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);
1350 raw_notifier_chain_unregister(&netdev_chain, nb);
1353 EXPORT_SYMBOL(register_netdevice_notifier);
1356 * unregister_netdevice_notifier - unregister a network notifier block
1359 * Unregister a notifier previously registered by
1360 * register_netdevice_notifier(). The notifier is unlinked into the
1361 * kernel structures and may then be reused. A negative errno code
1362 * is returned on a failure.
1365 int unregister_netdevice_notifier(struct notifier_block *nb)
1370 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1374 EXPORT_SYMBOL(unregister_netdevice_notifier);
1377 * call_netdevice_notifiers - call all network notifier blocks
1378 * @val: value passed unmodified to notifier function
1379 * @dev: net_device pointer passed unmodified to notifier function
1381 * Call all network notifier blocks. Parameters and return value
1382 * are as for raw_notifier_call_chain().
1385 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1387 return raw_notifier_call_chain(&netdev_chain, val, dev);
1390 /* When > 0 there are consumers of rx skb time stamps */
1391 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1393 void net_enable_timestamp(void)
1395 atomic_inc(&netstamp_needed);
1397 EXPORT_SYMBOL(net_enable_timestamp);
1399 void net_disable_timestamp(void)
1401 atomic_dec(&netstamp_needed);
1403 EXPORT_SYMBOL(net_disable_timestamp);
1405 static inline void net_timestamp(struct sk_buff *skb)
1407 if (atomic_read(&netstamp_needed))
1408 __net_timestamp(skb);
1410 skb->tstamp.tv64 = 0;
1414 * Support routine. Sends outgoing frames to any network
1415 * taps currently in use.
1418 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1420 struct packet_type *ptype;
1422 #ifdef CONFIG_NET_CLS_ACT
1423 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1430 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1431 /* Never send packets back to the socket
1432 * they originated from - MvS (miquels@drinkel.ow.org)
1434 if ((ptype->dev == dev || !ptype->dev) &&
1435 (ptype->af_packet_priv == NULL ||
1436 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1437 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1441 /* skb->nh should be correctly
1442 set by sender, so that the second statement is
1443 just protection against buggy protocols.
1445 skb_reset_mac_header(skb2);
1447 if (skb_network_header(skb2) < skb2->data ||
1448 skb2->network_header > skb2->tail) {
1449 if (net_ratelimit())
1450 printk(KERN_CRIT "protocol %04x is "
1452 skb2->protocol, dev->name);
1453 skb_reset_network_header(skb2);
1456 skb2->transport_header = skb2->network_header;
1457 skb2->pkt_type = PACKET_OUTGOING;
1458 ptype->func(skb2, skb->dev, ptype, skb->dev);
1465 static inline void __netif_reschedule(struct Qdisc *q)
1467 struct softnet_data *sd;
1468 unsigned long flags;
1470 local_irq_save(flags);
1471 sd = &__get_cpu_var(softnet_data);
1472 q->next_sched = sd->output_queue;
1473 sd->output_queue = q;
1474 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1475 local_irq_restore(flags);
1478 void __netif_schedule(struct Qdisc *q)
1480 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1481 __netif_reschedule(q);
1483 EXPORT_SYMBOL(__netif_schedule);
1485 void dev_kfree_skb_irq(struct sk_buff *skb)
1487 if (atomic_dec_and_test(&skb->users)) {
1488 struct softnet_data *sd;
1489 unsigned long flags;
1491 local_irq_save(flags);
1492 sd = &__get_cpu_var(softnet_data);
1493 skb->next = sd->completion_queue;
1494 sd->completion_queue = skb;
1495 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1496 local_irq_restore(flags);
1499 EXPORT_SYMBOL(dev_kfree_skb_irq);
1501 void dev_kfree_skb_any(struct sk_buff *skb)
1503 if (in_irq() || irqs_disabled())
1504 dev_kfree_skb_irq(skb);
1508 EXPORT_SYMBOL(dev_kfree_skb_any);
1512 * netif_device_detach - mark device as removed
1513 * @dev: network device
1515 * Mark device as removed from system and therefore no longer available.
1517 void netif_device_detach(struct net_device *dev)
1519 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1520 netif_running(dev)) {
1521 netif_tx_stop_all_queues(dev);
1524 EXPORT_SYMBOL(netif_device_detach);
1527 * netif_device_attach - mark device as attached
1528 * @dev: network device
1530 * Mark device as attached from system and restart if needed.
1532 void netif_device_attach(struct net_device *dev)
1534 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1535 netif_running(dev)) {
1536 netif_tx_wake_all_queues(dev);
1537 __netdev_watchdog_up(dev);
1540 EXPORT_SYMBOL(netif_device_attach);
1542 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1544 return ((features & NETIF_F_GEN_CSUM) ||
1545 ((features & NETIF_F_IP_CSUM) &&
1546 protocol == htons(ETH_P_IP)) ||
1547 ((features & NETIF_F_IPV6_CSUM) &&
1548 protocol == htons(ETH_P_IPV6)) ||
1549 ((features & NETIF_F_FCOE_CRC) &&
1550 protocol == htons(ETH_P_FCOE)));
1553 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1555 if (can_checksum_protocol(dev->features, skb->protocol))
1558 if (skb->protocol == htons(ETH_P_8021Q)) {
1559 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1560 if (can_checksum_protocol(dev->features & dev->vlan_features,
1561 veh->h_vlan_encapsulated_proto))
1569 * Invalidate hardware checksum when packet is to be mangled, and
1570 * complete checksum manually on outgoing path.
1572 int skb_checksum_help(struct sk_buff *skb)
1575 int ret = 0, offset;
1577 if (skb->ip_summed == CHECKSUM_COMPLETE)
1578 goto out_set_summed;
1580 if (unlikely(skb_shinfo(skb)->gso_size)) {
1581 /* Let GSO fix up the checksum. */
1582 goto out_set_summed;
1585 offset = skb->csum_start - skb_headroom(skb);
1586 BUG_ON(offset >= skb_headlen(skb));
1587 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1589 offset += skb->csum_offset;
1590 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1592 if (skb_cloned(skb) &&
1593 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1594 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1599 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1601 skb->ip_summed = CHECKSUM_NONE;
1605 EXPORT_SYMBOL(skb_checksum_help);
1608 * skb_gso_segment - Perform segmentation on skb.
1609 * @skb: buffer to segment
1610 * @features: features for the output path (see dev->features)
1612 * This function segments the given skb and returns a list of segments.
1614 * It may return NULL if the skb requires no segmentation. This is
1615 * only possible when GSO is used for verifying header integrity.
1617 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1619 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1620 struct packet_type *ptype;
1621 __be16 type = skb->protocol;
1624 skb_reset_mac_header(skb);
1625 skb->mac_len = skb->network_header - skb->mac_header;
1626 __skb_pull(skb, skb->mac_len);
1628 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1629 struct net_device *dev = skb->dev;
1630 struct ethtool_drvinfo info = {};
1632 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1633 dev->ethtool_ops->get_drvinfo(dev, &info);
1635 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1637 info.driver, dev ? dev->features : 0L,
1638 skb->sk ? skb->sk->sk_route_caps : 0L,
1639 skb->len, skb->data_len, skb->ip_summed);
1641 if (skb_header_cloned(skb) &&
1642 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1643 return ERR_PTR(err);
1647 list_for_each_entry_rcu(ptype,
1648 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1649 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1650 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1651 err = ptype->gso_send_check(skb);
1652 segs = ERR_PTR(err);
1653 if (err || skb_gso_ok(skb, features))
1655 __skb_push(skb, (skb->data -
1656 skb_network_header(skb)));
1658 segs = ptype->gso_segment(skb, features);
1664 __skb_push(skb, skb->data - skb_mac_header(skb));
1668 EXPORT_SYMBOL(skb_gso_segment);
1670 /* Take action when hardware reception checksum errors are detected. */
1672 void netdev_rx_csum_fault(struct net_device *dev)
1674 if (net_ratelimit()) {
1675 printk(KERN_ERR "%s: hw csum failure.\n",
1676 dev ? dev->name : "<unknown>");
1680 EXPORT_SYMBOL(netdev_rx_csum_fault);
1683 /* Actually, we should eliminate this check as soon as we know, that:
1684 * 1. IOMMU is present and allows to map all the memory.
1685 * 2. No high memory really exists on this machine.
1688 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1690 #ifdef CONFIG_HIGHMEM
1693 if (dev->features & NETIF_F_HIGHDMA)
1696 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1697 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1705 void (*destructor)(struct sk_buff *skb);
1708 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1710 static void dev_gso_skb_destructor(struct sk_buff *skb)
1712 struct dev_gso_cb *cb;
1715 struct sk_buff *nskb = skb->next;
1717 skb->next = nskb->next;
1720 } while (skb->next);
1722 cb = DEV_GSO_CB(skb);
1724 cb->destructor(skb);
1728 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1729 * @skb: buffer to segment
1731 * This function segments the given skb and stores the list of segments
1734 static int dev_gso_segment(struct sk_buff *skb)
1736 struct net_device *dev = skb->dev;
1737 struct sk_buff *segs;
1738 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1741 segs = skb_gso_segment(skb, features);
1743 /* Verifying header integrity only. */
1748 return PTR_ERR(segs);
1751 DEV_GSO_CB(skb)->destructor = skb->destructor;
1752 skb->destructor = dev_gso_skb_destructor;
1757 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1758 struct netdev_queue *txq)
1760 const struct net_device_ops *ops = dev->netdev_ops;
1761 int rc = NETDEV_TX_OK;
1763 if (likely(!skb->next)) {
1764 if (!list_empty(&ptype_all))
1765 dev_queue_xmit_nit(skb, dev);
1767 if (netif_needs_gso(dev, skb)) {
1768 if (unlikely(dev_gso_segment(skb)))
1775 * If device doesnt need skb->dst, release it right now while
1776 * its hot in this cpu cache
1778 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1781 rc = ops->ndo_start_xmit(skb, dev);
1782 if (rc == NETDEV_TX_OK)
1783 txq_trans_update(txq);
1785 * TODO: if skb_orphan() was called by
1786 * dev->hard_start_xmit() (for example, the unmodified
1787 * igb driver does that; bnx2 doesn't), then
1788 * skb_tx_software_timestamp() will be unable to send
1789 * back the time stamp.
1791 * How can this be prevented? Always create another
1792 * reference to the socket before calling
1793 * dev->hard_start_xmit()? Prevent that skb_orphan()
1794 * does anything in dev->hard_start_xmit() by clearing
1795 * the skb destructor before the call and restoring it
1796 * afterwards, then doing the skb_orphan() ourselves?
1803 struct sk_buff *nskb = skb->next;
1805 skb->next = nskb->next;
1807 rc = ops->ndo_start_xmit(nskb, dev);
1808 if (unlikely(rc != NETDEV_TX_OK)) {
1809 if (rc & ~NETDEV_TX_MASK)
1810 goto out_kfree_gso_skb;
1811 nskb->next = skb->next;
1815 txq_trans_update(txq);
1816 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1817 return NETDEV_TX_BUSY;
1818 } while (skb->next);
1821 if (likely(skb->next == NULL))
1822 skb->destructor = DEV_GSO_CB(skb)->destructor;
1828 static u32 skb_tx_hashrnd;
1830 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1834 if (skb_rx_queue_recorded(skb)) {
1835 hash = skb_get_rx_queue(skb);
1836 while (unlikely(hash >= dev->real_num_tx_queues))
1837 hash -= dev->real_num_tx_queues;
1841 if (skb->sk && skb->sk->sk_hash)
1842 hash = skb->sk->sk_hash;
1844 hash = skb->protocol;
1846 hash = jhash_1word(hash, skb_tx_hashrnd);
1848 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1850 EXPORT_SYMBOL(skb_tx_hash);
1852 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1854 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1855 if (net_ratelimit()) {
1856 WARN(1, "%s selects TX queue %d, but "
1857 "real number of TX queues is %d\n",
1858 dev->name, queue_index,
1859 dev->real_num_tx_queues);
1866 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1867 struct sk_buff *skb)
1870 struct sock *sk = skb->sk;
1872 if (sk_tx_queue_recorded(sk)) {
1873 queue_index = sk_tx_queue_get(sk);
1875 const struct net_device_ops *ops = dev->netdev_ops;
1877 if (ops->ndo_select_queue) {
1878 queue_index = ops->ndo_select_queue(dev, skb);
1879 queue_index = dev_cap_txqueue(dev, queue_index);
1882 if (dev->real_num_tx_queues > 1)
1883 queue_index = skb_tx_hash(dev, skb);
1885 if (sk && sk->sk_dst_cache)
1886 sk_tx_queue_set(sk, queue_index);
1890 skb_set_queue_mapping(skb, queue_index);
1891 return netdev_get_tx_queue(dev, queue_index);
1894 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
1895 struct net_device *dev,
1896 struct netdev_queue *txq)
1898 spinlock_t *root_lock = qdisc_lock(q);
1901 spin_lock(root_lock);
1902 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1905 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
1906 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
1908 * This is a work-conserving queue; there are no old skbs
1909 * waiting to be sent out; and the qdisc is not running -
1910 * xmit the skb directly.
1912 __qdisc_update_bstats(q, skb->len);
1913 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
1916 clear_bit(__QDISC_STATE_RUNNING, &q->state);
1918 rc = NET_XMIT_SUCCESS;
1920 rc = qdisc_enqueue_root(skb, q);
1923 spin_unlock(root_lock);
1929 * dev_queue_xmit - transmit a buffer
1930 * @skb: buffer to transmit
1932 * Queue a buffer for transmission to a network device. The caller must
1933 * have set the device and priority and built the buffer before calling
1934 * this function. The function can be called from an interrupt.
1936 * A negative errno code is returned on a failure. A success does not
1937 * guarantee the frame will be transmitted as it may be dropped due
1938 * to congestion or traffic shaping.
1940 * -----------------------------------------------------------------------------------
1941 * I notice this method can also return errors from the queue disciplines,
1942 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1945 * Regardless of the return value, the skb is consumed, so it is currently
1946 * difficult to retry a send to this method. (You can bump the ref count
1947 * before sending to hold a reference for retry if you are careful.)
1949 * When calling this method, interrupts MUST be enabled. This is because
1950 * the BH enable code must have IRQs enabled so that it will not deadlock.
1953 int dev_queue_xmit(struct sk_buff *skb)
1955 struct net_device *dev = skb->dev;
1956 struct netdev_queue *txq;
1960 /* GSO will handle the following emulations directly. */
1961 if (netif_needs_gso(dev, skb))
1964 if (skb_has_frags(skb) &&
1965 !(dev->features & NETIF_F_FRAGLIST) &&
1966 __skb_linearize(skb))
1969 /* Fragmented skb is linearized if device does not support SG,
1970 * or if at least one of fragments is in highmem and device
1971 * does not support DMA from it.
1973 if (skb_shinfo(skb)->nr_frags &&
1974 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1975 __skb_linearize(skb))
1978 /* If packet is not checksummed and device does not support
1979 * checksumming for this protocol, complete checksumming here.
1981 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1982 skb_set_transport_header(skb, skb->csum_start -
1984 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1989 /* Disable soft irqs for various locks below. Also
1990 * stops preemption for RCU.
1994 txq = dev_pick_tx(dev, skb);
1995 q = rcu_dereference(txq->qdisc);
1997 #ifdef CONFIG_NET_CLS_ACT
1998 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2001 rc = __dev_xmit_skb(skb, q, dev, txq);
2005 /* The device has no queue. Common case for software devices:
2006 loopback, all the sorts of tunnels...
2008 Really, it is unlikely that netif_tx_lock protection is necessary
2009 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2011 However, it is possible, that they rely on protection
2014 Check this and shot the lock. It is not prone from deadlocks.
2015 Either shot noqueue qdisc, it is even simpler 8)
2017 if (dev->flags & IFF_UP) {
2018 int cpu = smp_processor_id(); /* ok because BHs are off */
2020 if (txq->xmit_lock_owner != cpu) {
2022 HARD_TX_LOCK(dev, txq, cpu);
2024 if (!netif_tx_queue_stopped(txq)) {
2025 rc = dev_hard_start_xmit(skb, dev, txq);
2026 if (dev_xmit_complete(rc)) {
2027 HARD_TX_UNLOCK(dev, txq);
2031 HARD_TX_UNLOCK(dev, txq);
2032 if (net_ratelimit())
2033 printk(KERN_CRIT "Virtual device %s asks to "
2034 "queue packet!\n", dev->name);
2036 /* Recursion is detected! It is possible,
2038 if (net_ratelimit())
2039 printk(KERN_CRIT "Dead loop on virtual device "
2040 "%s, fix it urgently!\n", dev->name);
2045 rcu_read_unlock_bh();
2051 rcu_read_unlock_bh();
2054 EXPORT_SYMBOL(dev_queue_xmit);
2057 /*=======================================================================
2059 =======================================================================*/
2061 int netdev_max_backlog __read_mostly = 1000;
2062 int netdev_budget __read_mostly = 300;
2063 int weight_p __read_mostly = 64; /* old backlog weight */
2065 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2069 * netif_rx - post buffer to the network code
2070 * @skb: buffer to post
2072 * This function receives a packet from a device driver and queues it for
2073 * the upper (protocol) levels to process. It always succeeds. The buffer
2074 * may be dropped during processing for congestion control or by the
2078 * NET_RX_SUCCESS (no congestion)
2079 * NET_RX_DROP (packet was dropped)
2083 int netif_rx(struct sk_buff *skb)
2085 struct softnet_data *queue;
2086 unsigned long flags;
2088 /* if netpoll wants it, pretend we never saw it */
2089 if (netpoll_rx(skb))
2092 if (!skb->tstamp.tv64)
2096 * The code is rearranged so that the path is the most
2097 * short when CPU is congested, but is still operating.
2099 local_irq_save(flags);
2100 queue = &__get_cpu_var(softnet_data);
2102 __get_cpu_var(netdev_rx_stat).total++;
2103 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2104 if (queue->input_pkt_queue.qlen) {
2106 __skb_queue_tail(&queue->input_pkt_queue, skb);
2107 local_irq_restore(flags);
2108 return NET_RX_SUCCESS;
2111 napi_schedule(&queue->backlog);
2115 __get_cpu_var(netdev_rx_stat).dropped++;
2116 local_irq_restore(flags);
2121 EXPORT_SYMBOL(netif_rx);
2123 int netif_rx_ni(struct sk_buff *skb)
2128 err = netif_rx(skb);
2129 if (local_softirq_pending())
2135 EXPORT_SYMBOL(netif_rx_ni);
2137 static void net_tx_action(struct softirq_action *h)
2139 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2141 if (sd->completion_queue) {
2142 struct sk_buff *clist;
2144 local_irq_disable();
2145 clist = sd->completion_queue;
2146 sd->completion_queue = NULL;
2150 struct sk_buff *skb = clist;
2151 clist = clist->next;
2153 WARN_ON(atomic_read(&skb->users));
2158 if (sd->output_queue) {
2161 local_irq_disable();
2162 head = sd->output_queue;
2163 sd->output_queue = NULL;
2167 struct Qdisc *q = head;
2168 spinlock_t *root_lock;
2170 head = head->next_sched;
2172 root_lock = qdisc_lock(q);
2173 if (spin_trylock(root_lock)) {
2174 smp_mb__before_clear_bit();
2175 clear_bit(__QDISC_STATE_SCHED,
2178 spin_unlock(root_lock);
2180 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2182 __netif_reschedule(q);
2184 smp_mb__before_clear_bit();
2185 clear_bit(__QDISC_STATE_SCHED,
2193 static inline int deliver_skb(struct sk_buff *skb,
2194 struct packet_type *pt_prev,
2195 struct net_device *orig_dev)
2197 atomic_inc(&skb->users);
2198 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2201 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2203 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2204 /* This hook is defined here for ATM LANE */
2205 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2206 unsigned char *addr) __read_mostly;
2207 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2211 * If bridge module is loaded call bridging hook.
2212 * returns NULL if packet was consumed.
2214 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2215 struct sk_buff *skb) __read_mostly;
2216 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2218 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2219 struct packet_type **pt_prev, int *ret,
2220 struct net_device *orig_dev)
2222 struct net_bridge_port *port;
2224 if (skb->pkt_type == PACKET_LOOPBACK ||
2225 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2229 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2233 return br_handle_frame_hook(port, skb);
2236 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2239 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2240 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2241 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2243 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2244 struct packet_type **pt_prev,
2246 struct net_device *orig_dev)
2248 if (skb->dev->macvlan_port == NULL)
2252 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2255 return macvlan_handle_frame_hook(skb);
2258 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2261 #ifdef CONFIG_NET_CLS_ACT
2262 /* TODO: Maybe we should just force sch_ingress to be compiled in
2263 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2264 * a compare and 2 stores extra right now if we dont have it on
2265 * but have CONFIG_NET_CLS_ACT
2266 * NOTE: This doesnt stop any functionality; if you dont have
2267 * the ingress scheduler, you just cant add policies on ingress.
2270 static int ing_filter(struct sk_buff *skb)
2272 struct net_device *dev = skb->dev;
2273 u32 ttl = G_TC_RTTL(skb->tc_verd);
2274 struct netdev_queue *rxq;
2275 int result = TC_ACT_OK;
2278 if (MAX_RED_LOOP < ttl++) {
2280 "Redir loop detected Dropping packet (%d->%d)\n",
2281 skb->iif, dev->ifindex);
2285 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2286 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2288 rxq = &dev->rx_queue;
2291 if (q != &noop_qdisc) {
2292 spin_lock(qdisc_lock(q));
2293 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2294 result = qdisc_enqueue_root(skb, q);
2295 spin_unlock(qdisc_lock(q));
2301 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2302 struct packet_type **pt_prev,
2303 int *ret, struct net_device *orig_dev)
2305 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2309 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2312 /* Huh? Why does turning on AF_PACKET affect this? */
2313 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2316 switch (ing_filter(skb)) {
2330 * netif_nit_deliver - deliver received packets to network taps
2333 * This function is used to deliver incoming packets to network
2334 * taps. It should be used when the normal netif_receive_skb path
2335 * is bypassed, for example because of VLAN acceleration.
2337 void netif_nit_deliver(struct sk_buff *skb)
2339 struct packet_type *ptype;
2341 if (list_empty(&ptype_all))
2344 skb_reset_network_header(skb);
2345 skb_reset_transport_header(skb);
2346 skb->mac_len = skb->network_header - skb->mac_header;
2349 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2350 if (!ptype->dev || ptype->dev == skb->dev)
2351 deliver_skb(skb, ptype, skb->dev);
2357 * netif_receive_skb - process receive buffer from network
2358 * @skb: buffer to process
2360 * netif_receive_skb() is the main receive data processing function.
2361 * It always succeeds. The buffer may be dropped during processing
2362 * for congestion control or by the protocol layers.
2364 * This function may only be called from softirq context and interrupts
2365 * should be enabled.
2367 * Return values (usually ignored):
2368 * NET_RX_SUCCESS: no congestion
2369 * NET_RX_DROP: packet was dropped
2371 int netif_receive_skb(struct sk_buff *skb)
2373 struct packet_type *ptype, *pt_prev;
2374 struct net_device *orig_dev;
2375 struct net_device *null_or_orig;
2376 int ret = NET_RX_DROP;
2379 if (!skb->tstamp.tv64)
2382 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2383 return NET_RX_SUCCESS;
2385 /* if we've gotten here through NAPI, check netpoll */
2386 if (netpoll_receive_skb(skb))
2390 skb->iif = skb->dev->ifindex;
2392 null_or_orig = NULL;
2393 orig_dev = skb->dev;
2394 if (orig_dev->master) {
2395 if (skb_bond_should_drop(skb))
2396 null_or_orig = orig_dev; /* deliver only exact match */
2398 skb->dev = orig_dev->master;
2401 __get_cpu_var(netdev_rx_stat).total++;
2403 skb_reset_network_header(skb);
2404 skb_reset_transport_header(skb);
2405 skb->mac_len = skb->network_header - skb->mac_header;
2411 #ifdef CONFIG_NET_CLS_ACT
2412 if (skb->tc_verd & TC_NCLS) {
2413 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2418 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2419 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2420 ptype->dev == orig_dev) {
2422 ret = deliver_skb(skb, pt_prev, orig_dev);
2427 #ifdef CONFIG_NET_CLS_ACT
2428 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2434 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2437 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2441 type = skb->protocol;
2442 list_for_each_entry_rcu(ptype,
2443 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2444 if (ptype->type == type &&
2445 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2446 ptype->dev == orig_dev)) {
2448 ret = deliver_skb(skb, pt_prev, orig_dev);
2454 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2457 /* Jamal, now you will not able to escape explaining
2458 * me how you were going to use this. :-)
2467 EXPORT_SYMBOL(netif_receive_skb);
2469 /* Network device is going away, flush any packets still pending */
2470 static void flush_backlog(void *arg)
2472 struct net_device *dev = arg;
2473 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2474 struct sk_buff *skb, *tmp;
2476 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2477 if (skb->dev == dev) {
2478 __skb_unlink(skb, &queue->input_pkt_queue);
2483 static int napi_gro_complete(struct sk_buff *skb)
2485 struct packet_type *ptype;
2486 __be16 type = skb->protocol;
2487 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2490 if (NAPI_GRO_CB(skb)->count == 1) {
2491 skb_shinfo(skb)->gso_size = 0;
2496 list_for_each_entry_rcu(ptype, head, list) {
2497 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2500 err = ptype->gro_complete(skb);
2506 WARN_ON(&ptype->list == head);
2508 return NET_RX_SUCCESS;
2512 return netif_receive_skb(skb);
2515 void napi_gro_flush(struct napi_struct *napi)
2517 struct sk_buff *skb, *next;
2519 for (skb = napi->gro_list; skb; skb = next) {
2522 napi_gro_complete(skb);
2525 napi->gro_count = 0;
2526 napi->gro_list = NULL;
2528 EXPORT_SYMBOL(napi_gro_flush);
2530 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2532 struct sk_buff **pp = NULL;
2533 struct packet_type *ptype;
2534 __be16 type = skb->protocol;
2535 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2538 enum gro_result ret;
2540 if (!(skb->dev->features & NETIF_F_GRO))
2543 if (skb_is_gso(skb) || skb_has_frags(skb))
2547 list_for_each_entry_rcu(ptype, head, list) {
2548 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2551 skb_set_network_header(skb, skb_gro_offset(skb));
2552 mac_len = skb->network_header - skb->mac_header;
2553 skb->mac_len = mac_len;
2554 NAPI_GRO_CB(skb)->same_flow = 0;
2555 NAPI_GRO_CB(skb)->flush = 0;
2556 NAPI_GRO_CB(skb)->free = 0;
2558 pp = ptype->gro_receive(&napi->gro_list, skb);
2563 if (&ptype->list == head)
2566 same_flow = NAPI_GRO_CB(skb)->same_flow;
2567 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2570 struct sk_buff *nskb = *pp;
2574 napi_gro_complete(nskb);
2581 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2585 NAPI_GRO_CB(skb)->count = 1;
2586 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2587 skb->next = napi->gro_list;
2588 napi->gro_list = skb;
2592 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2593 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2595 BUG_ON(skb->end - skb->tail < grow);
2597 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2600 skb->data_len -= grow;
2602 skb_shinfo(skb)->frags[0].page_offset += grow;
2603 skb_shinfo(skb)->frags[0].size -= grow;
2605 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2606 put_page(skb_shinfo(skb)->frags[0].page);
2607 memmove(skb_shinfo(skb)->frags,
2608 skb_shinfo(skb)->frags + 1,
2609 --skb_shinfo(skb)->nr_frags);
2620 EXPORT_SYMBOL(dev_gro_receive);
2623 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2627 if (netpoll_rx_on(skb))
2630 for (p = napi->gro_list; p; p = p->next) {
2631 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2632 && !compare_ether_header(skb_mac_header(p),
2633 skb_gro_mac_header(skb));
2634 NAPI_GRO_CB(p)->flush = 0;
2637 return dev_gro_receive(napi, skb);
2640 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
2644 if (netif_receive_skb(skb))
2649 case GRO_MERGED_FREE:
2660 EXPORT_SYMBOL(napi_skb_finish);
2662 void skb_gro_reset_offset(struct sk_buff *skb)
2664 NAPI_GRO_CB(skb)->data_offset = 0;
2665 NAPI_GRO_CB(skb)->frag0 = NULL;
2666 NAPI_GRO_CB(skb)->frag0_len = 0;
2668 if (skb->mac_header == skb->tail &&
2669 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2670 NAPI_GRO_CB(skb)->frag0 =
2671 page_address(skb_shinfo(skb)->frags[0].page) +
2672 skb_shinfo(skb)->frags[0].page_offset;
2673 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2676 EXPORT_SYMBOL(skb_gro_reset_offset);
2678 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2680 skb_gro_reset_offset(skb);
2682 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2684 EXPORT_SYMBOL(napi_gro_receive);
2686 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2688 __skb_pull(skb, skb_headlen(skb));
2689 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2693 EXPORT_SYMBOL(napi_reuse_skb);
2695 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2697 struct sk_buff *skb = napi->skb;
2700 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
2706 EXPORT_SYMBOL(napi_get_frags);
2708 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
2714 skb->protocol = eth_type_trans(skb, napi->dev);
2716 if (ret == GRO_HELD)
2717 skb_gro_pull(skb, -ETH_HLEN);
2718 else if (netif_receive_skb(skb))
2723 case GRO_MERGED_FREE:
2724 napi_reuse_skb(napi, skb);
2733 EXPORT_SYMBOL(napi_frags_finish);
2735 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2737 struct sk_buff *skb = napi->skb;
2744 skb_reset_mac_header(skb);
2745 skb_gro_reset_offset(skb);
2747 off = skb_gro_offset(skb);
2748 hlen = off + sizeof(*eth);
2749 eth = skb_gro_header_fast(skb, off);
2750 if (skb_gro_header_hard(skb, hlen)) {
2751 eth = skb_gro_header_slow(skb, hlen, off);
2752 if (unlikely(!eth)) {
2753 napi_reuse_skb(napi, skb);
2759 skb_gro_pull(skb, sizeof(*eth));
2762 * This works because the only protocols we care about don't require
2763 * special handling. We'll fix it up properly at the end.
2765 skb->protocol = eth->h_proto;
2770 EXPORT_SYMBOL(napi_frags_skb);
2772 gro_result_t napi_gro_frags(struct napi_struct *napi)
2774 struct sk_buff *skb = napi_frags_skb(napi);
2779 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2781 EXPORT_SYMBOL(napi_gro_frags);
2783 static int process_backlog(struct napi_struct *napi, int quota)
2786 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2787 unsigned long start_time = jiffies;
2789 napi->weight = weight_p;
2791 struct sk_buff *skb;
2793 local_irq_disable();
2794 skb = __skb_dequeue(&queue->input_pkt_queue);
2796 __napi_complete(napi);
2802 netif_receive_skb(skb);
2803 } while (++work < quota && jiffies == start_time);
2809 * __napi_schedule - schedule for receive
2810 * @n: entry to schedule
2812 * The entry's receive function will be scheduled to run
2814 void __napi_schedule(struct napi_struct *n)
2816 unsigned long flags;
2818 local_irq_save(flags);
2819 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2820 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2821 local_irq_restore(flags);
2823 EXPORT_SYMBOL(__napi_schedule);
2825 void __napi_complete(struct napi_struct *n)
2827 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2828 BUG_ON(n->gro_list);
2830 list_del(&n->poll_list);
2831 smp_mb__before_clear_bit();
2832 clear_bit(NAPI_STATE_SCHED, &n->state);
2834 EXPORT_SYMBOL(__napi_complete);
2836 void napi_complete(struct napi_struct *n)
2838 unsigned long flags;
2841 * don't let napi dequeue from the cpu poll list
2842 * just in case its running on a different cpu
2844 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2848 local_irq_save(flags);
2850 local_irq_restore(flags);
2852 EXPORT_SYMBOL(napi_complete);
2854 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2855 int (*poll)(struct napi_struct *, int), int weight)
2857 INIT_LIST_HEAD(&napi->poll_list);
2858 napi->gro_count = 0;
2859 napi->gro_list = NULL;
2862 napi->weight = weight;
2863 list_add(&napi->dev_list, &dev->napi_list);
2865 #ifdef CONFIG_NETPOLL
2866 spin_lock_init(&napi->poll_lock);
2867 napi->poll_owner = -1;
2869 set_bit(NAPI_STATE_SCHED, &napi->state);
2871 EXPORT_SYMBOL(netif_napi_add);
2873 void netif_napi_del(struct napi_struct *napi)
2875 struct sk_buff *skb, *next;
2877 list_del_init(&napi->dev_list);
2878 napi_free_frags(napi);
2880 for (skb = napi->gro_list; skb; skb = next) {
2886 napi->gro_list = NULL;
2887 napi->gro_count = 0;
2889 EXPORT_SYMBOL(netif_napi_del);
2892 static void net_rx_action(struct softirq_action *h)
2894 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2895 unsigned long time_limit = jiffies + 2;
2896 int budget = netdev_budget;
2899 local_irq_disable();
2901 while (!list_empty(list)) {
2902 struct napi_struct *n;
2905 /* If softirq window is exhuasted then punt.
2906 * Allow this to run for 2 jiffies since which will allow
2907 * an average latency of 1.5/HZ.
2909 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2914 /* Even though interrupts have been re-enabled, this
2915 * access is safe because interrupts can only add new
2916 * entries to the tail of this list, and only ->poll()
2917 * calls can remove this head entry from the list.
2919 n = list_entry(list->next, struct napi_struct, poll_list);
2921 have = netpoll_poll_lock(n);
2925 /* This NAPI_STATE_SCHED test is for avoiding a race
2926 * with netpoll's poll_napi(). Only the entity which
2927 * obtains the lock and sees NAPI_STATE_SCHED set will
2928 * actually make the ->poll() call. Therefore we avoid
2929 * accidently calling ->poll() when NAPI is not scheduled.
2932 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
2933 work = n->poll(n, weight);
2937 WARN_ON_ONCE(work > weight);
2941 local_irq_disable();
2943 /* Drivers must not modify the NAPI state if they
2944 * consume the entire weight. In such cases this code
2945 * still "owns" the NAPI instance and therefore can
2946 * move the instance around on the list at-will.
2948 if (unlikely(work == weight)) {
2949 if (unlikely(napi_disable_pending(n))) {
2952 local_irq_disable();
2954 list_move_tail(&n->poll_list, list);
2957 netpoll_poll_unlock(have);
2962 #ifdef CONFIG_NET_DMA
2964 * There may not be any more sk_buffs coming right now, so push
2965 * any pending DMA copies to hardware
2967 dma_issue_pending_all();
2973 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2974 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2978 static gifconf_func_t *gifconf_list[NPROTO];
2981 * register_gifconf - register a SIOCGIF handler
2982 * @family: Address family
2983 * @gifconf: Function handler
2985 * Register protocol dependent address dumping routines. The handler
2986 * that is passed must not be freed or reused until it has been replaced
2987 * by another handler.
2989 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
2991 if (family >= NPROTO)
2993 gifconf_list[family] = gifconf;
2996 EXPORT_SYMBOL(register_gifconf);
3000 * Map an interface index to its name (SIOCGIFNAME)
3004 * We need this ioctl for efficient implementation of the
3005 * if_indextoname() function required by the IPv6 API. Without
3006 * it, we would have to search all the interfaces to find a
3010 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3012 struct net_device *dev;
3016 * Fetch the caller's info block.
3019 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3023 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3029 strcpy(ifr.ifr_name, dev->name);
3032 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3038 * Perform a SIOCGIFCONF call. This structure will change
3039 * size eventually, and there is nothing I can do about it.
3040 * Thus we will need a 'compatibility mode'.
3043 static int dev_ifconf(struct net *net, char __user *arg)
3046 struct net_device *dev;
3053 * Fetch the caller's info block.
3056 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3063 * Loop over the interfaces, and write an info block for each.
3067 for_each_netdev(net, dev) {
3068 for (i = 0; i < NPROTO; i++) {
3069 if (gifconf_list[i]) {
3072 done = gifconf_list[i](dev, NULL, 0);
3074 done = gifconf_list[i](dev, pos + total,
3084 * All done. Write the updated control block back to the caller.
3086 ifc.ifc_len = total;
3089 * Both BSD and Solaris return 0 here, so we do too.
3091 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3094 #ifdef CONFIG_PROC_FS
3096 * This is invoked by the /proc filesystem handler to display a device
3099 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3102 struct net *net = seq_file_net(seq);
3104 struct net_device *dev;
3108 return SEQ_START_TOKEN;
3111 for_each_netdev_rcu(net, dev)
3118 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3120 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3121 first_net_device(seq_file_net(seq)) :
3122 next_net_device((struct net_device *)v);
3125 return rcu_dereference(dev);
3128 void dev_seq_stop(struct seq_file *seq, void *v)
3134 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3136 const struct net_device_stats *stats = dev_get_stats(dev);
3138 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3139 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3140 dev->name, stats->rx_bytes, stats->rx_packets,
3142 stats->rx_dropped + stats->rx_missed_errors,
3143 stats->rx_fifo_errors,
3144 stats->rx_length_errors + stats->rx_over_errors +
3145 stats->rx_crc_errors + stats->rx_frame_errors,
3146 stats->rx_compressed, stats->multicast,
3147 stats->tx_bytes, stats->tx_packets,
3148 stats->tx_errors, stats->tx_dropped,
3149 stats->tx_fifo_errors, stats->collisions,
3150 stats->tx_carrier_errors +
3151 stats->tx_aborted_errors +
3152 stats->tx_window_errors +
3153 stats->tx_heartbeat_errors,
3154 stats->tx_compressed);
3158 * Called from the PROCfs module. This now uses the new arbitrary sized
3159 * /proc/net interface to create /proc/net/dev
3161 static int dev_seq_show(struct seq_file *seq, void *v)
3163 if (v == SEQ_START_TOKEN)
3164 seq_puts(seq, "Inter-| Receive "
3166 " face |bytes packets errs drop fifo frame "
3167 "compressed multicast|bytes packets errs "
3168 "drop fifo colls carrier compressed\n");
3170 dev_seq_printf_stats(seq, v);
3174 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3176 struct netif_rx_stats *rc = NULL;
3178 while (*pos < nr_cpu_ids)
3179 if (cpu_online(*pos)) {
3180 rc = &per_cpu(netdev_rx_stat, *pos);
3187 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3189 return softnet_get_online(pos);
3192 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3195 return softnet_get_online(pos);
3198 static void softnet_seq_stop(struct seq_file *seq, void *v)
3202 static int softnet_seq_show(struct seq_file *seq, void *v)
3204 struct netif_rx_stats *s = v;
3206 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3207 s->total, s->dropped, s->time_squeeze, 0,
3208 0, 0, 0, 0, /* was fastroute */
3213 static const struct seq_operations dev_seq_ops = {
3214 .start = dev_seq_start,
3215 .next = dev_seq_next,
3216 .stop = dev_seq_stop,
3217 .show = dev_seq_show,
3220 static int dev_seq_open(struct inode *inode, struct file *file)
3222 return seq_open_net(inode, file, &dev_seq_ops,
3223 sizeof(struct seq_net_private));
3226 static const struct file_operations dev_seq_fops = {
3227 .owner = THIS_MODULE,
3228 .open = dev_seq_open,
3230 .llseek = seq_lseek,
3231 .release = seq_release_net,
3234 static const struct seq_operations softnet_seq_ops = {
3235 .start = softnet_seq_start,
3236 .next = softnet_seq_next,
3237 .stop = softnet_seq_stop,
3238 .show = softnet_seq_show,
3241 static int softnet_seq_open(struct inode *inode, struct file *file)
3243 return seq_open(file, &softnet_seq_ops);
3246 static const struct file_operations softnet_seq_fops = {
3247 .owner = THIS_MODULE,
3248 .open = softnet_seq_open,
3250 .llseek = seq_lseek,
3251 .release = seq_release,
3254 static void *ptype_get_idx(loff_t pos)
3256 struct packet_type *pt = NULL;
3260 list_for_each_entry_rcu(pt, &ptype_all, list) {
3266 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3267 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3276 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3280 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3283 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3285 struct packet_type *pt;
3286 struct list_head *nxt;
3290 if (v == SEQ_START_TOKEN)
3291 return ptype_get_idx(0);
3294 nxt = pt->list.next;
3295 if (pt->type == htons(ETH_P_ALL)) {
3296 if (nxt != &ptype_all)
3299 nxt = ptype_base[0].next;
3301 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3303 while (nxt == &ptype_base[hash]) {
3304 if (++hash >= PTYPE_HASH_SIZE)
3306 nxt = ptype_base[hash].next;
3309 return list_entry(nxt, struct packet_type, list);
3312 static void ptype_seq_stop(struct seq_file *seq, void *v)
3318 static int ptype_seq_show(struct seq_file *seq, void *v)
3320 struct packet_type *pt = v;
3322 if (v == SEQ_START_TOKEN)
3323 seq_puts(seq, "Type Device Function\n");
3324 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3325 if (pt->type == htons(ETH_P_ALL))
3326 seq_puts(seq, "ALL ");
3328 seq_printf(seq, "%04x", ntohs(pt->type));
3330 seq_printf(seq, " %-8s %pF\n",
3331 pt->dev ? pt->dev->name : "", pt->func);
3337 static const struct seq_operations ptype_seq_ops = {
3338 .start = ptype_seq_start,
3339 .next = ptype_seq_next,
3340 .stop = ptype_seq_stop,
3341 .show = ptype_seq_show,
3344 static int ptype_seq_open(struct inode *inode, struct file *file)
3346 return seq_open_net(inode, file, &ptype_seq_ops,
3347 sizeof(struct seq_net_private));
3350 static const struct file_operations ptype_seq_fops = {
3351 .owner = THIS_MODULE,
3352 .open = ptype_seq_open,
3354 .llseek = seq_lseek,
3355 .release = seq_release_net,
3359 static int __net_init dev_proc_net_init(struct net *net)
3363 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3365 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3367 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3370 if (wext_proc_init(net))
3376 proc_net_remove(net, "ptype");
3378 proc_net_remove(net, "softnet_stat");
3380 proc_net_remove(net, "dev");
3384 static void __net_exit dev_proc_net_exit(struct net *net)
3386 wext_proc_exit(net);
3388 proc_net_remove(net, "ptype");
3389 proc_net_remove(net, "softnet_stat");
3390 proc_net_remove(net, "dev");
3393 static struct pernet_operations __net_initdata dev_proc_ops = {
3394 .init = dev_proc_net_init,
3395 .exit = dev_proc_net_exit,
3398 static int __init dev_proc_init(void)
3400 return register_pernet_subsys(&dev_proc_ops);
3403 #define dev_proc_init() 0
3404 #endif /* CONFIG_PROC_FS */
3408 * netdev_set_master - set up master/slave pair
3409 * @slave: slave device
3410 * @master: new master device
3412 * Changes the master device of the slave. Pass %NULL to break the
3413 * bonding. The caller must hold the RTNL semaphore. On a failure
3414 * a negative errno code is returned. On success the reference counts
3415 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3416 * function returns zero.
3418 int netdev_set_master(struct net_device *slave, struct net_device *master)
3420 struct net_device *old = slave->master;
3430 slave->master = master;
3438 slave->flags |= IFF_SLAVE;
3440 slave->flags &= ~IFF_SLAVE;
3442 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3445 EXPORT_SYMBOL(netdev_set_master);
3447 static void dev_change_rx_flags(struct net_device *dev, int flags)
3449 const struct net_device_ops *ops = dev->netdev_ops;
3451 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3452 ops->ndo_change_rx_flags(dev, flags);
3455 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3457 unsigned short old_flags = dev->flags;
3463 dev->flags |= IFF_PROMISC;
3464 dev->promiscuity += inc;
3465 if (dev->promiscuity == 0) {
3468 * If inc causes overflow, untouch promisc and return error.
3471 dev->flags &= ~IFF_PROMISC;
3473 dev->promiscuity -= inc;
3474 printk(KERN_WARNING "%s: promiscuity touches roof, "
3475 "set promiscuity failed, promiscuity feature "
3476 "of device might be broken.\n", dev->name);
3480 if (dev->flags != old_flags) {
3481 printk(KERN_INFO "device %s %s promiscuous mode\n",
3482 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3484 if (audit_enabled) {
3485 current_uid_gid(&uid, &gid);
3486 audit_log(current->audit_context, GFP_ATOMIC,
3487 AUDIT_ANOM_PROMISCUOUS,
3488 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3489 dev->name, (dev->flags & IFF_PROMISC),
3490 (old_flags & IFF_PROMISC),
3491 audit_get_loginuid(current),
3493 audit_get_sessionid(current));
3496 dev_change_rx_flags(dev, IFF_PROMISC);
3502 * dev_set_promiscuity - update promiscuity count on a device
3506 * Add or remove promiscuity from a device. While the count in the device
3507 * remains above zero the interface remains promiscuous. Once it hits zero
3508 * the device reverts back to normal filtering operation. A negative inc
3509 * value is used to drop promiscuity on the device.
3510 * Return 0 if successful or a negative errno code on error.
3512 int dev_set_promiscuity(struct net_device *dev, int inc)
3514 unsigned short old_flags = dev->flags;
3517 err = __dev_set_promiscuity(dev, inc);
3520 if (dev->flags != old_flags)
3521 dev_set_rx_mode(dev);
3524 EXPORT_SYMBOL(dev_set_promiscuity);
3527 * dev_set_allmulti - update allmulti count on a device
3531 * Add or remove reception of all multicast frames to a device. While the
3532 * count in the device remains above zero the interface remains listening
3533 * to all interfaces. Once it hits zero the device reverts back to normal
3534 * filtering operation. A negative @inc value is used to drop the counter
3535 * when releasing a resource needing all multicasts.
3536 * Return 0 if successful or a negative errno code on error.
3539 int dev_set_allmulti(struct net_device *dev, int inc)
3541 unsigned short old_flags = dev->flags;
3545 dev->flags |= IFF_ALLMULTI;
3546 dev->allmulti += inc;
3547 if (dev->allmulti == 0) {
3550 * If inc causes overflow, untouch allmulti and return error.
3553 dev->flags &= ~IFF_ALLMULTI;
3555 dev->allmulti -= inc;
3556 printk(KERN_WARNING "%s: allmulti touches roof, "
3557 "set allmulti failed, allmulti feature of "
3558 "device might be broken.\n", dev->name);
3562 if (dev->flags ^ old_flags) {
3563 dev_change_rx_flags(dev, IFF_ALLMULTI);
3564 dev_set_rx_mode(dev);
3568 EXPORT_SYMBOL(dev_set_allmulti);
3571 * Upload unicast and multicast address lists to device and
3572 * configure RX filtering. When the device doesn't support unicast
3573 * filtering it is put in promiscuous mode while unicast addresses
3576 void __dev_set_rx_mode(struct net_device *dev)
3578 const struct net_device_ops *ops = dev->netdev_ops;
3580 /* dev_open will call this function so the list will stay sane. */
3581 if (!(dev->flags&IFF_UP))
3584 if (!netif_device_present(dev))
3587 if (ops->ndo_set_rx_mode)
3588 ops->ndo_set_rx_mode(dev);
3590 /* Unicast addresses changes may only happen under the rtnl,
3591 * therefore calling __dev_set_promiscuity here is safe.
3593 if (dev->uc.count > 0 && !dev->uc_promisc) {
3594 __dev_set_promiscuity(dev, 1);
3595 dev->uc_promisc = 1;
3596 } else if (dev->uc.count == 0 && dev->uc_promisc) {
3597 __dev_set_promiscuity(dev, -1);
3598 dev->uc_promisc = 0;
3601 if (ops->ndo_set_multicast_list)
3602 ops->ndo_set_multicast_list(dev);
3606 void dev_set_rx_mode(struct net_device *dev)
3608 netif_addr_lock_bh(dev);
3609 __dev_set_rx_mode(dev);
3610 netif_addr_unlock_bh(dev);
3613 /* hw addresses list handling functions */
3615 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3616 int addr_len, unsigned char addr_type)
3618 struct netdev_hw_addr *ha;
3621 if (addr_len > MAX_ADDR_LEN)
3624 list_for_each_entry(ha, &list->list, list) {
3625 if (!memcmp(ha->addr, addr, addr_len) &&
3626 ha->type == addr_type) {
3633 alloc_size = sizeof(*ha);
3634 if (alloc_size < L1_CACHE_BYTES)
3635 alloc_size = L1_CACHE_BYTES;
3636 ha = kmalloc(alloc_size, GFP_ATOMIC);
3639 memcpy(ha->addr, addr, addr_len);
3640 ha->type = addr_type;
3643 list_add_tail_rcu(&ha->list, &list->list);
3648 static void ha_rcu_free(struct rcu_head *head)
3650 struct netdev_hw_addr *ha;
3652 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3656 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3657 int addr_len, unsigned char addr_type)
3659 struct netdev_hw_addr *ha;
3661 list_for_each_entry(ha, &list->list, list) {
3662 if (!memcmp(ha->addr, addr, addr_len) &&
3663 (ha->type == addr_type || !addr_type)) {
3666 list_del_rcu(&ha->list);
3667 call_rcu(&ha->rcu_head, ha_rcu_free);
3675 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3676 struct netdev_hw_addr_list *from_list,
3678 unsigned char addr_type)
3681 struct netdev_hw_addr *ha, *ha2;
3684 list_for_each_entry(ha, &from_list->list, list) {
3685 type = addr_type ? addr_type : ha->type;
3686 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3693 list_for_each_entry(ha2, &from_list->list, list) {
3696 type = addr_type ? addr_type : ha2->type;
3697 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3702 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3703 struct netdev_hw_addr_list *from_list,
3705 unsigned char addr_type)
3707 struct netdev_hw_addr *ha;
3710 list_for_each_entry(ha, &from_list->list, list) {
3711 type = addr_type ? addr_type : ha->type;
3712 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3716 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3717 struct netdev_hw_addr_list *from_list,
3721 struct netdev_hw_addr *ha, *tmp;
3723 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3725 err = __hw_addr_add(to_list, ha->addr,
3726 addr_len, ha->type);
3731 } else if (ha->refcount == 1) {
3732 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3733 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3739 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3740 struct netdev_hw_addr_list *from_list,
3743 struct netdev_hw_addr *ha, *tmp;
3745 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3747 __hw_addr_del(to_list, ha->addr,
3748 addr_len, ha->type);
3750 __hw_addr_del(from_list, ha->addr,
3751 addr_len, ha->type);
3756 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3758 struct netdev_hw_addr *ha, *tmp;
3760 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3761 list_del_rcu(&ha->list);
3762 call_rcu(&ha->rcu_head, ha_rcu_free);
3767 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3769 INIT_LIST_HEAD(&list->list);
3773 /* Device addresses handling functions */
3775 static void dev_addr_flush(struct net_device *dev)
3777 /* rtnl_mutex must be held here */
3779 __hw_addr_flush(&dev->dev_addrs);
3780 dev->dev_addr = NULL;
3783 static int dev_addr_init(struct net_device *dev)
3785 unsigned char addr[MAX_ADDR_LEN];
3786 struct netdev_hw_addr *ha;
3789 /* rtnl_mutex must be held here */
3791 __hw_addr_init(&dev->dev_addrs);
3792 memset(addr, 0, sizeof(addr));
3793 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3794 NETDEV_HW_ADDR_T_LAN);
3797 * Get the first (previously created) address from the list
3798 * and set dev_addr pointer to this location.
3800 ha = list_first_entry(&dev->dev_addrs.list,
3801 struct netdev_hw_addr, list);
3802 dev->dev_addr = ha->addr;
3808 * dev_addr_add - Add a device address
3810 * @addr: address to add
3811 * @addr_type: address type
3813 * Add a device address to the device or increase the reference count if
3814 * it already exists.
3816 * The caller must hold the rtnl_mutex.
3818 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3819 unsigned char addr_type)
3825 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3827 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3830 EXPORT_SYMBOL(dev_addr_add);
3833 * dev_addr_del - Release a device address.
3835 * @addr: address to delete
3836 * @addr_type: address type
3838 * Release reference to a device address and remove it from the device
3839 * if the reference count drops to zero.
3841 * The caller must hold the rtnl_mutex.
3843 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3844 unsigned char addr_type)
3847 struct netdev_hw_addr *ha;
3852 * We can not remove the first address from the list because
3853 * dev->dev_addr points to that.
3855 ha = list_first_entry(&dev->dev_addrs.list,
3856 struct netdev_hw_addr, list);
3857 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3860 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3863 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3866 EXPORT_SYMBOL(dev_addr_del);
3869 * dev_addr_add_multiple - Add device addresses from another device
3870 * @to_dev: device to which addresses will be added
3871 * @from_dev: device from which addresses will be added
3872 * @addr_type: address type - 0 means type will be used from from_dev
3874 * Add device addresses of the one device to another.
3876 * The caller must hold the rtnl_mutex.
3878 int dev_addr_add_multiple(struct net_device *to_dev,
3879 struct net_device *from_dev,
3880 unsigned char addr_type)
3886 if (from_dev->addr_len != to_dev->addr_len)
3888 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3889 to_dev->addr_len, addr_type);
3891 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3894 EXPORT_SYMBOL(dev_addr_add_multiple);
3897 * dev_addr_del_multiple - Delete device addresses by another device
3898 * @to_dev: device where the addresses will be deleted
3899 * @from_dev: device by which addresses the addresses will be deleted
3900 * @addr_type: address type - 0 means type will used from from_dev
3902 * Deletes addresses in to device by the list of addresses in from device.
3904 * The caller must hold the rtnl_mutex.
3906 int dev_addr_del_multiple(struct net_device *to_dev,
3907 struct net_device *from_dev,
3908 unsigned char addr_type)
3912 if (from_dev->addr_len != to_dev->addr_len)
3914 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3915 to_dev->addr_len, addr_type);
3916 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3919 EXPORT_SYMBOL(dev_addr_del_multiple);
3921 /* multicast addresses handling functions */
3923 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3924 void *addr, int alen, int glbl)
3926 struct dev_addr_list *da;
3928 for (; (da = *list) != NULL; list = &da->next) {
3929 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3930 alen == da->da_addrlen) {
3932 int old_glbl = da->da_gusers;
3949 int __dev_addr_add(struct dev_addr_list **list, int *count,
3950 void *addr, int alen, int glbl)
3952 struct dev_addr_list *da;
3954 for (da = *list; da != NULL; da = da->next) {
3955 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3956 da->da_addrlen == alen) {
3958 int old_glbl = da->da_gusers;
3968 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3971 memcpy(da->da_addr, addr, alen);
3972 da->da_addrlen = alen;
3974 da->da_gusers = glbl ? 1 : 0;
3982 * dev_unicast_delete - Release secondary unicast address.
3984 * @addr: address to delete
3986 * Release reference to a secondary unicast address and remove it
3987 * from the device if the reference count drops to zero.
3989 * The caller must hold the rtnl_mutex.
3991 int dev_unicast_delete(struct net_device *dev, void *addr)
3997 netif_addr_lock_bh(dev);
3998 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
3999 NETDEV_HW_ADDR_T_UNICAST);
4001 __dev_set_rx_mode(dev);
4002 netif_addr_unlock_bh(dev);
4005 EXPORT_SYMBOL(dev_unicast_delete);
4008 * dev_unicast_add - add a secondary unicast address
4010 * @addr: address to add
4012 * Add a secondary unicast address to the device or increase
4013 * the reference count if it already exists.
4015 * The caller must hold the rtnl_mutex.
4017 int dev_unicast_add(struct net_device *dev, void *addr)
4023 netif_addr_lock_bh(dev);
4024 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
4025 NETDEV_HW_ADDR_T_UNICAST);
4027 __dev_set_rx_mode(dev);
4028 netif_addr_unlock_bh(dev);
4031 EXPORT_SYMBOL(dev_unicast_add);
4033 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
4034 struct dev_addr_list **from, int *from_count)
4036 struct dev_addr_list *da, *next;
4040 while (da != NULL) {
4042 if (!da->da_synced) {
4043 err = __dev_addr_add(to, to_count,
4044 da->da_addr, da->da_addrlen, 0);
4049 } else if (da->da_users == 1) {
4050 __dev_addr_delete(to, to_count,
4051 da->da_addr, da->da_addrlen, 0);
4052 __dev_addr_delete(from, from_count,
4053 da->da_addr, da->da_addrlen, 0);
4059 EXPORT_SYMBOL_GPL(__dev_addr_sync);
4061 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4062 struct dev_addr_list **from, int *from_count)
4064 struct dev_addr_list *da, *next;
4067 while (da != NULL) {
4069 if (da->da_synced) {
4070 __dev_addr_delete(to, to_count,
4071 da->da_addr, da->da_addrlen, 0);
4073 __dev_addr_delete(from, from_count,
4074 da->da_addr, da->da_addrlen, 0);
4079 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4082 * dev_unicast_sync - Synchronize device's unicast list to another device
4083 * @to: destination device
4084 * @from: source device
4086 * Add newly added addresses to the destination device and release
4087 * addresses that have no users left. The source device must be
4088 * locked by netif_tx_lock_bh.
4090 * This function is intended to be called from the dev->set_rx_mode
4091 * function of layered software devices.
4093 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4097 if (to->addr_len != from->addr_len)
4100 netif_addr_lock_bh(to);
4101 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4103 __dev_set_rx_mode(to);
4104 netif_addr_unlock_bh(to);
4107 EXPORT_SYMBOL(dev_unicast_sync);
4110 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4111 * @to: destination device
4112 * @from: source device
4114 * Remove all addresses that were added to the destination device by
4115 * dev_unicast_sync(). This function is intended to be called from the
4116 * dev->stop function of layered software devices.
4118 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4120 if (to->addr_len != from->addr_len)
4123 netif_addr_lock_bh(from);
4124 netif_addr_lock(to);
4125 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4126 __dev_set_rx_mode(to);
4127 netif_addr_unlock(to);
4128 netif_addr_unlock_bh(from);
4130 EXPORT_SYMBOL(dev_unicast_unsync);
4132 static void dev_unicast_flush(struct net_device *dev)
4134 netif_addr_lock_bh(dev);
4135 __hw_addr_flush(&dev->uc);
4136 netif_addr_unlock_bh(dev);
4139 static void dev_unicast_init(struct net_device *dev)
4141 __hw_addr_init(&dev->uc);
4145 static void __dev_addr_discard(struct dev_addr_list **list)
4147 struct dev_addr_list *tmp;
4149 while (*list != NULL) {
4152 if (tmp->da_users > tmp->da_gusers)
4153 printk("__dev_addr_discard: address leakage! "
4154 "da_users=%d\n", tmp->da_users);
4159 static void dev_addr_discard(struct net_device *dev)
4161 netif_addr_lock_bh(dev);
4163 __dev_addr_discard(&dev->mc_list);
4166 netif_addr_unlock_bh(dev);
4170 * dev_get_flags - get flags reported to userspace
4173 * Get the combination of flag bits exported through APIs to userspace.
4175 unsigned dev_get_flags(const struct net_device *dev)
4179 flags = (dev->flags & ~(IFF_PROMISC |
4184 (dev->gflags & (IFF_PROMISC |
4187 if (netif_running(dev)) {
4188 if (netif_oper_up(dev))
4189 flags |= IFF_RUNNING;
4190 if (netif_carrier_ok(dev))
4191 flags |= IFF_LOWER_UP;
4192 if (netif_dormant(dev))
4193 flags |= IFF_DORMANT;
4198 EXPORT_SYMBOL(dev_get_flags);
4201 * dev_change_flags - change device settings
4203 * @flags: device state flags
4205 * Change settings on device based state flags. The flags are
4206 * in the userspace exported format.
4208 int dev_change_flags(struct net_device *dev, unsigned flags)
4211 int old_flags = dev->flags;
4216 * Set the flags on our device.
4219 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4220 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4222 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4226 * Load in the correct multicast list now the flags have changed.
4229 if ((old_flags ^ flags) & IFF_MULTICAST)
4230 dev_change_rx_flags(dev, IFF_MULTICAST);
4232 dev_set_rx_mode(dev);
4235 * Have we downed the interface. We handle IFF_UP ourselves
4236 * according to user attempts to set it, rather than blindly
4241 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4242 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4245 dev_set_rx_mode(dev);
4248 if (dev->flags & IFF_UP &&
4249 ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
4251 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4253 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4254 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4256 dev->gflags ^= IFF_PROMISC;
4257 dev_set_promiscuity(dev, inc);
4260 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4261 is important. Some (broken) drivers set IFF_PROMISC, when
4262 IFF_ALLMULTI is requested not asking us and not reporting.
4264 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4265 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4267 dev->gflags ^= IFF_ALLMULTI;
4268 dev_set_allmulti(dev, inc);
4271 /* Exclude state transition flags, already notified */
4272 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4274 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4278 EXPORT_SYMBOL(dev_change_flags);
4281 * dev_set_mtu - Change maximum transfer unit
4283 * @new_mtu: new transfer unit
4285 * Change the maximum transfer size of the network device.
4287 int dev_set_mtu(struct net_device *dev, int new_mtu)
4289 const struct net_device_ops *ops = dev->netdev_ops;
4292 if (new_mtu == dev->mtu)
4295 /* MTU must be positive. */
4299 if (!netif_device_present(dev))
4303 if (ops->ndo_change_mtu)
4304 err = ops->ndo_change_mtu(dev, new_mtu);
4308 if (!err && dev->flags & IFF_UP)
4309 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4312 EXPORT_SYMBOL(dev_set_mtu);
4315 * dev_set_mac_address - Change Media Access Control Address
4319 * Change the hardware (MAC) address of the device
4321 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4323 const struct net_device_ops *ops = dev->netdev_ops;
4326 if (!ops->ndo_set_mac_address)
4328 if (sa->sa_family != dev->type)
4330 if (!netif_device_present(dev))
4332 err = ops->ndo_set_mac_address(dev, sa);
4334 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4337 EXPORT_SYMBOL(dev_set_mac_address);
4340 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4342 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4345 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4351 case SIOCGIFFLAGS: /* Get interface flags */
4352 ifr->ifr_flags = (short) dev_get_flags(dev);
4355 case SIOCGIFMETRIC: /* Get the metric on the interface
4356 (currently unused) */
4357 ifr->ifr_metric = 0;
4360 case SIOCGIFMTU: /* Get the MTU of a device */
4361 ifr->ifr_mtu = dev->mtu;
4366 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4368 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4369 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4370 ifr->ifr_hwaddr.sa_family = dev->type;
4378 ifr->ifr_map.mem_start = dev->mem_start;
4379 ifr->ifr_map.mem_end = dev->mem_end;
4380 ifr->ifr_map.base_addr = dev->base_addr;
4381 ifr->ifr_map.irq = dev->irq;
4382 ifr->ifr_map.dma = dev->dma;
4383 ifr->ifr_map.port = dev->if_port;
4387 ifr->ifr_ifindex = dev->ifindex;
4391 ifr->ifr_qlen = dev->tx_queue_len;
4395 /* dev_ioctl() should ensure this case
4407 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4409 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4412 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4413 const struct net_device_ops *ops;
4418 ops = dev->netdev_ops;
4421 case SIOCSIFFLAGS: /* Set interface flags */
4422 return dev_change_flags(dev, ifr->ifr_flags);
4424 case SIOCSIFMETRIC: /* Set the metric on the interface
4425 (currently unused) */
4428 case SIOCSIFMTU: /* Set the MTU of a device */
4429 return dev_set_mtu(dev, ifr->ifr_mtu);
4432 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4434 case SIOCSIFHWBROADCAST:
4435 if (ifr->ifr_hwaddr.sa_family != dev->type)
4437 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4438 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4439 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4443 if (ops->ndo_set_config) {
4444 if (!netif_device_present(dev))
4446 return ops->ndo_set_config(dev, &ifr->ifr_map);
4451 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4452 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4454 if (!netif_device_present(dev))
4456 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4460 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4461 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4463 if (!netif_device_present(dev))
4465 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4469 if (ifr->ifr_qlen < 0)
4471 dev->tx_queue_len = ifr->ifr_qlen;
4475 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4476 return dev_change_name(dev, ifr->ifr_newname);
4479 * Unknown or private ioctl
4482 if ((cmd >= SIOCDEVPRIVATE &&
4483 cmd <= SIOCDEVPRIVATE + 15) ||
4484 cmd == SIOCBONDENSLAVE ||
4485 cmd == SIOCBONDRELEASE ||
4486 cmd == SIOCBONDSETHWADDR ||
4487 cmd == SIOCBONDSLAVEINFOQUERY ||
4488 cmd == SIOCBONDINFOQUERY ||
4489 cmd == SIOCBONDCHANGEACTIVE ||
4490 cmd == SIOCGMIIPHY ||
4491 cmd == SIOCGMIIREG ||
4492 cmd == SIOCSMIIREG ||
4493 cmd == SIOCBRADDIF ||
4494 cmd == SIOCBRDELIF ||
4495 cmd == SIOCSHWTSTAMP ||
4496 cmd == SIOCWANDEV) {
4498 if (ops->ndo_do_ioctl) {
4499 if (netif_device_present(dev))
4500 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4512 * This function handles all "interface"-type I/O control requests. The actual
4513 * 'doing' part of this is dev_ifsioc above.
4517 * dev_ioctl - network device ioctl
4518 * @net: the applicable net namespace
4519 * @cmd: command to issue
4520 * @arg: pointer to a struct ifreq in user space
4522 * Issue ioctl functions to devices. This is normally called by the
4523 * user space syscall interfaces but can sometimes be useful for
4524 * other purposes. The return value is the return from the syscall if
4525 * positive or a negative errno code on error.
4528 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4534 /* One special case: SIOCGIFCONF takes ifconf argument
4535 and requires shared lock, because it sleeps writing
4539 if (cmd == SIOCGIFCONF) {
4541 ret = dev_ifconf(net, (char __user *) arg);
4545 if (cmd == SIOCGIFNAME)
4546 return dev_ifname(net, (struct ifreq __user *)arg);
4548 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4551 ifr.ifr_name[IFNAMSIZ-1] = 0;
4553 colon = strchr(ifr.ifr_name, ':');
4558 * See which interface the caller is talking about.
4563 * These ioctl calls:
4564 * - can be done by all.
4565 * - atomic and do not require locking.
4576 dev_load(net, ifr.ifr_name);
4578 ret = dev_ifsioc_locked(net, &ifr, cmd);
4583 if (copy_to_user(arg, &ifr,
4584 sizeof(struct ifreq)))
4590 dev_load(net, ifr.ifr_name);
4592 ret = dev_ethtool(net, &ifr);
4597 if (copy_to_user(arg, &ifr,
4598 sizeof(struct ifreq)))
4604 * These ioctl calls:
4605 * - require superuser power.
4606 * - require strict serialization.
4612 if (!capable(CAP_NET_ADMIN))
4614 dev_load(net, ifr.ifr_name);
4616 ret = dev_ifsioc(net, &ifr, cmd);
4621 if (copy_to_user(arg, &ifr,
4622 sizeof(struct ifreq)))
4628 * These ioctl calls:
4629 * - require superuser power.
4630 * - require strict serialization.
4631 * - do not return a value
4641 case SIOCSIFHWBROADCAST:
4644 case SIOCBONDENSLAVE:
4645 case SIOCBONDRELEASE:
4646 case SIOCBONDSETHWADDR:
4647 case SIOCBONDCHANGEACTIVE:
4651 if (!capable(CAP_NET_ADMIN))
4654 case SIOCBONDSLAVEINFOQUERY:
4655 case SIOCBONDINFOQUERY:
4656 dev_load(net, ifr.ifr_name);
4658 ret = dev_ifsioc(net, &ifr, cmd);
4663 /* Get the per device memory space. We can add this but
4664 * currently do not support it */
4666 /* Set the per device memory buffer space.
4667 * Not applicable in our case */
4672 * Unknown or private ioctl.
4675 if (cmd == SIOCWANDEV ||
4676 (cmd >= SIOCDEVPRIVATE &&
4677 cmd <= SIOCDEVPRIVATE + 15)) {
4678 dev_load(net, ifr.ifr_name);
4680 ret = dev_ifsioc(net, &ifr, cmd);
4682 if (!ret && copy_to_user(arg, &ifr,
4683 sizeof(struct ifreq)))
4687 /* Take care of Wireless Extensions */
4688 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4689 return wext_handle_ioctl(net, &ifr, cmd, arg);
4696 * dev_new_index - allocate an ifindex
4697 * @net: the applicable net namespace
4699 * Returns a suitable unique value for a new device interface
4700 * number. The caller must hold the rtnl semaphore or the
4701 * dev_base_lock to be sure it remains unique.
4703 static int dev_new_index(struct net *net)
4709 if (!__dev_get_by_index(net, ifindex))
4714 /* Delayed registration/unregisteration */
4715 static LIST_HEAD(net_todo_list);
4717 static void net_set_todo(struct net_device *dev)
4719 list_add_tail(&dev->todo_list, &net_todo_list);
4722 static void rollback_registered_many(struct list_head *head)
4724 struct net_device *dev;
4726 BUG_ON(dev_boot_phase);
4729 list_for_each_entry(dev, head, unreg_list) {
4730 /* Some devices call without registering
4731 * for initialization unwind.
4733 if (dev->reg_state == NETREG_UNINITIALIZED) {
4734 pr_debug("unregister_netdevice: device %s/%p never "
4735 "was registered\n", dev->name, dev);
4741 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4743 /* If device is running, close it first. */
4746 /* And unlink it from device chain. */
4747 unlist_netdevice(dev);
4749 dev->reg_state = NETREG_UNREGISTERING;
4754 list_for_each_entry(dev, head, unreg_list) {
4755 /* Shutdown queueing discipline. */
4759 /* Notify protocols, that we are about to destroy
4760 this device. They should clean all the things.
4762 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4765 * Flush the unicast and multicast chains
4767 dev_unicast_flush(dev);
4768 dev_addr_discard(dev);
4770 if (dev->netdev_ops->ndo_uninit)
4771 dev->netdev_ops->ndo_uninit(dev);
4773 /* Notifier chain MUST detach us from master device. */
4774 WARN_ON(dev->master);
4776 /* Remove entries from kobject tree */
4777 netdev_unregister_kobject(dev);
4782 list_for_each_entry(dev, head, unreg_list)
4786 static void rollback_registered(struct net_device *dev)
4790 list_add(&dev->unreg_list, &single);
4791 rollback_registered_many(&single);
4794 static void __netdev_init_queue_locks_one(struct net_device *dev,
4795 struct netdev_queue *dev_queue,
4798 spin_lock_init(&dev_queue->_xmit_lock);
4799 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4800 dev_queue->xmit_lock_owner = -1;
4803 static void netdev_init_queue_locks(struct net_device *dev)
4805 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4806 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4809 unsigned long netdev_fix_features(unsigned long features, const char *name)
4811 /* Fix illegal SG+CSUM combinations. */
4812 if ((features & NETIF_F_SG) &&
4813 !(features & NETIF_F_ALL_CSUM)) {
4815 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4816 "checksum feature.\n", name);
4817 features &= ~NETIF_F_SG;
4820 /* TSO requires that SG is present as well. */
4821 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4823 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4824 "SG feature.\n", name);
4825 features &= ~NETIF_F_TSO;
4828 if (features & NETIF_F_UFO) {
4829 if (!(features & NETIF_F_GEN_CSUM)) {
4831 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4832 "since no NETIF_F_HW_CSUM feature.\n",
4834 features &= ~NETIF_F_UFO;
4837 if (!(features & NETIF_F_SG)) {
4839 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4840 "since no NETIF_F_SG feature.\n", name);
4841 features &= ~NETIF_F_UFO;
4847 EXPORT_SYMBOL(netdev_fix_features);
4850 * register_netdevice - register a network device
4851 * @dev: device to register
4853 * Take a completed network device structure and add it to the kernel
4854 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4855 * chain. 0 is returned on success. A negative errno code is returned
4856 * on a failure to set up the device, or if the name is a duplicate.
4858 * Callers must hold the rtnl semaphore. You may want
4859 * register_netdev() instead of this.
4862 * The locking appears insufficient to guarantee two parallel registers
4863 * will not get the same name.
4866 int register_netdevice(struct net_device *dev)
4868 struct hlist_head *head;
4869 struct hlist_node *p;
4871 struct net *net = dev_net(dev);
4873 BUG_ON(dev_boot_phase);
4878 /* When net_device's are persistent, this will be fatal. */
4879 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4882 spin_lock_init(&dev->addr_list_lock);
4883 netdev_set_addr_lockdep_class(dev);
4884 netdev_init_queue_locks(dev);
4888 /* Init, if this function is available */
4889 if (dev->netdev_ops->ndo_init) {
4890 ret = dev->netdev_ops->ndo_init(dev);
4898 if (!dev_valid_name(dev->name)) {
4903 dev->ifindex = dev_new_index(net);
4904 if (dev->iflink == -1)
4905 dev->iflink = dev->ifindex;
4907 /* Check for existence of name */
4908 head = dev_name_hash(net, dev->name);
4909 hlist_for_each(p, head) {
4910 struct net_device *d
4911 = hlist_entry(p, struct net_device, name_hlist);
4912 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4918 /* Fix illegal checksum combinations */
4919 if ((dev->features & NETIF_F_HW_CSUM) &&
4920 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4921 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4923 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4926 if ((dev->features & NETIF_F_NO_CSUM) &&
4927 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4928 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4930 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4933 dev->features = netdev_fix_features(dev->features, dev->name);
4935 /* Enable software GSO if SG is supported. */
4936 if (dev->features & NETIF_F_SG)
4937 dev->features |= NETIF_F_GSO;
4939 netdev_initialize_kobject(dev);
4941 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4942 ret = notifier_to_errno(ret);
4946 ret = netdev_register_kobject(dev);
4949 dev->reg_state = NETREG_REGISTERED;
4952 * Default initial state at registry is that the
4953 * device is present.
4956 set_bit(__LINK_STATE_PRESENT, &dev->state);
4958 dev_init_scheduler(dev);
4960 list_netdevice(dev);
4962 /* Notify protocols, that a new device appeared. */
4963 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4964 ret = notifier_to_errno(ret);
4966 rollback_registered(dev);
4967 dev->reg_state = NETREG_UNREGISTERED;
4974 if (dev->netdev_ops->ndo_uninit)
4975 dev->netdev_ops->ndo_uninit(dev);
4978 EXPORT_SYMBOL(register_netdevice);
4981 * init_dummy_netdev - init a dummy network device for NAPI
4982 * @dev: device to init
4984 * This takes a network device structure and initialize the minimum
4985 * amount of fields so it can be used to schedule NAPI polls without
4986 * registering a full blown interface. This is to be used by drivers
4987 * that need to tie several hardware interfaces to a single NAPI
4988 * poll scheduler due to HW limitations.
4990 int init_dummy_netdev(struct net_device *dev)
4992 /* Clear everything. Note we don't initialize spinlocks
4993 * are they aren't supposed to be taken by any of the
4994 * NAPI code and this dummy netdev is supposed to be
4995 * only ever used for NAPI polls
4997 memset(dev, 0, sizeof(struct net_device));
4999 /* make sure we BUG if trying to hit standard
5000 * register/unregister code path
5002 dev->reg_state = NETREG_DUMMY;
5004 /* initialize the ref count */
5005 atomic_set(&dev->refcnt, 1);
5007 /* NAPI wants this */
5008 INIT_LIST_HEAD(&dev->napi_list);
5010 /* a dummy interface is started by default */
5011 set_bit(__LINK_STATE_PRESENT, &dev->state);
5012 set_bit(__LINK_STATE_START, &dev->state);
5016 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5020 * register_netdev - register a network device
5021 * @dev: device to register
5023 * Take a completed network device structure and add it to the kernel
5024 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5025 * chain. 0 is returned on success. A negative errno code is returned
5026 * on a failure to set up the device, or if the name is a duplicate.
5028 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5029 * and expands the device name if you passed a format string to
5032 int register_netdev(struct net_device *dev)
5039 * If the name is a format string the caller wants us to do a
5042 if (strchr(dev->name, '%')) {
5043 err = dev_alloc_name(dev, dev->name);
5048 err = register_netdevice(dev);
5053 EXPORT_SYMBOL(register_netdev);
5056 * netdev_wait_allrefs - wait until all references are gone.
5058 * This is called when unregistering network devices.
5060 * Any protocol or device that holds a reference should register
5061 * for netdevice notification, and cleanup and put back the
5062 * reference if they receive an UNREGISTER event.
5063 * We can get stuck here if buggy protocols don't correctly
5066 static void netdev_wait_allrefs(struct net_device *dev)
5068 unsigned long rebroadcast_time, warning_time;
5070 rebroadcast_time = warning_time = jiffies;
5071 while (atomic_read(&dev->refcnt) != 0) {
5072 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5075 /* Rebroadcast unregister notification */
5076 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5078 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5080 /* We must not have linkwatch events
5081 * pending on unregister. If this
5082 * happens, we simply run the queue
5083 * unscheduled, resulting in a noop
5086 linkwatch_run_queue();
5091 rebroadcast_time = jiffies;
5096 if (time_after(jiffies, warning_time + 10 * HZ)) {
5097 printk(KERN_EMERG "unregister_netdevice: "
5098 "waiting for %s to become free. Usage "
5100 dev->name, atomic_read(&dev->refcnt));
5101 warning_time = jiffies;
5110 * register_netdevice(x1);
5111 * register_netdevice(x2);
5113 * unregister_netdevice(y1);
5114 * unregister_netdevice(y2);
5120 * We are invoked by rtnl_unlock().
5121 * This allows us to deal with problems:
5122 * 1) We can delete sysfs objects which invoke hotplug
5123 * without deadlocking with linkwatch via keventd.
5124 * 2) Since we run with the RTNL semaphore not held, we can sleep
5125 * safely in order to wait for the netdev refcnt to drop to zero.
5127 * We must not return until all unregister events added during
5128 * the interval the lock was held have been completed.
5130 void netdev_run_todo(void)
5132 struct list_head list;
5134 /* Snapshot list, allow later requests */
5135 list_replace_init(&net_todo_list, &list);
5139 while (!list_empty(&list)) {
5140 struct net_device *dev
5141 = list_entry(list.next, struct net_device, todo_list);
5142 list_del(&dev->todo_list);
5144 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5145 printk(KERN_ERR "network todo '%s' but state %d\n",
5146 dev->name, dev->reg_state);
5151 dev->reg_state = NETREG_UNREGISTERED;
5153 on_each_cpu(flush_backlog, dev, 1);
5155 netdev_wait_allrefs(dev);
5158 BUG_ON(atomic_read(&dev->refcnt));
5159 WARN_ON(dev->ip_ptr);
5160 WARN_ON(dev->ip6_ptr);
5161 WARN_ON(dev->dn_ptr);
5163 if (dev->destructor)
5164 dev->destructor(dev);
5166 /* Free network device */
5167 kobject_put(&dev->dev.kobj);
5172 * dev_get_stats - get network device statistics
5173 * @dev: device to get statistics from
5175 * Get network statistics from device. The device driver may provide
5176 * its own method by setting dev->netdev_ops->get_stats; otherwise
5177 * the internal statistics structure is used.
5179 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5181 const struct net_device_ops *ops = dev->netdev_ops;
5183 if (ops->ndo_get_stats)
5184 return ops->ndo_get_stats(dev);
5186 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5187 struct net_device_stats *stats = &dev->stats;
5189 struct netdev_queue *txq;
5191 for (i = 0; i < dev->num_tx_queues; i++) {
5192 txq = netdev_get_tx_queue(dev, i);
5193 tx_bytes += txq->tx_bytes;
5194 tx_packets += txq->tx_packets;
5195 tx_dropped += txq->tx_dropped;
5197 if (tx_bytes || tx_packets || tx_dropped) {
5198 stats->tx_bytes = tx_bytes;
5199 stats->tx_packets = tx_packets;
5200 stats->tx_dropped = tx_dropped;
5205 EXPORT_SYMBOL(dev_get_stats);
5207 static void netdev_init_one_queue(struct net_device *dev,
5208 struct netdev_queue *queue,
5214 static void netdev_init_queues(struct net_device *dev)
5216 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5217 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5218 spin_lock_init(&dev->tx_global_lock);
5222 * alloc_netdev_mq - allocate network device
5223 * @sizeof_priv: size of private data to allocate space for
5224 * @name: device name format string
5225 * @setup: callback to initialize device
5226 * @queue_count: the number of subqueues to allocate
5228 * Allocates a struct net_device with private data area for driver use
5229 * and performs basic initialization. Also allocates subquue structs
5230 * for each queue on the device at the end of the netdevice.
5232 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5233 void (*setup)(struct net_device *), unsigned int queue_count)
5235 struct netdev_queue *tx;
5236 struct net_device *dev;
5238 struct net_device *p;
5240 BUG_ON(strlen(name) >= sizeof(dev->name));
5242 alloc_size = sizeof(struct net_device);
5244 /* ensure 32-byte alignment of private area */
5245 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5246 alloc_size += sizeof_priv;
5248 /* ensure 32-byte alignment of whole construct */
5249 alloc_size += NETDEV_ALIGN - 1;
5251 p = kzalloc(alloc_size, GFP_KERNEL);
5253 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5257 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5259 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5264 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5265 dev->padded = (char *)dev - (char *)p;
5267 if (dev_addr_init(dev))
5270 dev_unicast_init(dev);
5272 dev_net_set(dev, &init_net);
5275 dev->num_tx_queues = queue_count;
5276 dev->real_num_tx_queues = queue_count;
5278 dev->gso_max_size = GSO_MAX_SIZE;
5280 netdev_init_queues(dev);
5282 INIT_LIST_HEAD(&dev->napi_list);
5283 INIT_LIST_HEAD(&dev->unreg_list);
5284 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5286 strcpy(dev->name, name);
5296 EXPORT_SYMBOL(alloc_netdev_mq);
5299 * free_netdev - free network device
5302 * This function does the last stage of destroying an allocated device
5303 * interface. The reference to the device object is released.
5304 * If this is the last reference then it will be freed.
5306 void free_netdev(struct net_device *dev)
5308 struct napi_struct *p, *n;
5310 release_net(dev_net(dev));
5314 /* Flush device addresses */
5315 dev_addr_flush(dev);
5317 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5320 /* Compatibility with error handling in drivers */
5321 if (dev->reg_state == NETREG_UNINITIALIZED) {
5322 kfree((char *)dev - dev->padded);
5326 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5327 dev->reg_state = NETREG_RELEASED;
5329 /* will free via device release */
5330 put_device(&dev->dev);
5332 EXPORT_SYMBOL(free_netdev);
5335 * synchronize_net - Synchronize with packet receive processing
5337 * Wait for packets currently being received to be done.
5338 * Does not block later packets from starting.
5340 void synchronize_net(void)
5345 EXPORT_SYMBOL(synchronize_net);
5348 * unregister_netdevice_queue - remove device from the kernel
5352 * This function shuts down a device interface and removes it
5353 * from the kernel tables.
5354 * If head not NULL, device is queued to be unregistered later.
5356 * Callers must hold the rtnl semaphore. You may want
5357 * unregister_netdev() instead of this.
5360 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5365 list_move_tail(&dev->unreg_list, head);
5367 rollback_registered(dev);
5368 /* Finish processing unregister after unlock */
5372 EXPORT_SYMBOL(unregister_netdevice_queue);
5375 * unregister_netdevice_many - unregister many devices
5376 * @head: list of devices
5379 void unregister_netdevice_many(struct list_head *head)
5381 struct net_device *dev;
5383 if (!list_empty(head)) {
5384 rollback_registered_many(head);
5385 list_for_each_entry(dev, head, unreg_list)
5389 EXPORT_SYMBOL(unregister_netdevice_many);
5392 * unregister_netdev - remove device from the kernel
5395 * This function shuts down a device interface and removes it
5396 * from the kernel tables.
5398 * This is just a wrapper for unregister_netdevice that takes
5399 * the rtnl semaphore. In general you want to use this and not
5400 * unregister_netdevice.
5402 void unregister_netdev(struct net_device *dev)
5405 unregister_netdevice(dev);
5408 EXPORT_SYMBOL(unregister_netdev);
5411 * dev_change_net_namespace - move device to different nethost namespace
5413 * @net: network namespace
5414 * @pat: If not NULL name pattern to try if the current device name
5415 * is already taken in the destination network namespace.
5417 * This function shuts down a device interface and moves it
5418 * to a new network namespace. On success 0 is returned, on
5419 * a failure a netagive errno code is returned.
5421 * Callers must hold the rtnl semaphore.
5424 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5427 const char *destname;
5432 /* Don't allow namespace local devices to be moved. */
5434 if (dev->features & NETIF_F_NETNS_LOCAL)
5438 /* Don't allow real devices to be moved when sysfs
5442 if (dev->dev.parent)
5446 /* Ensure the device has been registrered */
5448 if (dev->reg_state != NETREG_REGISTERED)
5451 /* Get out if there is nothing todo */
5453 if (net_eq(dev_net(dev), net))
5456 /* Pick the destination device name, and ensure
5457 * we can use it in the destination network namespace.
5460 destname = dev->name;
5461 if (__dev_get_by_name(net, destname)) {
5462 /* We get here if we can't use the current device name */
5465 if (!dev_valid_name(pat))
5467 if (strchr(pat, '%')) {
5468 if (__dev_alloc_name(net, pat, buf) < 0)
5473 if (__dev_get_by_name(net, destname))
5478 * And now a mini version of register_netdevice unregister_netdevice.
5481 /* If device is running close it first. */
5484 /* And unlink it from device chain */
5486 unlist_netdevice(dev);
5490 /* Shutdown queueing discipline. */
5493 /* Notify protocols, that we are about to destroy
5494 this device. They should clean all the things.
5496 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5499 * Flush the unicast and multicast chains
5501 dev_unicast_flush(dev);
5502 dev_addr_discard(dev);
5504 netdev_unregister_kobject(dev);
5506 /* Actually switch the network namespace */
5507 dev_net_set(dev, net);
5509 /* Assign the new device name */
5510 if (destname != dev->name)
5511 strcpy(dev->name, destname);
5513 /* If there is an ifindex conflict assign a new one */
5514 if (__dev_get_by_index(net, dev->ifindex)) {
5515 int iflink = (dev->iflink == dev->ifindex);
5516 dev->ifindex = dev_new_index(net);
5518 dev->iflink = dev->ifindex;
5521 /* Fixup kobjects */
5522 err = netdev_register_kobject(dev);
5525 /* Add the device back in the hashes */
5526 list_netdevice(dev);
5528 /* Notify protocols, that a new device appeared. */
5529 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5536 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5538 static int dev_cpu_callback(struct notifier_block *nfb,
5539 unsigned long action,
5542 struct sk_buff **list_skb;
5543 struct Qdisc **list_net;
5544 struct sk_buff *skb;
5545 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5546 struct softnet_data *sd, *oldsd;
5548 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5551 local_irq_disable();
5552 cpu = smp_processor_id();
5553 sd = &per_cpu(softnet_data, cpu);
5554 oldsd = &per_cpu(softnet_data, oldcpu);
5556 /* Find end of our completion_queue. */
5557 list_skb = &sd->completion_queue;
5559 list_skb = &(*list_skb)->next;
5560 /* Append completion queue from offline CPU. */
5561 *list_skb = oldsd->completion_queue;
5562 oldsd->completion_queue = NULL;
5564 /* Find end of our output_queue. */
5565 list_net = &sd->output_queue;
5567 list_net = &(*list_net)->next_sched;
5568 /* Append output queue from offline CPU. */
5569 *list_net = oldsd->output_queue;
5570 oldsd->output_queue = NULL;
5572 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5575 /* Process offline CPU's input_pkt_queue */
5576 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5584 * netdev_increment_features - increment feature set by one
5585 * @all: current feature set
5586 * @one: new feature set
5587 * @mask: mask feature set
5589 * Computes a new feature set after adding a device with feature set
5590 * @one to the master device with current feature set @all. Will not
5591 * enable anything that is off in @mask. Returns the new feature set.
5593 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5596 /* If device needs checksumming, downgrade to it. */
5597 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5598 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5599 else if (mask & NETIF_F_ALL_CSUM) {
5600 /* If one device supports v4/v6 checksumming, set for all. */
5601 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5602 !(all & NETIF_F_GEN_CSUM)) {
5603 all &= ~NETIF_F_ALL_CSUM;
5604 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5607 /* If one device supports hw checksumming, set for all. */
5608 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5609 all &= ~NETIF_F_ALL_CSUM;
5610 all |= NETIF_F_HW_CSUM;
5614 one |= NETIF_F_ALL_CSUM;
5616 one |= all & NETIF_F_ONE_FOR_ALL;
5617 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5618 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5622 EXPORT_SYMBOL(netdev_increment_features);
5624 static struct hlist_head *netdev_create_hash(void)
5627 struct hlist_head *hash;
5629 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5631 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5632 INIT_HLIST_HEAD(&hash[i]);
5637 /* Initialize per network namespace state */
5638 static int __net_init netdev_init(struct net *net)
5640 INIT_LIST_HEAD(&net->dev_base_head);
5642 net->dev_name_head = netdev_create_hash();
5643 if (net->dev_name_head == NULL)
5646 net->dev_index_head = netdev_create_hash();
5647 if (net->dev_index_head == NULL)
5653 kfree(net->dev_name_head);
5659 * netdev_drivername - network driver for the device
5660 * @dev: network device
5661 * @buffer: buffer for resulting name
5662 * @len: size of buffer
5664 * Determine network driver for device.
5666 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5668 const struct device_driver *driver;
5669 const struct device *parent;
5671 if (len <= 0 || !buffer)
5675 parent = dev->dev.parent;
5680 driver = parent->driver;
5681 if (driver && driver->name)
5682 strlcpy(buffer, driver->name, len);
5686 static void __net_exit netdev_exit(struct net *net)
5688 kfree(net->dev_name_head);
5689 kfree(net->dev_index_head);
5692 static struct pernet_operations __net_initdata netdev_net_ops = {
5693 .init = netdev_init,
5694 .exit = netdev_exit,
5697 static void __net_exit default_device_exit(struct net *net)
5699 struct net_device *dev;
5701 * Push all migratable of the network devices back to the
5702 * initial network namespace
5706 for_each_netdev(net, dev) {
5708 char fb_name[IFNAMSIZ];
5710 /* Ignore unmoveable devices (i.e. loopback) */
5711 if (dev->features & NETIF_F_NETNS_LOCAL)
5714 /* Delete virtual devices */
5715 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5716 dev->rtnl_link_ops->dellink(dev, NULL);
5720 /* Push remaing network devices to init_net */
5721 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5722 err = dev_change_net_namespace(dev, &init_net, fb_name);
5724 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5725 __func__, dev->name, err);
5733 static struct pernet_operations __net_initdata default_device_ops = {
5734 .exit = default_device_exit,
5738 * Initialize the DEV module. At boot time this walks the device list and
5739 * unhooks any devices that fail to initialise (normally hardware not
5740 * present) and leaves us with a valid list of present and active devices.
5745 * This is called single threaded during boot, so no need
5746 * to take the rtnl semaphore.
5748 static int __init net_dev_init(void)
5750 int i, rc = -ENOMEM;
5752 BUG_ON(!dev_boot_phase);
5754 if (dev_proc_init())
5757 if (netdev_kobject_init())
5760 INIT_LIST_HEAD(&ptype_all);
5761 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5762 INIT_LIST_HEAD(&ptype_base[i]);
5764 if (register_pernet_subsys(&netdev_net_ops))
5768 * Initialise the packet receive queues.
5771 for_each_possible_cpu(i) {
5772 struct softnet_data *queue;
5774 queue = &per_cpu(softnet_data, i);
5775 skb_queue_head_init(&queue->input_pkt_queue);
5776 queue->completion_queue = NULL;
5777 INIT_LIST_HEAD(&queue->poll_list);
5779 queue->backlog.poll = process_backlog;
5780 queue->backlog.weight = weight_p;
5781 queue->backlog.gro_list = NULL;
5782 queue->backlog.gro_count = 0;
5787 /* The loopback device is special if any other network devices
5788 * is present in a network namespace the loopback device must
5789 * be present. Since we now dynamically allocate and free the
5790 * loopback device ensure this invariant is maintained by
5791 * keeping the loopback device as the first device on the
5792 * list of network devices. Ensuring the loopback devices
5793 * is the first device that appears and the last network device
5796 if (register_pernet_device(&loopback_net_ops))
5799 if (register_pernet_device(&default_device_ops))
5802 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5803 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5805 hotcpu_notifier(dev_cpu_callback, 0);
5813 subsys_initcall(net_dev_init);
5815 static int __init initialize_hashrnd(void)
5817 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5821 late_initcall_sync(initialize_hashrnd);