V4L/DVB: dvb: fix sparse warnings
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / core / dev.c
blobec874218b20638887a54140ee3612f50659743b1
1 /*
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
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
14 * Additional Authors:
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>
22 * Changes:
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
34 * drivers
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
44 * call a packet.
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
50 * changes.
51 * Rudi Cilibrasi : Pass the right thing to
52 * set_mac_address()
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
58 * 1 device.
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
66 * the backlog queue.
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>
86 #include <linux/mm.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>
98 #include <net/sock.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>
105 #include <net/dst.h>
106 #include <net/pkt_sched.h>
107 #include <net/checksum.h>
108 #include <net/xfrm.h>
109 #include <linux/highmem.h>
110 #include <linux/init.h>
111 #include <linux/kmod.h>
112 #include <linux/module.h>
113 #include <linux/netpoll.h>
114 #include <linux/rcupdate.h>
115 #include <linux/delay.h>
116 #include <net/wext.h>
117 #include <net/iw_handler.h>
118 #include <asm/current.h>
119 #include <linux/audit.h>
120 #include <linux/dmaengine.h>
121 #include <linux/err.h>
122 #include <linux/ctype.h>
123 #include <linux/if_arp.h>
124 #include <linux/if_vlan.h>
125 #include <linux/ip.h>
126 #include <net/ip.h>
127 #include <linux/ipv6.h>
128 #include <linux/in.h>
129 #include <linux/jhash.h>
130 #include <linux/random.h>
131 #include <trace/events/napi.h>
133 #include "net-sysfs.h"
135 /* Instead of increasing this, you should create a hash table. */
136 #define MAX_GRO_SKBS 8
138 /* This should be increased if a protocol with a bigger head is added. */
139 #define GRO_MAX_HEAD (MAX_HEADER + 128)
142 * The list of packet types we will receive (as opposed to discard)
143 * and the routines to invoke.
145 * Why 16. Because with 16 the only overlap we get on a hash of the
146 * low nibble of the protocol value is RARP/SNAP/X.25.
148 * NOTE: That is no longer true with the addition of VLAN tags. Not
149 * sure which should go first, but I bet it won't make much
150 * difference if we are running VLANs. The good news is that
151 * this protocol won't be in the list unless compiled in, so
152 * the average user (w/out VLANs) will not be adversely affected.
153 * --BLG
155 * 0800 IP
156 * 8100 802.1Q VLAN
157 * 0001 802.3
158 * 0002 AX.25
159 * 0004 802.2
160 * 8035 RARP
161 * 0005 SNAP
162 * 0805 X.25
163 * 0806 ARP
164 * 8137 IPX
165 * 0009 Localtalk
166 * 86DD IPv6
169 #define PTYPE_HASH_SIZE (16)
170 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
172 static DEFINE_SPINLOCK(ptype_lock);
173 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
174 static struct list_head ptype_all __read_mostly; /* Taps */
177 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
178 * semaphore.
180 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
182 * Writers must hold the rtnl semaphore while they loop through the
183 * dev_base_head list, and hold dev_base_lock for writing when they do the
184 * actual updates. This allows pure readers to access the list even
185 * while a writer is preparing to update it.
187 * To put it another way, dev_base_lock is held for writing only to
188 * protect against pure readers; the rtnl semaphore provides the
189 * protection against other writers.
191 * See, for example usages, register_netdevice() and
192 * unregister_netdevice(), which must be called with the rtnl
193 * semaphore held.
195 DEFINE_RWLOCK(dev_base_lock);
196 EXPORT_SYMBOL(dev_base_lock);
198 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
200 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
201 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
204 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
206 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
209 /* Device list insertion */
210 static int list_netdevice(struct net_device *dev)
212 struct net *net = dev_net(dev);
214 ASSERT_RTNL();
216 write_lock_bh(&dev_base_lock);
217 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
218 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
219 hlist_add_head_rcu(&dev->index_hlist,
220 dev_index_hash(net, dev->ifindex));
221 write_unlock_bh(&dev_base_lock);
222 return 0;
225 /* Device list removal
226 * caller must respect a RCU grace period before freeing/reusing dev
228 static void unlist_netdevice(struct net_device *dev)
230 ASSERT_RTNL();
232 /* Unlink dev from the device chain */
233 write_lock_bh(&dev_base_lock);
234 list_del_rcu(&dev->dev_list);
235 hlist_del_rcu(&dev->name_hlist);
236 hlist_del_rcu(&dev->index_hlist);
237 write_unlock_bh(&dev_base_lock);
241 * Our notifier list
244 static RAW_NOTIFIER_HEAD(netdev_chain);
247 * Device drivers call our routines to queue packets here. We empty the
248 * queue in the local softnet handler.
251 DEFINE_PER_CPU(struct softnet_data, softnet_data);
252 EXPORT_PER_CPU_SYMBOL(softnet_data);
254 #ifdef CONFIG_LOCKDEP
256 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
257 * according to dev->type
259 static const unsigned short netdev_lock_type[] =
260 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
261 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
262 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
263 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
264 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
265 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
266 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
267 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
268 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
269 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
270 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
271 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
272 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
273 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
274 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
275 ARPHRD_VOID, ARPHRD_NONE};
277 static const char *const netdev_lock_name[] =
278 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
279 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
280 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
281 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
282 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
283 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
284 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
285 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
286 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
287 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
288 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
289 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
290 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
291 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
292 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
293 "_xmit_VOID", "_xmit_NONE"};
295 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
296 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
298 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
300 int i;
302 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
303 if (netdev_lock_type[i] == dev_type)
304 return i;
305 /* the last key is used by default */
306 return ARRAY_SIZE(netdev_lock_type) - 1;
309 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
310 unsigned short dev_type)
312 int i;
314 i = netdev_lock_pos(dev_type);
315 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
316 netdev_lock_name[i]);
319 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
321 int i;
323 i = netdev_lock_pos(dev->type);
324 lockdep_set_class_and_name(&dev->addr_list_lock,
325 &netdev_addr_lock_key[i],
326 netdev_lock_name[i]);
328 #else
329 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
330 unsigned short dev_type)
333 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
336 #endif
338 /*******************************************************************************
340 Protocol management and registration routines
342 *******************************************************************************/
345 * Add a protocol ID to the list. Now that the input handler is
346 * smarter we can dispense with all the messy stuff that used to be
347 * here.
349 * BEWARE!!! Protocol handlers, mangling input packets,
350 * MUST BE last in hash buckets and checking protocol handlers
351 * MUST start from promiscuous ptype_all chain in net_bh.
352 * It is true now, do not change it.
353 * Explanation follows: if protocol handler, mangling packet, will
354 * be the first on list, it is not able to sense, that packet
355 * is cloned and should be copied-on-write, so that it will
356 * change it and subsequent readers will get broken packet.
357 * --ANK (980803)
361 * dev_add_pack - add packet handler
362 * @pt: packet type declaration
364 * Add a protocol handler to the networking stack. The passed &packet_type
365 * is linked into kernel lists and may not be freed until it has been
366 * removed from the kernel lists.
368 * This call does not sleep therefore it can not
369 * guarantee all CPU's that are in middle of receiving packets
370 * will see the new packet type (until the next received packet).
373 void dev_add_pack(struct packet_type *pt)
375 int hash;
377 spin_lock_bh(&ptype_lock);
378 if (pt->type == htons(ETH_P_ALL))
379 list_add_rcu(&pt->list, &ptype_all);
380 else {
381 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
382 list_add_rcu(&pt->list, &ptype_base[hash]);
384 spin_unlock_bh(&ptype_lock);
386 EXPORT_SYMBOL(dev_add_pack);
389 * __dev_remove_pack - remove packet handler
390 * @pt: packet type declaration
392 * Remove a protocol handler that was previously added to the kernel
393 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
394 * from the kernel lists and can be freed or reused once this function
395 * returns.
397 * The packet type might still be in use by receivers
398 * and must not be freed until after all the CPU's have gone
399 * through a quiescent state.
401 void __dev_remove_pack(struct packet_type *pt)
403 struct list_head *head;
404 struct packet_type *pt1;
406 spin_lock_bh(&ptype_lock);
408 if (pt->type == htons(ETH_P_ALL))
409 head = &ptype_all;
410 else
411 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
413 list_for_each_entry(pt1, head, list) {
414 if (pt == pt1) {
415 list_del_rcu(&pt->list);
416 goto out;
420 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
421 out:
422 spin_unlock_bh(&ptype_lock);
424 EXPORT_SYMBOL(__dev_remove_pack);
427 * dev_remove_pack - remove packet handler
428 * @pt: packet type declaration
430 * Remove a protocol handler that was previously added to the kernel
431 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
432 * from the kernel lists and can be freed or reused once this function
433 * returns.
435 * This call sleeps to guarantee that no CPU is looking at the packet
436 * type after return.
438 void dev_remove_pack(struct packet_type *pt)
440 __dev_remove_pack(pt);
442 synchronize_net();
444 EXPORT_SYMBOL(dev_remove_pack);
446 /******************************************************************************
448 Device Boot-time Settings Routines
450 *******************************************************************************/
452 /* Boot time configuration table */
453 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
456 * netdev_boot_setup_add - add new setup entry
457 * @name: name of the device
458 * @map: configured settings for the device
460 * Adds new setup entry to the dev_boot_setup list. The function
461 * returns 0 on error and 1 on success. This is a generic routine to
462 * all netdevices.
464 static int netdev_boot_setup_add(char *name, struct ifmap *map)
466 struct netdev_boot_setup *s;
467 int i;
469 s = dev_boot_setup;
470 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
471 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
472 memset(s[i].name, 0, sizeof(s[i].name));
473 strlcpy(s[i].name, name, IFNAMSIZ);
474 memcpy(&s[i].map, map, sizeof(s[i].map));
475 break;
479 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
483 * netdev_boot_setup_check - check boot time settings
484 * @dev: the netdevice
486 * Check boot time settings for the device.
487 * The found settings are set for the device to be used
488 * later in the device probing.
489 * Returns 0 if no settings found, 1 if they are.
491 int netdev_boot_setup_check(struct net_device *dev)
493 struct netdev_boot_setup *s = dev_boot_setup;
494 int i;
496 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
497 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
498 !strcmp(dev->name, s[i].name)) {
499 dev->irq = s[i].map.irq;
500 dev->base_addr = s[i].map.base_addr;
501 dev->mem_start = s[i].map.mem_start;
502 dev->mem_end = s[i].map.mem_end;
503 return 1;
506 return 0;
508 EXPORT_SYMBOL(netdev_boot_setup_check);
512 * netdev_boot_base - get address from boot time settings
513 * @prefix: prefix for network device
514 * @unit: id for network device
516 * Check boot time settings for the base address of device.
517 * The found settings are set for the device to be used
518 * later in the device probing.
519 * Returns 0 if no settings found.
521 unsigned long netdev_boot_base(const char *prefix, int unit)
523 const struct netdev_boot_setup *s = dev_boot_setup;
524 char name[IFNAMSIZ];
525 int i;
527 sprintf(name, "%s%d", prefix, unit);
530 * If device already registered then return base of 1
531 * to indicate not to probe for this interface
533 if (__dev_get_by_name(&init_net, name))
534 return 1;
536 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
537 if (!strcmp(name, s[i].name))
538 return s[i].map.base_addr;
539 return 0;
543 * Saves at boot time configured settings for any netdevice.
545 int __init netdev_boot_setup(char *str)
547 int ints[5];
548 struct ifmap map;
550 str = get_options(str, ARRAY_SIZE(ints), ints);
551 if (!str || !*str)
552 return 0;
554 /* Save settings */
555 memset(&map, 0, sizeof(map));
556 if (ints[0] > 0)
557 map.irq = ints[1];
558 if (ints[0] > 1)
559 map.base_addr = ints[2];
560 if (ints[0] > 2)
561 map.mem_start = ints[3];
562 if (ints[0] > 3)
563 map.mem_end = ints[4];
565 /* Add new entry to the list */
566 return netdev_boot_setup_add(str, &map);
569 __setup("netdev=", netdev_boot_setup);
571 /*******************************************************************************
573 Device Interface Subroutines
575 *******************************************************************************/
578 * __dev_get_by_name - find a device by its name
579 * @net: the applicable net namespace
580 * @name: name to find
582 * Find an interface by name. Must be called under RTNL semaphore
583 * or @dev_base_lock. If the name is found a pointer to the device
584 * is returned. If the name is not found then %NULL is returned. The
585 * reference counters are not incremented so the caller must be
586 * careful with locks.
589 struct net_device *__dev_get_by_name(struct net *net, const char *name)
591 struct hlist_node *p;
592 struct net_device *dev;
593 struct hlist_head *head = dev_name_hash(net, name);
595 hlist_for_each_entry(dev, p, head, name_hlist)
596 if (!strncmp(dev->name, name, IFNAMSIZ))
597 return dev;
599 return NULL;
601 EXPORT_SYMBOL(__dev_get_by_name);
604 * dev_get_by_name_rcu - find a device by its name
605 * @net: the applicable net namespace
606 * @name: name to find
608 * Find an interface by name.
609 * If the name is found a pointer to the device is returned.
610 * If the name is not found then %NULL is returned.
611 * The reference counters are not incremented so the caller must be
612 * careful with locks. The caller must hold RCU lock.
615 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
617 struct hlist_node *p;
618 struct net_device *dev;
619 struct hlist_head *head = dev_name_hash(net, name);
621 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
622 if (!strncmp(dev->name, name, IFNAMSIZ))
623 return dev;
625 return NULL;
627 EXPORT_SYMBOL(dev_get_by_name_rcu);
630 * dev_get_by_name - find a device by its name
631 * @net: the applicable net namespace
632 * @name: name to find
634 * Find an interface by name. This can be called from any
635 * context and does its own locking. The returned handle has
636 * the usage count incremented and the caller must use dev_put() to
637 * release it when it is no longer needed. %NULL is returned if no
638 * matching device is found.
641 struct net_device *dev_get_by_name(struct net *net, const char *name)
643 struct net_device *dev;
645 rcu_read_lock();
646 dev = dev_get_by_name_rcu(net, name);
647 if (dev)
648 dev_hold(dev);
649 rcu_read_unlock();
650 return dev;
652 EXPORT_SYMBOL(dev_get_by_name);
655 * __dev_get_by_index - find a device by its ifindex
656 * @net: the applicable net namespace
657 * @ifindex: index of device
659 * Search for an interface by index. Returns %NULL if the device
660 * is not found or a pointer to the device. The device has not
661 * had its reference counter increased so the caller must be careful
662 * about locking. The caller must hold either the RTNL semaphore
663 * or @dev_base_lock.
666 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
668 struct hlist_node *p;
669 struct net_device *dev;
670 struct hlist_head *head = dev_index_hash(net, ifindex);
672 hlist_for_each_entry(dev, p, head, index_hlist)
673 if (dev->ifindex == ifindex)
674 return dev;
676 return NULL;
678 EXPORT_SYMBOL(__dev_get_by_index);
681 * dev_get_by_index_rcu - find a device by its ifindex
682 * @net: the applicable net namespace
683 * @ifindex: index of device
685 * Search for an interface by index. Returns %NULL if the device
686 * is not found or a pointer to the device. The device has not
687 * had its reference counter increased so the caller must be careful
688 * about locking. The caller must hold RCU lock.
691 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
693 struct hlist_node *p;
694 struct net_device *dev;
695 struct hlist_head *head = dev_index_hash(net, ifindex);
697 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
698 if (dev->ifindex == ifindex)
699 return dev;
701 return NULL;
703 EXPORT_SYMBOL(dev_get_by_index_rcu);
707 * dev_get_by_index - find a device by its ifindex
708 * @net: the applicable net namespace
709 * @ifindex: index of device
711 * Search for an interface by index. Returns NULL if the device
712 * is not found or a pointer to the device. The device returned has
713 * had a reference added and the pointer is safe until the user calls
714 * dev_put to indicate they have finished with it.
717 struct net_device *dev_get_by_index(struct net *net, int ifindex)
719 struct net_device *dev;
721 rcu_read_lock();
722 dev = dev_get_by_index_rcu(net, ifindex);
723 if (dev)
724 dev_hold(dev);
725 rcu_read_unlock();
726 return dev;
728 EXPORT_SYMBOL(dev_get_by_index);
731 * dev_getbyhwaddr - find a device by its hardware address
732 * @net: the applicable net namespace
733 * @type: media type of device
734 * @ha: hardware address
736 * Search for an interface by MAC address. Returns NULL if the device
737 * is not found or a pointer to the device. The caller must hold the
738 * rtnl semaphore. The returned device has not had its ref count increased
739 * and the caller must therefore be careful about locking
741 * BUGS:
742 * If the API was consistent this would be __dev_get_by_hwaddr
745 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
747 struct net_device *dev;
749 ASSERT_RTNL();
751 for_each_netdev(net, dev)
752 if (dev->type == type &&
753 !memcmp(dev->dev_addr, ha, dev->addr_len))
754 return dev;
756 return NULL;
758 EXPORT_SYMBOL(dev_getbyhwaddr);
760 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
762 struct net_device *dev;
764 ASSERT_RTNL();
765 for_each_netdev(net, dev)
766 if (dev->type == type)
767 return dev;
769 return NULL;
771 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
773 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
775 struct net_device *dev;
777 rtnl_lock();
778 dev = __dev_getfirstbyhwtype(net, type);
779 if (dev)
780 dev_hold(dev);
781 rtnl_unlock();
782 return dev;
784 EXPORT_SYMBOL(dev_getfirstbyhwtype);
787 * dev_get_by_flags - find any device with given flags
788 * @net: the applicable net namespace
789 * @if_flags: IFF_* values
790 * @mask: bitmask of bits in if_flags to check
792 * Search for any interface with the given flags. Returns NULL if a device
793 * is not found or a pointer to the device. The device returned has
794 * had a reference added and the pointer is safe until the user calls
795 * dev_put to indicate they have finished with it.
798 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
799 unsigned short mask)
801 struct net_device *dev, *ret;
803 ret = NULL;
804 rcu_read_lock();
805 for_each_netdev_rcu(net, dev) {
806 if (((dev->flags ^ if_flags) & mask) == 0) {
807 dev_hold(dev);
808 ret = dev;
809 break;
812 rcu_read_unlock();
813 return ret;
815 EXPORT_SYMBOL(dev_get_by_flags);
818 * dev_valid_name - check if name is okay for network device
819 * @name: name string
821 * Network device names need to be valid file names to
822 * to allow sysfs to work. We also disallow any kind of
823 * whitespace.
825 int dev_valid_name(const char *name)
827 if (*name == '\0')
828 return 0;
829 if (strlen(name) >= IFNAMSIZ)
830 return 0;
831 if (!strcmp(name, ".") || !strcmp(name, ".."))
832 return 0;
834 while (*name) {
835 if (*name == '/' || isspace(*name))
836 return 0;
837 name++;
839 return 1;
841 EXPORT_SYMBOL(dev_valid_name);
844 * __dev_alloc_name - allocate a name for a device
845 * @net: network namespace to allocate the device name in
846 * @name: name format string
847 * @buf: scratch buffer and result name string
849 * Passed a format string - eg "lt%d" it will try and find a suitable
850 * id. It scans list of devices to build up a free map, then chooses
851 * the first empty slot. The caller must hold the dev_base or rtnl lock
852 * while allocating the name and adding the device in order to avoid
853 * duplicates.
854 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
855 * Returns the number of the unit assigned or a negative errno code.
858 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
860 int i = 0;
861 const char *p;
862 const int max_netdevices = 8*PAGE_SIZE;
863 unsigned long *inuse;
864 struct net_device *d;
866 p = strnchr(name, IFNAMSIZ-1, '%');
867 if (p) {
869 * Verify the string as this thing may have come from
870 * the user. There must be either one "%d" and no other "%"
871 * characters.
873 if (p[1] != 'd' || strchr(p + 2, '%'))
874 return -EINVAL;
876 /* Use one page as a bit array of possible slots */
877 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
878 if (!inuse)
879 return -ENOMEM;
881 for_each_netdev(net, d) {
882 if (!sscanf(d->name, name, &i))
883 continue;
884 if (i < 0 || i >= max_netdevices)
885 continue;
887 /* avoid cases where sscanf is not exact inverse of printf */
888 snprintf(buf, IFNAMSIZ, name, i);
889 if (!strncmp(buf, d->name, IFNAMSIZ))
890 set_bit(i, inuse);
893 i = find_first_zero_bit(inuse, max_netdevices);
894 free_page((unsigned long) inuse);
897 if (buf != name)
898 snprintf(buf, IFNAMSIZ, name, i);
899 if (!__dev_get_by_name(net, buf))
900 return i;
902 /* It is possible to run out of possible slots
903 * when the name is long and there isn't enough space left
904 * for the digits, or if all bits are used.
906 return -ENFILE;
910 * dev_alloc_name - allocate a name for a device
911 * @dev: device
912 * @name: name format string
914 * Passed a format string - eg "lt%d" it will try and find a suitable
915 * id. It scans list of devices to build up a free map, then chooses
916 * the first empty slot. The caller must hold the dev_base or rtnl lock
917 * while allocating the name and adding the device in order to avoid
918 * duplicates.
919 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
920 * Returns the number of the unit assigned or a negative errno code.
923 int dev_alloc_name(struct net_device *dev, const char *name)
925 char buf[IFNAMSIZ];
926 struct net *net;
927 int ret;
929 BUG_ON(!dev_net(dev));
930 net = dev_net(dev);
931 ret = __dev_alloc_name(net, name, buf);
932 if (ret >= 0)
933 strlcpy(dev->name, buf, IFNAMSIZ);
934 return ret;
936 EXPORT_SYMBOL(dev_alloc_name);
938 static int dev_get_valid_name(struct net *net, const char *name, char *buf,
939 bool fmt)
941 if (!dev_valid_name(name))
942 return -EINVAL;
944 if (fmt && strchr(name, '%'))
945 return __dev_alloc_name(net, name, buf);
946 else if (__dev_get_by_name(net, name))
947 return -EEXIST;
948 else if (buf != name)
949 strlcpy(buf, name, IFNAMSIZ);
951 return 0;
955 * dev_change_name - change name of a device
956 * @dev: device
957 * @newname: name (or format string) must be at least IFNAMSIZ
959 * Change name of a device, can pass format strings "eth%d".
960 * for wildcarding.
962 int dev_change_name(struct net_device *dev, const char *newname)
964 char oldname[IFNAMSIZ];
965 int err = 0;
966 int ret;
967 struct net *net;
969 ASSERT_RTNL();
970 BUG_ON(!dev_net(dev));
972 net = dev_net(dev);
973 if (dev->flags & IFF_UP)
974 return -EBUSY;
976 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
977 return 0;
979 memcpy(oldname, dev->name, IFNAMSIZ);
981 err = dev_get_valid_name(net, newname, dev->name, 1);
982 if (err < 0)
983 return err;
985 rollback:
986 /* For now only devices in the initial network namespace
987 * are in sysfs.
989 if (net_eq(net, &init_net)) {
990 ret = device_rename(&dev->dev, dev->name);
991 if (ret) {
992 memcpy(dev->name, oldname, IFNAMSIZ);
993 return ret;
997 write_lock_bh(&dev_base_lock);
998 hlist_del(&dev->name_hlist);
999 write_unlock_bh(&dev_base_lock);
1001 synchronize_rcu();
1003 write_lock_bh(&dev_base_lock);
1004 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1005 write_unlock_bh(&dev_base_lock);
1007 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1008 ret = notifier_to_errno(ret);
1010 if (ret) {
1011 /* err >= 0 after dev_alloc_name() or stores the first errno */
1012 if (err >= 0) {
1013 err = ret;
1014 memcpy(dev->name, oldname, IFNAMSIZ);
1015 goto rollback;
1016 } else {
1017 printk(KERN_ERR
1018 "%s: name change rollback failed: %d.\n",
1019 dev->name, ret);
1023 return err;
1027 * dev_set_alias - change ifalias of a device
1028 * @dev: device
1029 * @alias: name up to IFALIASZ
1030 * @len: limit of bytes to copy from info
1032 * Set ifalias for a device,
1034 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1036 ASSERT_RTNL();
1038 if (len >= IFALIASZ)
1039 return -EINVAL;
1041 if (!len) {
1042 if (dev->ifalias) {
1043 kfree(dev->ifalias);
1044 dev->ifalias = NULL;
1046 return 0;
1049 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1050 if (!dev->ifalias)
1051 return -ENOMEM;
1053 strlcpy(dev->ifalias, alias, len+1);
1054 return len;
1059 * netdev_features_change - device changes features
1060 * @dev: device to cause notification
1062 * Called to indicate a device has changed features.
1064 void netdev_features_change(struct net_device *dev)
1066 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1068 EXPORT_SYMBOL(netdev_features_change);
1071 * netdev_state_change - device changes state
1072 * @dev: device to cause notification
1074 * Called to indicate a device has changed state. This function calls
1075 * the notifier chains for netdev_chain and sends a NEWLINK message
1076 * to the routing socket.
1078 void netdev_state_change(struct net_device *dev)
1080 if (dev->flags & IFF_UP) {
1081 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1082 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1085 EXPORT_SYMBOL(netdev_state_change);
1087 void netdev_bonding_change(struct net_device *dev, unsigned long event)
1089 call_netdevice_notifiers(event, dev);
1091 EXPORT_SYMBOL(netdev_bonding_change);
1094 * dev_load - load a network module
1095 * @net: the applicable net namespace
1096 * @name: name of interface
1098 * If a network interface is not present and the process has suitable
1099 * privileges this function loads the module. If module loading is not
1100 * available in this kernel then it becomes a nop.
1103 void dev_load(struct net *net, const char *name)
1105 struct net_device *dev;
1107 rcu_read_lock();
1108 dev = dev_get_by_name_rcu(net, name);
1109 rcu_read_unlock();
1111 if (!dev && capable(CAP_NET_ADMIN))
1112 request_module("%s", name);
1114 EXPORT_SYMBOL(dev_load);
1117 * dev_open - prepare an interface for use.
1118 * @dev: device to open
1120 * Takes a device from down to up state. The device's private open
1121 * function is invoked and then the multicast lists are loaded. Finally
1122 * the device is moved into the up state and a %NETDEV_UP message is
1123 * sent to the netdev notifier chain.
1125 * Calling this function on an active interface is a nop. On a failure
1126 * a negative errno code is returned.
1128 int dev_open(struct net_device *dev)
1130 const struct net_device_ops *ops = dev->netdev_ops;
1131 int ret;
1133 ASSERT_RTNL();
1136 * Is it already up?
1139 if (dev->flags & IFF_UP)
1140 return 0;
1143 * Is it even present?
1145 if (!netif_device_present(dev))
1146 return -ENODEV;
1148 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1149 ret = notifier_to_errno(ret);
1150 if (ret)
1151 return ret;
1154 * Call device private open method
1156 set_bit(__LINK_STATE_START, &dev->state);
1158 if (ops->ndo_validate_addr)
1159 ret = ops->ndo_validate_addr(dev);
1161 if (!ret && ops->ndo_open)
1162 ret = ops->ndo_open(dev);
1165 * If it went open OK then:
1168 if (ret)
1169 clear_bit(__LINK_STATE_START, &dev->state);
1170 else {
1172 * Set the flags.
1174 dev->flags |= IFF_UP;
1177 * Enable NET_DMA
1179 net_dmaengine_get();
1182 * Initialize multicasting status
1184 dev_set_rx_mode(dev);
1187 * Wakeup transmit queue engine
1189 dev_activate(dev);
1192 * ... and announce new interface.
1194 call_netdevice_notifiers(NETDEV_UP, dev);
1197 return ret;
1199 EXPORT_SYMBOL(dev_open);
1202 * dev_close - shutdown an interface.
1203 * @dev: device to shutdown
1205 * This function moves an active device into down state. A
1206 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1207 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1208 * chain.
1210 int dev_close(struct net_device *dev)
1212 const struct net_device_ops *ops = dev->netdev_ops;
1213 ASSERT_RTNL();
1215 might_sleep();
1217 if (!(dev->flags & IFF_UP))
1218 return 0;
1221 * Tell people we are going down, so that they can
1222 * prepare to death, when device is still operating.
1224 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1226 clear_bit(__LINK_STATE_START, &dev->state);
1228 /* Synchronize to scheduled poll. We cannot touch poll list,
1229 * it can be even on different cpu. So just clear netif_running().
1231 * dev->stop() will invoke napi_disable() on all of it's
1232 * napi_struct instances on this device.
1234 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1236 dev_deactivate(dev);
1239 * Call the device specific close. This cannot fail.
1240 * Only if device is UP
1242 * We allow it to be called even after a DETACH hot-plug
1243 * event.
1245 if (ops->ndo_stop)
1246 ops->ndo_stop(dev);
1249 * Device is now down.
1252 dev->flags &= ~IFF_UP;
1255 * Tell people we are down
1257 call_netdevice_notifiers(NETDEV_DOWN, dev);
1260 * Shutdown NET_DMA
1262 net_dmaengine_put();
1264 return 0;
1266 EXPORT_SYMBOL(dev_close);
1270 * dev_disable_lro - disable Large Receive Offload on a device
1271 * @dev: device
1273 * Disable Large Receive Offload (LRO) on a net device. Must be
1274 * called under RTNL. This is needed if received packets may be
1275 * forwarded to another interface.
1277 void dev_disable_lro(struct net_device *dev)
1279 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1280 dev->ethtool_ops->set_flags) {
1281 u32 flags = dev->ethtool_ops->get_flags(dev);
1282 if (flags & ETH_FLAG_LRO) {
1283 flags &= ~ETH_FLAG_LRO;
1284 dev->ethtool_ops->set_flags(dev, flags);
1287 WARN_ON(dev->features & NETIF_F_LRO);
1289 EXPORT_SYMBOL(dev_disable_lro);
1292 static int dev_boot_phase = 1;
1295 * Device change register/unregister. These are not inline or static
1296 * as we export them to the world.
1300 * register_netdevice_notifier - register a network notifier block
1301 * @nb: notifier
1303 * Register a notifier to be called when network device events occur.
1304 * The notifier passed is linked into the kernel structures and must
1305 * not be reused until it has been unregistered. A negative errno code
1306 * is returned on a failure.
1308 * When registered all registration and up events are replayed
1309 * to the new notifier to allow device to have a race free
1310 * view of the network device list.
1313 int register_netdevice_notifier(struct notifier_block *nb)
1315 struct net_device *dev;
1316 struct net_device *last;
1317 struct net *net;
1318 int err;
1320 rtnl_lock();
1321 err = raw_notifier_chain_register(&netdev_chain, nb);
1322 if (err)
1323 goto unlock;
1324 if (dev_boot_phase)
1325 goto unlock;
1326 for_each_net(net) {
1327 for_each_netdev(net, dev) {
1328 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1329 err = notifier_to_errno(err);
1330 if (err)
1331 goto rollback;
1333 if (!(dev->flags & IFF_UP))
1334 continue;
1336 nb->notifier_call(nb, NETDEV_UP, dev);
1340 unlock:
1341 rtnl_unlock();
1342 return err;
1344 rollback:
1345 last = dev;
1346 for_each_net(net) {
1347 for_each_netdev(net, dev) {
1348 if (dev == last)
1349 break;
1351 if (dev->flags & IFF_UP) {
1352 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1353 nb->notifier_call(nb, NETDEV_DOWN, dev);
1355 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1356 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1360 raw_notifier_chain_unregister(&netdev_chain, nb);
1361 goto unlock;
1363 EXPORT_SYMBOL(register_netdevice_notifier);
1366 * unregister_netdevice_notifier - unregister a network notifier block
1367 * @nb: notifier
1369 * Unregister a notifier previously registered by
1370 * register_netdevice_notifier(). The notifier is unlinked into the
1371 * kernel structures and may then be reused. A negative errno code
1372 * is returned on a failure.
1375 int unregister_netdevice_notifier(struct notifier_block *nb)
1377 int err;
1379 rtnl_lock();
1380 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1381 rtnl_unlock();
1382 return err;
1384 EXPORT_SYMBOL(unregister_netdevice_notifier);
1387 * call_netdevice_notifiers - call all network notifier blocks
1388 * @val: value passed unmodified to notifier function
1389 * @dev: net_device pointer passed unmodified to notifier function
1391 * Call all network notifier blocks. Parameters and return value
1392 * are as for raw_notifier_call_chain().
1395 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1397 return raw_notifier_call_chain(&netdev_chain, val, dev);
1400 /* When > 0 there are consumers of rx skb time stamps */
1401 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1403 void net_enable_timestamp(void)
1405 atomic_inc(&netstamp_needed);
1407 EXPORT_SYMBOL(net_enable_timestamp);
1409 void net_disable_timestamp(void)
1411 atomic_dec(&netstamp_needed);
1413 EXPORT_SYMBOL(net_disable_timestamp);
1415 static inline void net_timestamp(struct sk_buff *skb)
1417 if (atomic_read(&netstamp_needed))
1418 __net_timestamp(skb);
1419 else
1420 skb->tstamp.tv64 = 0;
1424 * dev_forward_skb - loopback an skb to another netif
1426 * @dev: destination network device
1427 * @skb: buffer to forward
1429 * return values:
1430 * NET_RX_SUCCESS (no congestion)
1431 * NET_RX_DROP (packet was dropped)
1433 * dev_forward_skb can be used for injecting an skb from the
1434 * start_xmit function of one device into the receive queue
1435 * of another device.
1437 * The receiving device may be in another namespace, so
1438 * we have to clear all information in the skb that could
1439 * impact namespace isolation.
1441 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1443 skb_orphan(skb);
1445 if (!(dev->flags & IFF_UP))
1446 return NET_RX_DROP;
1448 if (skb->len > (dev->mtu + dev->hard_header_len))
1449 return NET_RX_DROP;
1451 skb_dst_drop(skb);
1452 skb->tstamp.tv64 = 0;
1453 skb->pkt_type = PACKET_HOST;
1454 skb->protocol = eth_type_trans(skb, dev);
1455 skb->mark = 0;
1456 secpath_reset(skb);
1457 nf_reset(skb);
1458 return netif_rx(skb);
1460 EXPORT_SYMBOL_GPL(dev_forward_skb);
1463 * Support routine. Sends outgoing frames to any network
1464 * taps currently in use.
1467 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1469 struct packet_type *ptype;
1471 #ifdef CONFIG_NET_CLS_ACT
1472 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1473 net_timestamp(skb);
1474 #else
1475 net_timestamp(skb);
1476 #endif
1478 rcu_read_lock();
1479 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1480 /* Never send packets back to the socket
1481 * they originated from - MvS (miquels@drinkel.ow.org)
1483 if ((ptype->dev == dev || !ptype->dev) &&
1484 (ptype->af_packet_priv == NULL ||
1485 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1486 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1487 if (!skb2)
1488 break;
1490 /* skb->nh should be correctly
1491 set by sender, so that the second statement is
1492 just protection against buggy protocols.
1494 skb_reset_mac_header(skb2);
1496 if (skb_network_header(skb2) < skb2->data ||
1497 skb2->network_header > skb2->tail) {
1498 if (net_ratelimit())
1499 printk(KERN_CRIT "protocol %04x is "
1500 "buggy, dev %s\n",
1501 skb2->protocol, dev->name);
1502 skb_reset_network_header(skb2);
1505 skb2->transport_header = skb2->network_header;
1506 skb2->pkt_type = PACKET_OUTGOING;
1507 ptype->func(skb2, skb->dev, ptype, skb->dev);
1510 rcu_read_unlock();
1514 static inline void __netif_reschedule(struct Qdisc *q)
1516 struct softnet_data *sd;
1517 unsigned long flags;
1519 local_irq_save(flags);
1520 sd = &__get_cpu_var(softnet_data);
1521 q->next_sched = sd->output_queue;
1522 sd->output_queue = q;
1523 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1524 local_irq_restore(flags);
1527 void __netif_schedule(struct Qdisc *q)
1529 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1530 __netif_reschedule(q);
1532 EXPORT_SYMBOL(__netif_schedule);
1534 void dev_kfree_skb_irq(struct sk_buff *skb)
1536 if (atomic_dec_and_test(&skb->users)) {
1537 struct softnet_data *sd;
1538 unsigned long flags;
1540 local_irq_save(flags);
1541 sd = &__get_cpu_var(softnet_data);
1542 skb->next = sd->completion_queue;
1543 sd->completion_queue = skb;
1544 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1545 local_irq_restore(flags);
1548 EXPORT_SYMBOL(dev_kfree_skb_irq);
1550 void dev_kfree_skb_any(struct sk_buff *skb)
1552 if (in_irq() || irqs_disabled())
1553 dev_kfree_skb_irq(skb);
1554 else
1555 dev_kfree_skb(skb);
1557 EXPORT_SYMBOL(dev_kfree_skb_any);
1561 * netif_device_detach - mark device as removed
1562 * @dev: network device
1564 * Mark device as removed from system and therefore no longer available.
1566 void netif_device_detach(struct net_device *dev)
1568 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1569 netif_running(dev)) {
1570 netif_tx_stop_all_queues(dev);
1573 EXPORT_SYMBOL(netif_device_detach);
1576 * netif_device_attach - mark device as attached
1577 * @dev: network device
1579 * Mark device as attached from system and restart if needed.
1581 void netif_device_attach(struct net_device *dev)
1583 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1584 netif_running(dev)) {
1585 netif_tx_wake_all_queues(dev);
1586 __netdev_watchdog_up(dev);
1589 EXPORT_SYMBOL(netif_device_attach);
1591 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1593 return ((features & NETIF_F_GEN_CSUM) ||
1594 ((features & NETIF_F_IP_CSUM) &&
1595 protocol == htons(ETH_P_IP)) ||
1596 ((features & NETIF_F_IPV6_CSUM) &&
1597 protocol == htons(ETH_P_IPV6)) ||
1598 ((features & NETIF_F_FCOE_CRC) &&
1599 protocol == htons(ETH_P_FCOE)));
1602 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1604 if (can_checksum_protocol(dev->features, skb->protocol))
1605 return true;
1607 if (skb->protocol == htons(ETH_P_8021Q)) {
1608 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1609 if (can_checksum_protocol(dev->features & dev->vlan_features,
1610 veh->h_vlan_encapsulated_proto))
1611 return true;
1614 return false;
1618 * Invalidate hardware checksum when packet is to be mangled, and
1619 * complete checksum manually on outgoing path.
1621 int skb_checksum_help(struct sk_buff *skb)
1623 __wsum csum;
1624 int ret = 0, offset;
1626 if (skb->ip_summed == CHECKSUM_COMPLETE)
1627 goto out_set_summed;
1629 if (unlikely(skb_shinfo(skb)->gso_size)) {
1630 /* Let GSO fix up the checksum. */
1631 goto out_set_summed;
1634 offset = skb->csum_start - skb_headroom(skb);
1635 BUG_ON(offset >= skb_headlen(skb));
1636 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1638 offset += skb->csum_offset;
1639 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1641 if (skb_cloned(skb) &&
1642 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1643 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1644 if (ret)
1645 goto out;
1648 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1649 out_set_summed:
1650 skb->ip_summed = CHECKSUM_NONE;
1651 out:
1652 return ret;
1654 EXPORT_SYMBOL(skb_checksum_help);
1657 * skb_gso_segment - Perform segmentation on skb.
1658 * @skb: buffer to segment
1659 * @features: features for the output path (see dev->features)
1661 * This function segments the given skb and returns a list of segments.
1663 * It may return NULL if the skb requires no segmentation. This is
1664 * only possible when GSO is used for verifying header integrity.
1666 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1668 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1669 struct packet_type *ptype;
1670 __be16 type = skb->protocol;
1671 int err;
1673 skb_reset_mac_header(skb);
1674 skb->mac_len = skb->network_header - skb->mac_header;
1675 __skb_pull(skb, skb->mac_len);
1677 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1678 struct net_device *dev = skb->dev;
1679 struct ethtool_drvinfo info = {};
1681 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1682 dev->ethtool_ops->get_drvinfo(dev, &info);
1684 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1685 "ip_summed=%d",
1686 info.driver, dev ? dev->features : 0L,
1687 skb->sk ? skb->sk->sk_route_caps : 0L,
1688 skb->len, skb->data_len, skb->ip_summed);
1690 if (skb_header_cloned(skb) &&
1691 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1692 return ERR_PTR(err);
1695 rcu_read_lock();
1696 list_for_each_entry_rcu(ptype,
1697 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1698 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1699 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1700 err = ptype->gso_send_check(skb);
1701 segs = ERR_PTR(err);
1702 if (err || skb_gso_ok(skb, features))
1703 break;
1704 __skb_push(skb, (skb->data -
1705 skb_network_header(skb)));
1707 segs = ptype->gso_segment(skb, features);
1708 break;
1711 rcu_read_unlock();
1713 __skb_push(skb, skb->data - skb_mac_header(skb));
1715 return segs;
1717 EXPORT_SYMBOL(skb_gso_segment);
1719 /* Take action when hardware reception checksum errors are detected. */
1720 #ifdef CONFIG_BUG
1721 void netdev_rx_csum_fault(struct net_device *dev)
1723 if (net_ratelimit()) {
1724 printk(KERN_ERR "%s: hw csum failure.\n",
1725 dev ? dev->name : "<unknown>");
1726 dump_stack();
1729 EXPORT_SYMBOL(netdev_rx_csum_fault);
1730 #endif
1732 /* Actually, we should eliminate this check as soon as we know, that:
1733 * 1. IOMMU is present and allows to map all the memory.
1734 * 2. No high memory really exists on this machine.
1737 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1739 #ifdef CONFIG_HIGHMEM
1740 int i;
1742 if (dev->features & NETIF_F_HIGHDMA)
1743 return 0;
1745 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1746 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1747 return 1;
1749 #endif
1750 return 0;
1753 struct dev_gso_cb {
1754 void (*destructor)(struct sk_buff *skb);
1757 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1759 static void dev_gso_skb_destructor(struct sk_buff *skb)
1761 struct dev_gso_cb *cb;
1763 do {
1764 struct sk_buff *nskb = skb->next;
1766 skb->next = nskb->next;
1767 nskb->next = NULL;
1768 kfree_skb(nskb);
1769 } while (skb->next);
1771 cb = DEV_GSO_CB(skb);
1772 if (cb->destructor)
1773 cb->destructor(skb);
1777 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1778 * @skb: buffer to segment
1780 * This function segments the given skb and stores the list of segments
1781 * in skb->next.
1783 static int dev_gso_segment(struct sk_buff *skb)
1785 struct net_device *dev = skb->dev;
1786 struct sk_buff *segs;
1787 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1788 NETIF_F_SG : 0);
1790 segs = skb_gso_segment(skb, features);
1792 /* Verifying header integrity only. */
1793 if (!segs)
1794 return 0;
1796 if (IS_ERR(segs))
1797 return PTR_ERR(segs);
1799 skb->next = segs;
1800 DEV_GSO_CB(skb)->destructor = skb->destructor;
1801 skb->destructor = dev_gso_skb_destructor;
1803 return 0;
1806 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1807 struct netdev_queue *txq)
1809 const struct net_device_ops *ops = dev->netdev_ops;
1810 int rc = NETDEV_TX_OK;
1812 if (likely(!skb->next)) {
1813 if (!list_empty(&ptype_all))
1814 dev_queue_xmit_nit(skb, dev);
1816 if (netif_needs_gso(dev, skb)) {
1817 if (unlikely(dev_gso_segment(skb)))
1818 goto out_kfree_skb;
1819 if (skb->next)
1820 goto gso;
1824 * If device doesnt need skb->dst, release it right now while
1825 * its hot in this cpu cache
1827 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1828 skb_dst_drop(skb);
1830 rc = ops->ndo_start_xmit(skb, dev);
1831 if (rc == NETDEV_TX_OK)
1832 txq_trans_update(txq);
1834 * TODO: if skb_orphan() was called by
1835 * dev->hard_start_xmit() (for example, the unmodified
1836 * igb driver does that; bnx2 doesn't), then
1837 * skb_tx_software_timestamp() will be unable to send
1838 * back the time stamp.
1840 * How can this be prevented? Always create another
1841 * reference to the socket before calling
1842 * dev->hard_start_xmit()? Prevent that skb_orphan()
1843 * does anything in dev->hard_start_xmit() by clearing
1844 * the skb destructor before the call and restoring it
1845 * afterwards, then doing the skb_orphan() ourselves?
1847 return rc;
1850 gso:
1851 do {
1852 struct sk_buff *nskb = skb->next;
1854 skb->next = nskb->next;
1855 nskb->next = NULL;
1856 rc = ops->ndo_start_xmit(nskb, dev);
1857 if (unlikely(rc != NETDEV_TX_OK)) {
1858 if (rc & ~NETDEV_TX_MASK)
1859 goto out_kfree_gso_skb;
1860 nskb->next = skb->next;
1861 skb->next = nskb;
1862 return rc;
1864 txq_trans_update(txq);
1865 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1866 return NETDEV_TX_BUSY;
1867 } while (skb->next);
1869 out_kfree_gso_skb:
1870 if (likely(skb->next == NULL))
1871 skb->destructor = DEV_GSO_CB(skb)->destructor;
1872 out_kfree_skb:
1873 kfree_skb(skb);
1874 return rc;
1877 static u32 skb_tx_hashrnd;
1879 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1881 u32 hash;
1883 if (skb_rx_queue_recorded(skb)) {
1884 hash = skb_get_rx_queue(skb);
1885 while (unlikely(hash >= dev->real_num_tx_queues))
1886 hash -= dev->real_num_tx_queues;
1887 return hash;
1890 if (skb->sk && skb->sk->sk_hash)
1891 hash = skb->sk->sk_hash;
1892 else
1893 hash = skb->protocol;
1895 hash = jhash_1word(hash, skb_tx_hashrnd);
1897 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1899 EXPORT_SYMBOL(skb_tx_hash);
1901 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1903 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1904 if (net_ratelimit()) {
1905 WARN(1, "%s selects TX queue %d, but "
1906 "real number of TX queues is %d\n",
1907 dev->name, queue_index,
1908 dev->real_num_tx_queues);
1910 return 0;
1912 return queue_index;
1915 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1916 struct sk_buff *skb)
1918 u16 queue_index;
1919 struct sock *sk = skb->sk;
1921 if (sk_tx_queue_recorded(sk)) {
1922 queue_index = sk_tx_queue_get(sk);
1923 } else {
1924 const struct net_device_ops *ops = dev->netdev_ops;
1926 if (ops->ndo_select_queue) {
1927 queue_index = ops->ndo_select_queue(dev, skb);
1928 queue_index = dev_cap_txqueue(dev, queue_index);
1929 } else {
1930 queue_index = 0;
1931 if (dev->real_num_tx_queues > 1)
1932 queue_index = skb_tx_hash(dev, skb);
1934 if (sk && sk->sk_dst_cache)
1935 sk_tx_queue_set(sk, queue_index);
1939 skb_set_queue_mapping(skb, queue_index);
1940 return netdev_get_tx_queue(dev, queue_index);
1943 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
1944 struct net_device *dev,
1945 struct netdev_queue *txq)
1947 spinlock_t *root_lock = qdisc_lock(q);
1948 int rc;
1950 spin_lock(root_lock);
1951 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1952 kfree_skb(skb);
1953 rc = NET_XMIT_DROP;
1954 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
1955 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
1957 * This is a work-conserving queue; there are no old skbs
1958 * waiting to be sent out; and the qdisc is not running -
1959 * xmit the skb directly.
1961 __qdisc_update_bstats(q, skb->len);
1962 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
1963 __qdisc_run(q);
1964 else
1965 clear_bit(__QDISC_STATE_RUNNING, &q->state);
1967 rc = NET_XMIT_SUCCESS;
1968 } else {
1969 rc = qdisc_enqueue_root(skb, q);
1970 qdisc_run(q);
1972 spin_unlock(root_lock);
1974 return rc;
1978 * dev_queue_xmit - transmit a buffer
1979 * @skb: buffer to transmit
1981 * Queue a buffer for transmission to a network device. The caller must
1982 * have set the device and priority and built the buffer before calling
1983 * this function. The function can be called from an interrupt.
1985 * A negative errno code is returned on a failure. A success does not
1986 * guarantee the frame will be transmitted as it may be dropped due
1987 * to congestion or traffic shaping.
1989 * -----------------------------------------------------------------------------------
1990 * I notice this method can also return errors from the queue disciplines,
1991 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1992 * be positive.
1994 * Regardless of the return value, the skb is consumed, so it is currently
1995 * difficult to retry a send to this method. (You can bump the ref count
1996 * before sending to hold a reference for retry if you are careful.)
1998 * When calling this method, interrupts MUST be enabled. This is because
1999 * the BH enable code must have IRQs enabled so that it will not deadlock.
2000 * --BLG
2002 int dev_queue_xmit(struct sk_buff *skb)
2004 struct net_device *dev = skb->dev;
2005 struct netdev_queue *txq;
2006 struct Qdisc *q;
2007 int rc = -ENOMEM;
2009 /* GSO will handle the following emulations directly. */
2010 if (netif_needs_gso(dev, skb))
2011 goto gso;
2013 if (skb_has_frags(skb) &&
2014 !(dev->features & NETIF_F_FRAGLIST) &&
2015 __skb_linearize(skb))
2016 goto out_kfree_skb;
2018 /* Fragmented skb is linearized if device does not support SG,
2019 * or if at least one of fragments is in highmem and device
2020 * does not support DMA from it.
2022 if (skb_shinfo(skb)->nr_frags &&
2023 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
2024 __skb_linearize(skb))
2025 goto out_kfree_skb;
2027 /* If packet is not checksummed and device does not support
2028 * checksumming for this protocol, complete checksumming here.
2030 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2031 skb_set_transport_header(skb, skb->csum_start -
2032 skb_headroom(skb));
2033 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2034 goto out_kfree_skb;
2037 gso:
2038 /* Disable soft irqs for various locks below. Also
2039 * stops preemption for RCU.
2041 rcu_read_lock_bh();
2043 txq = dev_pick_tx(dev, skb);
2044 q = rcu_dereference(txq->qdisc);
2046 #ifdef CONFIG_NET_CLS_ACT
2047 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2048 #endif
2049 if (q->enqueue) {
2050 rc = __dev_xmit_skb(skb, q, dev, txq);
2051 goto out;
2054 /* The device has no queue. Common case for software devices:
2055 loopback, all the sorts of tunnels...
2057 Really, it is unlikely that netif_tx_lock protection is necessary
2058 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2059 counters.)
2060 However, it is possible, that they rely on protection
2061 made by us here.
2063 Check this and shot the lock. It is not prone from deadlocks.
2064 Either shot noqueue qdisc, it is even simpler 8)
2066 if (dev->flags & IFF_UP) {
2067 int cpu = smp_processor_id(); /* ok because BHs are off */
2069 if (txq->xmit_lock_owner != cpu) {
2071 HARD_TX_LOCK(dev, txq, cpu);
2073 if (!netif_tx_queue_stopped(txq)) {
2074 rc = dev_hard_start_xmit(skb, dev, txq);
2075 if (dev_xmit_complete(rc)) {
2076 HARD_TX_UNLOCK(dev, txq);
2077 goto out;
2080 HARD_TX_UNLOCK(dev, txq);
2081 if (net_ratelimit())
2082 printk(KERN_CRIT "Virtual device %s asks to "
2083 "queue packet!\n", dev->name);
2084 } else {
2085 /* Recursion is detected! It is possible,
2086 * unfortunately */
2087 if (net_ratelimit())
2088 printk(KERN_CRIT "Dead loop on virtual device "
2089 "%s, fix it urgently!\n", dev->name);
2093 rc = -ENETDOWN;
2094 rcu_read_unlock_bh();
2096 out_kfree_skb:
2097 kfree_skb(skb);
2098 return rc;
2099 out:
2100 rcu_read_unlock_bh();
2101 return rc;
2103 EXPORT_SYMBOL(dev_queue_xmit);
2106 /*=======================================================================
2107 Receiver routines
2108 =======================================================================*/
2110 int netdev_max_backlog __read_mostly = 1000;
2111 int netdev_budget __read_mostly = 300;
2112 int weight_p __read_mostly = 64; /* old backlog weight */
2114 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2118 * netif_rx - post buffer to the network code
2119 * @skb: buffer to post
2121 * This function receives a packet from a device driver and queues it for
2122 * the upper (protocol) levels to process. It always succeeds. The buffer
2123 * may be dropped during processing for congestion control or by the
2124 * protocol layers.
2126 * return values:
2127 * NET_RX_SUCCESS (no congestion)
2128 * NET_RX_DROP (packet was dropped)
2132 int netif_rx(struct sk_buff *skb)
2134 struct softnet_data *queue;
2135 unsigned long flags;
2137 /* if netpoll wants it, pretend we never saw it */
2138 if (netpoll_rx(skb))
2139 return NET_RX_DROP;
2141 if (!skb->tstamp.tv64)
2142 net_timestamp(skb);
2145 * The code is rearranged so that the path is the most
2146 * short when CPU is congested, but is still operating.
2148 local_irq_save(flags);
2149 queue = &__get_cpu_var(softnet_data);
2151 __get_cpu_var(netdev_rx_stat).total++;
2152 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2153 if (queue->input_pkt_queue.qlen) {
2154 enqueue:
2155 __skb_queue_tail(&queue->input_pkt_queue, skb);
2156 local_irq_restore(flags);
2157 return NET_RX_SUCCESS;
2160 napi_schedule(&queue->backlog);
2161 goto enqueue;
2164 __get_cpu_var(netdev_rx_stat).dropped++;
2165 local_irq_restore(flags);
2167 kfree_skb(skb);
2168 return NET_RX_DROP;
2170 EXPORT_SYMBOL(netif_rx);
2172 int netif_rx_ni(struct sk_buff *skb)
2174 int err;
2176 preempt_disable();
2177 err = netif_rx(skb);
2178 if (local_softirq_pending())
2179 do_softirq();
2180 preempt_enable();
2182 return err;
2184 EXPORT_SYMBOL(netif_rx_ni);
2186 static void net_tx_action(struct softirq_action *h)
2188 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2190 if (sd->completion_queue) {
2191 struct sk_buff *clist;
2193 local_irq_disable();
2194 clist = sd->completion_queue;
2195 sd->completion_queue = NULL;
2196 local_irq_enable();
2198 while (clist) {
2199 struct sk_buff *skb = clist;
2200 clist = clist->next;
2202 WARN_ON(atomic_read(&skb->users));
2203 __kfree_skb(skb);
2207 if (sd->output_queue) {
2208 struct Qdisc *head;
2210 local_irq_disable();
2211 head = sd->output_queue;
2212 sd->output_queue = NULL;
2213 local_irq_enable();
2215 while (head) {
2216 struct Qdisc *q = head;
2217 spinlock_t *root_lock;
2219 head = head->next_sched;
2221 root_lock = qdisc_lock(q);
2222 if (spin_trylock(root_lock)) {
2223 smp_mb__before_clear_bit();
2224 clear_bit(__QDISC_STATE_SCHED,
2225 &q->state);
2226 qdisc_run(q);
2227 spin_unlock(root_lock);
2228 } else {
2229 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2230 &q->state)) {
2231 __netif_reschedule(q);
2232 } else {
2233 smp_mb__before_clear_bit();
2234 clear_bit(__QDISC_STATE_SCHED,
2235 &q->state);
2242 static inline int deliver_skb(struct sk_buff *skb,
2243 struct packet_type *pt_prev,
2244 struct net_device *orig_dev)
2246 atomic_inc(&skb->users);
2247 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2250 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2252 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2253 /* This hook is defined here for ATM LANE */
2254 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2255 unsigned char *addr) __read_mostly;
2256 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2257 #endif
2260 * If bridge module is loaded call bridging hook.
2261 * returns NULL if packet was consumed.
2263 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2264 struct sk_buff *skb) __read_mostly;
2265 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2267 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2268 struct packet_type **pt_prev, int *ret,
2269 struct net_device *orig_dev)
2271 struct net_bridge_port *port;
2273 if (skb->pkt_type == PACKET_LOOPBACK ||
2274 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2275 return skb;
2277 if (*pt_prev) {
2278 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2279 *pt_prev = NULL;
2282 return br_handle_frame_hook(port, skb);
2284 #else
2285 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2286 #endif
2288 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2289 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2290 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2292 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2293 struct packet_type **pt_prev,
2294 int *ret,
2295 struct net_device *orig_dev)
2297 if (skb->dev->macvlan_port == NULL)
2298 return skb;
2300 if (*pt_prev) {
2301 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2302 *pt_prev = NULL;
2304 return macvlan_handle_frame_hook(skb);
2306 #else
2307 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2308 #endif
2310 #ifdef CONFIG_NET_CLS_ACT
2311 /* TODO: Maybe we should just force sch_ingress to be compiled in
2312 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2313 * a compare and 2 stores extra right now if we dont have it on
2314 * but have CONFIG_NET_CLS_ACT
2315 * NOTE: This doesnt stop any functionality; if you dont have
2316 * the ingress scheduler, you just cant add policies on ingress.
2319 static int ing_filter(struct sk_buff *skb)
2321 struct net_device *dev = skb->dev;
2322 u32 ttl = G_TC_RTTL(skb->tc_verd);
2323 struct netdev_queue *rxq;
2324 int result = TC_ACT_OK;
2325 struct Qdisc *q;
2327 if (MAX_RED_LOOP < ttl++) {
2328 printk(KERN_WARNING
2329 "Redir loop detected Dropping packet (%d->%d)\n",
2330 skb->skb_iif, dev->ifindex);
2331 return TC_ACT_SHOT;
2334 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2335 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2337 rxq = &dev->rx_queue;
2339 q = rxq->qdisc;
2340 if (q != &noop_qdisc) {
2341 spin_lock(qdisc_lock(q));
2342 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2343 result = qdisc_enqueue_root(skb, q);
2344 spin_unlock(qdisc_lock(q));
2347 return result;
2350 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2351 struct packet_type **pt_prev,
2352 int *ret, struct net_device *orig_dev)
2354 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2355 goto out;
2357 if (*pt_prev) {
2358 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2359 *pt_prev = NULL;
2360 } else {
2361 /* Huh? Why does turning on AF_PACKET affect this? */
2362 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2365 switch (ing_filter(skb)) {
2366 case TC_ACT_SHOT:
2367 case TC_ACT_STOLEN:
2368 kfree_skb(skb);
2369 return NULL;
2372 out:
2373 skb->tc_verd = 0;
2374 return skb;
2376 #endif
2379 * netif_nit_deliver - deliver received packets to network taps
2380 * @skb: buffer
2382 * This function is used to deliver incoming packets to network
2383 * taps. It should be used when the normal netif_receive_skb path
2384 * is bypassed, for example because of VLAN acceleration.
2386 void netif_nit_deliver(struct sk_buff *skb)
2388 struct packet_type *ptype;
2390 if (list_empty(&ptype_all))
2391 return;
2393 skb_reset_network_header(skb);
2394 skb_reset_transport_header(skb);
2395 skb->mac_len = skb->network_header - skb->mac_header;
2397 rcu_read_lock();
2398 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2399 if (!ptype->dev || ptype->dev == skb->dev)
2400 deliver_skb(skb, ptype, skb->dev);
2402 rcu_read_unlock();
2406 * netif_receive_skb - process receive buffer from network
2407 * @skb: buffer to process
2409 * netif_receive_skb() is the main receive data processing function.
2410 * It always succeeds. The buffer may be dropped during processing
2411 * for congestion control or by the protocol layers.
2413 * This function may only be called from softirq context and interrupts
2414 * should be enabled.
2416 * Return values (usually ignored):
2417 * NET_RX_SUCCESS: no congestion
2418 * NET_RX_DROP: packet was dropped
2420 int netif_receive_skb(struct sk_buff *skb)
2422 struct packet_type *ptype, *pt_prev;
2423 struct net_device *orig_dev;
2424 struct net_device *null_or_orig;
2425 int ret = NET_RX_DROP;
2426 __be16 type;
2428 if (!skb->tstamp.tv64)
2429 net_timestamp(skb);
2431 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2432 return NET_RX_SUCCESS;
2434 /* if we've gotten here through NAPI, check netpoll */
2435 if (netpoll_receive_skb(skb))
2436 return NET_RX_DROP;
2438 if (!skb->skb_iif)
2439 skb->skb_iif = skb->dev->ifindex;
2441 null_or_orig = NULL;
2442 orig_dev = skb->dev;
2443 if (orig_dev->master) {
2444 if (skb_bond_should_drop(skb))
2445 null_or_orig = orig_dev; /* deliver only exact match */
2446 else
2447 skb->dev = orig_dev->master;
2450 __get_cpu_var(netdev_rx_stat).total++;
2452 skb_reset_network_header(skb);
2453 skb_reset_transport_header(skb);
2454 skb->mac_len = skb->network_header - skb->mac_header;
2456 pt_prev = NULL;
2458 rcu_read_lock();
2460 #ifdef CONFIG_NET_CLS_ACT
2461 if (skb->tc_verd & TC_NCLS) {
2462 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2463 goto ncls;
2465 #endif
2467 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2468 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2469 ptype->dev == orig_dev) {
2470 if (pt_prev)
2471 ret = deliver_skb(skb, pt_prev, orig_dev);
2472 pt_prev = ptype;
2476 #ifdef CONFIG_NET_CLS_ACT
2477 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2478 if (!skb)
2479 goto out;
2480 ncls:
2481 #endif
2483 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2484 if (!skb)
2485 goto out;
2486 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2487 if (!skb)
2488 goto out;
2490 type = skb->protocol;
2491 list_for_each_entry_rcu(ptype,
2492 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2493 if (ptype->type == type &&
2494 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2495 ptype->dev == orig_dev)) {
2496 if (pt_prev)
2497 ret = deliver_skb(skb, pt_prev, orig_dev);
2498 pt_prev = ptype;
2502 if (pt_prev) {
2503 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2504 } else {
2505 kfree_skb(skb);
2506 /* Jamal, now you will not able to escape explaining
2507 * me how you were going to use this. :-)
2509 ret = NET_RX_DROP;
2512 out:
2513 rcu_read_unlock();
2514 return ret;
2516 EXPORT_SYMBOL(netif_receive_skb);
2518 /* Network device is going away, flush any packets still pending */
2519 static void flush_backlog(void *arg)
2521 struct net_device *dev = arg;
2522 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2523 struct sk_buff *skb, *tmp;
2525 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2526 if (skb->dev == dev) {
2527 __skb_unlink(skb, &queue->input_pkt_queue);
2528 kfree_skb(skb);
2532 static int napi_gro_complete(struct sk_buff *skb)
2534 struct packet_type *ptype;
2535 __be16 type = skb->protocol;
2536 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2537 int err = -ENOENT;
2539 if (NAPI_GRO_CB(skb)->count == 1) {
2540 skb_shinfo(skb)->gso_size = 0;
2541 goto out;
2544 rcu_read_lock();
2545 list_for_each_entry_rcu(ptype, head, list) {
2546 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2547 continue;
2549 err = ptype->gro_complete(skb);
2550 break;
2552 rcu_read_unlock();
2554 if (err) {
2555 WARN_ON(&ptype->list == head);
2556 kfree_skb(skb);
2557 return NET_RX_SUCCESS;
2560 out:
2561 return netif_receive_skb(skb);
2564 void napi_gro_flush(struct napi_struct *napi)
2566 struct sk_buff *skb, *next;
2568 for (skb = napi->gro_list; skb; skb = next) {
2569 next = skb->next;
2570 skb->next = NULL;
2571 napi_gro_complete(skb);
2574 napi->gro_count = 0;
2575 napi->gro_list = NULL;
2577 EXPORT_SYMBOL(napi_gro_flush);
2579 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2581 struct sk_buff **pp = NULL;
2582 struct packet_type *ptype;
2583 __be16 type = skb->protocol;
2584 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2585 int same_flow;
2586 int mac_len;
2587 enum gro_result ret;
2589 if (!(skb->dev->features & NETIF_F_GRO))
2590 goto normal;
2592 if (skb_is_gso(skb) || skb_has_frags(skb))
2593 goto normal;
2595 rcu_read_lock();
2596 list_for_each_entry_rcu(ptype, head, list) {
2597 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2598 continue;
2600 skb_set_network_header(skb, skb_gro_offset(skb));
2601 mac_len = skb->network_header - skb->mac_header;
2602 skb->mac_len = mac_len;
2603 NAPI_GRO_CB(skb)->same_flow = 0;
2604 NAPI_GRO_CB(skb)->flush = 0;
2605 NAPI_GRO_CB(skb)->free = 0;
2607 pp = ptype->gro_receive(&napi->gro_list, skb);
2608 break;
2610 rcu_read_unlock();
2612 if (&ptype->list == head)
2613 goto normal;
2615 same_flow = NAPI_GRO_CB(skb)->same_flow;
2616 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2618 if (pp) {
2619 struct sk_buff *nskb = *pp;
2621 *pp = nskb->next;
2622 nskb->next = NULL;
2623 napi_gro_complete(nskb);
2624 napi->gro_count--;
2627 if (same_flow)
2628 goto ok;
2630 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2631 goto normal;
2633 napi->gro_count++;
2634 NAPI_GRO_CB(skb)->count = 1;
2635 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2636 skb->next = napi->gro_list;
2637 napi->gro_list = skb;
2638 ret = GRO_HELD;
2640 pull:
2641 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2642 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2644 BUG_ON(skb->end - skb->tail < grow);
2646 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2648 skb->tail += grow;
2649 skb->data_len -= grow;
2651 skb_shinfo(skb)->frags[0].page_offset += grow;
2652 skb_shinfo(skb)->frags[0].size -= grow;
2654 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2655 put_page(skb_shinfo(skb)->frags[0].page);
2656 memmove(skb_shinfo(skb)->frags,
2657 skb_shinfo(skb)->frags + 1,
2658 --skb_shinfo(skb)->nr_frags);
2663 return ret;
2665 normal:
2666 ret = GRO_NORMAL;
2667 goto pull;
2669 EXPORT_SYMBOL(dev_gro_receive);
2671 static gro_result_t
2672 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2674 struct sk_buff *p;
2676 if (netpoll_rx_on(skb))
2677 return GRO_NORMAL;
2679 for (p = napi->gro_list; p; p = p->next) {
2680 NAPI_GRO_CB(p)->same_flow =
2681 (p->dev == skb->dev) &&
2682 !compare_ether_header(skb_mac_header(p),
2683 skb_gro_mac_header(skb));
2684 NAPI_GRO_CB(p)->flush = 0;
2687 return dev_gro_receive(napi, skb);
2690 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
2692 switch (ret) {
2693 case GRO_NORMAL:
2694 if (netif_receive_skb(skb))
2695 ret = GRO_DROP;
2696 break;
2698 case GRO_DROP:
2699 case GRO_MERGED_FREE:
2700 kfree_skb(skb);
2701 break;
2703 case GRO_HELD:
2704 case GRO_MERGED:
2705 break;
2708 return ret;
2710 EXPORT_SYMBOL(napi_skb_finish);
2712 void skb_gro_reset_offset(struct sk_buff *skb)
2714 NAPI_GRO_CB(skb)->data_offset = 0;
2715 NAPI_GRO_CB(skb)->frag0 = NULL;
2716 NAPI_GRO_CB(skb)->frag0_len = 0;
2718 if (skb->mac_header == skb->tail &&
2719 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2720 NAPI_GRO_CB(skb)->frag0 =
2721 page_address(skb_shinfo(skb)->frags[0].page) +
2722 skb_shinfo(skb)->frags[0].page_offset;
2723 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2726 EXPORT_SYMBOL(skb_gro_reset_offset);
2728 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2730 skb_gro_reset_offset(skb);
2732 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2734 EXPORT_SYMBOL(napi_gro_receive);
2736 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2738 __skb_pull(skb, skb_headlen(skb));
2739 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2741 napi->skb = skb;
2743 EXPORT_SYMBOL(napi_reuse_skb);
2745 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2747 struct sk_buff *skb = napi->skb;
2749 if (!skb) {
2750 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
2751 if (skb)
2752 napi->skb = skb;
2754 return skb;
2756 EXPORT_SYMBOL(napi_get_frags);
2758 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
2759 gro_result_t ret)
2761 switch (ret) {
2762 case GRO_NORMAL:
2763 case GRO_HELD:
2764 skb->protocol = eth_type_trans(skb, skb->dev);
2766 if (ret == GRO_HELD)
2767 skb_gro_pull(skb, -ETH_HLEN);
2768 else if (netif_receive_skb(skb))
2769 ret = GRO_DROP;
2770 break;
2772 case GRO_DROP:
2773 case GRO_MERGED_FREE:
2774 napi_reuse_skb(napi, skb);
2775 break;
2777 case GRO_MERGED:
2778 break;
2781 return ret;
2783 EXPORT_SYMBOL(napi_frags_finish);
2785 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2787 struct sk_buff *skb = napi->skb;
2788 struct ethhdr *eth;
2789 unsigned int hlen;
2790 unsigned int off;
2792 napi->skb = NULL;
2794 skb_reset_mac_header(skb);
2795 skb_gro_reset_offset(skb);
2797 off = skb_gro_offset(skb);
2798 hlen = off + sizeof(*eth);
2799 eth = skb_gro_header_fast(skb, off);
2800 if (skb_gro_header_hard(skb, hlen)) {
2801 eth = skb_gro_header_slow(skb, hlen, off);
2802 if (unlikely(!eth)) {
2803 napi_reuse_skb(napi, skb);
2804 skb = NULL;
2805 goto out;
2809 skb_gro_pull(skb, sizeof(*eth));
2812 * This works because the only protocols we care about don't require
2813 * special handling. We'll fix it up properly at the end.
2815 skb->protocol = eth->h_proto;
2817 out:
2818 return skb;
2820 EXPORT_SYMBOL(napi_frags_skb);
2822 gro_result_t napi_gro_frags(struct napi_struct *napi)
2824 struct sk_buff *skb = napi_frags_skb(napi);
2826 if (!skb)
2827 return GRO_DROP;
2829 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2831 EXPORT_SYMBOL(napi_gro_frags);
2833 static int process_backlog(struct napi_struct *napi, int quota)
2835 int work = 0;
2836 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2837 unsigned long start_time = jiffies;
2839 napi->weight = weight_p;
2840 do {
2841 struct sk_buff *skb;
2843 local_irq_disable();
2844 skb = __skb_dequeue(&queue->input_pkt_queue);
2845 if (!skb) {
2846 __napi_complete(napi);
2847 local_irq_enable();
2848 break;
2850 local_irq_enable();
2852 netif_receive_skb(skb);
2853 } while (++work < quota && jiffies == start_time);
2855 return work;
2859 * __napi_schedule - schedule for receive
2860 * @n: entry to schedule
2862 * The entry's receive function will be scheduled to run
2864 void __napi_schedule(struct napi_struct *n)
2866 unsigned long flags;
2868 local_irq_save(flags);
2869 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2870 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2871 local_irq_restore(flags);
2873 EXPORT_SYMBOL(__napi_schedule);
2875 void __napi_complete(struct napi_struct *n)
2877 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2878 BUG_ON(n->gro_list);
2880 list_del(&n->poll_list);
2881 smp_mb__before_clear_bit();
2882 clear_bit(NAPI_STATE_SCHED, &n->state);
2884 EXPORT_SYMBOL(__napi_complete);
2886 void napi_complete(struct napi_struct *n)
2888 unsigned long flags;
2891 * don't let napi dequeue from the cpu poll list
2892 * just in case its running on a different cpu
2894 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2895 return;
2897 napi_gro_flush(n);
2898 local_irq_save(flags);
2899 __napi_complete(n);
2900 local_irq_restore(flags);
2902 EXPORT_SYMBOL(napi_complete);
2904 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2905 int (*poll)(struct napi_struct *, int), int weight)
2907 INIT_LIST_HEAD(&napi->poll_list);
2908 napi->gro_count = 0;
2909 napi->gro_list = NULL;
2910 napi->skb = NULL;
2911 napi->poll = poll;
2912 napi->weight = weight;
2913 list_add(&napi->dev_list, &dev->napi_list);
2914 napi->dev = dev;
2915 #ifdef CONFIG_NETPOLL
2916 spin_lock_init(&napi->poll_lock);
2917 napi->poll_owner = -1;
2918 #endif
2919 set_bit(NAPI_STATE_SCHED, &napi->state);
2921 EXPORT_SYMBOL(netif_napi_add);
2923 void netif_napi_del(struct napi_struct *napi)
2925 struct sk_buff *skb, *next;
2927 list_del_init(&napi->dev_list);
2928 napi_free_frags(napi);
2930 for (skb = napi->gro_list; skb; skb = next) {
2931 next = skb->next;
2932 skb->next = NULL;
2933 kfree_skb(skb);
2936 napi->gro_list = NULL;
2937 napi->gro_count = 0;
2939 EXPORT_SYMBOL(netif_napi_del);
2942 static void net_rx_action(struct softirq_action *h)
2944 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2945 unsigned long time_limit = jiffies + 2;
2946 int budget = netdev_budget;
2947 void *have;
2949 local_irq_disable();
2951 while (!list_empty(list)) {
2952 struct napi_struct *n;
2953 int work, weight;
2955 /* If softirq window is exhuasted then punt.
2956 * Allow this to run for 2 jiffies since which will allow
2957 * an average latency of 1.5/HZ.
2959 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2960 goto softnet_break;
2962 local_irq_enable();
2964 /* Even though interrupts have been re-enabled, this
2965 * access is safe because interrupts can only add new
2966 * entries to the tail of this list, and only ->poll()
2967 * calls can remove this head entry from the list.
2969 n = list_entry(list->next, struct napi_struct, poll_list);
2971 have = netpoll_poll_lock(n);
2973 weight = n->weight;
2975 /* This NAPI_STATE_SCHED test is for avoiding a race
2976 * with netpoll's poll_napi(). Only the entity which
2977 * obtains the lock and sees NAPI_STATE_SCHED set will
2978 * actually make the ->poll() call. Therefore we avoid
2979 * accidently calling ->poll() when NAPI is not scheduled.
2981 work = 0;
2982 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
2983 work = n->poll(n, weight);
2984 trace_napi_poll(n);
2987 WARN_ON_ONCE(work > weight);
2989 budget -= work;
2991 local_irq_disable();
2993 /* Drivers must not modify the NAPI state if they
2994 * consume the entire weight. In such cases this code
2995 * still "owns" the NAPI instance and therefore can
2996 * move the instance around on the list at-will.
2998 if (unlikely(work == weight)) {
2999 if (unlikely(napi_disable_pending(n))) {
3000 local_irq_enable();
3001 napi_complete(n);
3002 local_irq_disable();
3003 } else
3004 list_move_tail(&n->poll_list, list);
3007 netpoll_poll_unlock(have);
3009 out:
3010 local_irq_enable();
3012 #ifdef CONFIG_NET_DMA
3014 * There may not be any more sk_buffs coming right now, so push
3015 * any pending DMA copies to hardware
3017 dma_issue_pending_all();
3018 #endif
3020 return;
3022 softnet_break:
3023 __get_cpu_var(netdev_rx_stat).time_squeeze++;
3024 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3025 goto out;
3028 static gifconf_func_t *gifconf_list[NPROTO];
3031 * register_gifconf - register a SIOCGIF handler
3032 * @family: Address family
3033 * @gifconf: Function handler
3035 * Register protocol dependent address dumping routines. The handler
3036 * that is passed must not be freed or reused until it has been replaced
3037 * by another handler.
3039 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3041 if (family >= NPROTO)
3042 return -EINVAL;
3043 gifconf_list[family] = gifconf;
3044 return 0;
3046 EXPORT_SYMBOL(register_gifconf);
3050 * Map an interface index to its name (SIOCGIFNAME)
3054 * We need this ioctl for efficient implementation of the
3055 * if_indextoname() function required by the IPv6 API. Without
3056 * it, we would have to search all the interfaces to find a
3057 * match. --pb
3060 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3062 struct net_device *dev;
3063 struct ifreq ifr;
3066 * Fetch the caller's info block.
3069 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3070 return -EFAULT;
3072 rcu_read_lock();
3073 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3074 if (!dev) {
3075 rcu_read_unlock();
3076 return -ENODEV;
3079 strcpy(ifr.ifr_name, dev->name);
3080 rcu_read_unlock();
3082 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3083 return -EFAULT;
3084 return 0;
3088 * Perform a SIOCGIFCONF call. This structure will change
3089 * size eventually, and there is nothing I can do about it.
3090 * Thus we will need a 'compatibility mode'.
3093 static int dev_ifconf(struct net *net, char __user *arg)
3095 struct ifconf ifc;
3096 struct net_device *dev;
3097 char __user *pos;
3098 int len;
3099 int total;
3100 int i;
3103 * Fetch the caller's info block.
3106 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3107 return -EFAULT;
3109 pos = ifc.ifc_buf;
3110 len = ifc.ifc_len;
3113 * Loop over the interfaces, and write an info block for each.
3116 total = 0;
3117 for_each_netdev(net, dev) {
3118 for (i = 0; i < NPROTO; i++) {
3119 if (gifconf_list[i]) {
3120 int done;
3121 if (!pos)
3122 done = gifconf_list[i](dev, NULL, 0);
3123 else
3124 done = gifconf_list[i](dev, pos + total,
3125 len - total);
3126 if (done < 0)
3127 return -EFAULT;
3128 total += done;
3134 * All done. Write the updated control block back to the caller.
3136 ifc.ifc_len = total;
3139 * Both BSD and Solaris return 0 here, so we do too.
3141 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3144 #ifdef CONFIG_PROC_FS
3146 * This is invoked by the /proc filesystem handler to display a device
3147 * in detail.
3149 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3150 __acquires(RCU)
3152 struct net *net = seq_file_net(seq);
3153 loff_t off;
3154 struct net_device *dev;
3156 rcu_read_lock();
3157 if (!*pos)
3158 return SEQ_START_TOKEN;
3160 off = 1;
3161 for_each_netdev_rcu(net, dev)
3162 if (off++ == *pos)
3163 return dev;
3165 return NULL;
3168 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3170 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3171 first_net_device(seq_file_net(seq)) :
3172 next_net_device((struct net_device *)v);
3174 ++*pos;
3175 return rcu_dereference(dev);
3178 void dev_seq_stop(struct seq_file *seq, void *v)
3179 __releases(RCU)
3181 rcu_read_unlock();
3184 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3186 const struct net_device_stats *stats = dev_get_stats(dev);
3188 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3189 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3190 dev->name, stats->rx_bytes, stats->rx_packets,
3191 stats->rx_errors,
3192 stats->rx_dropped + stats->rx_missed_errors,
3193 stats->rx_fifo_errors,
3194 stats->rx_length_errors + stats->rx_over_errors +
3195 stats->rx_crc_errors + stats->rx_frame_errors,
3196 stats->rx_compressed, stats->multicast,
3197 stats->tx_bytes, stats->tx_packets,
3198 stats->tx_errors, stats->tx_dropped,
3199 stats->tx_fifo_errors, stats->collisions,
3200 stats->tx_carrier_errors +
3201 stats->tx_aborted_errors +
3202 stats->tx_window_errors +
3203 stats->tx_heartbeat_errors,
3204 stats->tx_compressed);
3208 * Called from the PROCfs module. This now uses the new arbitrary sized
3209 * /proc/net interface to create /proc/net/dev
3211 static int dev_seq_show(struct seq_file *seq, void *v)
3213 if (v == SEQ_START_TOKEN)
3214 seq_puts(seq, "Inter-| Receive "
3215 " | Transmit\n"
3216 " face |bytes packets errs drop fifo frame "
3217 "compressed multicast|bytes packets errs "
3218 "drop fifo colls carrier compressed\n");
3219 else
3220 dev_seq_printf_stats(seq, v);
3221 return 0;
3224 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3226 struct netif_rx_stats *rc = NULL;
3228 while (*pos < nr_cpu_ids)
3229 if (cpu_online(*pos)) {
3230 rc = &per_cpu(netdev_rx_stat, *pos);
3231 break;
3232 } else
3233 ++*pos;
3234 return rc;
3237 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3239 return softnet_get_online(pos);
3242 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3244 ++*pos;
3245 return softnet_get_online(pos);
3248 static void softnet_seq_stop(struct seq_file *seq, void *v)
3252 static int softnet_seq_show(struct seq_file *seq, void *v)
3254 struct netif_rx_stats *s = v;
3256 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3257 s->total, s->dropped, s->time_squeeze, 0,
3258 0, 0, 0, 0, /* was fastroute */
3259 s->cpu_collision);
3260 return 0;
3263 static const struct seq_operations dev_seq_ops = {
3264 .start = dev_seq_start,
3265 .next = dev_seq_next,
3266 .stop = dev_seq_stop,
3267 .show = dev_seq_show,
3270 static int dev_seq_open(struct inode *inode, struct file *file)
3272 return seq_open_net(inode, file, &dev_seq_ops,
3273 sizeof(struct seq_net_private));
3276 static const struct file_operations dev_seq_fops = {
3277 .owner = THIS_MODULE,
3278 .open = dev_seq_open,
3279 .read = seq_read,
3280 .llseek = seq_lseek,
3281 .release = seq_release_net,
3284 static const struct seq_operations softnet_seq_ops = {
3285 .start = softnet_seq_start,
3286 .next = softnet_seq_next,
3287 .stop = softnet_seq_stop,
3288 .show = softnet_seq_show,
3291 static int softnet_seq_open(struct inode *inode, struct file *file)
3293 return seq_open(file, &softnet_seq_ops);
3296 static const struct file_operations softnet_seq_fops = {
3297 .owner = THIS_MODULE,
3298 .open = softnet_seq_open,
3299 .read = seq_read,
3300 .llseek = seq_lseek,
3301 .release = seq_release,
3304 static void *ptype_get_idx(loff_t pos)
3306 struct packet_type *pt = NULL;
3307 loff_t i = 0;
3308 int t;
3310 list_for_each_entry_rcu(pt, &ptype_all, list) {
3311 if (i == pos)
3312 return pt;
3313 ++i;
3316 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3317 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3318 if (i == pos)
3319 return pt;
3320 ++i;
3323 return NULL;
3326 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3327 __acquires(RCU)
3329 rcu_read_lock();
3330 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3333 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3335 struct packet_type *pt;
3336 struct list_head *nxt;
3337 int hash;
3339 ++*pos;
3340 if (v == SEQ_START_TOKEN)
3341 return ptype_get_idx(0);
3343 pt = v;
3344 nxt = pt->list.next;
3345 if (pt->type == htons(ETH_P_ALL)) {
3346 if (nxt != &ptype_all)
3347 goto found;
3348 hash = 0;
3349 nxt = ptype_base[0].next;
3350 } else
3351 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3353 while (nxt == &ptype_base[hash]) {
3354 if (++hash >= PTYPE_HASH_SIZE)
3355 return NULL;
3356 nxt = ptype_base[hash].next;
3358 found:
3359 return list_entry(nxt, struct packet_type, list);
3362 static void ptype_seq_stop(struct seq_file *seq, void *v)
3363 __releases(RCU)
3365 rcu_read_unlock();
3368 static int ptype_seq_show(struct seq_file *seq, void *v)
3370 struct packet_type *pt = v;
3372 if (v == SEQ_START_TOKEN)
3373 seq_puts(seq, "Type Device Function\n");
3374 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3375 if (pt->type == htons(ETH_P_ALL))
3376 seq_puts(seq, "ALL ");
3377 else
3378 seq_printf(seq, "%04x", ntohs(pt->type));
3380 seq_printf(seq, " %-8s %pF\n",
3381 pt->dev ? pt->dev->name : "", pt->func);
3384 return 0;
3387 static const struct seq_operations ptype_seq_ops = {
3388 .start = ptype_seq_start,
3389 .next = ptype_seq_next,
3390 .stop = ptype_seq_stop,
3391 .show = ptype_seq_show,
3394 static int ptype_seq_open(struct inode *inode, struct file *file)
3396 return seq_open_net(inode, file, &ptype_seq_ops,
3397 sizeof(struct seq_net_private));
3400 static const struct file_operations ptype_seq_fops = {
3401 .owner = THIS_MODULE,
3402 .open = ptype_seq_open,
3403 .read = seq_read,
3404 .llseek = seq_lseek,
3405 .release = seq_release_net,
3409 static int __net_init dev_proc_net_init(struct net *net)
3411 int rc = -ENOMEM;
3413 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3414 goto out;
3415 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3416 goto out_dev;
3417 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3418 goto out_softnet;
3420 if (wext_proc_init(net))
3421 goto out_ptype;
3422 rc = 0;
3423 out:
3424 return rc;
3425 out_ptype:
3426 proc_net_remove(net, "ptype");
3427 out_softnet:
3428 proc_net_remove(net, "softnet_stat");
3429 out_dev:
3430 proc_net_remove(net, "dev");
3431 goto out;
3434 static void __net_exit dev_proc_net_exit(struct net *net)
3436 wext_proc_exit(net);
3438 proc_net_remove(net, "ptype");
3439 proc_net_remove(net, "softnet_stat");
3440 proc_net_remove(net, "dev");
3443 static struct pernet_operations __net_initdata dev_proc_ops = {
3444 .init = dev_proc_net_init,
3445 .exit = dev_proc_net_exit,
3448 static int __init dev_proc_init(void)
3450 return register_pernet_subsys(&dev_proc_ops);
3452 #else
3453 #define dev_proc_init() 0
3454 #endif /* CONFIG_PROC_FS */
3458 * netdev_set_master - set up master/slave pair
3459 * @slave: slave device
3460 * @master: new master device
3462 * Changes the master device of the slave. Pass %NULL to break the
3463 * bonding. The caller must hold the RTNL semaphore. On a failure
3464 * a negative errno code is returned. On success the reference counts
3465 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3466 * function returns zero.
3468 int netdev_set_master(struct net_device *slave, struct net_device *master)
3470 struct net_device *old = slave->master;
3472 ASSERT_RTNL();
3474 if (master) {
3475 if (old)
3476 return -EBUSY;
3477 dev_hold(master);
3480 slave->master = master;
3482 synchronize_net();
3484 if (old)
3485 dev_put(old);
3487 if (master)
3488 slave->flags |= IFF_SLAVE;
3489 else
3490 slave->flags &= ~IFF_SLAVE;
3492 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3493 return 0;
3495 EXPORT_SYMBOL(netdev_set_master);
3497 static void dev_change_rx_flags(struct net_device *dev, int flags)
3499 const struct net_device_ops *ops = dev->netdev_ops;
3501 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3502 ops->ndo_change_rx_flags(dev, flags);
3505 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3507 unsigned short old_flags = dev->flags;
3508 uid_t uid;
3509 gid_t gid;
3511 ASSERT_RTNL();
3513 dev->flags |= IFF_PROMISC;
3514 dev->promiscuity += inc;
3515 if (dev->promiscuity == 0) {
3517 * Avoid overflow.
3518 * If inc causes overflow, untouch promisc and return error.
3520 if (inc < 0)
3521 dev->flags &= ~IFF_PROMISC;
3522 else {
3523 dev->promiscuity -= inc;
3524 printk(KERN_WARNING "%s: promiscuity touches roof, "
3525 "set promiscuity failed, promiscuity feature "
3526 "of device might be broken.\n", dev->name);
3527 return -EOVERFLOW;
3530 if (dev->flags != old_flags) {
3531 printk(KERN_INFO "device %s %s promiscuous mode\n",
3532 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3533 "left");
3534 if (audit_enabled) {
3535 current_uid_gid(&uid, &gid);
3536 audit_log(current->audit_context, GFP_ATOMIC,
3537 AUDIT_ANOM_PROMISCUOUS,
3538 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3539 dev->name, (dev->flags & IFF_PROMISC),
3540 (old_flags & IFF_PROMISC),
3541 audit_get_loginuid(current),
3542 uid, gid,
3543 audit_get_sessionid(current));
3546 dev_change_rx_flags(dev, IFF_PROMISC);
3548 return 0;
3552 * dev_set_promiscuity - update promiscuity count on a device
3553 * @dev: device
3554 * @inc: modifier
3556 * Add or remove promiscuity from a device. While the count in the device
3557 * remains above zero the interface remains promiscuous. Once it hits zero
3558 * the device reverts back to normal filtering operation. A negative inc
3559 * value is used to drop promiscuity on the device.
3560 * Return 0 if successful or a negative errno code on error.
3562 int dev_set_promiscuity(struct net_device *dev, int inc)
3564 unsigned short old_flags = dev->flags;
3565 int err;
3567 err = __dev_set_promiscuity(dev, inc);
3568 if (err < 0)
3569 return err;
3570 if (dev->flags != old_flags)
3571 dev_set_rx_mode(dev);
3572 return err;
3574 EXPORT_SYMBOL(dev_set_promiscuity);
3577 * dev_set_allmulti - update allmulti count on a device
3578 * @dev: device
3579 * @inc: modifier
3581 * Add or remove reception of all multicast frames to a device. While the
3582 * count in the device remains above zero the interface remains listening
3583 * to all interfaces. Once it hits zero the device reverts back to normal
3584 * filtering operation. A negative @inc value is used to drop the counter
3585 * when releasing a resource needing all multicasts.
3586 * Return 0 if successful or a negative errno code on error.
3589 int dev_set_allmulti(struct net_device *dev, int inc)
3591 unsigned short old_flags = dev->flags;
3593 ASSERT_RTNL();
3595 dev->flags |= IFF_ALLMULTI;
3596 dev->allmulti += inc;
3597 if (dev->allmulti == 0) {
3599 * Avoid overflow.
3600 * If inc causes overflow, untouch allmulti and return error.
3602 if (inc < 0)
3603 dev->flags &= ~IFF_ALLMULTI;
3604 else {
3605 dev->allmulti -= inc;
3606 printk(KERN_WARNING "%s: allmulti touches roof, "
3607 "set allmulti failed, allmulti feature of "
3608 "device might be broken.\n", dev->name);
3609 return -EOVERFLOW;
3612 if (dev->flags ^ old_flags) {
3613 dev_change_rx_flags(dev, IFF_ALLMULTI);
3614 dev_set_rx_mode(dev);
3616 return 0;
3618 EXPORT_SYMBOL(dev_set_allmulti);
3621 * Upload unicast and multicast address lists to device and
3622 * configure RX filtering. When the device doesn't support unicast
3623 * filtering it is put in promiscuous mode while unicast addresses
3624 * are present.
3626 void __dev_set_rx_mode(struct net_device *dev)
3628 const struct net_device_ops *ops = dev->netdev_ops;
3630 /* dev_open will call this function so the list will stay sane. */
3631 if (!(dev->flags&IFF_UP))
3632 return;
3634 if (!netif_device_present(dev))
3635 return;
3637 if (ops->ndo_set_rx_mode)
3638 ops->ndo_set_rx_mode(dev);
3639 else {
3640 /* Unicast addresses changes may only happen under the rtnl,
3641 * therefore calling __dev_set_promiscuity here is safe.
3643 if (dev->uc.count > 0 && !dev->uc_promisc) {
3644 __dev_set_promiscuity(dev, 1);
3645 dev->uc_promisc = 1;
3646 } else if (dev->uc.count == 0 && dev->uc_promisc) {
3647 __dev_set_promiscuity(dev, -1);
3648 dev->uc_promisc = 0;
3651 if (ops->ndo_set_multicast_list)
3652 ops->ndo_set_multicast_list(dev);
3656 void dev_set_rx_mode(struct net_device *dev)
3658 netif_addr_lock_bh(dev);
3659 __dev_set_rx_mode(dev);
3660 netif_addr_unlock_bh(dev);
3663 /* hw addresses list handling functions */
3665 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3666 int addr_len, unsigned char addr_type)
3668 struct netdev_hw_addr *ha;
3669 int alloc_size;
3671 if (addr_len > MAX_ADDR_LEN)
3672 return -EINVAL;
3674 list_for_each_entry(ha, &list->list, list) {
3675 if (!memcmp(ha->addr, addr, addr_len) &&
3676 ha->type == addr_type) {
3677 ha->refcount++;
3678 return 0;
3683 alloc_size = sizeof(*ha);
3684 if (alloc_size < L1_CACHE_BYTES)
3685 alloc_size = L1_CACHE_BYTES;
3686 ha = kmalloc(alloc_size, GFP_ATOMIC);
3687 if (!ha)
3688 return -ENOMEM;
3689 memcpy(ha->addr, addr, addr_len);
3690 ha->type = addr_type;
3691 ha->refcount = 1;
3692 ha->synced = false;
3693 list_add_tail_rcu(&ha->list, &list->list);
3694 list->count++;
3695 return 0;
3698 static void ha_rcu_free(struct rcu_head *head)
3700 struct netdev_hw_addr *ha;
3702 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3703 kfree(ha);
3706 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3707 int addr_len, unsigned char addr_type)
3709 struct netdev_hw_addr *ha;
3711 list_for_each_entry(ha, &list->list, list) {
3712 if (!memcmp(ha->addr, addr, addr_len) &&
3713 (ha->type == addr_type || !addr_type)) {
3714 if (--ha->refcount)
3715 return 0;
3716 list_del_rcu(&ha->list);
3717 call_rcu(&ha->rcu_head, ha_rcu_free);
3718 list->count--;
3719 return 0;
3722 return -ENOENT;
3725 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3726 struct netdev_hw_addr_list *from_list,
3727 int addr_len,
3728 unsigned char addr_type)
3730 int err;
3731 struct netdev_hw_addr *ha, *ha2;
3732 unsigned char type;
3734 list_for_each_entry(ha, &from_list->list, list) {
3735 type = addr_type ? addr_type : ha->type;
3736 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3737 if (err)
3738 goto unroll;
3740 return 0;
3742 unroll:
3743 list_for_each_entry(ha2, &from_list->list, list) {
3744 if (ha2 == ha)
3745 break;
3746 type = addr_type ? addr_type : ha2->type;
3747 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3749 return err;
3752 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3753 struct netdev_hw_addr_list *from_list,
3754 int addr_len,
3755 unsigned char addr_type)
3757 struct netdev_hw_addr *ha;
3758 unsigned char type;
3760 list_for_each_entry(ha, &from_list->list, list) {
3761 type = addr_type ? addr_type : ha->type;
3762 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3766 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3767 struct netdev_hw_addr_list *from_list,
3768 int addr_len)
3770 int err = 0;
3771 struct netdev_hw_addr *ha, *tmp;
3773 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3774 if (!ha->synced) {
3775 err = __hw_addr_add(to_list, ha->addr,
3776 addr_len, ha->type);
3777 if (err)
3778 break;
3779 ha->synced = true;
3780 ha->refcount++;
3781 } else if (ha->refcount == 1) {
3782 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3783 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3786 return err;
3789 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3790 struct netdev_hw_addr_list *from_list,
3791 int addr_len)
3793 struct netdev_hw_addr *ha, *tmp;
3795 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3796 if (ha->synced) {
3797 __hw_addr_del(to_list, ha->addr,
3798 addr_len, ha->type);
3799 ha->synced = false;
3800 __hw_addr_del(from_list, ha->addr,
3801 addr_len, ha->type);
3806 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3808 struct netdev_hw_addr *ha, *tmp;
3810 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3811 list_del_rcu(&ha->list);
3812 call_rcu(&ha->rcu_head, ha_rcu_free);
3814 list->count = 0;
3817 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3819 INIT_LIST_HEAD(&list->list);
3820 list->count = 0;
3823 /* Device addresses handling functions */
3825 static void dev_addr_flush(struct net_device *dev)
3827 /* rtnl_mutex must be held here */
3829 __hw_addr_flush(&dev->dev_addrs);
3830 dev->dev_addr = NULL;
3833 static int dev_addr_init(struct net_device *dev)
3835 unsigned char addr[MAX_ADDR_LEN];
3836 struct netdev_hw_addr *ha;
3837 int err;
3839 /* rtnl_mutex must be held here */
3841 __hw_addr_init(&dev->dev_addrs);
3842 memset(addr, 0, sizeof(addr));
3843 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3844 NETDEV_HW_ADDR_T_LAN);
3845 if (!err) {
3847 * Get the first (previously created) address from the list
3848 * and set dev_addr pointer to this location.
3850 ha = list_first_entry(&dev->dev_addrs.list,
3851 struct netdev_hw_addr, list);
3852 dev->dev_addr = ha->addr;
3854 return err;
3858 * dev_addr_add - Add a device address
3859 * @dev: device
3860 * @addr: address to add
3861 * @addr_type: address type
3863 * Add a device address to the device or increase the reference count if
3864 * it already exists.
3866 * The caller must hold the rtnl_mutex.
3868 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3869 unsigned char addr_type)
3871 int err;
3873 ASSERT_RTNL();
3875 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3876 if (!err)
3877 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3878 return err;
3880 EXPORT_SYMBOL(dev_addr_add);
3883 * dev_addr_del - Release a device address.
3884 * @dev: device
3885 * @addr: address to delete
3886 * @addr_type: address type
3888 * Release reference to a device address and remove it from the device
3889 * if the reference count drops to zero.
3891 * The caller must hold the rtnl_mutex.
3893 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3894 unsigned char addr_type)
3896 int err;
3897 struct netdev_hw_addr *ha;
3899 ASSERT_RTNL();
3902 * We can not remove the first address from the list because
3903 * dev->dev_addr points to that.
3905 ha = list_first_entry(&dev->dev_addrs.list,
3906 struct netdev_hw_addr, list);
3907 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3908 return -ENOENT;
3910 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3911 addr_type);
3912 if (!err)
3913 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3914 return err;
3916 EXPORT_SYMBOL(dev_addr_del);
3919 * dev_addr_add_multiple - Add device addresses from another device
3920 * @to_dev: device to which addresses will be added
3921 * @from_dev: device from which addresses will be added
3922 * @addr_type: address type - 0 means type will be used from from_dev
3924 * Add device addresses of the one device to another.
3926 * The caller must hold the rtnl_mutex.
3928 int dev_addr_add_multiple(struct net_device *to_dev,
3929 struct net_device *from_dev,
3930 unsigned char addr_type)
3932 int err;
3934 ASSERT_RTNL();
3936 if (from_dev->addr_len != to_dev->addr_len)
3937 return -EINVAL;
3938 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3939 to_dev->addr_len, addr_type);
3940 if (!err)
3941 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3942 return err;
3944 EXPORT_SYMBOL(dev_addr_add_multiple);
3947 * dev_addr_del_multiple - Delete device addresses by another device
3948 * @to_dev: device where the addresses will be deleted
3949 * @from_dev: device by which addresses the addresses will be deleted
3950 * @addr_type: address type - 0 means type will used from from_dev
3952 * Deletes addresses in to device by the list of addresses in from device.
3954 * The caller must hold the rtnl_mutex.
3956 int dev_addr_del_multiple(struct net_device *to_dev,
3957 struct net_device *from_dev,
3958 unsigned char addr_type)
3960 ASSERT_RTNL();
3962 if (from_dev->addr_len != to_dev->addr_len)
3963 return -EINVAL;
3964 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3965 to_dev->addr_len, addr_type);
3966 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3967 return 0;
3969 EXPORT_SYMBOL(dev_addr_del_multiple);
3971 /* multicast addresses handling functions */
3973 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3974 void *addr, int alen, int glbl)
3976 struct dev_addr_list *da;
3978 for (; (da = *list) != NULL; list = &da->next) {
3979 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3980 alen == da->da_addrlen) {
3981 if (glbl) {
3982 int old_glbl = da->da_gusers;
3983 da->da_gusers = 0;
3984 if (old_glbl == 0)
3985 break;
3987 if (--da->da_users)
3988 return 0;
3990 *list = da->next;
3991 kfree(da);
3992 (*count)--;
3993 return 0;
3996 return -ENOENT;
3999 int __dev_addr_add(struct dev_addr_list **list, int *count,
4000 void *addr, int alen, int glbl)
4002 struct dev_addr_list *da;
4004 for (da = *list; da != NULL; da = da->next) {
4005 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
4006 da->da_addrlen == alen) {
4007 if (glbl) {
4008 int old_glbl = da->da_gusers;
4009 da->da_gusers = 1;
4010 if (old_glbl)
4011 return 0;
4013 da->da_users++;
4014 return 0;
4018 da = kzalloc(sizeof(*da), GFP_ATOMIC);
4019 if (da == NULL)
4020 return -ENOMEM;
4021 memcpy(da->da_addr, addr, alen);
4022 da->da_addrlen = alen;
4023 da->da_users = 1;
4024 da->da_gusers = glbl ? 1 : 0;
4025 da->next = *list;
4026 *list = da;
4027 (*count)++;
4028 return 0;
4032 * dev_unicast_delete - Release secondary unicast address.
4033 * @dev: device
4034 * @addr: address to delete
4036 * Release reference to a secondary unicast address and remove it
4037 * from the device if the reference count drops to zero.
4039 * The caller must hold the rtnl_mutex.
4041 int dev_unicast_delete(struct net_device *dev, void *addr)
4043 int err;
4045 ASSERT_RTNL();
4047 netif_addr_lock_bh(dev);
4048 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
4049 NETDEV_HW_ADDR_T_UNICAST);
4050 if (!err)
4051 __dev_set_rx_mode(dev);
4052 netif_addr_unlock_bh(dev);
4053 return err;
4055 EXPORT_SYMBOL(dev_unicast_delete);
4058 * dev_unicast_add - add a secondary unicast address
4059 * @dev: device
4060 * @addr: address to add
4062 * Add a secondary unicast address to the device or increase
4063 * the reference count if it already exists.
4065 * The caller must hold the rtnl_mutex.
4067 int dev_unicast_add(struct net_device *dev, void *addr)
4069 int err;
4071 ASSERT_RTNL();
4073 netif_addr_lock_bh(dev);
4074 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
4075 NETDEV_HW_ADDR_T_UNICAST);
4076 if (!err)
4077 __dev_set_rx_mode(dev);
4078 netif_addr_unlock_bh(dev);
4079 return err;
4081 EXPORT_SYMBOL(dev_unicast_add);
4083 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
4084 struct dev_addr_list **from, int *from_count)
4086 struct dev_addr_list *da, *next;
4087 int err = 0;
4089 da = *from;
4090 while (da != NULL) {
4091 next = da->next;
4092 if (!da->da_synced) {
4093 err = __dev_addr_add(to, to_count,
4094 da->da_addr, da->da_addrlen, 0);
4095 if (err < 0)
4096 break;
4097 da->da_synced = 1;
4098 da->da_users++;
4099 } else if (da->da_users == 1) {
4100 __dev_addr_delete(to, to_count,
4101 da->da_addr, da->da_addrlen, 0);
4102 __dev_addr_delete(from, from_count,
4103 da->da_addr, da->da_addrlen, 0);
4105 da = next;
4107 return err;
4109 EXPORT_SYMBOL_GPL(__dev_addr_sync);
4111 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4112 struct dev_addr_list **from, int *from_count)
4114 struct dev_addr_list *da, *next;
4116 da = *from;
4117 while (da != NULL) {
4118 next = da->next;
4119 if (da->da_synced) {
4120 __dev_addr_delete(to, to_count,
4121 da->da_addr, da->da_addrlen, 0);
4122 da->da_synced = 0;
4123 __dev_addr_delete(from, from_count,
4124 da->da_addr, da->da_addrlen, 0);
4126 da = next;
4129 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4132 * dev_unicast_sync - Synchronize device's unicast list to another device
4133 * @to: destination device
4134 * @from: source device
4136 * Add newly added addresses to the destination device and release
4137 * addresses that have no users left. The source device must be
4138 * locked by netif_tx_lock_bh.
4140 * This function is intended to be called from the dev->set_rx_mode
4141 * function of layered software devices.
4143 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4145 int err = 0;
4147 if (to->addr_len != from->addr_len)
4148 return -EINVAL;
4150 netif_addr_lock_bh(to);
4151 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4152 if (!err)
4153 __dev_set_rx_mode(to);
4154 netif_addr_unlock_bh(to);
4155 return err;
4157 EXPORT_SYMBOL(dev_unicast_sync);
4160 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4161 * @to: destination device
4162 * @from: source device
4164 * Remove all addresses that were added to the destination device by
4165 * dev_unicast_sync(). This function is intended to be called from the
4166 * dev->stop function of layered software devices.
4168 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4170 if (to->addr_len != from->addr_len)
4171 return;
4173 netif_addr_lock_bh(from);
4174 netif_addr_lock(to);
4175 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4176 __dev_set_rx_mode(to);
4177 netif_addr_unlock(to);
4178 netif_addr_unlock_bh(from);
4180 EXPORT_SYMBOL(dev_unicast_unsync);
4182 static void dev_unicast_flush(struct net_device *dev)
4184 netif_addr_lock_bh(dev);
4185 __hw_addr_flush(&dev->uc);
4186 netif_addr_unlock_bh(dev);
4189 static void dev_unicast_init(struct net_device *dev)
4191 __hw_addr_init(&dev->uc);
4195 static void __dev_addr_discard(struct dev_addr_list **list)
4197 struct dev_addr_list *tmp;
4199 while (*list != NULL) {
4200 tmp = *list;
4201 *list = tmp->next;
4202 if (tmp->da_users > tmp->da_gusers)
4203 printk("__dev_addr_discard: address leakage! "
4204 "da_users=%d\n", tmp->da_users);
4205 kfree(tmp);
4209 static void dev_addr_discard(struct net_device *dev)
4211 netif_addr_lock_bh(dev);
4213 __dev_addr_discard(&dev->mc_list);
4214 dev->mc_count = 0;
4216 netif_addr_unlock_bh(dev);
4220 * dev_get_flags - get flags reported to userspace
4221 * @dev: device
4223 * Get the combination of flag bits exported through APIs to userspace.
4225 unsigned dev_get_flags(const struct net_device *dev)
4227 unsigned flags;
4229 flags = (dev->flags & ~(IFF_PROMISC |
4230 IFF_ALLMULTI |
4231 IFF_RUNNING |
4232 IFF_LOWER_UP |
4233 IFF_DORMANT)) |
4234 (dev->gflags & (IFF_PROMISC |
4235 IFF_ALLMULTI));
4237 if (netif_running(dev)) {
4238 if (netif_oper_up(dev))
4239 flags |= IFF_RUNNING;
4240 if (netif_carrier_ok(dev))
4241 flags |= IFF_LOWER_UP;
4242 if (netif_dormant(dev))
4243 flags |= IFF_DORMANT;
4246 return flags;
4248 EXPORT_SYMBOL(dev_get_flags);
4251 * dev_change_flags - change device settings
4252 * @dev: device
4253 * @flags: device state flags
4255 * Change settings on device based state flags. The flags are
4256 * in the userspace exported format.
4258 int dev_change_flags(struct net_device *dev, unsigned flags)
4260 int ret, changes;
4261 int old_flags = dev->flags;
4263 ASSERT_RTNL();
4266 * Set the flags on our device.
4269 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4270 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4271 IFF_AUTOMEDIA)) |
4272 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4273 IFF_ALLMULTI));
4276 * Load in the correct multicast list now the flags have changed.
4279 if ((old_flags ^ flags) & IFF_MULTICAST)
4280 dev_change_rx_flags(dev, IFF_MULTICAST);
4282 dev_set_rx_mode(dev);
4285 * Have we downed the interface. We handle IFF_UP ourselves
4286 * according to user attempts to set it, rather than blindly
4287 * setting it.
4290 ret = 0;
4291 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4292 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4294 if (!ret)
4295 dev_set_rx_mode(dev);
4298 if (dev->flags & IFF_UP &&
4299 ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
4300 IFF_VOLATILE)))
4301 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4303 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4304 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4306 dev->gflags ^= IFF_PROMISC;
4307 dev_set_promiscuity(dev, inc);
4310 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4311 is important. Some (broken) drivers set IFF_PROMISC, when
4312 IFF_ALLMULTI is requested not asking us and not reporting.
4314 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4315 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4317 dev->gflags ^= IFF_ALLMULTI;
4318 dev_set_allmulti(dev, inc);
4321 /* Exclude state transition flags, already notified */
4322 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4323 if (changes)
4324 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4326 return ret;
4328 EXPORT_SYMBOL(dev_change_flags);
4331 * dev_set_mtu - Change maximum transfer unit
4332 * @dev: device
4333 * @new_mtu: new transfer unit
4335 * Change the maximum transfer size of the network device.
4337 int dev_set_mtu(struct net_device *dev, int new_mtu)
4339 const struct net_device_ops *ops = dev->netdev_ops;
4340 int err;
4342 if (new_mtu == dev->mtu)
4343 return 0;
4345 /* MTU must be positive. */
4346 if (new_mtu < 0)
4347 return -EINVAL;
4349 if (!netif_device_present(dev))
4350 return -ENODEV;
4352 err = 0;
4353 if (ops->ndo_change_mtu)
4354 err = ops->ndo_change_mtu(dev, new_mtu);
4355 else
4356 dev->mtu = new_mtu;
4358 if (!err && dev->flags & IFF_UP)
4359 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4360 return err;
4362 EXPORT_SYMBOL(dev_set_mtu);
4365 * dev_set_mac_address - Change Media Access Control Address
4366 * @dev: device
4367 * @sa: new address
4369 * Change the hardware (MAC) address of the device
4371 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4373 const struct net_device_ops *ops = dev->netdev_ops;
4374 int err;
4376 if (!ops->ndo_set_mac_address)
4377 return -EOPNOTSUPP;
4378 if (sa->sa_family != dev->type)
4379 return -EINVAL;
4380 if (!netif_device_present(dev))
4381 return -ENODEV;
4382 err = ops->ndo_set_mac_address(dev, sa);
4383 if (!err)
4384 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4385 return err;
4387 EXPORT_SYMBOL(dev_set_mac_address);
4390 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4392 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4394 int err;
4395 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4397 if (!dev)
4398 return -ENODEV;
4400 switch (cmd) {
4401 case SIOCGIFFLAGS: /* Get interface flags */
4402 ifr->ifr_flags = (short) dev_get_flags(dev);
4403 return 0;
4405 case SIOCGIFMETRIC: /* Get the metric on the interface
4406 (currently unused) */
4407 ifr->ifr_metric = 0;
4408 return 0;
4410 case SIOCGIFMTU: /* Get the MTU of a device */
4411 ifr->ifr_mtu = dev->mtu;
4412 return 0;
4414 case SIOCGIFHWADDR:
4415 if (!dev->addr_len)
4416 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4417 else
4418 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4419 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4420 ifr->ifr_hwaddr.sa_family = dev->type;
4421 return 0;
4423 case SIOCGIFSLAVE:
4424 err = -EINVAL;
4425 break;
4427 case SIOCGIFMAP:
4428 ifr->ifr_map.mem_start = dev->mem_start;
4429 ifr->ifr_map.mem_end = dev->mem_end;
4430 ifr->ifr_map.base_addr = dev->base_addr;
4431 ifr->ifr_map.irq = dev->irq;
4432 ifr->ifr_map.dma = dev->dma;
4433 ifr->ifr_map.port = dev->if_port;
4434 return 0;
4436 case SIOCGIFINDEX:
4437 ifr->ifr_ifindex = dev->ifindex;
4438 return 0;
4440 case SIOCGIFTXQLEN:
4441 ifr->ifr_qlen = dev->tx_queue_len;
4442 return 0;
4444 default:
4445 /* dev_ioctl() should ensure this case
4446 * is never reached
4448 WARN_ON(1);
4449 err = -EINVAL;
4450 break;
4453 return err;
4457 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4459 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4461 int err;
4462 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4463 const struct net_device_ops *ops;
4465 if (!dev)
4466 return -ENODEV;
4468 ops = dev->netdev_ops;
4470 switch (cmd) {
4471 case SIOCSIFFLAGS: /* Set interface flags */
4472 return dev_change_flags(dev, ifr->ifr_flags);
4474 case SIOCSIFMETRIC: /* Set the metric on the interface
4475 (currently unused) */
4476 return -EOPNOTSUPP;
4478 case SIOCSIFMTU: /* Set the MTU of a device */
4479 return dev_set_mtu(dev, ifr->ifr_mtu);
4481 case SIOCSIFHWADDR:
4482 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4484 case SIOCSIFHWBROADCAST:
4485 if (ifr->ifr_hwaddr.sa_family != dev->type)
4486 return -EINVAL;
4487 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4488 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4489 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4490 return 0;
4492 case SIOCSIFMAP:
4493 if (ops->ndo_set_config) {
4494 if (!netif_device_present(dev))
4495 return -ENODEV;
4496 return ops->ndo_set_config(dev, &ifr->ifr_map);
4498 return -EOPNOTSUPP;
4500 case SIOCADDMULTI:
4501 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4502 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4503 return -EINVAL;
4504 if (!netif_device_present(dev))
4505 return -ENODEV;
4506 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4507 dev->addr_len, 1);
4509 case SIOCDELMULTI:
4510 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4511 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4512 return -EINVAL;
4513 if (!netif_device_present(dev))
4514 return -ENODEV;
4515 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4516 dev->addr_len, 1);
4518 case SIOCSIFTXQLEN:
4519 if (ifr->ifr_qlen < 0)
4520 return -EINVAL;
4521 dev->tx_queue_len = ifr->ifr_qlen;
4522 return 0;
4524 case SIOCSIFNAME:
4525 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4526 return dev_change_name(dev, ifr->ifr_newname);
4529 * Unknown or private ioctl
4531 default:
4532 if ((cmd >= SIOCDEVPRIVATE &&
4533 cmd <= SIOCDEVPRIVATE + 15) ||
4534 cmd == SIOCBONDENSLAVE ||
4535 cmd == SIOCBONDRELEASE ||
4536 cmd == SIOCBONDSETHWADDR ||
4537 cmd == SIOCBONDSLAVEINFOQUERY ||
4538 cmd == SIOCBONDINFOQUERY ||
4539 cmd == SIOCBONDCHANGEACTIVE ||
4540 cmd == SIOCGMIIPHY ||
4541 cmd == SIOCGMIIREG ||
4542 cmd == SIOCSMIIREG ||
4543 cmd == SIOCBRADDIF ||
4544 cmd == SIOCBRDELIF ||
4545 cmd == SIOCSHWTSTAMP ||
4546 cmd == SIOCWANDEV) {
4547 err = -EOPNOTSUPP;
4548 if (ops->ndo_do_ioctl) {
4549 if (netif_device_present(dev))
4550 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4551 else
4552 err = -ENODEV;
4554 } else
4555 err = -EINVAL;
4558 return err;
4562 * This function handles all "interface"-type I/O control requests. The actual
4563 * 'doing' part of this is dev_ifsioc above.
4567 * dev_ioctl - network device ioctl
4568 * @net: the applicable net namespace
4569 * @cmd: command to issue
4570 * @arg: pointer to a struct ifreq in user space
4572 * Issue ioctl functions to devices. This is normally called by the
4573 * user space syscall interfaces but can sometimes be useful for
4574 * other purposes. The return value is the return from the syscall if
4575 * positive or a negative errno code on error.
4578 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4580 struct ifreq ifr;
4581 int ret;
4582 char *colon;
4584 /* One special case: SIOCGIFCONF takes ifconf argument
4585 and requires shared lock, because it sleeps writing
4586 to user space.
4589 if (cmd == SIOCGIFCONF) {
4590 rtnl_lock();
4591 ret = dev_ifconf(net, (char __user *) arg);
4592 rtnl_unlock();
4593 return ret;
4595 if (cmd == SIOCGIFNAME)
4596 return dev_ifname(net, (struct ifreq __user *)arg);
4598 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4599 return -EFAULT;
4601 ifr.ifr_name[IFNAMSIZ-1] = 0;
4603 colon = strchr(ifr.ifr_name, ':');
4604 if (colon)
4605 *colon = 0;
4608 * See which interface the caller is talking about.
4611 switch (cmd) {
4613 * These ioctl calls:
4614 * - can be done by all.
4615 * - atomic and do not require locking.
4616 * - return a value
4618 case SIOCGIFFLAGS:
4619 case SIOCGIFMETRIC:
4620 case SIOCGIFMTU:
4621 case SIOCGIFHWADDR:
4622 case SIOCGIFSLAVE:
4623 case SIOCGIFMAP:
4624 case SIOCGIFINDEX:
4625 case SIOCGIFTXQLEN:
4626 dev_load(net, ifr.ifr_name);
4627 rcu_read_lock();
4628 ret = dev_ifsioc_locked(net, &ifr, cmd);
4629 rcu_read_unlock();
4630 if (!ret) {
4631 if (colon)
4632 *colon = ':';
4633 if (copy_to_user(arg, &ifr,
4634 sizeof(struct ifreq)))
4635 ret = -EFAULT;
4637 return ret;
4639 case SIOCETHTOOL:
4640 dev_load(net, ifr.ifr_name);
4641 rtnl_lock();
4642 ret = dev_ethtool(net, &ifr);
4643 rtnl_unlock();
4644 if (!ret) {
4645 if (colon)
4646 *colon = ':';
4647 if (copy_to_user(arg, &ifr,
4648 sizeof(struct ifreq)))
4649 ret = -EFAULT;
4651 return ret;
4654 * These ioctl calls:
4655 * - require superuser power.
4656 * - require strict serialization.
4657 * - return a value
4659 case SIOCGMIIPHY:
4660 case SIOCGMIIREG:
4661 case SIOCSIFNAME:
4662 if (!capable(CAP_NET_ADMIN))
4663 return -EPERM;
4664 dev_load(net, ifr.ifr_name);
4665 rtnl_lock();
4666 ret = dev_ifsioc(net, &ifr, cmd);
4667 rtnl_unlock();
4668 if (!ret) {
4669 if (colon)
4670 *colon = ':';
4671 if (copy_to_user(arg, &ifr,
4672 sizeof(struct ifreq)))
4673 ret = -EFAULT;
4675 return ret;
4678 * These ioctl calls:
4679 * - require superuser power.
4680 * - require strict serialization.
4681 * - do not return a value
4683 case SIOCSIFFLAGS:
4684 case SIOCSIFMETRIC:
4685 case SIOCSIFMTU:
4686 case SIOCSIFMAP:
4687 case SIOCSIFHWADDR:
4688 case SIOCSIFSLAVE:
4689 case SIOCADDMULTI:
4690 case SIOCDELMULTI:
4691 case SIOCSIFHWBROADCAST:
4692 case SIOCSIFTXQLEN:
4693 case SIOCSMIIREG:
4694 case SIOCBONDENSLAVE:
4695 case SIOCBONDRELEASE:
4696 case SIOCBONDSETHWADDR:
4697 case SIOCBONDCHANGEACTIVE:
4698 case SIOCBRADDIF:
4699 case SIOCBRDELIF:
4700 case SIOCSHWTSTAMP:
4701 if (!capable(CAP_NET_ADMIN))
4702 return -EPERM;
4703 /* fall through */
4704 case SIOCBONDSLAVEINFOQUERY:
4705 case SIOCBONDINFOQUERY:
4706 dev_load(net, ifr.ifr_name);
4707 rtnl_lock();
4708 ret = dev_ifsioc(net, &ifr, cmd);
4709 rtnl_unlock();
4710 return ret;
4712 case SIOCGIFMEM:
4713 /* Get the per device memory space. We can add this but
4714 * currently do not support it */
4715 case SIOCSIFMEM:
4716 /* Set the per device memory buffer space.
4717 * Not applicable in our case */
4718 case SIOCSIFLINK:
4719 return -EINVAL;
4722 * Unknown or private ioctl.
4724 default:
4725 if (cmd == SIOCWANDEV ||
4726 (cmd >= SIOCDEVPRIVATE &&
4727 cmd <= SIOCDEVPRIVATE + 15)) {
4728 dev_load(net, ifr.ifr_name);
4729 rtnl_lock();
4730 ret = dev_ifsioc(net, &ifr, cmd);
4731 rtnl_unlock();
4732 if (!ret && copy_to_user(arg, &ifr,
4733 sizeof(struct ifreq)))
4734 ret = -EFAULT;
4735 return ret;
4737 /* Take care of Wireless Extensions */
4738 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4739 return wext_handle_ioctl(net, &ifr, cmd, arg);
4740 return -EINVAL;
4746 * dev_new_index - allocate an ifindex
4747 * @net: the applicable net namespace
4749 * Returns a suitable unique value for a new device interface
4750 * number. The caller must hold the rtnl semaphore or the
4751 * dev_base_lock to be sure it remains unique.
4753 static int dev_new_index(struct net *net)
4755 static int ifindex;
4756 for (;;) {
4757 if (++ifindex <= 0)
4758 ifindex = 1;
4759 if (!__dev_get_by_index(net, ifindex))
4760 return ifindex;
4764 /* Delayed registration/unregisteration */
4765 static LIST_HEAD(net_todo_list);
4767 static void net_set_todo(struct net_device *dev)
4769 list_add_tail(&dev->todo_list, &net_todo_list);
4772 static void rollback_registered_many(struct list_head *head)
4774 struct net_device *dev, *tmp;
4776 BUG_ON(dev_boot_phase);
4777 ASSERT_RTNL();
4779 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4780 /* Some devices call without registering
4781 * for initialization unwind. Remove those
4782 * devices and proceed with the remaining.
4784 if (dev->reg_state == NETREG_UNINITIALIZED) {
4785 pr_debug("unregister_netdevice: device %s/%p never "
4786 "was registered\n", dev->name, dev);
4788 WARN_ON(1);
4789 list_del(&dev->unreg_list);
4790 continue;
4793 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4795 /* If device is running, close it first. */
4796 dev_close(dev);
4798 /* And unlink it from device chain. */
4799 unlist_netdevice(dev);
4801 dev->reg_state = NETREG_UNREGISTERING;
4804 synchronize_net();
4806 list_for_each_entry(dev, head, unreg_list) {
4807 /* Shutdown queueing discipline. */
4808 dev_shutdown(dev);
4811 /* Notify protocols, that we are about to destroy
4812 this device. They should clean all the things.
4814 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4817 * Flush the unicast and multicast chains
4819 dev_unicast_flush(dev);
4820 dev_addr_discard(dev);
4822 if (dev->netdev_ops->ndo_uninit)
4823 dev->netdev_ops->ndo_uninit(dev);
4825 /* Notifier chain MUST detach us from master device. */
4826 WARN_ON(dev->master);
4828 /* Remove entries from kobject tree */
4829 netdev_unregister_kobject(dev);
4832 /* Process any work delayed until the end of the batch */
4833 dev = list_entry(head->next, struct net_device, unreg_list);
4834 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4836 synchronize_net();
4838 list_for_each_entry(dev, head, unreg_list)
4839 dev_put(dev);
4842 static void rollback_registered(struct net_device *dev)
4844 LIST_HEAD(single);
4846 list_add(&dev->unreg_list, &single);
4847 rollback_registered_many(&single);
4850 static void __netdev_init_queue_locks_one(struct net_device *dev,
4851 struct netdev_queue *dev_queue,
4852 void *_unused)
4854 spin_lock_init(&dev_queue->_xmit_lock);
4855 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4856 dev_queue->xmit_lock_owner = -1;
4859 static void netdev_init_queue_locks(struct net_device *dev)
4861 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4862 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4865 unsigned long netdev_fix_features(unsigned long features, const char *name)
4867 /* Fix illegal SG+CSUM combinations. */
4868 if ((features & NETIF_F_SG) &&
4869 !(features & NETIF_F_ALL_CSUM)) {
4870 if (name)
4871 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4872 "checksum feature.\n", name);
4873 features &= ~NETIF_F_SG;
4876 /* TSO requires that SG is present as well. */
4877 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4878 if (name)
4879 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4880 "SG feature.\n", name);
4881 features &= ~NETIF_F_TSO;
4884 if (features & NETIF_F_UFO) {
4885 if (!(features & NETIF_F_GEN_CSUM)) {
4886 if (name)
4887 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4888 "since no NETIF_F_HW_CSUM feature.\n",
4889 name);
4890 features &= ~NETIF_F_UFO;
4893 if (!(features & NETIF_F_SG)) {
4894 if (name)
4895 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4896 "since no NETIF_F_SG feature.\n", name);
4897 features &= ~NETIF_F_UFO;
4901 return features;
4903 EXPORT_SYMBOL(netdev_fix_features);
4906 * netif_stacked_transfer_operstate - transfer operstate
4907 * @rootdev: the root or lower level device to transfer state from
4908 * @dev: the device to transfer operstate to
4910 * Transfer operational state from root to device. This is normally
4911 * called when a stacking relationship exists between the root
4912 * device and the device(a leaf device).
4914 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4915 struct net_device *dev)
4917 if (rootdev->operstate == IF_OPER_DORMANT)
4918 netif_dormant_on(dev);
4919 else
4920 netif_dormant_off(dev);
4922 if (netif_carrier_ok(rootdev)) {
4923 if (!netif_carrier_ok(dev))
4924 netif_carrier_on(dev);
4925 } else {
4926 if (netif_carrier_ok(dev))
4927 netif_carrier_off(dev);
4930 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4933 * register_netdevice - register a network device
4934 * @dev: device to register
4936 * Take a completed network device structure and add it to the kernel
4937 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4938 * chain. 0 is returned on success. A negative errno code is returned
4939 * on a failure to set up the device, or if the name is a duplicate.
4941 * Callers must hold the rtnl semaphore. You may want
4942 * register_netdev() instead of this.
4944 * BUGS:
4945 * The locking appears insufficient to guarantee two parallel registers
4946 * will not get the same name.
4949 int register_netdevice(struct net_device *dev)
4951 int ret;
4952 struct net *net = dev_net(dev);
4954 BUG_ON(dev_boot_phase);
4955 ASSERT_RTNL();
4957 might_sleep();
4959 /* When net_device's are persistent, this will be fatal. */
4960 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4961 BUG_ON(!net);
4963 spin_lock_init(&dev->addr_list_lock);
4964 netdev_set_addr_lockdep_class(dev);
4965 netdev_init_queue_locks(dev);
4967 dev->iflink = -1;
4969 /* Init, if this function is available */
4970 if (dev->netdev_ops->ndo_init) {
4971 ret = dev->netdev_ops->ndo_init(dev);
4972 if (ret) {
4973 if (ret > 0)
4974 ret = -EIO;
4975 goto out;
4979 ret = dev_get_valid_name(net, dev->name, dev->name, 0);
4980 if (ret)
4981 goto err_uninit;
4983 dev->ifindex = dev_new_index(net);
4984 if (dev->iflink == -1)
4985 dev->iflink = dev->ifindex;
4987 /* Fix illegal checksum combinations */
4988 if ((dev->features & NETIF_F_HW_CSUM) &&
4989 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4990 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4991 dev->name);
4992 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4995 if ((dev->features & NETIF_F_NO_CSUM) &&
4996 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4997 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4998 dev->name);
4999 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5002 dev->features = netdev_fix_features(dev->features, dev->name);
5004 /* Enable software GSO if SG is supported. */
5005 if (dev->features & NETIF_F_SG)
5006 dev->features |= NETIF_F_GSO;
5008 netdev_initialize_kobject(dev);
5010 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5011 ret = notifier_to_errno(ret);
5012 if (ret)
5013 goto err_uninit;
5015 ret = netdev_register_kobject(dev);
5016 if (ret)
5017 goto err_uninit;
5018 dev->reg_state = NETREG_REGISTERED;
5021 * Default initial state at registry is that the
5022 * device is present.
5025 set_bit(__LINK_STATE_PRESENT, &dev->state);
5027 dev_init_scheduler(dev);
5028 dev_hold(dev);
5029 list_netdevice(dev);
5031 /* Notify protocols, that a new device appeared. */
5032 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5033 ret = notifier_to_errno(ret);
5034 if (ret) {
5035 rollback_registered(dev);
5036 dev->reg_state = NETREG_UNREGISTERED;
5039 * Prevent userspace races by waiting until the network
5040 * device is fully setup before sending notifications.
5042 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5044 out:
5045 return ret;
5047 err_uninit:
5048 if (dev->netdev_ops->ndo_uninit)
5049 dev->netdev_ops->ndo_uninit(dev);
5050 goto out;
5052 EXPORT_SYMBOL(register_netdevice);
5055 * init_dummy_netdev - init a dummy network device for NAPI
5056 * @dev: device to init
5058 * This takes a network device structure and initialize the minimum
5059 * amount of fields so it can be used to schedule NAPI polls without
5060 * registering a full blown interface. This is to be used by drivers
5061 * that need to tie several hardware interfaces to a single NAPI
5062 * poll scheduler due to HW limitations.
5064 int init_dummy_netdev(struct net_device *dev)
5066 /* Clear everything. Note we don't initialize spinlocks
5067 * are they aren't supposed to be taken by any of the
5068 * NAPI code and this dummy netdev is supposed to be
5069 * only ever used for NAPI polls
5071 memset(dev, 0, sizeof(struct net_device));
5073 /* make sure we BUG if trying to hit standard
5074 * register/unregister code path
5076 dev->reg_state = NETREG_DUMMY;
5078 /* initialize the ref count */
5079 atomic_set(&dev->refcnt, 1);
5081 /* NAPI wants this */
5082 INIT_LIST_HEAD(&dev->napi_list);
5084 /* a dummy interface is started by default */
5085 set_bit(__LINK_STATE_PRESENT, &dev->state);
5086 set_bit(__LINK_STATE_START, &dev->state);
5088 return 0;
5090 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5094 * register_netdev - register a network device
5095 * @dev: device to register
5097 * Take a completed network device structure and add it to the kernel
5098 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5099 * chain. 0 is returned on success. A negative errno code is returned
5100 * on a failure to set up the device, or if the name is a duplicate.
5102 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5103 * and expands the device name if you passed a format string to
5104 * alloc_netdev.
5106 int register_netdev(struct net_device *dev)
5108 int err;
5110 rtnl_lock();
5113 * If the name is a format string the caller wants us to do a
5114 * name allocation.
5116 if (strchr(dev->name, '%')) {
5117 err = dev_alloc_name(dev, dev->name);
5118 if (err < 0)
5119 goto out;
5122 err = register_netdevice(dev);
5123 out:
5124 rtnl_unlock();
5125 return err;
5127 EXPORT_SYMBOL(register_netdev);
5130 * netdev_wait_allrefs - wait until all references are gone.
5132 * This is called when unregistering network devices.
5134 * Any protocol or device that holds a reference should register
5135 * for netdevice notification, and cleanup and put back the
5136 * reference if they receive an UNREGISTER event.
5137 * We can get stuck here if buggy protocols don't correctly
5138 * call dev_put.
5140 static void netdev_wait_allrefs(struct net_device *dev)
5142 unsigned long rebroadcast_time, warning_time;
5144 linkwatch_forget_dev(dev);
5146 rebroadcast_time = warning_time = jiffies;
5147 while (atomic_read(&dev->refcnt) != 0) {
5148 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5149 rtnl_lock();
5151 /* Rebroadcast unregister notification */
5152 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5153 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5154 * should have already handle it the first time */
5156 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5157 &dev->state)) {
5158 /* We must not have linkwatch events
5159 * pending on unregister. If this
5160 * happens, we simply run the queue
5161 * unscheduled, resulting in a noop
5162 * for this device.
5164 linkwatch_run_queue();
5167 __rtnl_unlock();
5169 rebroadcast_time = jiffies;
5172 msleep(250);
5174 if (time_after(jiffies, warning_time + 10 * HZ)) {
5175 printk(KERN_EMERG "unregister_netdevice: "
5176 "waiting for %s to become free. Usage "
5177 "count = %d\n",
5178 dev->name, atomic_read(&dev->refcnt));
5179 warning_time = jiffies;
5184 /* The sequence is:
5186 * rtnl_lock();
5187 * ...
5188 * register_netdevice(x1);
5189 * register_netdevice(x2);
5190 * ...
5191 * unregister_netdevice(y1);
5192 * unregister_netdevice(y2);
5193 * ...
5194 * rtnl_unlock();
5195 * free_netdev(y1);
5196 * free_netdev(y2);
5198 * We are invoked by rtnl_unlock().
5199 * This allows us to deal with problems:
5200 * 1) We can delete sysfs objects which invoke hotplug
5201 * without deadlocking with linkwatch via keventd.
5202 * 2) Since we run with the RTNL semaphore not held, we can sleep
5203 * safely in order to wait for the netdev refcnt to drop to zero.
5205 * We must not return until all unregister events added during
5206 * the interval the lock was held have been completed.
5208 void netdev_run_todo(void)
5210 struct list_head list;
5212 /* Snapshot list, allow later requests */
5213 list_replace_init(&net_todo_list, &list);
5215 __rtnl_unlock();
5217 while (!list_empty(&list)) {
5218 struct net_device *dev
5219 = list_entry(list.next, struct net_device, todo_list);
5220 list_del(&dev->todo_list);
5222 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5223 printk(KERN_ERR "network todo '%s' but state %d\n",
5224 dev->name, dev->reg_state);
5225 dump_stack();
5226 continue;
5229 dev->reg_state = NETREG_UNREGISTERED;
5231 on_each_cpu(flush_backlog, dev, 1);
5233 netdev_wait_allrefs(dev);
5235 /* paranoia */
5236 BUG_ON(atomic_read(&dev->refcnt));
5237 WARN_ON(dev->ip_ptr);
5238 WARN_ON(dev->ip6_ptr);
5239 WARN_ON(dev->dn_ptr);
5241 if (dev->destructor)
5242 dev->destructor(dev);
5244 /* Free network device */
5245 kobject_put(&dev->dev.kobj);
5250 * dev_txq_stats_fold - fold tx_queues stats
5251 * @dev: device to get statistics from
5252 * @stats: struct net_device_stats to hold results
5254 void dev_txq_stats_fold(const struct net_device *dev,
5255 struct net_device_stats *stats)
5257 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5258 unsigned int i;
5259 struct netdev_queue *txq;
5261 for (i = 0; i < dev->num_tx_queues; i++) {
5262 txq = netdev_get_tx_queue(dev, i);
5263 tx_bytes += txq->tx_bytes;
5264 tx_packets += txq->tx_packets;
5265 tx_dropped += txq->tx_dropped;
5267 if (tx_bytes || tx_packets || tx_dropped) {
5268 stats->tx_bytes = tx_bytes;
5269 stats->tx_packets = tx_packets;
5270 stats->tx_dropped = tx_dropped;
5273 EXPORT_SYMBOL(dev_txq_stats_fold);
5276 * dev_get_stats - get network device statistics
5277 * @dev: device to get statistics from
5279 * Get network statistics from device. The device driver may provide
5280 * its own method by setting dev->netdev_ops->get_stats; otherwise
5281 * the internal statistics structure is used.
5283 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5285 const struct net_device_ops *ops = dev->netdev_ops;
5287 if (ops->ndo_get_stats)
5288 return ops->ndo_get_stats(dev);
5290 dev_txq_stats_fold(dev, &dev->stats);
5291 return &dev->stats;
5293 EXPORT_SYMBOL(dev_get_stats);
5295 static void netdev_init_one_queue(struct net_device *dev,
5296 struct netdev_queue *queue,
5297 void *_unused)
5299 queue->dev = dev;
5302 static void netdev_init_queues(struct net_device *dev)
5304 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5305 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5306 spin_lock_init(&dev->tx_global_lock);
5310 * alloc_netdev_mq - allocate network device
5311 * @sizeof_priv: size of private data to allocate space for
5312 * @name: device name format string
5313 * @setup: callback to initialize device
5314 * @queue_count: the number of subqueues to allocate
5316 * Allocates a struct net_device with private data area for driver use
5317 * and performs basic initialization. Also allocates subquue structs
5318 * for each queue on the device at the end of the netdevice.
5320 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5321 void (*setup)(struct net_device *), unsigned int queue_count)
5323 struct netdev_queue *tx;
5324 struct net_device *dev;
5325 size_t alloc_size;
5326 struct net_device *p;
5328 BUG_ON(strlen(name) >= sizeof(dev->name));
5330 alloc_size = sizeof(struct net_device);
5331 if (sizeof_priv) {
5332 /* ensure 32-byte alignment of private area */
5333 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5334 alloc_size += sizeof_priv;
5336 /* ensure 32-byte alignment of whole construct */
5337 alloc_size += NETDEV_ALIGN - 1;
5339 p = kzalloc(alloc_size, GFP_KERNEL);
5340 if (!p) {
5341 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5342 return NULL;
5345 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5346 if (!tx) {
5347 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5348 "tx qdiscs.\n");
5349 goto free_p;
5352 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5353 dev->padded = (char *)dev - (char *)p;
5355 if (dev_addr_init(dev))
5356 goto free_tx;
5358 dev_unicast_init(dev);
5360 dev_net_set(dev, &init_net);
5362 dev->_tx = tx;
5363 dev->num_tx_queues = queue_count;
5364 dev->real_num_tx_queues = queue_count;
5366 dev->gso_max_size = GSO_MAX_SIZE;
5368 netdev_init_queues(dev);
5370 INIT_LIST_HEAD(&dev->napi_list);
5371 INIT_LIST_HEAD(&dev->unreg_list);
5372 INIT_LIST_HEAD(&dev->link_watch_list);
5373 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5374 setup(dev);
5375 strcpy(dev->name, name);
5376 return dev;
5378 free_tx:
5379 kfree(tx);
5381 free_p:
5382 kfree(p);
5383 return NULL;
5385 EXPORT_SYMBOL(alloc_netdev_mq);
5388 * free_netdev - free network device
5389 * @dev: device
5391 * This function does the last stage of destroying an allocated device
5392 * interface. The reference to the device object is released.
5393 * If this is the last reference then it will be freed.
5395 void free_netdev(struct net_device *dev)
5397 struct napi_struct *p, *n;
5399 release_net(dev_net(dev));
5401 kfree(dev->_tx);
5403 /* Flush device addresses */
5404 dev_addr_flush(dev);
5406 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5407 netif_napi_del(p);
5409 /* Compatibility with error handling in drivers */
5410 if (dev->reg_state == NETREG_UNINITIALIZED) {
5411 kfree((char *)dev - dev->padded);
5412 return;
5415 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5416 dev->reg_state = NETREG_RELEASED;
5418 /* will free via device release */
5419 put_device(&dev->dev);
5421 EXPORT_SYMBOL(free_netdev);
5424 * synchronize_net - Synchronize with packet receive processing
5426 * Wait for packets currently being received to be done.
5427 * Does not block later packets from starting.
5429 void synchronize_net(void)
5431 might_sleep();
5432 synchronize_rcu();
5434 EXPORT_SYMBOL(synchronize_net);
5437 * unregister_netdevice_queue - remove device from the kernel
5438 * @dev: device
5439 * @head: list
5441 * This function shuts down a device interface and removes it
5442 * from the kernel tables.
5443 * If head not NULL, device is queued to be unregistered later.
5445 * Callers must hold the rtnl semaphore. You may want
5446 * unregister_netdev() instead of this.
5449 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5451 ASSERT_RTNL();
5453 if (head) {
5454 list_move_tail(&dev->unreg_list, head);
5455 } else {
5456 rollback_registered(dev);
5457 /* Finish processing unregister after unlock */
5458 net_set_todo(dev);
5461 EXPORT_SYMBOL(unregister_netdevice_queue);
5464 * unregister_netdevice_many - unregister many devices
5465 * @head: list of devices
5467 void unregister_netdevice_many(struct list_head *head)
5469 struct net_device *dev;
5471 if (!list_empty(head)) {
5472 rollback_registered_many(head);
5473 list_for_each_entry(dev, head, unreg_list)
5474 net_set_todo(dev);
5477 EXPORT_SYMBOL(unregister_netdevice_many);
5480 * unregister_netdev - remove device from the kernel
5481 * @dev: device
5483 * This function shuts down a device interface and removes it
5484 * from the kernel tables.
5486 * This is just a wrapper for unregister_netdevice that takes
5487 * the rtnl semaphore. In general you want to use this and not
5488 * unregister_netdevice.
5490 void unregister_netdev(struct net_device *dev)
5492 rtnl_lock();
5493 unregister_netdevice(dev);
5494 rtnl_unlock();
5496 EXPORT_SYMBOL(unregister_netdev);
5499 * dev_change_net_namespace - move device to different nethost namespace
5500 * @dev: device
5501 * @net: network namespace
5502 * @pat: If not NULL name pattern to try if the current device name
5503 * is already taken in the destination network namespace.
5505 * This function shuts down a device interface and moves it
5506 * to a new network namespace. On success 0 is returned, on
5507 * a failure a netagive errno code is returned.
5509 * Callers must hold the rtnl semaphore.
5512 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5514 int err;
5516 ASSERT_RTNL();
5518 /* Don't allow namespace local devices to be moved. */
5519 err = -EINVAL;
5520 if (dev->features & NETIF_F_NETNS_LOCAL)
5521 goto out;
5523 #ifdef CONFIG_SYSFS
5524 /* Don't allow real devices to be moved when sysfs
5525 * is enabled.
5527 err = -EINVAL;
5528 if (dev->dev.parent)
5529 goto out;
5530 #endif
5532 /* Ensure the device has been registrered */
5533 err = -EINVAL;
5534 if (dev->reg_state != NETREG_REGISTERED)
5535 goto out;
5537 /* Get out if there is nothing todo */
5538 err = 0;
5539 if (net_eq(dev_net(dev), net))
5540 goto out;
5542 /* Pick the destination device name, and ensure
5543 * we can use it in the destination network namespace.
5545 err = -EEXIST;
5546 if (__dev_get_by_name(net, dev->name)) {
5547 /* We get here if we can't use the current device name */
5548 if (!pat)
5549 goto out;
5550 if (dev_get_valid_name(net, pat, dev->name, 1))
5551 goto out;
5555 * And now a mini version of register_netdevice unregister_netdevice.
5558 /* If device is running close it first. */
5559 dev_close(dev);
5561 /* And unlink it from device chain */
5562 err = -ENODEV;
5563 unlist_netdevice(dev);
5565 synchronize_net();
5567 /* Shutdown queueing discipline. */
5568 dev_shutdown(dev);
5570 /* Notify protocols, that we are about to destroy
5571 this device. They should clean all the things.
5573 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5574 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5577 * Flush the unicast and multicast chains
5579 dev_unicast_flush(dev);
5580 dev_addr_discard(dev);
5582 netdev_unregister_kobject(dev);
5584 /* Actually switch the network namespace */
5585 dev_net_set(dev, net);
5587 /* If there is an ifindex conflict assign a new one */
5588 if (__dev_get_by_index(net, dev->ifindex)) {
5589 int iflink = (dev->iflink == dev->ifindex);
5590 dev->ifindex = dev_new_index(net);
5591 if (iflink)
5592 dev->iflink = dev->ifindex;
5595 /* Fixup kobjects */
5596 err = netdev_register_kobject(dev);
5597 WARN_ON(err);
5599 /* Add the device back in the hashes */
5600 list_netdevice(dev);
5602 /* Notify protocols, that a new device appeared. */
5603 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5606 * Prevent userspace races by waiting until the network
5607 * device is fully setup before sending notifications.
5609 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5611 synchronize_net();
5612 err = 0;
5613 out:
5614 return err;
5616 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5618 static int dev_cpu_callback(struct notifier_block *nfb,
5619 unsigned long action,
5620 void *ocpu)
5622 struct sk_buff **list_skb;
5623 struct Qdisc **list_net;
5624 struct sk_buff *skb;
5625 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5626 struct softnet_data *sd, *oldsd;
5628 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5629 return NOTIFY_OK;
5631 local_irq_disable();
5632 cpu = smp_processor_id();
5633 sd = &per_cpu(softnet_data, cpu);
5634 oldsd = &per_cpu(softnet_data, oldcpu);
5636 /* Find end of our completion_queue. */
5637 list_skb = &sd->completion_queue;
5638 while (*list_skb)
5639 list_skb = &(*list_skb)->next;
5640 /* Append completion queue from offline CPU. */
5641 *list_skb = oldsd->completion_queue;
5642 oldsd->completion_queue = NULL;
5644 /* Find end of our output_queue. */
5645 list_net = &sd->output_queue;
5646 while (*list_net)
5647 list_net = &(*list_net)->next_sched;
5648 /* Append output queue from offline CPU. */
5649 *list_net = oldsd->output_queue;
5650 oldsd->output_queue = NULL;
5652 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5653 local_irq_enable();
5655 /* Process offline CPU's input_pkt_queue */
5656 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5657 netif_rx(skb);
5659 return NOTIFY_OK;
5664 * netdev_increment_features - increment feature set by one
5665 * @all: current feature set
5666 * @one: new feature set
5667 * @mask: mask feature set
5669 * Computes a new feature set after adding a device with feature set
5670 * @one to the master device with current feature set @all. Will not
5671 * enable anything that is off in @mask. Returns the new feature set.
5673 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5674 unsigned long mask)
5676 /* If device needs checksumming, downgrade to it. */
5677 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5678 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5679 else if (mask & NETIF_F_ALL_CSUM) {
5680 /* If one device supports v4/v6 checksumming, set for all. */
5681 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5682 !(all & NETIF_F_GEN_CSUM)) {
5683 all &= ~NETIF_F_ALL_CSUM;
5684 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5687 /* If one device supports hw checksumming, set for all. */
5688 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5689 all &= ~NETIF_F_ALL_CSUM;
5690 all |= NETIF_F_HW_CSUM;
5694 one |= NETIF_F_ALL_CSUM;
5696 one |= all & NETIF_F_ONE_FOR_ALL;
5697 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5698 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5700 return all;
5702 EXPORT_SYMBOL(netdev_increment_features);
5704 static struct hlist_head *netdev_create_hash(void)
5706 int i;
5707 struct hlist_head *hash;
5709 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5710 if (hash != NULL)
5711 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5712 INIT_HLIST_HEAD(&hash[i]);
5714 return hash;
5717 /* Initialize per network namespace state */
5718 static int __net_init netdev_init(struct net *net)
5720 INIT_LIST_HEAD(&net->dev_base_head);
5722 net->dev_name_head = netdev_create_hash();
5723 if (net->dev_name_head == NULL)
5724 goto err_name;
5726 net->dev_index_head = netdev_create_hash();
5727 if (net->dev_index_head == NULL)
5728 goto err_idx;
5730 return 0;
5732 err_idx:
5733 kfree(net->dev_name_head);
5734 err_name:
5735 return -ENOMEM;
5739 * netdev_drivername - network driver for the device
5740 * @dev: network device
5741 * @buffer: buffer for resulting name
5742 * @len: size of buffer
5744 * Determine network driver for device.
5746 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5748 const struct device_driver *driver;
5749 const struct device *parent;
5751 if (len <= 0 || !buffer)
5752 return buffer;
5753 buffer[0] = 0;
5755 parent = dev->dev.parent;
5757 if (!parent)
5758 return buffer;
5760 driver = parent->driver;
5761 if (driver && driver->name)
5762 strlcpy(buffer, driver->name, len);
5763 return buffer;
5766 static void __net_exit netdev_exit(struct net *net)
5768 kfree(net->dev_name_head);
5769 kfree(net->dev_index_head);
5772 static struct pernet_operations __net_initdata netdev_net_ops = {
5773 .init = netdev_init,
5774 .exit = netdev_exit,
5777 static void __net_exit default_device_exit(struct net *net)
5779 struct net_device *dev, *aux;
5781 * Push all migratable network devices back to the
5782 * initial network namespace
5784 rtnl_lock();
5785 for_each_netdev_safe(net, dev, aux) {
5786 int err;
5787 char fb_name[IFNAMSIZ];
5789 /* Ignore unmoveable devices (i.e. loopback) */
5790 if (dev->features & NETIF_F_NETNS_LOCAL)
5791 continue;
5793 /* Leave virtual devices for the generic cleanup */
5794 if (dev->rtnl_link_ops)
5795 continue;
5797 /* Push remaing network devices to init_net */
5798 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5799 err = dev_change_net_namespace(dev, &init_net, fb_name);
5800 if (err) {
5801 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5802 __func__, dev->name, err);
5803 BUG();
5806 rtnl_unlock();
5809 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5811 /* At exit all network devices most be removed from a network
5812 * namespace. Do this in the reverse order of registeration.
5813 * Do this across as many network namespaces as possible to
5814 * improve batching efficiency.
5816 struct net_device *dev;
5817 struct net *net;
5818 LIST_HEAD(dev_kill_list);
5820 rtnl_lock();
5821 list_for_each_entry(net, net_list, exit_list) {
5822 for_each_netdev_reverse(net, dev) {
5823 if (dev->rtnl_link_ops)
5824 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5825 else
5826 unregister_netdevice_queue(dev, &dev_kill_list);
5829 unregister_netdevice_many(&dev_kill_list);
5830 rtnl_unlock();
5833 static struct pernet_operations __net_initdata default_device_ops = {
5834 .exit = default_device_exit,
5835 .exit_batch = default_device_exit_batch,
5839 * Initialize the DEV module. At boot time this walks the device list and
5840 * unhooks any devices that fail to initialise (normally hardware not
5841 * present) and leaves us with a valid list of present and active devices.
5846 * This is called single threaded during boot, so no need
5847 * to take the rtnl semaphore.
5849 static int __init net_dev_init(void)
5851 int i, rc = -ENOMEM;
5853 BUG_ON(!dev_boot_phase);
5855 if (dev_proc_init())
5856 goto out;
5858 if (netdev_kobject_init())
5859 goto out;
5861 INIT_LIST_HEAD(&ptype_all);
5862 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5863 INIT_LIST_HEAD(&ptype_base[i]);
5865 if (register_pernet_subsys(&netdev_net_ops))
5866 goto out;
5869 * Initialise the packet receive queues.
5872 for_each_possible_cpu(i) {
5873 struct softnet_data *queue;
5875 queue = &per_cpu(softnet_data, i);
5876 skb_queue_head_init(&queue->input_pkt_queue);
5877 queue->completion_queue = NULL;
5878 INIT_LIST_HEAD(&queue->poll_list);
5880 queue->backlog.poll = process_backlog;
5881 queue->backlog.weight = weight_p;
5882 queue->backlog.gro_list = NULL;
5883 queue->backlog.gro_count = 0;
5886 dev_boot_phase = 0;
5888 /* The loopback device is special if any other network devices
5889 * is present in a network namespace the loopback device must
5890 * be present. Since we now dynamically allocate and free the
5891 * loopback device ensure this invariant is maintained by
5892 * keeping the loopback device as the first device on the
5893 * list of network devices. Ensuring the loopback devices
5894 * is the first device that appears and the last network device
5895 * that disappears.
5897 if (register_pernet_device(&loopback_net_ops))
5898 goto out;
5900 if (register_pernet_device(&default_device_ops))
5901 goto out;
5903 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5904 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5906 hotcpu_notifier(dev_cpu_callback, 0);
5907 dst_init();
5908 dev_mcast_init();
5909 rc = 0;
5910 out:
5911 return rc;
5914 subsys_initcall(net_dev_init);
5916 static int __init initialize_hashrnd(void)
5918 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5919 return 0;
5922 late_initcall_sync(initialize_hashrnd);