev ? dev->name : "<unknown>");
1681 EXPORT_SYMBOL(netdev_rx_csum_fault);
1684 /* Actually, we should eliminate this check as soon as we know, that:
1685 * 1. IOMMU is present and allows to map all the memory.
1686 * 2. No high memory really exists on this machine.
1689 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1691 #ifdef CONFIG_HIGHMEM
1694 if (dev->features & NETIF_F_HIGHDMA)
1697 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1698 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1706 void (*destructor)(struct sk_buff *skb);
1709 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1711 static void dev_gso_skb_destructor(struct sk_buff *skb)
1713 struct dev_gso_cb *cb;
1716 struct sk_buff *nskb = skb->next;
1718 skb->next = nskb->next;
1721 } while (skb->next);
1723 cb = DEV_GSO_CB(skb);
1725 cb->destructor(skb);
1729 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1730 * @skb: buffer to segment
1732 * This function segments the given skb and stores the list of segments
1735 static int dev_gso_segment(struct sk_buff *skb)
1737 struct net_device *dev = skb->dev;
1738 struct sk_buff *segs;
1739 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1742 segs = skb_gso_segment(skb, features);
1744 /* Verifying header integrity only. */
1749 return PTR_ERR(segs);
1752 DEV_GSO_CB(skb)->destructor = skb->destructor;
1753 skb->destructor = dev_gso_skb_destructor;
1758 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1759 struct netdev_queue *txq)
1761 const struct net_device_ops *ops = dev->netdev_ops;
1762 int rc = NETDEV_TX_OK;
1764 if (likely(!skb->next)) {
1765 if (!list_empty(&ptype_all))
1766 dev_queue_xmit_nit(skb, dev);
1768 if (netif_needs_gso(dev, skb)) {
1769 if (unlikely(dev_gso_segment(skb)))
1776 * If device doesnt need skb->dst, release it right now while
1777 * its hot in this cpu cache
1779 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1782 rc = ops->ndo_start_xmit(skb, dev);
1783 if (rc == NETDEV_TX_OK)
1784 txq_trans_update(txq);
1786 * TODO: if skb_orphan() was called by
1787 * dev->hard_start_xmit() (for example, the unmodified
1788 * igb driver does that; bnx2 doesn't), then
1789 * skb_tx_software_timestamp() will be unable to send
1790 * back the time stamp.
1792 * How can this be prevented? Always create another
1793 * reference to the socket before calling
1794 * dev->hard_start_xmit()? Prevent that skb_orphan()
1795 * does anything in dev->hard_start_xmit() by clearing
1796 * the skb destructor before the call and restoring it
1797 * afterwards, then doing the skb_orphan() ourselves?
1804 struct sk_buff *nskb = skb->next;
1806 skb->next = nskb->next;
1808 rc = ops->ndo_start_xmit(nskb, dev);
1809 if (unlikely(rc != NETDEV_TX_OK)) {
1810 if (rc & ~NETDEV_TX_MASK)
1811 goto out_kfree_gso_skb;
1812 nskb->next = skb->next;
1816 txq_trans_update(txq);
1817 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1818 return NETDEV_TX_BUSY;
1819 } while (skb->next);
1822 if (likely(skb->next == NULL))
1823 skb->destructor = DEV_GSO_CB(skb)->destructor;
1829 static u32 skb_tx_hashrnd;
1831 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1835 if (skb_rx_queue_recorded(skb)) {
1836 hash = skb_get_rx_queue(skb);
1837 while (unlikely(hash >= dev->real_num_tx_queues))
1838 hash -= dev->real_num_tx_queues;
1842 if (skb->sk && skb->sk->sk_hash)
1843 hash = skb->sk->sk_hash;
1845 hash = skb->protocol;
1847 hash = jhash_1word(hash, skb_tx_hashrnd);
1849 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1851 EXPORT_SYMBOL(skb_tx_hash);
1853 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1855 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1856 if (net_ratelimit()) {
1857 WARN(1, "%s selects TX queue %d, but "
1858 "real number of TX queues is %d\n",
1859 dev->name, queue_index,
1860 dev->real_num_tx_queues);
1867 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1868 struct sk_buff *skb)
1871 struct sock *sk = skb->sk;
1873 if (sk_tx_queue_recorded(sk)) {
1874 queue_index = sk_tx_queue_get(sk);
1876 const struct net_device_ops *ops = dev->netdev_ops;
1878 if (ops->ndo_select_queue) {
1879 queue_index = ops->ndo_select_queue(dev, skb);
1880 queue_index = dev_cap_txqueue(dev, queue_index);
1883 if (dev->real_num_tx_queues > 1)
1884 queue_index = skb_tx_hash(dev, skb);
1886 if (sk && sk->sk_dst_cache)
1887 sk_tx_queue_set(sk, queue_index);
1891 skb_set_queue_mapping(skb, queue_index);
1892 return netdev_get_tx_queue(dev, queue_index);
1895 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
1896 struct net_device *dev,
1897 struct netdev_queue *txq)
1899 spinlock_t *root_lock = qdisc_lock(q);
1902 spin_lock(root_lock);
1903 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1906 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
1907 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
1909 * This is a work-conserving queue; there are no old skbs
1910 * waiting to be sent out; and the qdisc is not running -
1911 * xmit the skb directly.
1913 __qdisc_update_bstats(q, skb->len);
1914 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
1917 clear_bit(__QDISC_STATE_RUNNING, &q->state);
1919 rc = NET_XMIT_SUCCESS;
1921 rc = qdisc_enqueue_root(skb, q);
1924 spin_unlock(root_lock);
1930 * dev_queue_xmit - transmit a buffer
1931 * @skb: buffer to transmit
1933 * Queue a buffer for transmission to a network device. The caller must
1934 * have set the device and priority and built the buffer before calling
1935 * this function. The function can be called from an interrupt.
1937 * A negative errno code is returned on a failure. A success does not
1938 * guarantee the frame will be transmitted as it may be dropped due
1939 * to congestion or traffic shaping.
1941 * -----------------------------------------------------------------------------------
1942 * I notice this method can also return errors from the queue disciplines,
1943 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1946 * Regardless of the return value, the skb is consumed, so it is currently
1947 * difficult to retry a send to this method. (You can bump the ref count
1948 * before sending to hold a reference for retry if you are careful.)
1950 * When calling this method, interrupts MUST be enabled. This is because
1951 * the BH enable code must have IRQs enabled so that it will not deadlock.
1954 int dev_queue_xmit(struct sk_buff *skb)
1956 struct net_device *dev = skb->dev;
1957 struct netdev_queue *txq;
1961 /* GSO will handle the following emulations directly. */
1962 if (netif_needs_gso(dev, skb))
1965 if (skb_has_frags(skb) &&
1966 !(dev->features & NETIF_F_FRAGLIST) &&
1967 __skb_linearize(skb))
1970 /* Fragmented skb is linearized if device does not support SG,
1971 * or if at least one of fragments is in highmem and device
1972 * does not support DMA from it.
1974 if (skb_shinfo(skb)->nr_frags &&
1975 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1976 __skb_linearize(skb))
1979 /* If packet is not checksummed and device does not support
1980 * checksumming for this protocol, complete checksumming here.
1982 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1983 skb_set_transport_header(skb, skb->csum_start -
1985 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1990 /* Disable soft irqs for various locks below. Also
1991 * stops preemption for RCU.
1995 txq = dev_pick_tx(dev, skb);
1996 q = rcu_dereference(txq->qdisc);
1998 #ifdef CONFIG_NET_CLS_ACT
1999 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2002 rc = __dev_xmit_skb(skb, q, dev, txq);
2006 /* The device has no queue. Common case for software devices:
2007 loopback, all the sorts of tunnels...
2009 Really, it is unlikely that netif_tx_lock protection is necessary
2010 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2012 However, it is possible, that they rely on protection
2015 Check this and shot the lock. It is not prone from deadlocks.
2016 Either shot noqueue qdisc, it is even simpler 8)
2018 if (dev->flags & IFF_UP) {
2019 int cpu = smp_processor_id(); /* ok because BHs are off */
2021 if (txq->xmit_lock_owner != cpu) {
2023 HARD_TX_LOCK(dev, txq, cpu);
2025 if (!netif_tx_queue_stopped(txq)) {
2026 rc = dev_hard_start_xmit(skb, dev, txq);
2027 if (dev_xmit_complete(rc)) {
2028 HARD_TX_UNLOCK(dev, txq);
2032 HARD_TX_UNLOCK(dev, txq);
2033 if (net_ratelimit())
2034 printk(KERN_CRIT "Virtual device %s asks to "
2035 "queue packet!\n", dev->name);
2037 /* Recursion is detected! It is possible,
2039 if (net_ratelimit())
2040 printk(KERN_CRIT "Dead loop on virtual device "
2041 "%s, fix it urgently!\n", dev->name);
2046 rcu_read_unlock_bh();
2052 rcu_read_unlock_bh();
2055 EXPORT_SYMBOL(dev_queue_xmit);
2058 /*=======================================================================
2060 =======================================================================*/
2062 int netdev_max_backlog __read_mostly = 1000;
2063 int netdev_budget __read_mostly = 300;
2064 int weight_p __read_mostly = 64; /* old backlog weight */
2066 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2070 * netif_rx - post buffer to the network code
2071 * @skb: buffer to post
2073 * This function receives a packet from a device driver and queues it for
2074 * the upper (protocol) levels to process. It always succeeds. The buffer
2075 * may be dropped during processing for congestion control or by the
2079 * NET_RX_SUCCESS (no congestion)
2080 * NET_RX_DROP (packet was dropped)
2084 int netif_rx(struct sk_buff *skb)
2086 struct softnet_data *queue;
2087 unsigned long flags;
2089 /* if netpoll wants it, pretend we never saw it */
2090 if (netpoll_rx(skb))
2093 if (!skb->tstamp.tv64)
2097 * The code is rearranged so that the path is the most
2098 * short when CPU is congested, but is still operating.
2100 local_irq_save(flags);
2101 queue = &__get_cpu_var(softnet_data);
2103 __get_cpu_var(netdev_rx_stat).total++;
2104 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2105 if (queue->input_pkt_queue.qlen) {
2107 __skb_queue_tail(&queue->input_pkt_queue, skb);
2108 local_irq_restore(flags);
2109 return NET_RX_SUCCESS;
2112 napi_schedule(&queue->backlog);
2116 __get_cpu_var(netdev_rx_stat).dropped++;
2117 local_irq_restore(flags);
2122 EXPORT_SYMBOL(netif_rx);
2124 int netif_rx_ni(struct sk_buff *skb)
2129 err = netif_rx(skb);
2130 if (local_softirq_pending())
2136 EXPORT_SYMBOL(netif_rx_ni);
2138 static void net_tx_action(struct softirq_action *h)
2140 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2142 if (sd->completion_queue) {
2143 struct sk_buff *clist;
2145 local_irq_disable();
2146 clist = sd->completion_queue;
2147 sd->completion_queue = NULL;
2151 struct sk_buff *skb = clist;
2152 clist = clist->next;
2154 WARN_ON(atomic_read(&skb->users));
2159 if (sd->output_queue) {
2162 local_irq_disable();
2163 head = sd->output_queue;
2164 sd->output_queue = NULL;
2168 struct Qdisc *q = head;
2169 spinlock_t *root_lock;
2171 head = head->next_sched;
2173 root_lock = qdisc_lock(q);
2174 if (spin_trylock(root_lock)) {
2175 smp_mb__before_clear_bit();
2176 clear_bit(__QDISC_STATE_SCHED,
2179 spin_unlock(root_lock);
2181 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2183 __netif_reschedule(q);
2185 smp_mb__before_clear_bit();
2186 clear_bit(__QDISC_STATE_SCHED,
2194 static inline int deliver_skb(struct sk_buff *skb,
2195 struct packet_type *pt_prev,
2196 struct net_device *orig_dev)
2198 atomic_inc(&skb->users);
2199 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2202 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2204 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2205 /* This hook is defined here for ATM LANE */
2206 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2207 unsigned char *addr) __read_mostly;
2208 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2212 * If bridge module is loaded call bridging hook.
2213 * returns NULL if packet was consumed.
2215 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2216 struct sk_buff *skb) __read_mostly;
2217 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2219 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2220 struct packet_type **pt_prev, int *ret,
2221 struct net_device *orig_dev)
2223 struct net_bridge_port *port;
2225 if (skb->pkt_type == PACKET_LOOPBACK ||
2226 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2230 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2234 return br_handle_frame_hook(port, skb);
2237 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2240 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2241 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2242 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2244 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2245 struct packet_type **pt_prev,
2247 struct net_device *orig_dev)
2249 if (skb->dev->macvlan_port == NULL)
2253 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2256 return macvlan_handle_frame_hook(skb);
2259 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2262 #ifdef CONFIG_NET_CLS_ACT
2263 /* TODO: Maybe we should just force sch_ingress to be compiled in
2264 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2265 * a compare and 2 stores extra right now if we dont have it on
2266 * but have CONFIG_NET_CLS_ACT
2267 * NOTE: This doesnt stop any functionality; if you dont have
2268 * the ingress scheduler, you just cant add policies on ingress.
2271 static int ing_filter(struct sk_buff *skb)
2273 struct net_device *dev = skb->dev;
2274 u32 ttl = G_TC_RTTL(skb->tc_verd);
2275 struct netdev_queue *rxq;
2276 int result = TC_ACT_OK;
2279 if (MAX_RED_LOOP < ttl++) {
2281 "Redir loop detected Dropping packet (%d->%d)\n",
2282 skb->iif, dev->ifindex);
2286 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2287 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2289 rxq = &dev->rx_queue;
2292 if (q != &noop_qdisc) {
2293 spin_lock(qdisc_lock(q));
2294 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2295 result = qdisc_enqueue_root(skb, q);
2296 spin_unlock(qdisc_lock(q));
2302 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2303 struct packet_type **pt_prev,
2304 int *ret, struct net_device *orig_dev)
2306 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2310 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2313 /* Huh? Why does turning on AF_PACKET affect this? */
2314 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2317 switch (ing_filter(skb)) {
2331 * netif_nit_deliver - deliver received packets to network taps
2334 * This function is used to deliver incoming packets to network
2335 * taps. It should be used when the normal netif_receive_skb path
2336 * is bypassed, for example because of VLAN acceleration.
2338 void netif_nit_deliver(struct sk_buff *skb)
2340 struct packet_type *ptype;
2342 if (list_empty(&ptype_all))
2345 skb_reset_network_header(skb);
2346 skb_reset_transport_header(skb);
2347 skb->mac_len = skb->network_header - skb->mac_header;
2350 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2351 if (!ptype->dev || ptype->dev == skb->dev)
2352 deliver_skb(skb, ptype, skb->dev);
2358 * netif_receive_skb - process receive buffer from network
2359 * @skb: buffer to process
2361 * netif_receive_skb() is the main receive data processing function.
2362 * It always succeeds. The buffer may be dropped during processing
2363 * for congestion control or by the protocol layers.
2365 * This function may only be called from softirq context and interrupts
2366 * should be enabled.
2368 * Return values (usually ignored):
2369 * NET_RX_SUCCESS: no congestion
2370 * NET_RX_DROP: packet was dropped
2372 int netif_receive_skb(struct sk_buff *skb)
2374 struct packet_type *ptype, *pt_prev;
2375 struct net_device *orig_dev;
2376 struct net_device *null_or_orig;
2377 int ret = NET_RX_DROP;
2380 if (!skb->tstamp.tv64)
2383 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2384 return NET_RX_SUCCESS;
2386 /* if we've gotten here through NAPI, check netpoll */
2387 if (netpoll_receive_skb(skb))
2391 skb->iif = skb->dev->ifindex;
2393 null_or_orig = NULL;
2394 orig_dev = skb->dev;
2395 if (orig_dev->master) {
2396 if (skb_bond_should_drop(skb))
2397 null_or_orig = orig_dev; /* deliver only exact match */
2399 skb->dev = orig_dev->master;
2402 __get_cpu_var(netdev_rx_stat).total++;
2404 skb_reset_network_header(skb);
2405 skb_reset_transport_header(skb);
2406 skb->mac_len = skb->network_header - skb->mac_header;
2412 #ifdef CONFIG_NET_CLS_ACT
2413 if (skb->tc_verd & TC_NCLS) {
2414 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2419 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2420 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2421 ptype->dev == orig_dev) {
2423 ret = deliver_skb(skb, pt_prev, orig_dev);
2428 #ifdef CONFIG_NET_CLS_ACT
2429 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2435 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2438 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2442 type = skb->protocol;
2443 list_for_each_entry_rcu(ptype,
2444 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2445 if (ptype->type == type &&
2446 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2447 ptype->dev == orig_dev)) {
2449 ret = deliver_skb(skb, pt_prev, orig_dev);
2455 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2458 /* Jamal, now you will not able to escape explaining
2459 * me how you were going to use this. :-)
2468 EXPORT_SYMBOL(netif_receive_skb);
2470 /* Network device is going away, flush any packets still pending */
2471 static void flush_backlog(void *arg)
2473 struct net_device *dev = arg;
2474 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2475 struct sk_buff *skb, *tmp;
2477 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2478 if (skb->dev == dev) {
2479 __skb_unlink(skb, &queue->input_pkt_queue);
2484 static int napi_gro_complete(struct sk_buff *skb)
2486 struct packet_type *ptype;
2487 __be16 type = skb->protocol;
2488 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2491 if (NAPI_GRO_CB(skb)->count == 1) {
2492 skb_shinfo(skb)->gso_size = 0;
2497 list_for_each_entry_rcu(ptype, head, list) {
2498 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2501 err = ptype->gro_complete(skb);
2507 WARN_ON(&ptype->list == head);
2509 return NET_RX_SUCCESS;
2513 return netif_receive_skb(skb);
2516 void napi_gro_flush(struct napi_struct *napi)
2518 struct sk_buff *skb, *next;
2520 for (skb = napi->gro_list; skb; skb = next) {
2523 napi_gro_complete(skb);
2526 napi->gro_count = 0;
2527 napi->gro_list = NULL;
2529 EXPORT_SYMBOL(napi_gro_flush);
2531 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2533 struct sk_buff **pp = NULL;
2534 struct packet_type *ptype;
2535 __be16 type = skb->protocol;
2536 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2539 enum gro_result ret;
2541 if (!(skb->dev->features & NETIF_F_GRO))
2544 if (skb_is_gso(skb) || skb_has_frags(skb))
2548 list_for_each_entry_rcu(ptype, head, list) {
2549 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2552 skb_set_network_header(skb, skb_gro_offset(skb));
2553 mac_len = skb->network_header - skb->mac_header;
2554 skb->mac_len = mac_len;
2555 NAPI_GRO_CB(skb)->same_flow = 0;
2556 NAPI_GRO_CB(skb)->flush = 0;
2557 NAPI_GRO_CB(skb)->free = 0;
2559 pp = ptype->gro_receive(&napi->gro_list, skb);
2564 if (&ptype->list == head)
2567 same_flow = NAPI_GRO_CB(skb)->same_flow;
2568 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2571 struct sk_buff *nskb = *pp;
2575 napi_gro_complete(nskb);
2582 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2586 NAPI_GRO_CB(skb)->count = 1;
2587 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2588 skb->next = napi->gro_list;
2589 napi->gro_list = skb;
2593 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2594 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2596 BUG_ON(skb->end - skb->tail < grow);
2598 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2601 skb->data_len -= grow;
2603 skb_shinfo(skb)->frags[0].page_offset += grow;
2604 skb_shinfo(skb)->frags[0].size -= grow;
2606 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2607 put_page(skb_shinfo(skb)->frags[0].page);
2608 memmove(skb_shinfo(skb)->frags,
2609 skb_shinfo(skb)->frags + 1,
2610 --skb_shinfo(skb)->nr_frags);
2621 EXPORT_SYMBOL(dev_gro_receive);
2624 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2628 if (netpoll_rx_on(skb))
2631 for (p = napi->gro_list; p; p = p->next) {
2632 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2633 && !compare_ether_header(skb_mac_header(p),
2634 skb_gro_mac_header(skb));
2635 NAPI_GRO_CB(p)->flush = 0;
2638 return dev_gro_receive(napi, skb);
2641 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
2645 if (netif_receive_skb(skb))
2650 case GRO_MERGED_FREE:
2661 EXPORT_SYMBOL(napi_skb_finish);
2663 void skb_gro_reset_offset(struct sk_buff *skb)
2665 NAPI_GRO_CB(skb)->data_offset = 0;
2666 NAPI_GRO_CB(skb)->frag0 = NULL;
2667 NAPI_GRO_CB(skb)->frag0_len = 0;
2669 if (skb->mac_header == skb->tail &&
2670 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2671 NAPI_GRO_CB(skb)->frag0 =
2672 page_address(skb_shinfo(skb)->frags[0].page) +
2673 skb_shinfo(skb)->frags[0].page_offset;
2674 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2677 EXPORT_SYMBOL(skb_gro_reset_offset);
2679 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2681 skb_gro_reset_offset(skb);
2683 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2685 EXPORT_SYMBOL(napi_gro_receive);
2687 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2689 __skb_pull(skb, skb_headlen(skb));
2690 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2694 EXPORT_SYMBOL(napi_reuse_skb);
2696 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2698 struct sk_buff *skb = napi->skb;
2701 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
2707 EXPORT_SYMBOL(napi_get_frags);
2709 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
2715 skb->protocol = eth_type_trans(skb, napi->dev);
2717 if (ret == GRO_HELD)
2718 skb_gro_pull(skb, -ETH_HLEN);
2719 else if (netif_receive_skb(skb))
2724 case GRO_MERGED_FREE:
2725 napi_reuse_skb(napi, skb);
2734 EXPORT_SYMBOL(napi_frags_finish);
2736 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2738 struct sk_buff *skb = napi->skb;
2745 skb_reset_mac_header(skb);
2746 skb_gro_reset_offset(skb);
2748 off = skb_gro_offset(skb);
2749 hlen = off + sizeof(*eth);
2750 eth = skb_gro_header_fast(skb, off);
2751 if (skb_gro_header_hard(skb, hlen)) {
2752 eth = skb_gro_header_slow(skb, hlen, off);
2753 if (unlikely(!eth)) {
2754 napi_reuse_skb(napi, skb);
2760 skb_gro_pull(skb, sizeof(*eth));
2763 * This works because the only protocols we care about don't require
2764 * special handling. We'll fix it up properly at the end.
2766 skb->protocol = eth->h_proto;
2771 EXPORT_SYMBOL(napi_frags_skb);
2773 gro_result_t napi_gro_frags(struct napi_struct *napi)
2775 struct sk_buff *skb = napi_frags_skb(napi);
2780 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2782 EXPORT_SYMBOL(napi_gro_frags);
2784 static int process_backlog(struct napi_struct *napi, int quota)
2787 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2788 unsigned long start_time = jiffies;
2790 napi->weight = weight_p;
2792 struct sk_buff *skb;
2794 local_irq_disable();
2795 skb = __skb_dequeue(&queue->input_pkt_queue);
2797 __napi_complete(napi);
2803 netif_receive_skb(skb);
2804 } while (++work < quota && jiffies == start_time);
2810 * __napi_schedule - schedule for receive
2811 * @n: entry to schedule
2813 * The entry's receive function will be scheduled to run
2815 void __napi_schedule(struct napi_struct *n)
2817 unsigned long flags;
2819 local_irq_save(flags);
2820 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2821 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2822 local_irq_restore(flags);
2824 EXPORT_SYMBOL(__napi_schedule);
2826 void __napi_complete(struct napi_struct *n)
2828 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2829 BUG_ON(n->gro_list);
2831 list_del(&n->poll_list);
2832 smp_mb__before_clear_bit();
2833 clear_bit(NAPI_STATE_SCHED, &n->state);
2835 EXPORT_SYMBOL(__napi_complete);
2837 void napi_complete(struct napi_struct *n)
2839 unsigned long flags;
2842 * don't let napi dequeue from the cpu poll list
2843 * just in case its running on a different cpu
2845 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2849 local_irq_save(flags);
2851 local_irq_restore(flags);
2853 EXPORT_SYMBOL(napi_complete);
2855 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2856 int (*poll)(struct napi_struct *, int), int weight)
2858 INIT_LIST_HEAD(&napi->poll_list);
2859 napi->gro_count = 0;
2860 napi->gro_list = NULL;
2863 napi->weight = weight;
2864 list_add(&napi->dev_list, &dev->napi_list);
2866 #ifdef CONFIG_NETPOLL
2867 spin_lock_init(&napi->poll_lock);
2868 napi->poll_owner = -1;
2870 set_bit(NAPI_STATE_SCHED, &napi->state);
2872 EXPORT_SYMBOL(netif_napi_add);
2874 void netif_napi_del(struct napi_struct *napi)
2876 struct sk_buff *skb, *next;
2878 list_del_init(&napi->dev_list);
2879 napi_free_frags(napi);
2881 for (skb = napi->gro_list; skb; skb = next) {
2887 napi->gro_list = NULL;
2888 napi->gro_count = 0;
2890 EXPORT_SYMBOL(netif_napi_del);
2893 static void net_rx_action(struct softirq_action *h)
2895 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2896 unsigned long time_limit = jiffies + 2;
2897 int budget = netdev_budget;
2900 local_irq_disable();
2902 while (!list_empty(list)) {
2903 struct napi_struct *n;
2906 /* If softirq window is exhuasted then punt.
2907 * Allow this to run for 2 jiffies since which will allow
2908 * an average latency of 1.5/HZ.
2910 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2915 /* Even though interrupts have been re-enabled, this
2916 * access is safe because interrupts can only add new
2917 * entries to the tail of this list, and only ->poll()
2918 * calls can remove this head entry from the list.
2920 n = list_entry(list->next, struct napi_struct, poll_list);
2922 have = netpoll_poll_lock(n);
2926 /* This NAPI_STATE_SCHED test is for avoiding a race
2927 * with netpoll's poll_napi(). Only the entity which
2928 * obtains the lock and sees NAPI_STATE_SCHED set will
2929 * actually make the ->poll() call. Therefore we avoid
2930 * accidently calling ->poll() when NAPI is not scheduled.
2933 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
2934 work = n->poll(n, weight);
2938 WARN_ON_ONCE(work > weight);
2942 local_irq_disable();
2944 /* Drivers must not modify the NAPI state if they
2945 * consume the entire weight. In such cases this code
2946 * still "owns" the NAPI instance and therefore can
2947 * move the instance around on the list at-will.
2949 if (unlikely(work == weight)) {
2950 if (unlikely(napi_disable_pending(n))) {
2953 local_irq_disable();
2955 list_move_tail(&n->poll_list, list);
2958 netpoll_poll_unlock(have);
2963 #ifdef CONFIG_NET_DMA
2965 * There may not be any more sk_buffs coming right now, so push
2966 * any pending DMA copies to hardware
2968 dma_issue_pending_all();
2974 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2975 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2979 static gifconf_func_t *gifconf_list[NPROTO];
2982 * register_gifconf - register a SIOCGIF handler
2983 * @family: Address family
2984 * @gifconf: Function handler
2986 * Register protocol dependent address dumping routines. The handler
2987 * that is passed must not be freed or reused until it has been replaced
2988 * by another handler.
2990 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
2992 if (family >= NPROTO)
2994 gifconf_list[family] = gifconf;
2997 EXPORT_SYMBOL(register_gifconf);
3001 * Map an interface index to its name (SIOCGIFNAME)
3005 * We need this ioctl for efficient implementation of the
3006 * if_indextoname() function required by the IPv6 API. Without
3007 * it, we would have to search all the interfaces to find a
3011 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3013 struct net_device *dev;
3017 * Fetch the caller's info block.
3020 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3024 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3030 strcpy(ifr.ifr_name, dev->name);
3033 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3039 * Perform a SIOCGIFCONF call. This structure will change
3040 * size eventually, and there is nothing I can do about it.
3041 * Thus we will need a 'compatibility mode'.
3044 static int dev_ifconf(struct net *net, char __user *arg)
3047 struct net_device *dev;
3054 * Fetch the caller's info block.
3057 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3064 * Loop over the interfaces, and write an info block for each.
3068 for_each_netdev(net, dev) {
3069 for (i = 0; i < NPROTO; i++) {
3070 if (gifconf_list[i]) {
3073 done = gifconf_list[i](dev, NULL, 0);
3075 done = gifconf_list[i](dev, pos + total,
3085 * All done. Write the updated control block back to the caller.
3087 ifc.ifc_len = total;
3090 * Both BSD and Solaris return 0 here, so we do too.
3092 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3095 #ifdef CONFIG_PROC_FS
3097 * This is invoked by the /proc filesystem handler to display a device
3100 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3103 struct net *net = seq_file_net(seq);
3105 struct net_device *dev;
3109 return SEQ_START_TOKEN;
3112 for_each_netdev_rcu(net, dev)
3119 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3121 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3122 first_net_device(seq_file_net(seq)) :
3123 next_net_device((struct net_device *)v);
3126 return rcu_dereference(dev);
3129 void dev_seq_stop(struct seq_file *seq, void *v)
3135 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3137 const struct net_device_stats *stats = dev_get_stats(dev);
3139 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3140 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3141 dev->name, stats->rx_bytes, stats->rx_packets,
3143 stats->rx_dropped + stats->rx_missed_errors,
3144 stats->rx_fifo_errors,
3145 stats->rx_length_errors + stats->rx_over_errors +
3146 stats->rx_crc_errors + stats->rx_frame_errors,
3147 stats->rx_compressed, stats->multicast,
3148 stats->tx_bytes, stats->tx_packets,
3149 stats->tx_errors, stats->tx_dropped,
3150 stats->tx_fifo_errors, stats->collisions,
3151 stats->tx_carrier_errors +
3152 stats->tx_aborted_errors +
3153 stats->tx_window_errors +
3154 stats->tx_heartbeat_errors,
3155 stats->tx_compressed);
3159 * Called from the PROCfs module. This now uses the new arbitrary sized
3160 * /proc/net interface to create /proc/net/dev
3162 static int dev_seq_show(struct seq_file *seq, void *v)
3164 if (v == SEQ_START_TOKEN)
3165 seq_puts(seq, "Inter-| Receive "
3167 " face |bytes packets errs drop fifo frame "
3168 "compressed multicast|bytes packets errs "
3169 "drop fifo colls carrier compressed\n");
3171 dev_seq_printf_stats(seq, v);
3175 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3177 struct netif_rx_stats *rc = NULL;
3179 while (*pos < nr_cpu_ids)
3180 if (cpu_online(*pos)) {
3181 rc = &per_cpu(netdev_rx_stat, *pos);
3188 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3190 return softnet_get_online(pos);
3193 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3196 return softnet_get_online(pos);
3199 static void softnet_seq_stop(struct seq_file *seq, void *v)
3203 static int softnet_seq_show(struct seq_file *seq, void *v)
3205 struct netif_rx_stats *s = v;
3207 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3208 s->total, s->dropped, s->time_squeeze, 0,
3209 0, 0, 0, 0, /* was fastroute */
3214 static const struct seq_operations dev_seq_ops = {
3215 .start = dev_seq_start,
3216 .next = dev_seq_next,
3217 .stop = dev_seq_stop,
3218 .show = dev_seq_show,
3221 static int dev_seq_open(struct inode *inode, struct file *file)
3223 return seq_open_net(inode, file, &dev_seq_ops,
3224 sizeof(struct seq_net_private));
3227 static const struct file_operations dev_seq_fops = {
3228 .owner = THIS_MODULE,
3229 .open = dev_seq_open,
3231 .llseek = seq_lseek,
3232 .release = seq_release_net,
3235 static const struct seq_operations softnet_seq_ops = {
3236 .start = softnet_seq_start,
3237 .next = softnet_seq_next,
3238 .stop = softnet_seq_stop,
3239 .show = softnet_seq_show,
3242 static int softnet_seq_open(struct inode *inode, struct file *file)
3244 return seq_open(file, &softnet_seq_ops);
3247 static const struct file_operations softnet_seq_fops = {
3248 .owner = THIS_MODULE,
3249 .open = softnet_seq_open,
3251 .llseek = seq_lseek,
3252 .release = seq_release,
3255 static void *ptype_get_idx(loff_t pos)
3257 struct packet_type *pt = NULL;
3261 list_for_each_entry_rcu(pt, &ptype_all, list) {
3267 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3268 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3277 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3281 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3284 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3286 struct packet_type *pt;
3287 struct list_head *nxt;
3291 if (v == SEQ_START_TOKEN)
3292 return ptype_get_idx(0);
3295 nxt = pt->list.next;
3296 if (pt->type == htons(ETH_P_ALL)) {
3297 if (nxt != &ptype_all)
3300 nxt = ptype_base[0].next;
3302 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3304 while (nxt == &ptype_base[hash]) {
3305 if (++hash >= PTYPE_HASH_SIZE)
3307 nxt = ptype_base[hash].next;
3310 return list_entry(nxt, struct packet_type, list);
3313 static void ptype_seq_stop(struct seq_file *seq, void *v)
3319 static int ptype_seq_show(struct seq_file *seq, void *v)
3321 struct packet_type *pt = v;
3323 if (v == SEQ_START_TOKEN)
3324 seq_puts(seq, "Type Device Function\n");
3325 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3326 if (pt->type == htons(ETH_P_ALL))
3327 seq_puts(seq, "ALL ");
3329 seq_printf(seq, "%04x", ntohs(pt->type));
3331 seq_printf(seq, " %-8s %pF\n",
3332 pt->dev ? pt->dev->name : "", pt->func);
3338 static const struct seq_operations ptype_seq_ops = {
3339 .start = ptype_seq_start,
3340 .next = ptype_seq_next,
3341 .stop = ptype_seq_stop,
3342 .show = ptype_seq_show,
3345 static int ptype_seq_open(struct inode *inode, struct file *file)
3347 return seq_open_net(inode, file, &ptype_seq_ops,
3348 sizeof(struct seq_net_private));
3351 static const struct file_operations ptype_seq_fops = {
3352 .owner = THIS_MODULE,
3353 .open = ptype_seq_open,
3355 .llseek = seq_lseek,
3356 .release = seq_release_net,
3360 static int __net_init dev_proc_net_init(struct net *net)
3364 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3366 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3368 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3371 if (wext_proc_init(net))
3377 proc_net_remove(net, "ptype");
3379 proc_net_remove(net, "softnet_stat");
3381 proc_net_remove(net, "dev");
3385 static void __net_exit dev_proc_net_exit(struct net *net)
3387 wext_proc_exit(net);
3389 proc_net_remove(net, "ptype");
3390 proc_net_remove(net, "softnet_stat");
3391 proc_net_remove(net, "dev");
3394 static struct pernet_operations __net_initdata dev_proc_ops = {
3395 .init = dev_proc_net_init,
3396 .exit = dev_proc_net_exit,
3399 static int __init dev_proc_init(void)
3401 return register_pernet_subsys(&dev_proc_ops);
3404 #define dev_proc_init() 0
3405 #endif /* CONFIG_PROC_FS */
3409 * netdev_set_master - set up master/slave pair
3410 * @slave: slave device
3411 * @master: new master device
3413 * Changes the master device of the slave. Pass %NULL to break the
3414 * bonding. The caller must hold the RTNL semaphore. On a failure
3415 * a negative errno code is returned. On success the reference counts
3416 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3417 * function returns zero.
3419 int netdev_set_master(struct net_device *slave, struct net_device *master)
3421 struct net_device *old = slave->master;
3431 slave->master = master;
3439 slave->flags |= IFF_SLAVE;
3441 slave->flags &= ~IFF_SLAVE;
3443 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3446 EXPORT_SYMBOL(netdev_set_master);
3448 static void dev_change_rx_flags(struct net_device *dev, int flags)
3450 const struct net_device_ops *ops = dev->netdev_ops;
3452 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3453 ops->ndo_change_rx_flags(dev, flags);
3456 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3458 unsigned short old_flags = dev->flags;
3464 dev->flags |= IFF_PROMISC;
3465 dev->promiscuity += inc;
3466 if (dev->promiscuity == 0) {
3469 * If inc causes overflow, untouch promisc and return error.
3472 dev->flags &= ~IFF_PROMISC;
3474 dev->promiscuity -= inc;
3475 printk(KERN_WARNING "%s: promiscuity touches roof, "
3476 "set promiscuity failed, promiscuity feature "
3477 "of device might be broken.\n", dev->name);
3481 if (dev->flags != old_flags) {
3482 printk(KERN_INFO "device %s %s promiscuous mode\n",
3483 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3485 if (audit_enabled) {
3486 current_uid_gid(&uid, &gid);
3487 audit_log(current->audit_context, GFP_ATOMIC,
3488 AUDIT_ANOM_PROMISCUOUS,
3489 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3490 dev->name, (dev->flags & IFF_PROMISC),
3491 (old_flags & IFF_PROMISC),
3492 audit_get_loginuid(current),
3494 audit_get_sessionid(current));
3497 dev_change_rx_flags(dev, IFF_PROMISC);
3503 * dev_set_promiscuity - update promiscuity count on a device
3507 * Add or remove promiscuity from a device. While the count in the device
3508 * remains above zero the interface remains promiscuous. Once it hits zero
3509 * the device reverts back to normal filtering operation. A negative inc
3510 * value is used to drop promiscuity on the device.
3511 * Return 0 if successful or a negative errno code on error.
3513 int dev_set_promiscuity(struct net_device *dev, int inc)
3515 unsigned short old_flags = dev->flags;
3518 err = __dev_set_promiscuity(dev, inc);
3521 if (dev->flags != old_flags)
3522 dev_set_rx_mode(dev);
3525 EXPORT_SYMBOL(dev_set_promiscuity);
3528 * dev_set_allmulti - update allmulti count on a device
3532 * Add or remove reception of all multicast frames to a device. While the
3533 * count in the device remains above zero the interface remains listening
3534 * to all interfaces. Once it hits zero the device reverts back to normal
3535 * filtering operation. A negative @inc value is used to drop the counter
3536 * when releasing a resource needing all multicasts.
3537 * Return 0 if successful or a negative errno code on error.
3540 int dev_set_allmulti(struct net_device *dev, int inc)
3542 unsigned short old_flags = dev->flags;
3546 dev->flags |= IFF_ALLMULTI;
3547 dev->allmulti += inc;
3548 if (dev->allmulti == 0) {
3551 * If inc causes overflow, untouch allmulti and return error.
3554 dev->flags &= ~IFF_ALLMULTI;
3556 dev->allmulti -= inc;
3557 printk(KERN_WARNING "%s: allmulti touches roof, "
3558 "set allmulti failed, allmulti feature of "
3559 "device might be broken.\n", dev->name);
3563 if (dev->flags ^ old_flags) {
3564 dev_change_rx_flags(dev, IFF_ALLMULTI);
3565 dev_set_rx_mode(dev);
3569 EXPORT_SYMBOL(dev_set_allmulti);
3572 * Upload unicast and multicast address lists to device and
3573 * configure RX filtering. When the device doesn't support unicast
3574 * filtering it is put in promiscuous mode while unicast addresses
3577 void __dev_set_rx_mode(struct net_device *dev)
3579 const struct net_device_ops *ops = dev->netdev_ops;
3581 /* dev_open will call this function so the list will stay sane. */
3582 if (!(dev->flags&IFF_UP))
3585 if (!netif_device_present(dev))
3588 if (ops->ndo_set_rx_mode)
3589 ops->ndo_set_rx_mode(dev);
3591 /* Unicast addresses changes may only happen under the rtnl,
3592 * therefore calling __dev_set_promiscuity here is safe.
3594 if (dev->uc.count > 0 && !dev->uc_promisc) {
3595 __dev_set_promiscuity(dev, 1);
3596 dev->uc_promisc = 1;
3597 } else if (dev->uc.count == 0 && dev->uc_promisc) {
3598 __dev_set_promiscuity(dev, -1);
3599 dev->uc_promisc = 0;
3602 if (ops->ndo_set_multicast_list)
3603 ops->ndo_set_multicast_list(dev);
3607 void dev_set_rx_mode(struct net_device *dev)
3609 netif_addr_lock_bh(dev);
3610 __dev_set_rx_mode(dev);
3611 netif_addr_unlock_bh(dev);
3614 /* hw addresses list handling functions */
3616 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3617 int addr_len, unsigned char addr_type)
3619 struct netdev_hw_addr *ha;
3622 if (addr_len > MAX_ADDR_LEN)
3625 list_for_each_entry(ha, &list->list, list) {
3626 if (!memcmp(ha->addr, addr, addr_len) &&
3627 ha->type == addr_type) {
3634 alloc_size = sizeof(*ha);
3635 if (alloc_size < L1_CACHE_BYTES)
3636 alloc_size = L1_CACHE_BYTES;
3637 ha = kmalloc(alloc_size, GFP_ATOMIC);
3640 memcpy(ha->addr, addr, addr_len);
3641 ha->type = addr_type;
3644 list_add_tail_rcu(&ha->list, &list->list);
3649 static void ha_rcu_free(struct rcu_head *head)
3651 struct netdev_hw_addr *ha;
3653 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3657 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3658 int addr_len, unsigned char addr_type)
3660 struct netdev_hw_addr *ha;
3662 list_for_each_entry(ha, &list->list, list) {
3663 if (!memcmp(ha->addr, addr, addr_len) &&
3664 (ha->type == addr_type || !addr_type)) {
3667 list_del_rcu(&ha->list);
3668 call_rcu(&ha->rcu_head, ha_rcu_free);
3676 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3677 struct netdev_hw_addr_list *from_list,
3679 unsigned char addr_type)
3682 struct netdev_hw_addr *ha, *ha2;
3685 list_for_each_entry(ha, &from_list->list, list) {
3686 type = addr_type ? addr_type : ha->type;
3687 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3694 list_for_each_entry(ha2, &from_list->list, list) {
3697 type = addr_type ? addr_type : ha2->type;
3698 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3703 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3704 struct netdev_hw_addr_list *from_list,
3706 unsigned char addr_type)
3708 struct netdev_hw_addr *ha;
3711 list_for_each_entry(ha, &from_list->list, list) {
3712 type = addr_type ? addr_type : ha->type;
3713 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3717 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3718 struct netdev_hw_addr_list *from_list,
3722 struct netdev_hw_addr *ha, *tmp;
3724 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3726 err = __hw_addr_add(to_list, ha->addr,
3727 addr_len, ha->type);
3732 } else if (ha->refcount == 1) {
3733 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3734 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3740 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3741 struct netdev_hw_addr_list *from_list,
3744 struct netdev_hw_addr *ha, *tmp;
3746 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3748 __hw_addr_del(to_list, ha->addr,
3749 addr_len, ha->type);
3751 __hw_addr_del(from_list, ha->addr,
3752 addr_len, ha->type);
3757 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3759 struct netdev_hw_addr *ha, *tmp;
3761 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3762 list_del_rcu(&ha->list);
3763 call_rcu(&ha->rcu_head, ha_rcu_free);
3768 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3770 INIT_LIST_HEAD(&list->list);
3774 /* Device addresses handling functions */
3776 static void dev_addr_flush(struct net_device *dev)
3778 /* rtnl_mutex must be held here */
3780 __hw_addr_flush(&dev->dev_addrs);
3781 dev->dev_addr = NULL;
3784 static int dev_addr_init(struct net_device *dev)
3786 unsigned char addr[MAX_ADDR_LEN];
3787 struct netdev_hw_addr *ha;
3790 /* rtnl_mutex must be held here */
3792 __hw_addr_init(&dev->dev_addrs);
3793 memset(addr, 0, sizeof(addr));
3794 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3795 NETDEV_HW_ADDR_T_LAN);
3798 * Get the first (previously created) address from the list
3799 * and set dev_addr pointer to this location.
3801 ha = list_first_entry(&dev->dev_addrs.list,
3802 struct netdev_hw_addr, list);
3803 dev->dev_addr = ha->addr;
3809 * dev_addr_add - Add a device address
3811 * @addr: address to add
3812 * @addr_type: address type
3814 * Add a device address to the device or increase the reference count if
3815 * it already exists.
3817 * The caller must hold the rtnl_mutex.
3819 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3820 unsigned char addr_type)
3826 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3828 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3831 EXPORT_SYMBOL(dev_addr_add);
3834 * dev_addr_del - Release a device address.
3836 * @addr: address to delete
3837 * @addr_type: address type
3839 * Release reference to a device address and remove it from the device
3840 * if the reference count drops to zero.
3842 * The caller must hold the rtnl_mutex.
3844 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3845 unsigned char addr_type)
3848 struct netdev_hw_addr *ha;
3853 * We can not remove the first address from the list because
3854 * dev->dev_addr points to that.
3856 ha = list_first_entry(&dev->dev_addrs.list,
3857 struct netdev_hw_addr, list);
3858 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3861 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3864 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3867 EXPORT_SYMBOL(dev_addr_del);
3870 * dev_addr_add_multiple - Add device addresses from another device
3871 * @to_dev: device to which addresses will be added
3872 * @from_dev: device from which addresses will be added
3873 * @addr_type: address type - 0 means type will be used from from_dev
3875 * Add device addresses of the one device to another.
3877 * The caller must hold the rtnl_mutex.
3879 int dev_addr_add_multiple(struct net_device *to_dev,
3880 struct net_device *from_dev,
3881 unsigned char addr_type)
3887 if (from_dev->addr_len != to_dev->addr_len)
3889 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3890 to_dev->addr_len, addr_type);
3892 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3895 EXPORT_SYMBOL(dev_addr_add_multiple);
3898 * dev_addr_del_multiple - Delete device addresses by another device
3899 * @to_dev: device where the addresses will be deleted
3900 * @from_dev: device by which addresses the addresses will be deleted
3901 * @addr_type: address type - 0 means type will used from from_dev
3903 * Deletes addresses in to device by the list of addresses in from device.
3905 * The caller must hold the rtnl_mutex.
3907 int dev_addr_del_multiple(struct net_device *to_dev,
3908 struct net_device *from_dev,
3909 unsigned char addr_type)
3913 if (from_dev->addr_len != to_dev->addr_len)
3915 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3916 to_dev->addr_len, addr_type);
3917 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3920 EXPORT_SYMBOL(dev_addr_del_multiple);
3922 /* multicast addresses handling functions */
3924 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3925 void *addr, int alen, int glbl)
3927 struct dev_addr_list *da;
3929 for (; (da = *list) != NULL; list = &da->next) {
3930 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3931 alen == da->da_addrlen) {
3933 int old_glbl = da->da_gusers;
3950 int __dev_addr_add(struct dev_addr_list **list, int *count,
3951 void *addr, int alen, int glbl)
3953 struct dev_addr_list *da;
3955 for (da = *list; da != NULL; da = da->next) {
3956 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3957 da->da_addrlen == alen) {
3959 int old_glbl = da->da_gusers;
3969 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3972 memcpy(da->da_addr, addr, alen);
3973 da->da_addrlen = alen;
3975 da->da_gusers = glbl ? 1 : 0;
3983 * dev_unicast_delete - Release secondary unicast address.
3985 * @addr: address to delete
3987 * Release reference to a secondary unicast address and remove it
3988 * from the device if the reference count drops to zero.
3990 * The caller must hold the rtnl_mutex.
3992 int dev_unicast_delete(struct net_device *dev, void *addr)
3998 netif_addr_lock_bh(dev);
3999 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
4000 NETDEV_HW_ADDR_T_UNICAST);
4002 __dev_set_rx_mode(dev);
4003 netif_addr_unlock_bh(dev);
4006 EXPORT_SYMBOL(dev_unicast_delete);
4009 * dev_unicast_add - add a secondary unicast address
4011 * @addr: address to add
4013 * Add a secondary unicast address to the device or increase
4014 * the reference count if it already exists.
4016 * The caller must hold the rtnl_mutex.
4018 int dev_unicast_add(struct net_device *dev, void *addr)
4024 netif_addr_lock_bh(dev);
4025 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
4026 NETDEV_HW_ADDR_T_UNICAST);
4028 __dev_set_rx_mode(dev);
4029 netif_addr_unlock_bh(dev);
4032 EXPORT_SYMBOL(dev_unicast_add);
4034 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
4035 struct dev_addr_list **from, int *from_count)
4037 struct dev_addr_list *da, *next;
4041 while (da != NULL) {
4043 if (!da->da_synced) {
4044 err = __dev_addr_add(to, to_count,
4045 da->da_addr, da->da_addrlen, 0);
4050 } else if (da->da_users == 1) {
4051 __dev_addr_delete(to, to_count,
4052 da->da_addr, da->da_addrlen, 0);
4053 __dev_addr_delete(from, from_count,
4054 da->da_addr, da->da_addrlen, 0);
4060 EXPORT_SYMBOL_GPL(__dev_addr_sync);
4062 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4063 struct dev_addr_list **from, int *from_count)
4065 struct dev_addr_list *da, *next;
4068 while (da != NULL) {
4070 if (da->da_synced) {
4071 __dev_addr_delete(to, to_count,
4072 da->da_addr, da->da_addrlen, 0);
4074 __dev_addr_delete(from, from_count,
4075 da->da_addr, da->da_addrlen, 0);
4080 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4083 * dev_unicast_sync - Synchronize device's unicast list to another device
4084 * @to: destination device
4085 * @from: source device
4087 * Add newly added addresses to the destination device and release
4088 * addresses that have no users left. The source device must be
4089 * locked by netif_tx_lock_bh.
4091 * This function is intended to be called from the dev->set_rx_mode
4092 * function of layered software devices.
4094 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4098 if (to->addr_len != from->addr_len)
4101 netif_addr_lock_bh(to);
4102 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4104 __dev_set_rx_mode(to);
4105 netif_addr_unlock_bh(to);
4108 EXPORT_SYMBOL(dev_unicast_sync);
4111 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4112 * @to: destination device
4113 * @from: source device
4115 * Remove all addresses that were added to the destination device by
4116 * dev_unicast_sync(). This function is intended to be called from the
4117 * dev->stop function of layered software devices.
4119 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4121 if (to->addr_len != from->addr_len)
4124 netif_addr_lock_bh(from);
4125 netif_addr_lock(to);
4126 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4127 __dev_set_rx_mode(to);
4128 netif_addr_unlock(to);
4129 netif_addr_unlock_bh(from);
4131 EXPORT_SYMBOL(dev_unicast_unsync);
4133 static void dev_unicast_flush(struct net_device *dev)
4135 netif_addr_lock_bh(dev);
4136 __hw_addr_flush(&dev->uc);
4137 netif_addr_unlock_bh(dev);
4140 static void dev_unicast_init(struct net_device *dev)
4142 __hw_addr_init(&dev->uc);
4146 static void __dev_addr_discard(struct dev_addr_list **list)
4148 struct dev_addr_list *tmp;
4150 while (*list != NULL) {
4153 if (tmp->da_users > tmp->da_gusers)
4154 printk("__dev_addr_discard: address leakage! "
4155 "da_users=%d\n", tmp->da_users);
4160 static void dev_addr_discard(struct net_device *dev)
4162 netif_addr_lock_bh(dev);
4164 __dev_addr_discard(&dev->mc_list);
4167 netif_addr_unlock_bh(dev);
4171 * dev_get_flags - get flags reported to userspace
4174 * Get the combination of flag bits exported through APIs to userspace.
4176 unsigned dev_get_flags(const struct net_device *dev)
4180 flags = (dev->flags & ~(IFF_PROMISC |
4185 (dev->gflags & (IFF_PROMISC |
4188 if (netif_running(dev)) {
4189 if (netif_oper_up(dev))
4190 flags |= IFF_RUNNING;
4191 if (netif_carrier_ok(dev))
4192 flags |= IFF_LOWER_UP;
4193 if (netif_dormant(dev))
4194 flags |= IFF_DORMANT;
4199 EXPORT_SYMBOL(dev_get_flags);
4202 * dev_change_flags - change device settings
4204 * @flags: device state flags
4206 * Change settings on device based state flags. The flags are
4207 * in the userspace exported format.
4209 int dev_change_flags(struct net_device *dev, unsigned flags)
4212 int old_flags = dev->flags;
4217 * Set the flags on our device.
4220 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4221 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4223 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4227 * Load in the correct multicast list now the flags have changed.
4230 if ((old_flags ^ flags) & IFF_MULTICAST)
4231 dev_change_rx_flags(dev, IFF_MULTICAST);
4233 dev_set_rx_mode(dev);
4236 * Have we downed the interface. We handle IFF_UP ourselves
4237 * according to user attempts to set it, rather than blindly
4242 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4243 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4246 dev_set_rx_mode(dev);
4249 if (dev->flags & IFF_UP &&
4250 ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
4252 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4254 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4255 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4257 dev->gflags ^= IFF_PROMISC;
4258 dev_set_promiscuity(dev, inc);
4261 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4262 is important. Some (broken) drivers set IFF_PROMISC, when
4263 IFF_ALLMULTI is requested not asking us and not reporting.
4265 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4266 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4268 dev->gflags ^= IFF_ALLMULTI;
4269 dev_set_allmulti(dev, inc);
4272 /* Exclude state transition flags, already notified */
4273 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4275 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4279 EXPORT_SYMBOL(dev_change_flags);
4282 * dev_set_mtu - Change maximum transfer unit
4284 * @new_mtu: new transfer unit
4286 * Change the maximum transfer size of the network device.
4288 int dev_set_mtu(struct net_device *dev, int new_mtu)
4290 const struct net_device_ops *ops = dev->netdev_ops;
4293 if (new_mtu == dev->mtu)
4296 /* MTU must be positive. */
4300 if (!netif_device_present(dev))
4304 if (ops->ndo_change_mtu)
4305 err = ops->ndo_change_mtu(dev, new_mtu);
4309 if (!err && dev->flags & IFF_UP)
4310 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4313 EXPORT_SYMBOL(dev_set_mtu);
4316 * dev_set_mac_address - Change Media Access Control Address
4320 * Change the hardware (MAC) address of the device
4322 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4324 const struct net_device_ops *ops = dev->netdev_ops;
4327 if (!ops->ndo_set_mac_address)
4329 if (sa->sa_family != dev->type)
4331 if (!netif_device_present(dev))
4333 err = ops->ndo_set_mac_address(dev, sa);
4335 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4338 EXPORT_SYMBOL(dev_set_mac_address);
4341 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4343 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4346 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4352 case SIOCGIFFLAGS: /* Get interface flags */
4353 ifr->ifr_flags = (short) dev_get_flags(dev);
4356 case SIOCGIFMETRIC: /* Get the metric on the interface
4357 (currently unused) */
4358 ifr->ifr_metric = 0;
4361 case SIOCGIFMTU: /* Get the MTU of a device */
4362 ifr->ifr_mtu = dev->mtu;
4367 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4369 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4370 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4371 ifr->ifr_hwaddr.sa_family = dev->type;
4379 ifr->ifr_map.mem_start = dev->mem_start;
4380 ifr->ifr_map.mem_end = dev->mem_end;
4381 ifr->ifr_map.base_addr = dev->base_addr;
4382 ifr->ifr_map.irq = dev->irq;
4383 ifr->ifr_map.dma = dev->dma;
4384 ifr->ifr_map.port = dev->if_port;
4388 ifr->ifr_ifindex = dev->ifindex;
4392 ifr->ifr_qlen = dev->tx_queue_len;
4396 /* dev_ioctl() should ensure this case
4408 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4410 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4413 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4414 const struct net_device_ops *ops;
4419 ops = dev->netdev_ops;
4422 case SIOCSIFFLAGS: /* Set interface flags */
4423 return dev_change_flags(dev, ifr->ifr_flags);
4425 case SIOCSIFMETRIC: /* Set the metric on the interface
4426 (currently unused) */
4429 case SIOCSIFMTU: /* Set the MTU of a device */
4430 return dev_set_mtu(dev, ifr->ifr_mtu);
4433 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4435 case SIOCSIFHWBROADCAST:
4436 if (ifr->ifr_hwaddr.sa_family != dev->type)
4438 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4439 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4440 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4444 if (ops->ndo_set_config) {
4445 if (!netif_device_present(dev))
4447 return ops->ndo_set_config(dev, &ifr->ifr_map);
4452 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4453 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4455 if (!netif_device_present(dev))
4457 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4461 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4462 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4464 if (!netif_device_present(dev))
4466 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4470 if (ifr->ifr_qlen < 0)
4472 dev->tx_queue_len = ifr->ifr_qlen;
4476 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4477 return dev_change_name(dev, ifr->ifr_newname);
4480 * Unknown or private ioctl
4483 if ((cmd >= SIOCDEVPRIVATE &&
4484 cmd <= SIOCDEVPRIVATE + 15) ||
4485 cmd == SIOCBONDENSLAVE ||
4486 cmd == SIOCBONDRELEASE ||
4487 cmd == SIOCBONDSETHWADDR ||
4488 cmd == SIOCBONDSLAVEINFOQUERY ||
4489 cmd == SIOCBONDINFOQUERY ||
4490 cmd == SIOCBONDCHANGEACTIVE ||
4491 cmd == SIOCGMIIPHY ||
4492 cmd == SIOCGMIIREG ||
4493 cmd == SIOCSMIIREG ||
4494 cmd == SIOCBRADDIF ||
4495 cmd == SIOCBRDELIF ||
4496 cmd == SIOCSHWTSTAMP ||
4497 cmd == SIOCWANDEV) {
4499 if (ops->ndo_do_ioctl) {
4500 if (netif_device_present(dev))
4501 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4513 * This function handles all "interface"-type I/O control requests. The actual
4514 * 'doing' part of this is dev_ifsioc above.
4518 * dev_ioctl - network device ioctl
4519 * @net: the applicable net namespace
4520 * @cmd: command to issue
4521 * @arg: pointer to a struct ifreq in user space
4523 * Issue ioctl functions to devices. This is normally called by the
4524 * user space syscall interfaces but can sometimes be useful for
4525 * other purposes. The return value is the return from the syscall if
4526 * positive or a negative errno code on error.
4529 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4535 /* One special case: SIOCGIFCONF takes ifconf argument
4536 and requires shared lock, because it sleeps writing
4540 if (cmd == SIOCGIFCONF) {
4542 ret = dev_ifconf(net, (char __user *) arg);
4546 if (cmd == SIOCGIFNAME)
4547 return dev_ifname(net, (struct ifreq __user *)arg);
4549 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4552 ifr.ifr_name[IFNAMSIZ-1] = 0;
4554 colon = strchr(ifr.ifr_name, ':');
4559 * See which interface the caller is talking about.
4564 * These ioctl calls:
4565 * - can be done by all.
4566 * - atomic and do not require locking.
4577 dev_load(net, ifr.ifr_name);
4579 ret = dev_ifsioc_locked(net, &ifr, cmd);
4584 if (copy_to_user(arg, &ifr,
4585 sizeof(struct ifreq)))
4591 dev_load(net, ifr.ifr_name);
4593 ret = dev_ethtool(net, &ifr);
4598 if (copy_to_user(arg, &ifr,
4599 sizeof(struct ifreq)))
4605 * These ioctl calls:
4606 * - require superuser power.
4607 * - require strict serialization.
4613 if (!capable(CAP_NET_ADMIN))
4615 dev_load(net, ifr.ifr_name);
4617 ret = dev_ifsioc(net, &ifr, cmd);
4622 if (copy_to_user(arg, &ifr,
4623 sizeof(struct ifreq)))
4629 * These ioctl calls:
4630 * - require superuser power.
4631 * - require strict serialization.
4632 * - do not return a value
4642 case SIOCSIFHWBROADCAST:
4645 case SIOCBONDENSLAVE:
4646 case SIOCBONDRELEASE:
4647 case SIOCBONDSETHWADDR:
4648 case SIOCBONDCHANGEACTIVE:
4652 if (!capable(CAP_NET_ADMIN))
4655 case SIOCBONDSLAVEINFOQUERY:
4656 case SIOCBONDINFOQUERY:
4657 dev_load(net, ifr.ifr_name);
4659 ret = dev_ifsioc(net, &ifr, cmd);
4664 /* Get the per device memory space. We can add this but
4665 * currently do not support it */
4667 /* Set the per device memory buffer space.
4668 * Not applicable in our case */
4673 * Unknown or private ioctl.
4676 if (cmd == SIOCWANDEV ||
4677 (cmd >= SIOCDEVPRIVATE &&
4678 cmd <= SIOCDEVPRIVATE + 15)) {
4679 dev_load(net, ifr.ifr_name);
4681 ret = dev_ifsioc(net, &ifr, cmd);
4683 if (!ret && copy_to_user(arg, &ifr,
4684 sizeof(struct ifreq)))
4688 /* Take care of Wireless Extensions */
4689 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4690 return wext_handle_ioctl(net, &ifr, cmd, arg);
4697 * dev_new_index - allocate an ifindex
4698 * @net: the applicable net namespace
4700 * Returns a suitable unique value for a new device interface
4701 * number. The caller must hold the rtnl semaphore or the
4702 * dev_base_lock to be sure it remains unique.
4704 static int dev_new_index(struct net *net)
4710 if (!__dev_get_by_index(net, ifindex))
4715 /* Delayed registration/unregisteration */
4716 static LIST_HEAD(net_todo_list);
4718 static void net_set_todo(struct net_device *dev)
4720 list_add_tail(&dev->todo_list, &net_todo_list);
4723 static void rollback_registered_many(struct list_head *head)
4725 struct net_device *dev, *aux, *fdev;
4726 LIST_HEAD(pernet_list);
4728 BUG_ON(dev_boot_phase);
4731 list_for_each_entry(dev, head, unreg_list) {
4732 /* Some devices call without registering
4733 * for initialization unwind.
4735 if (dev->reg_state == NETREG_UNINITIALIZED) {
4736 pr_debug("unregister_netdevice: device %s/%p never "
4737 "was registered\n", dev->name, dev);
4743 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4745 /* If device is running, close it first. */
4748 /* And unlink it from device chain. */
4749 unlist_netdevice(dev);
4751 dev->reg_state = NETREG_UNREGISTERING;
4756 list_for_each_entry(dev, head, unreg_list) {
4757 /* Shutdown queueing discipline. */
4761 /* Notify protocols, that we are about to destroy
4762 this device. They should clean all the things.
4764 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4767 * Flush the unicast and multicast chains
4769 dev_unicast_flush(dev);
4770 dev_addr_discard(dev);
4772 if (dev->netdev_ops->ndo_uninit)
4773 dev->netdev_ops->ndo_uninit(dev);
4775 /* Notifier chain MUST detach us from master device. */
4776 WARN_ON(dev->master);
4778 /* Remove entries from kobject tree */
4779 netdev_unregister_kobject(dev);
4784 list_for_each_entry_safe(dev, aux, head, unreg_list) {
4786 list_for_each_entry(fdev, &pernet_list, unreg_list) {
4787 if (dev_net(dev) == dev_net(fdev)) {
4794 list_move(&dev->unreg_list, &pernet_list);
4797 list_for_each_entry_safe(dev, aux, &pernet_list, unreg_list) {
4798 call_netdevice_notifiers(NETDEV_UNREGISTER_PERNET, dev);
4799 list_move(&dev->unreg_list, head);
4804 static void rollback_registered(struct net_device *dev)
4808 list_add(&dev->unreg_list, &single);
4809 rollback_registered_many(&single);
4812 static void __netdev_init_queue_locks_one(struct net_device *dev,
4813 struct netdev_queue *dev_queue,
4816 spin_lock_init(&dev_queue->_xmit_lock);
4817 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4818 dev_queue->xmit_lock_owner = -1;
4821 static void netdev_init_queue_locks(struct net_device *dev)
4823 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4824 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4827 unsigned long netdev_fix_features(unsigned long features, const char *name)
4829 /* Fix illegal SG+CSUM combinations. */
4830 if ((features & NETIF_F_SG) &&
4831 !(features & NETIF_F_ALL_CSUM)) {
4833 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4834 "checksum feature.\n", name);
4835 features &= ~NETIF_F_SG;
4838 /* TSO requires that SG is present as well. */
4839 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4841 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4842 "SG feature.\n", name);
4843 features &= ~NETIF_F_TSO;
4846 if (features & NETIF_F_UFO) {
4847 if (!(features & NETIF_F_GEN_CSUM)) {
4849 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4850 "since no NETIF_F_HW_CSUM feature.\n",
4852 features &= ~NETIF_F_UFO;
4855 if (!(features & NETIF_F_SG)) {
4857 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4858 "since no NETIF_F_SG feature.\n", name);
4859 features &= ~NETIF_F_UFO;
4865 EXPORT_SYMBOL(netdev_fix_features);
4868 * register_netdevice - register a network device
4869 * @dev: device to register
4871 * Take a completed network device structure and add it to the kernel
4872 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4873 * chain. 0 is returned on success. A negative errno code is returned
4874 * on a failure to set up the device, or if the name is a duplicate.
4876 * Callers must hold the rtnl semaphore. You may want
4877 * register_netdev() instead of this.
4880 * The locking appears insufficient to guarantee two parallel registers
4881 * will not get the same name.
4884 int register_netdevice(struct net_device *dev)
4886 struct hlist_head *head;
4887 struct hlist_node *p;
4889 struct net *net = dev_net(dev);
4891 BUG_ON(dev_boot_phase);
4896 /* When net_device's are persistent, this will be fatal. */
4897 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4900 spin_lock_init(&dev->addr_list_lock);
4901 netdev_set_addr_lockdep_class(dev);
4902 netdev_init_queue_locks(dev);
4906 /* Init, if this function is available */
4907 if (dev->netdev_ops->ndo_init) {
4908 ret = dev->netdev_ops->ndo_init(dev);
4916 if (!dev_valid_name(dev->name)) {
4921 dev->ifindex = dev_new_index(net);
4922 if (dev->iflink == -1)
4923 dev->iflink = dev->ifindex;
4925 /* Check for existence of name */
4926 head = dev_name_hash(net, dev->name);
4927 hlist_for_each(p, head) {
4928 struct net_device *d
4929 = hlist_entry(p, struct net_device, name_hlist);
4930 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4936 /* Fix illegal checksum combinations */
4937 if ((dev->features & NETIF_F_HW_CSUM) &&
4938 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4939 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4941 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4944 if ((dev->features & NETIF_F_NO_CSUM) &&
4945 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4946 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4948 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4951 dev->features = netdev_fix_features(dev->features, dev->name);
4953 /* Enable software GSO if SG is supported. */
4954 if (dev->features & NETIF_F_SG)
4955 dev->features |= NETIF_F_GSO;
4957 netdev_initialize_kobject(dev);
4959 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4960 ret = notifier_to_errno(ret);
4964 ret = netdev_register_kobject(dev);
4967 dev->reg_state = NETREG_REGISTERED;
4970 * Default initial state at registry is that the
4971 * device is present.
4974 set_bit(__LINK_STATE_PRESENT, &dev->state);
4976 dev_init_scheduler(dev);
4978 list_netdevice(dev);
4980 /* Notify protocols, that a new device appeared. */
4981 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4982 ret = notifier_to_errno(ret);
4984 rollback_registered(dev);
4985 dev->reg_state = NETREG_UNREGISTERED;
4992 if (dev->netdev_ops->ndo_uninit)
4993 dev->netdev_ops->ndo_uninit(dev);
4996 EXPORT_SYMBOL(register_netdevice);
4999 * init_dummy_netdev - init a dummy network device for NAPI
5000 * @dev: device to init
5002 * This takes a network device structure and initialize the minimum
5003 * amount of fields so it can be used to schedule NAPI polls without
5004 * registering a full blown interface. This is to be used by drivers
5005 * that need to tie several hardware interfaces to a single NAPI
5006 * poll scheduler due to HW limitations.
5008 int init_dummy_netdev(struct net_device *dev)
5010 /* Clear everything. Note we don't initialize spinlocks
5011 * are they aren't supposed to be taken by any of the
5012 * NAPI code and this dummy netdev is supposed to be
5013 * only ever used for NAPI polls
5015 memset(dev, 0, sizeof(struct net_device));
5017 /* make sure we BUG if trying to hit standard
5018 * register/unregister code path
5020 dev->reg_state = NETREG_DUMMY;
5022 /* initialize the ref count */
5023 atomic_set(&dev->refcnt, 1);
5025 /* NAPI wants this */
5026 INIT_LIST_HEAD(&dev->napi_list);
5028 /* a dummy interface is started by default */
5029 set_bit(__LINK_STATE_PRESENT, &dev->state);
5030 set_bit(__LINK_STATE_START, &dev->state);
5034 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5038 * register_netdev - register a network device
5039 * @dev: device to register
5041 * Take a completed network device structure and add it to the kernel
5042 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5043 * chain. 0 is returned on success. A negative errno code is returned
5044 * on a failure to set up the device, or if the name is a duplicate.
5046 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5047 * and expands the device name if you passed a format string to
5050 int register_netdev(struct net_device *dev)
5057 * If the name is a format string the caller wants us to do a
5060 if (strchr(dev->name, '%')) {
5061 err = dev_alloc_name(dev, dev->name);
5066 err = register_netdevice(dev);
5071 EXPORT_SYMBOL(register_netdev);
5074 * netdev_wait_allrefs - wait until all references are gone.
5076 * This is called when unregistering network devices.
5078 * Any protocol or device that holds a reference should register
5079 * for netdevice notification, and cleanup and put back the
5080 * reference if they receive an UNREGISTER event.
5081 * We can get stuck here if buggy protocols don't correctly
5084 static void netdev_wait_allrefs(struct net_device *dev)
5086 unsigned long rebroadcast_time, warning_time;
5088 rebroadcast_time = warning_time = jiffies;
5089 while (atomic_read(&dev->refcnt) != 0) {
5090 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5093 /* Rebroadcast unregister notification */
5094 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5095 /* don't resend NETDEV_UNREGISTER_PERNET, _PERNET users
5096 * should have already handle it the first time */
5098 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5100 /* We must not have linkwatch events
5101 * pending on unregister. If this
5102 * happens, we simply run the queue
5103 * unscheduled, resulting in a noop
5106 linkwatch_run_queue();
5111 rebroadcast_time = jiffies;
5116 if (time_after(jiffies, warning_time + 10 * HZ)) {
5117 printk(KERN_EMERG "unregister_netdevice: "
5118 "waiting for %s to become free. Usage "
5120 dev->name, atomic_read(&dev->refcnt));
5121 warning_time = jiffies;
5130 * register_netdevice(x1);
5131 * register_netdevice(x2);
5133 * unregister_netdevice(y1);
5134 * unregister_netdevice(y2);
5140 * We are invoked by rtnl_unlock().
5141 * This allows us to deal with problems:
5142 * 1) We can delete sysfs objects which invoke hotplug
5143 * without deadlocking with linkwatch via keventd.
5144 * 2) Since we run with the RTNL semaphore not held, we can sleep
5145 * safely in order to wait for the netdev refcnt to drop to zero.
5147 * We must not return until all unregister events added during
5148 * the interval the lock was held have been completed.
5150 void netdev_run_todo(void)
5152 struct list_head list;
5154 /* Snapshot list, allow later requests */
5155 list_replace_init(&net_todo_list, &list);
5159 while (!list_empty(&list)) {
5160 struct net_device *dev
5161 = list_entry(list.next, struct net_device, todo_list);
5162 list_del(&dev->todo_list);
5164 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5165 printk(KERN_ERR "network todo '%s' but state %d\n",
5166 dev->name, dev->reg_state);
5171 dev->reg_state = NETREG_UNREGISTERED;
5173 on_each_cpu(flush_backlog, dev, 1);
5175 netdev_wait_allrefs(dev);
5178 BUG_ON(atomic_read(&dev->refcnt));
5179 WARN_ON(dev->ip_ptr);
5180 WARN_ON(dev->ip6_ptr);
5181 WARN_ON(dev->dn_ptr);
5183 if (dev->destructor)
5184 dev->destructor(dev);
5186 /* Free network device */
5187 kobject_put(&dev->dev.kobj);
5192 * dev_txq_stats_fold - fold tx_queues stats
5193 * @dev: device to get statistics from
5194 * @stats: struct net_device_stats to hold results
5196 void dev_txq_stats_fold(const struct net_device *dev,
5197 struct net_device_stats *stats)
5199 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5201 struct netdev_queue *txq;
5203 for (i = 0; i < dev->num_tx_queues; i++) {
5204 txq = netdev_get_tx_queue(dev, i);
5205 tx_bytes += txq->tx_bytes;
5206 tx_packets += txq->tx_packets;
5207 tx_dropped += txq->tx_dropped;
5209 if (tx_bytes || tx_packets || tx_dropped) {
5210 stats->tx_bytes = tx_bytes;
5211 stats->tx_packets = tx_packets;
5212 stats->tx_dropped = tx_dropped;
5215 EXPORT_SYMBOL(dev_txq_stats_fold);
5218 * dev_get_stats - get network device statistics
5219 * @dev: device to get statistics from
5221 * Get network statistics from device. The device driver may provide
5222 * its own method by setting dev->netdev_ops->get_stats; otherwise
5223 * the internal statistics structure is used.
5225 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5227 const struct net_device_ops *ops = dev->netdev_ops;
5229 if (ops->ndo_get_stats)
5230 return ops->ndo_get_stats(dev);
5232 dev_txq_stats_fold(dev, &dev->stats);
5235 EXPORT_SYMBOL(dev_get_stats);
5237 static void netdev_init_one_queue(struct net_device *dev,
5238 struct netdev_queue *queue,
5244 static void netdev_init_queues(struct net_device *dev)
5246 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5247 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5248 spin_lock_init(&dev->tx_global_lock);
5252 * alloc_netdev_mq - allocate network device
5253 * @sizeof_priv: size of private data to allocate space for
5254 * @name: device name format string
5255 * @setup: callback to initialize device
5256 * @queue_count: the number of subqueues to allocate
5258 * Allocates a struct net_device with private data area for driver use
5259 * and performs basic initialization. Also allocates subquue structs
5260 * for each queue on the device at the end of the netdevice.
5262 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5263 void (*setup)(struct net_device *), unsigned int queue_count)
5265 struct netdev_queue *tx;
5266 struct net_device *dev;
5268 struct net_device *p;
5270 BUG_ON(strlen(name) >= sizeof(dev->name));
5272 alloc_size = sizeof(struct net_device);
5274 /* ensure 32-byte alignment of private area */
5275 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5276 alloc_size += sizeof_priv;
5278 /* ensure 32-byte alignment of whole construct */
5279 alloc_size += NETDEV_ALIGN - 1;
5281 p = kzalloc(alloc_size, GFP_KERNEL);
5283 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5287 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5289 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5294 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5295 dev->padded = (char *)dev - (char *)p;
5297 if (dev_addr_init(dev))
5300 dev_unicast_init(dev);
5302 dev_net_set(dev, &init_net);
5305 dev->num_tx_queues = queue_count;
5306 dev->real_num_tx_queues = queue_count;
5308 dev->gso_max_size = GSO_MAX_SIZE;
5310 netdev_init_queues(dev);
5312 INIT_LIST_HEAD(&dev->napi_list);
5313 INIT_LIST_HEAD(&dev->unreg_list);
5314 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5316 strcpy(dev->name, name);
5326 EXPORT_SYMBOL(alloc_netdev_mq);
5329 * free_netdev - free network device
5332 * This function does the last stage of destroying an allocated device
5333 * interface. The reference to the device object is released.
5334 * If this is the last reference then it will be freed.
5336 void free_netdev(struct net_device *dev)
5338 struct napi_struct *p, *n;
5340 release_net(dev_net(dev));
5344 /* Flush device addresses */
5345 dev_addr_flush(dev);
5347 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5350 /* Compatibility with error handling in drivers */
5351 if (dev->reg_state == NETREG_UNINITIALIZED) {
5352 kfree((char *)dev - dev->padded);
5356 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5357 dev->reg_state = NETREG_RELEASED;
5359 /* will free via device release */
5360 put_device(&dev->dev);
5362 EXPORT_SYMBOL(free_netdev);
5365 * synchronize_net - Synchronize with packet receive processing
5367 * Wait for packets currently being received to be done.
5368 * Does not block later packets from starting.
5370 void synchronize_net(void)
5375 EXPORT_SYMBOL(synchronize_net);
5378 * unregister_netdevice_queue - remove device from the kernel
5382 * This function shuts down a device interface and removes it
5383 * from the kernel tables.
5384 * If head not NULL, device is queued to be unregistered later.
5386 * Callers must hold the rtnl semaphore. You may want
5387 * unregister_netdev() instead of this.
5390 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5395 list_move_tail(&dev->unreg_list, head);
5397 rollback_registered(dev);
5398 /* Finish processing unregister after unlock */
5402 EXPORT_SYMBOL(unregister_netdevice_queue);
5405 * unregister_netdevice_many - unregister many devices
5406 * @head: list of devices
5408 * WARNING: Calling this modifies the given list
5409 * (in rollback_registered_many). It may change the order of the elements
5410 * in the list. However, you can assume it does not add or delete elements
5413 void unregister_netdevice_many(struct list_head *head)
5415 struct net_device *dev;
5417 if (!list_empty(head)) {
5418 rollback_registered_many(head);
5419 list_for_each_entry(dev, head, unreg_list)
5423 EXPORT_SYMBOL(unregister_netdevice_many);
5426 * unregister_netdev - remove device from the kernel
5429 * This function shuts down a device interface and removes it
5430 * from the kernel tables.
5432 * This is just a wrapper for unregister_netdevice that takes
5433 * the rtnl semaphore. In general you want to use this and not
5434 * unregister_netdevice.
5436 void unregister_netdev(struct net_device *dev)
5439 unregister_netdevice(dev);
5442 EXPORT_SYMBOL(unregister_netdev);
5445 * dev_change_net_namespace - move device to different nethost namespace
5447 * @net: network namespace
5448 * @pat: If not NULL name pattern to try if the current device name
5449 * is already taken in the destination network namespace.
5451 * This function shuts down a device interface and moves it
5452 * to a new network namespace. On success 0 is returned, on
5453 * a failure a netagive errno code is returned.
5455 * Callers must hold the rtnl semaphore.
5458 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5461 const char *destname;
5466 /* Don't allow namespace local devices to be moved. */
5468 if (dev->features & NETIF_F_NETNS_LOCAL)
5472 /* Don't allow real devices to be moved when sysfs
5476 if (dev->dev.parent)
5480 /* Ensure the device has been registrered */
5482 if (dev->reg_state != NETREG_REGISTERED)
5485 /* Get out if there is nothing todo */
5487 if (net_eq(dev_net(dev), net))
5490 /* Pick the destination device name, and ensure
5491 * we can use it in the destination network namespace.
5494 destname = dev->name;
5495 if (__dev_get_by_name(net, destname)) {
5496 /* We get here if we can't use the current device name */
5499 if (!dev_valid_name(pat))
5501 if (strchr(pat, '%')) {
5502 if (__dev_alloc_name(net, pat, buf) < 0)
5507 if (__dev_get_by_name(net, destname))
5512 * And now a mini version of register_netdevice unregister_netdevice.
5515 /* If device is running close it first. */
5518 /* And unlink it from device chain */
5520 unlist_netdevice(dev);
5524 /* Shutdown queueing discipline. */
5527 /* Notify protocols, that we are about to destroy
5528 this device. They should clean all the things.
5530 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5531 call_netdevice_notifiers(NETDEV_UNREGISTER_PERNET, dev);
5534 * Flush the unicast and multicast chains
5536 dev_unicast_flush(dev);
5537 dev_addr_discard(dev);
5539 netdev_unregister_kobject(dev);
5541 /* Actually switch the network namespace */
5542 dev_net_set(dev, net);
5544 /* Assign the new device name */
5545 if (destname != dev->name)
5546 strcpy(dev->name, destname);
5548 /* If there is an ifindex conflict assign a new one */
5549 if (__dev_get_by_index(net, dev->ifindex)) {
5550 int iflink = (dev->iflink == dev->ifindex);
5551 dev->ifindex = dev_new_index(net);
5553 dev->iflink = dev->ifindex;
5556 /* Fixup kobjects */
5557 err = netdev_register_kobject(dev);
5560 /* Add the device back in the hashes */
5561 list_netdevice(dev);
5563 /* Notify protocols, that a new device appeared. */
5564 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5571 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5573 static int dev_cpu_callback(struct notifier_block *nfb,
5574 unsigned long action,
5577 struct sk_buff **list_skb;
5578 struct Qdisc **list_net;
5579 struct sk_buff *skb;
5580 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5581 struct softnet_data *sd, *oldsd;
5583 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5586 local_irq_disable();
5587 cpu = smp_processor_id();
5588 sd = &per_cpu(softnet_data, cpu);
5589 oldsd = &per_cpu(softnet_data, oldcpu);
5591 /* Find end of our completion_queue. */
5592 list_skb = &sd->completion_queue;
5594 list_skb = &(*list_skb)->next;
5595 /* Append completion queue from offline CPU. */
5596 *list_skb = oldsd->completion_queue;
5597 oldsd->completion_queue = NULL;
5599 /* Find end of our output_queue. */
5600 list_net = &sd->output_queue;
5602 list_net = &(*list_net)->next_sched;
5603 /* Append output queue from offline CPU. */
5604 *list_net = oldsd->output_queue;
5605 oldsd->output_queue = NULL;
5607 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5610 /* Process offline CPU's input_pkt_queue */
5611 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5619 * netdev_increment_features - increment feature set by one
5620 * @all: current feature set
5621 * @one: new feature set
5622 * @mask: mask feature set
5624 * Computes a new feature set after adding a device with feature set
5625 * @one to the master device with current feature set @all. Will not
5626 * enable anything that is off in @mask. Returns the new feature set.
5628 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5631 /* If device needs checksumming, downgrade to it. */
5632 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5633 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5634 else if (mask & NETIF_F_ALL_CSUM) {
5635 /* If one device supports v4/v6 checksumming, set for all. */
5636 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5637 !(all & NETIF_F_GEN_CSUM)) {
5638 all &= ~NETIF_F_ALL_CSUM;
5639 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5642 /* If one device supports hw checksumming, set for all. */
5643 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5644 all &= ~NETIF_F_ALL_CSUM;
5645 all |= NETIF_F_HW_CSUM;
5649 one |= NETIF_F_ALL_CSUM;
5651 one |= all & NETIF_F_ONE_FOR_ALL;
5652 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5653 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5657 EXPORT_SYMBOL(netdev_increment_features);
5659 static struct hlist_head *netdev_create_hash(void)
5662 struct hlist_head *hash;
5664 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5666 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5667 INIT_HLIST_HEAD(&hash[i]);
5672 /* Initialize per network namespace state */
5673 static int __net_init netdev_init(struct net *net)
5675 INIT_LIST_HEAD(&net->dev_base_head);
5677 net->dev_name_head = netdev_create_hash();
5678 if (net->dev_name_head == NULL)
5681 net->dev_index_head = netdev_create_hash();
5682 if (net->dev_index_head == NULL)
5688 kfree(net->dev_name_head);
5694 * netdev_drivername - network driver for the device
5695 * @dev: network device
5696 * @buffer: buffer for resulting name
5697 * @len: size of buffer
5699 * Determine network driver for device.
5701 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5703 const struct device_driver *driver;
5704 const struct device *parent;
5706 if (len <= 0 || !buffer)
5710 parent = dev->dev.parent;
5715 driver = parent->driver;
5716 if (driver && driver->name)
5717 strlcpy(buffer, driver->name, len);
5721 static void __net_exit netdev_exit(struct net *net)
5723 kfree(net->dev_name_head);
5724 kfree(net->dev_index_head);
5727 static struct pernet_operations __net_initdata netdev_net_ops = {
5728 .init = netdev_init,
5729 .exit = netdev_exit,
5732 static void __net_exit default_device_exit(struct net *net)
5734 struct net_device *dev;
5736 * Push all migratable of the network devices back to the
5737 * initial network namespace
5741 for_each_netdev(net, dev) {
5743 char fb_name[IFNAMSIZ];
5745 /* Ignore unmoveable devices (i.e. loopback) */
5746 if (dev->features & NETIF_F_NETNS_LOCAL)
5749 /* Delete virtual devices */
5750 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5751 dev->rtnl_link_ops->dellink(dev, NULL);
5755 /* Push remaing network devices to init_net */
5756 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5757 err = dev_change_net_namespace(dev, &init_net, fb_name);
5759 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5760 __func__, dev->name, err);
5768 static struct pernet_operations __net_initdata default_device_ops = {
5769 .exit = default_device_exit,
5773 * Initialize the DEV module. At boot time this walks the device list and
5774 * unhooks any devices that fail to initialise (normally hardware not
5775 * present) and leaves us with a valid list of present and active devices.
5780 * This is called single threaded during boot, so no need
5781 * to take the rtnl semaphore.
5783 static int __init net_dev_init(void)
5785 int i, rc = -ENOMEM;
5787 BUG_ON(!dev_boot_phase);
5789 if (dev_proc_init())
5792 if (netdev_kobject_init())
5795 INIT_LIST_HEAD(&ptype_all);
5796 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5797 INIT_LIST_HEAD(&ptype_base[i]);
5799 if (register_pernet_subsys(&netdev_net_ops))
5803 * Initialise the packet receive queues.
5806 for_each_possible_cpu(i) {
5807 struct softnet_data *queue;
5809 queue = &per_cpu(softnet_data, i);
5810 skb_queue_head_init(&queue->input_pkt_queue);
5811 queue->completion_queue = NULL;
5812 INIT_LIST_HEAD(&queue->poll_list);
5814 queue->backlog.poll = process_backlog;
5815 queue->backlog.weight = weight_p;
5816 queue->backlog.gro_list = NULL;
5817 queue->backlog.gro_count = 0;
5822 /* The loopback device is special if any other network devices
5823 * is present in a network namespace the loopback device must
5824 * be present. Since we now dynamically allocate and free the
5825 * loopback device ensure this invariant is maintained by
5826 * keeping the loopback device as the first device on the
5827 * list of network devices. Ensuring the loopback devices
5828 * is the first device that appears and the last network device
5831 if (register_pernet_device(&loopback_net_ops))
5834 if (register_pernet_device(&default_device_ops))
5837 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5838 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5840 hotcpu_notifier(dev_cpu_callback, 0);
5848 subsys_initcall(net_dev_init);
5850 static int __init initialize_hashrnd(void)
5852 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5856 late_initcall_sync(initialize_hashrnd);